Telechargé par Abdessamed BENLAHBIB

Smart Grids Clouds Communications

publicité
SMART
GRIDS
Clouds , C ommuni c a t i o n s , O p e n
Sourc e, and Aut o m a t i o n
Devices, Circuits, and Systems
Series Editor
Krzysztof Iniewski
CMOS Emerging Technologies Research Inc.,
Vancouver, British Columbia, Canada
PUBLISHED TITLES:
Atomic Nanoscale Technology in the Nuclear Industry
Taeho Woo
Biological and Medical Sensor Technologies
Krzysztof Iniewski
Building Sensor Networks: From Design to Applications
Ioanis Nikolaidis and Krzysztof Iniewski
Circuits at the Nanoscale: Communications, Imaging, and Sensing
Krzysztof Iniewski
Electrical Solitons: Theory, Design, and Applications
David Ricketts and Donhee Ham
Electronics for Radiation Detection
Krzysztof Iniewski
Embedded and Networking Systems:
Design, Software, and Implementation
Gul N. Khan and Krzysztof Iniewski
Energy Harvesting with Functional Materials and Microsystems
Madhu Bhaskaran, Sharath Sriram, and Krzysztof Iniewski
Graphene, Carbon Nanotubes, and Nanostuctures:
Techniques and Applications
James E. Morris and Krzysztof Iniewski
High-Speed Photonics Interconnects
Lukas Chrostowski and Krzysztof Iniewski
Integrated Microsystems: Electronics, Photonics, and Biotechnology
Krzysztof Iniewski
Integrated Power Devices and TCAD Simulation
Yue Fu, Zhanming Li, Wai Tung Ng, and Johnny K.O. Sin
Internet Networks: Wired, Wireless, and Optical Technologies
Krzysztof Iniewski
Low Power Emerging Wireless Technologies
Reza Mahmoudi and Krzysztof Iniewski
Medical Imaging: Technology and Applications
Troy Farncombe and Krzysztof Iniewski
PUBLISHED TITLES:
MEMS: Fundamental Technology and Applications
Vikas Choudhary and Krzysztof Iniewski
Microluidics and Nanotechnology: Biosensing to the Single Molecule Limit
Eric Lagally and Krzysztof Iniewski
MIMO Power Line Communications: Narrow and Broadband Standards,
EMC, and Advanced Processing
Lars Torsten Berger, Andreas Schwager, Pascal Pagani, and Daniel Schneider
Nano-Semiconductors: Devices and Technology
Krzysztof Iniewski
Nanoelectronic Device Applications Handbook
James E. Morris and Krzysztof Iniewski
Nanoplasmonics: Advanced Device Applications
James W. M. Chon and Krzysztof Iniewski
Nanoscale Semiconductor Memories: Technology and Applications
Santosh K. Kurinec and Krzysztof Iniewski
Novel Advances in Microsystems Technologies and Their Applications
Laurent A. Francis and Krzysztof Iniewski
Optical, Acoustic, Magnetic, and Mechanical Sensor Technologies
Krzysztof Iniewski
Radiation Effects in Semiconductors
Krzysztof Iniewski
Semiconductor Radiation Detection Systems
Krzysztof Iniewski
Smart Grids: Clouds, Communications, Open Source, and Automation
David Bakken and Krzysztof Iniewski
Smart Sensors for Industrial Applications
Krzysztof Iniewski
Technologies for Smart Sensors and Sensor Fusion
Kevin Yallup and Krzysztof Iniewski
Telecommunication Networks
Eugenio Iannone
Testing for Small-Delay Defects in Nanoscale CMOS Integrated Circuits
Sandeep K. Goel and Krishnendu Chakrabarty
Wireless Technologies: Circuits, Systems, and Devices
Krzysztof Iniewski
FORTHCOMING TITLES:
3D Circuit and System Design: Multicore Architecture,
Thermal Management, and Reliability
Rohit Sharma and Krzysztof Iniewski
Circuits and Systems for Security and Privacy
Farhana Sheikh and Leonel Sousa
CMOS: Front-End Electronics for Radiation Sensors
Angelo Rivetti
Gallium Nitride (GaN): Physics, Devices, and Technology
Farid Medjdoub and Krzysztof Iniewski
High Frequency Communication and Sensing: Traveling-Wave Techniques
Ahmet Tekin and Ahmed Emira
High-Speed Devices and Circuits with THz Applications
Jung Han Choi and Krzysztof Iniewski
Labs-on-Chip: Physics, Design and Technology
Eugenio Iannone
Laser-Based Optical Detection of Explosives
Paul M. Pellegrino, Ellen L. Holthoff, and Mikella E. Farrell
Metallic Spintronic Devices
Xiaobin Wang
Mobile Point-of-Care Monitors and Diagnostic Device Design
Walter Karlen and Krzysztof Iniewski
Nanoelectronics: Devices, Circuits, and Systems
Nikos Konofaos
Nanomaterials: A Guide to Fabrication and Applications
Gordon Harling and Krzysztof Iniewski
Nanopatterning and Nanoscale Devices for Biological Applications
Krzysztof Iniewski and Seila Selimovic
Optical Fiber Sensors and Applications
Ginu Rajan and Krzysztof Iniewski
Organic Solar Cells: Materials, Devices, Interfaces, and Modeling
Qiquan Qiao and Krzysztof Iniewski
Power Management Integrated Circuits and Technologies
Mona M. Hella and Patrick Mercier
Radio Frequency Integrated Circuit Design
Sebastian Magierowski
Semiconductor Device Technology: Silicon and Materials
Tomasz Brozek and Krzysztof Iniewski
FORTHCOMING TITLES:
Soft Errors: From Particles to Circuits
Jean-Luc Autran and Daniela Munteanu
VLSI: Circuits for Emerging Applications
Tomasz Wojcicki and Krzysztof Iniewski
Wireless Transceiver Circuits: System Perspectives and Design Aspects
Woogeun Rhee and Krzysztof Iniewski
SMART
GRIDS
Clouds , C ommuni c a t i o n s , O p e n
Sourc e, and Aut o m a t i o n
EDITED BY
David Bakken
Wa s h in g to n Sta te University
School of Ele c tr ic a l E n g in e e r in g and C omputer S cience
MANAGING EDITOR
Krzysztof Iniewski
CMOS E m e r g in g Te c h n o lo g ies R esearch Inc.
Va n c o u v e r, Br itis h Co lu mbia, C anada
Boca Raton London New York
CRC Press is an imprint of the
Taylor & Francis Group, an informa business
MATLAB® is a trademark of The MathWorks, Inc. and is used with permission. The MathWorks does not
warrant the accuracy of the text or exercises in this book. This book’s use or discussion of MATLAB® software or related products does not constitute endorsement or sponsorship by The MathWorks of a particular
pedagogical approach or particular use of the MATLAB® software.
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2014 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works
Version Date: 20140305
International Standard Book Number-13: 978-1-4822-0612-8 (eBook - PDF)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts
have been made to publish reliable data and information, but the author and publisher cannot assume
responsibility for the validity of all materials or the consequences of their use. The authors and publishers
have attempted to trace the copyright holders of all material reproduced in this publication and apologize to
copyright holders if permission to publish in this form has not been obtained. If any copyright material has
not been acknowledged please write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented,
including photocopying, microfilming, and recording, or in any information storage or retrieval system,
without written permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.
com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood
Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and
registration for a variety of users. For organizations that have been granted a photocopy license by the CCC,
a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used
only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
Contents
Preface.................................................................................................................... xiii
Editors ...................................................................................................................... xv
Contributors ...........................................................................................................xvii
Chapter 1
Mission-Critical Cloud Computing for Critical Infrastructures .......... 1
Thoshitha Gamage, David Anderson, David Bakken, Kenneth
Birman, Anjan Bose, Carl Hauser, Ketan Maheshwari, and
Robbert van Renesse
Chapter 2
Power Application Possibilities with Mission-Critical Cloud
Computing .......................................................................................... 17
David Bakken, Pranavamoorthy Balasubramanian, Thoshitha
Gamage, Santiago Grijalva, Kory W. Hedman, Yilu Liu,
Vaithianathan Venkatasubramanian, and Hao Zho
Chapter 3
Emerging Wide-Area Power Applications with Mission-Critical
Data Delivery Requirements .............................................................. 33
Greg Zweigle
Chapter 4
GridStat: High Availability, Low Latency, and Adaptive Sensor
Data Delivery for Smart Generation and Transmission ..................... 55
David E. Bakken, Harald Gjermundrød, and Ioanna Dionysiou
Chapter 5
A Distributed Framework for Smart Grid Modeling,
Monitoring, and Control ................................................................... 115
Alfredo Vaccaro and Eugenio Zimeo
Chapter 6
Role of PLC Technology in Smart Grid Communication
Networks....................................................................................... 133
Angeliki M. Sarai, Artemis C. Voulkidis, Spiros Livieratos,
and Panayotis G. Cottis
Chapter 7
Power Grid Network Analysis for Smart Grid Applications............ 151
Zhifang Wang, Anna Scaglione, and Robert J. Thomas
ix
x
Chapter 8
Contents
Open Source Software, an Enabling Technology for Smart
Grid Evolution .................................................................................. 179
Russell Robertson, Fred Elmendorf, and Shawn Williams
Chapter 9
Contribution of Microgrids to the Development of
the Smart Grid .............................................................................. 191
Tine L. Vandoorn and Lieven Vandevelde
Chapter 10 Microgrids ........................................................................................ 213
Mietek Glinkowski, Adam Guglielmo, Alexandre Oudalov,
Gary Rackliffe, Bill Rose, Ernst Scholtz, Lokesh Verma, and
Fang Yang
Chapter 11 Integrating Consumer Advance Demand Data in Smart Grid
Energy Supply Chain ....................................................................... 251
Tongdan Jin, Chongqing Kang, and Heping Chen
Chapter 12 Photovoltaic Energy Generation and Control for an Autonomous
Shunt Active Power Filter................................................................. 275
Ayman Blorfan, Damien Flieller, Patrice Wira, Guy Sturtzer,
and Jean Mercklé
Chapter 13 Self-Tuning and Self-Diagnosing Simulation ................................... 311
Jin Ma
Chapter 14 A Consensus-Based Fully Distributed Load Management
Algorithm for Smart Grid ................................................................ 333
Yinliang Xu, Wei Zhang, and Wenxin Liu
Chapter 15 Expert Systems Application for the Reconiguration of Electric
Distribution Systems ........................................................................ 359
Horacio Tovar-Hernández and Guillermo Gutierrez-Alcaraz
Chapter 16 Load Data Cleansing and Bus Load Coincidence Factors ............... 375
Wenyuan Li, Ke Wang, and Wijarn Wangdee
Chapter 17 Smart Metering and Infrastructure .................................................. 399
Wenpeng Luan and Wenyuan Li
xi
Contents
Chapter 18 Vision of Future Control Centers in Smart Grids ............................ 421
Fangxing Li, Pei Zhang, Sarina Adhikari, Yanli Wei,
and Qinran Hu
Index ...................................................................................................................... 435
Preface
While electric interconnections have had different kinds and levels of intelligence in
them for many decades, in the last 6 years the notion of the “smart grid” has come
seemingly out of nowhere to be on the minds of not just power engineers but policy
makers, regulators, rate commissions, and the general public. Inherent in the notion
of the smart grid is the ability to communicate much more sensor data and have far
more computations at many more locations using these data.
The purpose of this book is to give power engineers, information technology workers
in the electric sector, and others a snapshot of the state of the art and practice today as
well as a peek into the future regarding the smart grid. There is a special focus on new
kinds of communications and computations enabled or necessitated by the smart grid.
This book is divided into four parts. Part I deals with cloud computing, whose use
is being seriously considered for planning and operational use in a number of utilities
and independent system operators/regional transmission organizations as of March
2014. Cloud computing has the potential to deploy massive amounts of computational
resources to help grid operations, especially under contingency situations. Chapter 1
describes the mission-critical features that cloud computing infrastructures must support in order to be appropriate for operational use in power grids. It also describes
the Advanced Research Projects Agency-Energy GridCloud project to develop such
technologies. Chapter 2 describes a handful of “killer apps” for cloud computing in
power grid operations. It has been written by leading power researchers.
Part II deals with wide-area communications for power grids. Chapter 3 describes a
wide range of power application programs that have extreme communications requirements over wide distances. Such applications are becoming more widely deployed as
grids come under more pressure with every passing year. Chapter 4 describes GridStat,
a middleware communications framework designed from the ground up to meet these
challenging requirements. The chapter includes a detailed analysis of how different
technologies used in today’s grids such as multiprotocol label switching, Internet protocol multicast, IEC 61850, and others are inadequate for the applications described in
Chapter 3 and the requirements derived from them in Chapter 4. Chapter 5 presents an
advanced framework based on the service-oriented architecture approach for integrated
modeling, monitoring, and control. Chapter 6 analyzes the role of power line communication, which is also called broadband over power lines, in the smart grid. Power line
communication/broadband over power lines technologies can provide additional redundant paths for data delivery in a grid, and ones that have failure characteristics other
than traditional network communications infrastructures. Finally, Chapter 7 describes a
novel approach for estimating the statistical properties of power grids. This is an important irst step toward having more broadly reusable power algorithms with greater conidence, as computer scientists and mathematicians have done for centuries.
Part III deals with open source, something common in other industries that is starting to draw great interest from utilities and has great potential to help stimulate innovation in power grids (which suffer from a far higher degree of “vendor lock-in” than
xiii
xiv
Preface
most other industries). Chapter 8 explains what open source software is and its history.
It then overviews a number of freely available open source power application programs.
Part IV deals with the broad category of automation. Chapter 9 explains how
microgrids it into the smart grid landscape and how they can contribute to its
operations. Chapter 10 describes in detail the design and operation of microgrids.
Chapter 11 introduces a virtual energy provisioning concept by which utilities can
collect and aggregate advanced demand information in order to better manage smart
grid supply chains. Chapter 12 describes a new technique for better managing photovoltaic energy while limiting harmonic pollution. Chapter 13 provides an approach
for two-way interactions between simulations and an operational wide-area measurement system that is both self-tuning and self-diagnosing. Chapter 14 describes an
approach for load management in smart grids that is stable, distributed and employs
multiagent techniques. Chapter 15 details the use of an expert system application to
enable electric distribution systems to be reconigured in new and advantageous way.
Chapter 16 describes an approach for both cleansing the load curve data and calculating bus load coincidence factors in order to better exploit smart meter data. Chapter 17
overviews an advanced metering infrastructure system and its components, discusses
its beneits, and summarizes a variety of applications by which smart metering and
infrastructure supports both planning and operations. Finally, Chapter 18 offers a
vision of how smart grid control centers may look in the future.
David E. Bakken
Pullman, Washington
Krzysztof (Kris) Iniewski
Vancouver, British Columbia
MATLAB® is a registered trademark of The MathWorks, Inc. For product information, please contact:
The MathWorks, Inc.
3 Apple Hill Drive
Natick, MA 01760-2098 USA
Tel: 508 647 7000
Fax: 508-647-7001
E-mail: [email protected]
Web: www.mathworks.com
MATLAB® and Simulink® are trademarks of the MathWorks, Inc. and are used with
permission. The MathWorks does not warrant the accuracy of the text or exercises in this
book. This book’s use or discussion of MATLAB® and Simulink® software or related
products does not constitute endorsement or sponsorship by the MathWorks of a particular pedagogical approach or particular use of the MATLAB® and Simulink® software.
Editors
David Bakken is a professor of computer science in the School of Electrical
Engineering and Computer Science at Washington State University and chief scientist at GridStat, Inc. His research interests include wide-area distributed computing
systems, middleware implementation, and dependable computing. Since 1999, he
has been working closely with researchers in his department’s very strong electric
power group on helping rethink the way data delivery is done in power grids over
the wide area, and is considered the world’s leading expert on this. His GridStat
data-delivery software has inluenced the shape of the emerging NASPInet. He is
a frequent visitor and lecturer at utilities, electrical engineering departments, and
power meetings worldwide.
Prior to Washington State University, Dr. Bakken was a research scientist at BBN
(Cambridge, MA), which built the irst Internet in 1969. There he was coinventor
of the Quality Objects middleware framework, in which the Defense Advanced
Research Projects Agency invested more than 50 BBN person-years, which was
integrated with approximately 10 other research projects in various demonstrations,
and which lew in Boeing experimental aircraft. Dr. Bakken has worked for Boeing
and consulted for Amazon.com, Harris Corp., Realtime Innovations, Intel, TriGeo
Network Security, and others. He holds a MS (1990) and a PhD (1994) in computer
science from the University of Arizona, and Bachelor of Science degrees in computer science and mathematics from Washington State University (1985). He is the
author of over 100 publications and coinventor of three patents.
Krzysztof (Kris) Iniewski manages R&D at Redlen Technologies Inc., a startup company in Vancouver, Canada. Redlen’s revolutionary production process for
advanced semiconductor materials enables a new generation of more accurate, alldigital, radiation-based imaging solutions. Kris is also president of CMOS Emerging
Technologies Research Inc. (www.cmosetr.com), an organization of high-tech events
covering communications, microsystems, optoelectronics, and sensors. In his career,
Dr. Iniewski has held numerous faculty and management positions at the University
of Toronto, the University of Alberta, Simon Fraser University, and PMC-Sierra
Inc. He has published over 100 research papers in international journals and conferences. He holds 18 international patents granted in the United States, Canada,
France, Germany, and Japan. He is a frequent invited speaker and has consulted for
multiple organizations internationally. He has written and edited several books for
CRC Press, Cambridge University Press, IEEE Press, Wiley, McGraw-Hill, Artech
House, and Springer. His personal goal is to contribute to healthy living and sustainability through innovative engineering solutions. In his leisure time, Kris can
be found hiking, sailing, skiing, or biking in beautiful British Columbia. He can be
reached at [email protected].
xv
Contributors
Sarina Adhikari
Department of Electrical Engineering
and Computer Science
University of Tennessee
Knoxville, Tennessee
Anjan Bose
School of Electrical Engineering and
Computer Science
Washington State University
Pullman, Washington
David Anderson
School of Electrical Engineering and
Computer Science
Washington State University
Pullman, Washington
Heping Chen
Ingram School of Engineering
Texas State University
San Marcos, Texas
David Bakken
School of Electrical Engineering and
Computer Science
Washington State University
Pullman, Washington
Pranavamoorthy Balasubramanian
School of Electrical, Computer, and
Energy Engineering
Arizona State University
Tempe, Arizona
Kenneth Birman
Department of Computer Science
Cornell University
Ithaca, New York
Ayman Blorfan
Modelling, Intelligence, Process and
Systems Laboratory
Université de Haute Alsace
Mulhouse, France
and
National Institute of Applied Science
Strasbourg, France
Panayotis G. Cottis
School of Electrical and Computer
Engineering
National Technical University of
Athens
Athens, Greece
Ioanna Dionysiou
Department of Computer Science
University of Nicosia
Nicosia, Cyprus
Fred Elmendorf
Grid Protection Alliance
Chattanooga, Tennessee
Damien Flieller
National Institute of Applied Science
Research Group of Electrical and
Electronics in Nancy
Strasbourg, France
Thoshitha Gamage
School of Electrical Engineering and
Computer Science
Washington State University
Pullman, Washington
xvii
xviii
Contributors
Harald Gjermundrød
Department of Computer Science
University of Nicosia
Nicosia, Cyprus
Tongdan Jin
Ingram School of Engineering
Texas State University
San Marcos, Texas
Mietek Glinkowski
ABB Inc.
Raleigh, North Carolina
Chongqing Kang
Department of Electrical Engineering
Tsinghua University
Beijing, China
Santiago Grijalva
School of Electrical and Computer
Engineering
Georgia Institute of Technology
Atlanta, Georgia
Fangxing Li
Department of Electrical Engineering
and Computer Science
University of Tennessee
Knoxville, Tennessee
Adam Guglielmo
ABB Inc.
Raleigh, North Carolina
Wenyuan Li
School of Electrical Engineering
Chongqing University
Chongqing, China
Guillermo Gutierrez-Alcaraz
Department of Electrical
Engineering
Instituto Tecnológico de Morelia
Morelia, Mexico
Carl Hauser
School of Electrical Engineering and
Computer Science
Washington State University
Pullman, Washington
Kory W. Hedman
School of Electrical, Computer, and
Energy Engineering
Arizona State University
Tempe, Arizona
Qinran Hu
Department of Electrical Engineering
and Computer Science
University of Tennessee
Knoxville, Tennessee
and
BC Hydro
Vancouver, Canada
Yilu Liu
Department of Electrical Engineering
and Computer Science
University of Tennessee
Knoxville, Tennessee
Wenxin Liu
Klipsch School of Electrical and
Computer Engineering
New Mexico State University
Las Cruces, New Mexico
Spiros Livieratos
School of Pedagogical and
Technological Education
Athens, Greece
Wenpeng Luan
State Grid Smart Grid Research
Institute
Beijing, China
xix
Contributors
Jin Ma
School of Electrical and Information
Engineering
The University of Sydney
New South Wales, Australia
Anna Scaglione
Department of Electrical and Computer
Engineering
University of California
Davis, California
Ketan Maheshwari
Argonne National Laboratory
Lemont, Illinois
Ernst Scholtz
ABB Inc.
Raleigh, North Carolina
Jean Mercklé
Modelling, Intelligence, Process and
Systems Laboratory
Université de Haute Alsace
Mulhouse, France
Guy Sturtzer
National Institute of Applied Science
Research Group of Electrical and
Electronics in Nancy
Strasbourg, France
Alexandre Oudalov
ABB Inc.
Raleigh, North Carolina
Robert J. Thomas
Department of Electrical and Computer
Engineering
Cornell University
Ithaca, New York
Gary Rackliffe
ABB Inc.
Raleigh, North Carolina
Robbert van Renesse
Department of Computer Science
Cornell University
Ithaca, New York
Russell Robertson
Grid Protection Alliance
Chattanooga, Tennessee
Bill Rose
ABB Inc.
Raleigh, North Carolina
Angeliki M. Sarai
School of Electrical and Computer
Engineering
National Technical University of Athens
Athens, Greece
Horacio Tovar-Hernández
Department of Electrical Engineering
Instituto Tecnológico de Morelia
Morelia, Mexico
Alfredo Vaccaro
Department of Engineering
University of Sannio
Benevento, Italy
Lieven Vandevelde
Department of Electrical Energy
Ghent University
Ghent, Belgium
Tine L. Vandoorn
Department of Electrical Energy
Ghent University
Ghent, Belgium
xx
Contributors
Vaithianathan Venkatasubramanian
School of Electrical Engineering and
Computer Science
Washington State University
Pullman, Washington
Patrice Wira
Modelling, Intelligence, Process and
Systems Laboratory
Université de Haute Alsace
Mulhouse, France
Lokesh Verma
ABB Inc.
Raleigh, North Carolina
Yinliang Xu
Klipsch School of Electrical and
Computer Engineering
New Mexico State University
Las Cruces, New Mexico
Artemis C. Voulkidis
School of Electrical and Computer
Engineering
National Technical University of Athens
Athens, Greece
Ke Wang
School of Computing Science
Simon Fraser University
Vancouver, Canada
Zhifang Wang
Department of Electrical and Computer
Engineering
Virginia Commonwealth University
Richmond, Virginia
Wijarn Wangdee
The Sirindhorn International
Thai–German Graduate School
of Engineering
King Mongkut’s University of
Technology North Bangkok
Bangkok, Thailand
Yanli Wei
Department of Electrical Engineering
and Computer Science
University of Tennessee
Knoxville, Tennessee
Shawn Williams
Grid Protection Alliance
Chattanooga, Tennessee
Fang Yang
ABB Inc.
Raleigh, North Carolina
Pei Zhang
Grid Operations and Planning Electric
Power Research Institute
Palo Alto, California
Wei Zhang
Klipsch School of Electrical and
Computer Engineering
New Mexico State University
Las Cruces, New Mexico
Hao Zhu
Department of Electrical and Computer
Engineering
University of Illinois
Champaign, Illinois
Eugenio Zimeo
Department of Engineering
University of Sannio
Benevento, Italy
Greg Zweigle
Schweitzer Engineering
Laboratories, Inc.
Pullman, Washington
1
Mission-Critical Cloud
Computing for Critical
Infrastructures
Thoshitha Gamage, David Anderson, David
Bakken, Kenneth Birman, Anjan Bose, Carl Hauser,
Ketan Maheshwari, and Robbert van Renesse
CONTENTS
1.1
Introduction ...................................................................................................... 1
1.1.1 Cloud Computing.................................................................................. 2
1.1.2 Advanced Power Grid ...........................................................................4
1.2 Cloud Computing’s Role in the Advanced Power Grid .................................... 5
1.2.1 Berkeley Grand Challenges and the Power Grid ..................................7
1.3 Model for Cloud-Based Power Grid Applications ............................................ 8
1.4 GridCloud: A Capability Demonstration Case Study ......................................9
1.4.1 GridStat.................................................................................................9
1.4.2 Isis2...................................................................................................... 10
1.4.3 TCP-R ................................................................................................. 11
1.4.4 GridSim .............................................................................................. 12
1.4.5 GridCloud Architecture ...................................................................... 12
1.5 Conclusions ..................................................................................................... 13
References ................................................................................................................ 15
1.1
INTRODUCTION
The term cloud is becoming prevalent in nearly every facet of day-to-day life, bringing up an imperative research question: how can the cloud improve future critical
infrastructures? Certainly, cloud computing has already made a huge impact on the
computing landscape and has permanently incorporated itself into almost all sectors of industry. The same, however, cannot be said of critical infrastructures. Most
notably, the power industry has been very cautious regarding cloud-based computing
capabilities.
This is not a total surprise: the power industry is notoriously conservative about
changing its operating model, and its rate commissions are generally focused on
short-term goals. With thousands of moving parts, owned and operated by just as
many stakeholders, even modest changes are dificult. Furthermore, continuing to
1
2
Smart Grids
operate while incorporating large paradigm shifts is neither a straightforward nor
a risk-free process. In addition to industry conservatism, progress is slowed by the
lack of comprehensive cloud-based solutions meeting current and future power grid
application requirements. Nevertheless, there are numerous opportunities on many
fronts—from bulk power generation, through wide-area transmission, to residential
distribution, including at the microgrid level—where cloud technologies can bolster
power grid operations and improve the grid’s eficiency, security, and reliability.
The impact of cloud computing is best exempliied by the recent boom in
e-commerce and online shopping. The cloud has empowered modern customers
with outstanding bargaining power in making their purchasing choices by providing up-to-date pricing information on products from a wide array of sources whose
computing infrastructure is cost-effective and scalable on demand. For example, not
long ago air travelers relied on local travel agents to get the best prices on their reservations. Cloud computing has revolutionized this market, allowing vendors to easily
provide customers with web-based reservation services.
In fact, a recent study shows that online travel e-commerce skyrocketed from a
mere $30 billion in 2002 to a staggering $103 billion, breaking the $100 billion mark
for the irst time in the United States in 2012 [1]. A similar phenomenon applies to
retail shopping. Nowadays, online retail shops offer a variety of products, ranging
from consumer electronics, clothing, books, jewelry, and video games to event tickets, digital media, and lots more at competitive prices. Mainstream online shops such
as Amazon, eBay, Etsy, and so on provide customers with an unprecedented global
marketplace to both buy and sell items. Almost all major US retail giants, such as
Walmart, Macy’s, BestBuy, Target, and so on, have adopted a hybrid sales model,
providing online shops to complement the traditional in-store shopping experience.
A more recent trend is lash sale sites (Fab, Woot, Deals2Buy, Totsy, MyHabit, etc.),
which offer limited-time deals and offers. All in all, retail e-commerce in the United
States increased by as much as 15% in 2012, totaling $289 billion. To put this into
perspective, the total was $72 billion 10 years earlier. Such rapid growth relied heavily on cloud-based technology to provide the massive computing resources behind
online shopping.
1.1.1 CLOUD COMPUTING
What truly characterizes cloud computing is its business model. The cloud provides
on-demand access to virtually limitless hardware and software resources meeting
the users’ requirements. Furthermore, users only pay for resources they use, based
on the time of use and capacity. The National Institute of Standards and Technology
(NIST) deines ive essential cloud characteristics: on-demand self-service, broad
network access, resource pooling, rapid elasticity, and measured service [2].
The computational model of the cloud features two key characteristics—
abstraction and virtualization. The cloud provides its end users with well-deined
application programming interfaces (APIs) that support requests to a wide range
of hardware and software resources. Cloud computing supports various conigurations (central processing unit [CPU], memory, platform, input/output [I/O], networking, storage, servers) and capacities (scale) while abstracting resource management
Mission-Critical Cloud Computing for Critical Infrastructures
3
(setup, startup, maintenance, etc.), underlying infrastructure technology, physical
space, and human labor requirements. The end users see only APIs when they access
services on the cloud. For example, users of Dropbox, the popular cloud-based
online storage, only need to know that their stored items are accessible through the
API; they do not need any knowledge of the underlying infrastructure supporting the
service. Furthermore, end users are relieved of owning large computing resources
that are often underused. Instead, resources are housed in large data centers as a
shared resource pool serving multiple users, thus optimizing their use and amortizing the cost of maintenance. At the same time, end users are unaware of where their
resources physically reside, effectively virtualizing the computing resources.
Cloud computing provides three service models: software as a service (SaaS),
platform as a service (PaaS), and infrastructure as a service (IaaS). Each of these
service models provides unique APIs. Services can be purchased separately, but are
typically purchased as a solution stack. The SaaS model offers end-point business
applications which are customizable and conigurable based on speciic needs. One
good example is the Google Apps framework, which offers a large suite of end-user
applications (email, online storage, streaming channels, domain names, messaging,
web hosting, etc.) that individuals, businesses, universities, and other organizations
can purchase individually or in combination. Software offered in this manner has a
shorter development life cycle, resulting in frequent updates and up-to-date versions.
The life-cycle maintenance is explicitly handled by the service provider, who offers
the software on a pay-per-use basis. Since the software is hosted in the cloud, there
is no explicit installation or maintenance process for the end users in their native
environment. Some of the prominent SaaS providers include Salesforce, Google,
Microsoft, Intuit, Oracle, and so on (Figure 1.1).
The PaaS model offers a development environment, middleware capabilities, and
a deployment stack for application developers to build tailor-made applications or
host prepurchased SaaS. Amazon Web Services (AWS), Google App Engine, and
Microsoft Azure are a few examples of PaaS. In contrast to SaaS, PaaS does not
abstract development life-cycle support, given that most end users in this model are
application developers. Nevertheless, the abstraction aspect of cloud computing is
Software as a service
(application)
Platform as a service
(operating system)
Infrastructure as a service
(hardware)
FIGURE 1.1 Cloud service models as a stack.
4
Smart Grids
still present in PaaS, where developers rely on underlying abstracted features such as
infrastructure, operating system, backup and version control features, development
and testing tools, runtime environment, worklow management, code security, and
collaborative facilities.
The IaaS model offers the fundamental hardware, networking, and storage capabilities needed to host PaaS or custom user platforms. Services offered in IaaS include
hardware-level provisioning, public and private network connectivity, (redundant)
load balancing, replication, data center space, and irewalls. IaaS relieves end users
of operational and capital expenses. While the other two models also provide these
features, here they are much more prominent, since IaaS is the closest model to actual
hardware. Moreover, since the actual hardware is virtualized in climate-controlled
data centers, IaaS can shield end users from eventual hardware failures, greatly
increasing availability and eliminating repair and maintenance costs. A popular IaaS
provider, Amazon Elastic Compute Cloud (EC2), offers 9 hardware instance families in 18 types [3]. Some of the other IaaS providers include GoGrid, Elastic Hosts,
AppNexus, and Mosso [4].
1.1.2 ADVANCED POWER GRID
Online shopping is just one of many instances where cloud computing is making
its mark on society. The power grid, in fact, is currently at an interesting crossroads in this technological space. One fundamental capability that engineers are
striving to improve is the grid’s situational awareness—its real-time knowledge of
grid state—through highly time-synchronized phasor measurement units (PMUs),
accurate digital fault recorders (DFRs), advanced metering infrastructure (AMI),
smart meters, and signiicantly better communication. The industry is also facing
a massive inlux of ubiquitous household devices that exchange information related
to energy consumption. In light of these new technologies, the traditional power
grid is being transformed into what is popularly known as the “smart grid” or the
“advanced power grid.”
The evolution of the power grid brings its own share of challenges. The newly
introduced data have the potential to dramatically increase accuracy, but only if processed quickly and correctly. True situational awareness and real-time control decisions go hand in hand. The feasibility of achieving these two objectives, however,
heavily depends on three key features:
1. The ability to capture the power grid state accurately and synchronously
2. The ability to deliver grid state data reliably and in a timely manner over a
(potentially) wide area
3. The ability to rapidly process large quantities of state data and redirect the
resulting information to appropriate power application(s), and, to a lesser
extent, the ability to rapidly acquire computing resources for on-demand
data processing
Emerging power applications are the direct beneiciaries of rapid data capture, delivery, processing, and retrieval. One such example is the transition from
Mission-Critical Cloud Computing for Critical Infrastructures
5
conventional state estimation to direct state calculation. Beginning in the early
1960s, the power grid has been employing supervisory control and data acquisition (SCADA) technology for many of its real-time requisites, such as balancing
load against supply, demand response, and contingency detection and analysis.
SCADA uses a slow, cyclic polling architecture in which decisions are based on
unsynchronized measurements that may be several seconds old. Consequently, the
estimated state lags the actual state most. Thus, state estimation gives very limited insight and visibility into the grid’s actual operational status. In contrast, tightly
time-synchronized PMU data streams deliver data under strict quality of service
(QoS) guarantees—low latency and high availability—allowing control centers to
perform direct state calculations and measurements. The capabilities that come with
the availability of status data make creating a real-time picture of the grid’s operational state much more realistic [5].
There are also many myths surrounding the operations of a power grid in conjunction with big data and its eficient use. The following is a nonexhaustive list of
some of these myths.
1. Timeliness: Real-time data is a relative term. Often the application requirements dictate the timeliness needs. Modern software and hardware technologies provide many workarounds on the timeliness of data availability
on wide area networks with average bandwidths. One of them is selective
packet dropping. This technique guarantees a minimum QoS while delivering information to recipients in a timely manner. Smart power grids will
greatly beneit from these techniques.
2. Security and Safety: Security and safety are concerns often cited by decision makers when considering new technologies. While absolute security
is impossible, most concerns arising from data security issues have been
technically addressed. One large factor that affects security is human
errors and oversights. Often, insuficient emphasis is given to this side of
security. More and more emphasis is given to the communication channels.
Securing an already secure channel only results in performance losses and
overheads.
3. Cost: The cost of maintaining information infrastructures has become a
major portion of budgets for large industries, and is a substantial challenge
in running a sustainable, data-centered architecture. Thanks to data centers
and cloud computing infrastructures, these challenges are being successfully addressed. Clouds facilitate outsourcing of large-scale computational
infrastructures while achieving provably reliable QoS.
1.2
CLOUD COMPUTING’S ROLE IN THE
ADVANCED POWER GRID
Cloud computing can play a vital role in improving the advanced power grid’s
situational awareness and the ability to derive better control decisions. As mentioned earlier, emerging power applications will leverage large amounts of data
in making control decisions affecting the stability and reliability of the grid.
6
Smart Grids
Analyzing and processing such large amounts of data require data parallelism and
massive computational capabilities well beyond general-purpose computing. Beyond
data analysis, the future grid can greatly beneit from much more extensive simulation and analysis to remediate stressful situations. These are spontaneous special
purpose applications (e.g., system integrity protection schemes [SIPS], also known as
remedial action schemes [RASs] or special protection schemes [SPSs]) [6], each with
different needs—real time, scaling, and computational—that are triggered by grid
disturbances such as frequency oscillations, voltage luctuations, line overloads, and
blackouts. Moreover, the number of power grid applications and their computational
needs can only be expected to increase as the grid evolves. Managing this variety
of applications and needs presents a challenge. Keeping these applications running
idle on dedicated hardware until the speciic condition is triggered is both ineficient
and expensive.
An elegant solution is presented here which utilizes cloud computing and its
rapid elasticity. Power grid applications can utilize the cloud to rapidly deploy an
application-speciic infrastructure using IaaS and PaaS to achieve new levels of
availability and scalability. Availability and scalability are properties that are much
harder to meet in a piecemeal fashion, but are inherent features of the cloud and
easily adoptable. Cloud-based solutions also beneit entities at different levels of
the control hierarchy, giving them the ability to perform an independent, replicated
analysis on the same sensor data. The ability to elastically manage resources in the
presence of a grid disturbance is extremely attractive in comparison with in-house
solutions, which could be overprovisioned or underprovisioned at the time of need.
Another area where cloud computing performs well is in supporting the varying
needs of the growing ecosystem of power applications. Both PaaS and SaaS will be
useful for developing and supporting power applications.
PaaS for the power grid will need to encompass industry best practices, engineering standards, compliance requirements, and data privacy and security requirements
as properties of the platform itself. The CAP theorem [7] argues that simultaneously
achieving three key properties—consistency, availability, and partition tolerance—
is impossible in distributed systems. As a result, and especially since their apps
are not mission critical, present-day commercial clouds often sacriice consistency
in favor of availability. Cloud environments that are used for power applications
must be able to guarantee high-assurance properties, including consistency, fault
tolerance, and real-time responsiveness, in order to support the anticipated needs of
power applications.
While PaaS enables power researchers and developers to expand the power application ecosystem, SaaS can abstract essential properties and requirements to provide
end-user application solutions. Grid incident-speciic applications can be offered as
SaaS, readily deployable by power utilities. The success of power grid SaaS depends
heavily on the completeness and the richness of power grid PaaS. The overarching
challenge lies in ensuring that power applications delivered across SaaS/PaaS models inherently carry the necessary high-assurance properties. The subtle intricacies
of high-assurance properties, which are often outside the power engineering realm,
will necessitate a different approach to cloud computing as well as a stronger mesh
between power engineering and computer science.
Mission-Critical Cloud Computing for Critical Infrastructures
7
1.2.1 BERKELEY GRAND CHALLENGES AND THE POWER GRID
The Berkeley view of the cloud [8,9] outlines 10 grand challenges and opportunities
for cloud computing. The following list reviews some of these challenges and their
implications for cloud-based power grid applications:
1. Ensuring High Service Availability: Consistency is arguably one of the
most critical requirements for cloud-based power applications [10], but
availability is a close second. Many of the early adopters of cloud technology support availability as a liveness property, while smart-grid applications depend on availability as a safety property. Availability also relates
to the issue of whether cloud-based power grid applications should follow
stateful or stateless models. The ability to design stateful applications often
depends on the availability of state information.
Achieving high availability requires avoiding single point of failure
scenarios and potential bottlenecks. The general consensus is that the
cloud promotes and provides high availability. However, using cloud services from a single service provider allows a single point of failure [9].
Interoperability between different cloud vendors for the sake of availability
for power grid applications merely because of the many proprietary and
market advantages is a far-fetched ambition. Perhaps one solution would
be for the power grid community to manage and operate its own cloud,
either overlaying existing commercial clouds or as a private infrastructure
with built-in replication at all levels. Such an initiative, however, would be
dictated by the many economic drivers.
2. Eliminating Data Transfer Bottlenecks: Large amounts of high-frequency
data must cross the cloud boundary to reach power applications running
within the cloud. Application responsiveness is directly tied to the time
lines with which data reach their destination. The outermost layer of the
cloud can have a dual role as a sensor data aggregator for sources outside the
cloud and as a multiplexer toward the applications within the cloud. Thus, a
suficiently large number of replicated cloud end points for sensor data must
be provided in order to prevent a potential data transfer bottleneck.
3. Assuring Data Conidentiality and Availability: For a community historically notorious for a conservative modus operandi, sharing sensor data is a
frightening proposition. Power grid entities operate under industry regulations and standards that can prevent data sharing in many circumstances.
Additionally, companies are reluctant to share market-sensitive data that
could give away economic advantage. Power application data traversing the cloud must be secured, meeting all compliance requirements, so
that they cannot be used by unintended parties. Thus, the cloud will need
to provide adequate data sanitization and iltering capabilities to protect
application data.
4. Performance Predictability under Scaling: The enormous amount of data
that some power applications require, combined with the impacts of virtualized I/O channels and interrupts, leads to unpredictable performance
8
Smart Grids
during elastic scaling. Different IaaS vendors exhibit different I/O performance, resource acquisition, and release time characteristics. The computational performance on current cloud computing infrastructures also shows
signs of strain under scaling [11]. High-end batch processing applications
will require improved resource sharing and scheduling capabilities for virtual machines to ensure strict QoS demands.
1.3 MODEL FOR CLOUD-BASED POWER GRID APPLICATIONS
Many of the cloud adoption challenges outlined in Section 1.2 are essentially about
supporting highly scalable, highly assured behaviors and stringent communication
guarantees. These are properties that are rarely found in today’s commercial cloud
infrastructures, which are optimized to support mobile applications and web-based
e-commerce applications. The notions of real-time responsiveness, guaranteed consistency, data security, and fault tolerance are signiicantly more forgiving in these
applications than in infrastructure control, supplying little incentive for current commercial clouds to embrace the type of changes necessary to support critical infrastructure systems.
Figure 1.2 visually represents an abstract architectural model for cloud-based
power applications. The architecture includes three basic components:
1. A real-time data collection and transport infrastructure
2. A soft state, elastic outer cloud tier that supports data collection, data preprocessing, temporary archiving, data sanitization and iltering, and multiplexing to services residing in interior tiers
3. Interior cloud tiers hosting services and applications, and supporting data
processing, analysis, batch processing, persistent storage, and visualization
functions
The data collection and transportation infrastructure sits between the physical
sensors and the outermost tier of the cloud, and is the communication backbone of
Outermost cloud tier
(data collectors)
Interior cloud tier
(applications)
Data
multiplexing
Computation
Sensors
Data collection
and
transportation
Data
aggregation
Hard
archiving
Batch processing Visualization
Data filtering
FIGURE 1.2
Analysis
Soft archiving
Data sanitization
An abstract architectural model for a cloud-based power grid application.
Mission-Critical Cloud Computing for Critical Infrastructures
9
the overall architecture. This component is responsible for delivering data that are
produced outside the cloud to the irst-tier cloud collectors with strong QoS guarantees such as guaranteed delivery, ultrahigh availability, ultralow latency, and guaranteed latency. The soft state, outermost cloud tier provides the interface to data
lowing to the applications hosted in the interior tiers. The primary objective of this
tier is to provide high availability, to exhibit rapid elasticity, and to forward correct
data to the appropriate applications. To aid in this process, this tier will also host
auxiliary applications that provide data sanitization, iltering of bad data, buffering
(or soft achieving), data preprocessing, and forwarding capabilities. Availability and
fault tolerance are heightened by replicated shards—nodes that collect data from a
group of sensors—and by mapping sensor data sources appropriately to the shards.
The interior cloud tiers host the actual applications that consume data from the
shards and perform analysis, computation, and batch processing tasks. Additionally,
the results of these deeper computations may be delivered at high rates to visualization applications residing inside and outside the cloud.
1.4 GRIDCLOUD: A CAPABILITY DEMONSTRATION CASE STUDY
An Advanced Research Projects Agency-Energy (ARPA-E)-funded, high-proile
research collaboration between Cornell University and Washington State University
is spearheading efforts to develop, prototype, and demonstrate a powerful and comprehensive software platform realizing the cloud computing needs of the future
power grid. Appropriately named GridCloud [12], this research project aims to bring
together best-of-breed, already existing high-assurance distributed system technologies as a basis to innovate new cloud architectural models for the monitoring, management, and control of power systems. The technologies integrated in this effort
include GridStat [13,14], Isis2 [15,16], TCP-R [17], and GridSim [18]. A brief description of each of these technologies is presented here.
1.4.1 GRIDSTAT
GridStat implements a data delivery overlay network framework designed from
the bottom up to meet the challenging requirements of the electric power grid (see
Chapter 4). Power grids today are increasingly populated with high-rate, time-synchronized sensors that include PMUs and DFRs, whose functionalities are actually blurring.
High-rate, time-synchronized data are expected to form the basis of many monitoring and control applications with a wide variety of delivery requirements and conigurations across such dimensions as geographic scope, latency, volume, and required
availability [19]. These needs cannot be met by Internet protocol (IP) multicast, which
forces all subscribers of a given sensor variable to get all updates at the highest rate
that any subscriber requires. They also cannot be met by multiprotocol label switching
(MPLS), which is not designed to provide per-message guarantees (only overall statistical guarantees) and also only has three bits (eight categories) with which to categorize
the millions of different sensor lows that will likely be deployed in 5–10 years.
GridStat delivers rate-based updates of sensor variables with a wide range of
QoS+ guarantees (latency, rate, availability) that include support for ultralow latency
10
Smart Grids
and ultrahigh availability, which are implemented by sending updates over redundant
disjoint paths, each of which meets the end-to-end latency requirements for the given
subscription. Additionally, GridStat enables different subscribers to a given sensor
variable to require different QoS+ guarantees, which can greatly reduce bandwidth
requirements and improve scalability.
GridStat’s data delivery plane is a lat graph of forwarding engines (FEs), each
of which stores the state for every subscription whose updates it forwards. FEs forward sensor updates on each outgoing link at the highest rate that any downstream
subscriber requires. They drop updates that are not needed downstream, based on
the expressed rate requirements of subscribers. GridStat’s management plane is
implemented as a hierarchy of QoS brokers that can be mapped onto the natural
hierarchy of the power grid. Each node in the hierarchy is designed to contain policies for resource permissions, security permissions, aggregation, and adaptations to
anomalies. With these policies, the management plane calculates the paths required
for the data delivery (with the given number of disjoint paths) and updates the forwarding tables in the FEs. Applications interact with GridStat using publisher and
subscriber software libraries through which the applications’ requirements for QoS
are conveyed to the management plane. GridStat incorporates mechanisms for securing communication between the management plane entities and those of the data
plane. Security mechanisms for end-to-end message security between publishers and
subscribers are modular and conigurable, allowing different data streams and applications to fulill different security and real-time requirements [20]. GridStat in the
power grid provides the opportunity to respond to different power system operating
conditions with different communication conigurations. GridStat provides a mechanism by which communication patterns can be rapidly changed among multiple preconigured modes in response to anticipated power system contingencies.
1.4.2 ISIS2
Isis2 is a high-assurance replication and coordination technology that makes it easy
to capture information at one location and share it in a consistent, fault-tolerant,
secure manner with applications running at other locations—perhaps great numbers of them. This system revisits a powerful and widely accepted technology for
replicating objects or computations, but with a new focus on running at cloud scale,
where the system might be deployed onto thousands of nodes and supporting new
styles of machine-learning algorithms. Isis2 enables massive parallelism, strong consistency, and automated fault tolerance, and requires little sophistication on the part
of its users. With Isis2, all computational nodes and applications sharing the same
data see it [the data?] evolve in the same manner and at nearly the same time, with
delays often measured in hundreds of microseconds. The system also supports replicated computation and coordination: with Isis2 one could marshal 10,000 machines
to jointly perform a computation, search a massive database, or simulate the consequences of control actions, all in a manner that is fast, secure against attack or intrusion, and correct even if some crashes occur.
The form of assurance offered by Isis2 is associated with a formal model that
merges two important prior models—virtual synchrony [21] and Paxos [22]. Isis2
Mission-Critical Cloud Computing for Critical Infrastructures
11
embeds these ideas into modern object-oriented programming languages. Isis2 is
used to create two new components for GridCloud: a version of the technology specialized for use in wide-area power systems networks, and support for high-assurance
smart-grid applications that are hosted in cloud computing data centers.
The GridCloud researchers believe that Isis2 can be used to support services that
run on standard cloud infrastructures and yet (unlike today’s cloud solutions) are
able to guarantee continuous availability, automatically adapting under attack so that
intruders cannot disrupt the grid even if a few nodes are compromised. They are
also analyzing and demonstrating the best options for building cloud services that
respond to requests in a time-critical manner.
1.4.3 TCP-R
GridCloud will tie together a very large number of components, including
sensors, actuators, forwarding elements and aggregators, cloud-based services,
and so on, using Internet standards. For best performance, it is important that
related components communicate using persistent, stateful connections. Stateful
connections reduce retransmissions and wasteful connections, and provide better
low control. The standard for stateful connections in the Internet is transmission
control protocol (TCP). TCP provides network connections that provide reliable irst
in, irst out (FIFO) communication as well as fair low provisioning using adaptive
congestion windows.
Consider a cloud service that sends commands to a large number of actuators.
The cloud service consists of a cluster of a few hundred servers. To keep actuators
simple, and also to allow lexibility in evolving the cloud service, the cloud service
should appear to the actuators as a single end point with a single TCP/IP address.
While an actuator will receive commands from a particular server machine in the
cluster, it appears to the actuators (and their software) as if the cloud service is a
single, highly reliable, and fast machine. It is desirable to maintain this illusion even
when connections migrate between server machines for load balancing, for hardware
or software upgrades, or when rebooting cloud servers. TCP connections, unfortunately, are between socket end points, and, using current operating systems abstractions, socket end points cannot migrate or survive process failures. Also, the cloud
service would have to maintain a TCP socket for every actuator. This does not scale
well, as each TCP socket involves storing a lot of state information.
Replacing TCP with a radically different protocol would not be feasible today.
Operating systems and even networking hardware implement TCP connections
very eficiently. TCP is the dominant communication protocol on the Internet, and
Internet routers have evolved to support TCP eficiently, easily scaling to many millions of simultaneous TCP connections. TCP-R proposes to support standard TCP
connections, but to extend them with a technology that addresses the shortcomings
mentioned above. The essential idea is to extend the cloud service with a ilter that
intercepts and preprocesses TCP packets. The ilter is scalable and maintains little
state per TCP connection (on the order of 32 bytes). It has only soft state (i.e., it does
not have to store its state persistently across crashes, greatly simplifying fault tolerance). The ilter allows servers to migrate TCP connections, and TCP connections
12
Smart Grids
to survive server failure and recovery. Originally developed to maintain TCP connections between border gateway protocol (BGP) (Internet routing) servers across
failures and subsequent recovery [23], TCP-R is extended into a scalable technology
for a cluster serving client end points and also to “park” connections that are not
currently live.
1.4.4
GRIDSIM
GridSim is a real-time, end-to-end power grid simulation package that is unique in
its integration of a real-time simulator, data delivery infrastructure, and multiple
applications all running in real time. The goal of this project is to simulate power
grid operation, control, and communications on a grid-wide scale (e.g., the Western
Interconnection), as well as to provide utilities with a way to explore new equipment
deployments and predict reactions to contingencies. The ability to simulate operation under different equipment deployment conigurations includes large-scale conigurations of PMUs. With the objective of simulating real-world equipment usage,
and usage in conjunction with readily available industry equipment, the GridSim
simulation package uses the industry standard C37.118 data format for all streaming
measurement data.
The irst element in the GridSim platform is a transient power stability simulator, specially modiied to output streaming data in real time. The output data are
encoded into C37.118 and sent to a huge number of substation processes. At each of
these processes, the data are aggregated, as would be done in a real power utility substation. The data are also sent to any of the substation-level power applications that
are running. Both the raw substation data as well as any power application outputs
are then published to GridStat.
GridStat allows the substation data to be distributed as they would be in the real
world. Published data can be sent via redundant paths, 1◊many communication
(publish-subscribe, whose degenerate version is network-level multicast), and so on.
The lexibility provided by the GridStat data delivery middleware allows subscription applications to be easily integrated into the system with minimal reconiguration. Published data are available to any subscribers of GridStat, including the two
applications included in the GridSim simulation, the hierarchical state estimator
(HSE), and the oscillation and damping monitor (Figure 1.3).
1.4.5 GRIDCLOUD ARCHITECTURE
GridCloud was designed with the expectation that the developers of the advanced
power grid will require easy access to large computing resources. Tasks may require
large-scale computation, or may involve such large amounts of data that simply hosting and accessing the data will pose a substantial scalability challenge. This leads
us to believe that cloud computing will play an important role in the future grid,
supplementing the roles played by existing data center architectures. The compelling economics of cloud computing, the ease of creating “apps” that might control
household power consumption (not a subject that has been mentioned yet), and the
remarkable scalability of the cloud all support this conclusion.
13
Mission-Critical Cloud Computing for Critical Infrastructures
Powertech
TSAT simulator
Measurement
generator
Static data
generator
Simulated
power
system
C37.118
generator
GridStat
FE
FE
Substationlevel
simulation
Substation
Substation N
O Substation
Su O Substation
SE Su OM
Substation
SE Sub OM
Substation
SE Subs OM
Substation
SE Substation gateway
SE
FE
FE
FE
FE
Substation 1
Controllevel
applications
OpenPDC
Oscillation
monitor
State
estimator
FIGURE 1.3 The GridSim architecture. (From Anderson, D., Zhao, C., Hauser, C.,
Venkatasubramanian, V., Bakken, D., and Bose, A., IEEE Power and Energy Magazine, 10,
49–57, 2012.)
Figure 1.4 shows the architecture of GridCloud. The representative application
used in this case is a HSE [18,24,25]. Data sources represent PMUs, which stream
data to data collectors across a wide area using GridStat. The HSE comprises several
substation-level state estimators that aggregate, ilter, and process PMU data before
forwarding them to a control center-level state estimator. The input and the irst-level
computation are inherently sharded at substation granularity. Furthermore, computations are inherently parallel between substations. Thus, the HSE has a natural mapping in GridCloud with substation state estimators residing in the outermost tier of
the cloud while the control center state estimator is moved to the interior tier. The
substation state estimators are replicated to increase fault tolerance and availability.
The consistency of the replicas is managed through Isis2. TCP-R is used to provide
fail-over capabilities for connections across the cloud.
1.5
CONCLUSIONS
This chapter presents a roadmap of how cloud computing can be used to support the
computational needs of the advanced power grid. Today’s commercial cloud computing infrastructure lacks the essential properties required by power grid applications.
These deiciencies are explained and a cloud-based power grid application architecture is presented which overcomes these dificulties using well-known distributed
system constructs. Furthermore, the GridSim project, which instantiates this model,
is presented as a case study example.
14
EC2 cloud
COL11
COL21
S1
COLN1
P1
S-SE11
S-SE21
Isis2
S-SEN1
TCP-R
FE
Control center SE
COL
FE
GridStat
(UDP)
P2
1
2
COL22
FE
COLN2
FE
S-SE12
S-SE22
S-SE1
S-SE2
S-SEN2
FE
FE
Results
S2
Computation
FE
Data origin
Localhost
Local client
and
visualizer
S-SEN
FE
PM
SK
COL1M
COL2M
COLNM
GridCloud
S-SE2M
S-SENM
Output
Smart Grids
FIGURE 1.4 GridCloud architecture.
S-SE1M
Mission-Critical Cloud Computing for Critical Infrastructures
15
REFERENCES
1. G. Fulgoni, A. Lipsman and I. Essling, State of the U.S. Online Retail Economy in Q4
2012, February 2013. Available at: http://goo.gl/BCnbn.
2. P. Mell and T. Grance, The NIST Deinition of Cloud Computing: Recommendations of
the National Institute of Standards and Technology. The National Institute of Standards
and Technology, Gaithersburg, MD, 2011.
3. A. W. Services, Amazon elastic compute cloud: User guide API version 2013-02-01.
http://awsdocs.s3.amazonaws.com/EC2/latest/ec2-ug.pdf (accessed 3 June 2013).
4. R. Prodan and S. Ostermann, A survey and taxonomy of infrastructure as a service
and web hosting cloud providers, in Proceedings of the 10th IEEE/ACM International
Conference on Grid Computing, October 13–15, 2009, Banff, AB.
5. D. Novosel, V. Madani, B. Bhargava, K. Vu and J. Cole, Dawn of the grid synchronization, IEEE Power and Energy Magazine, 6(1), 49–60, 2008.
6. S. Horowitz, D. Novosel, V. Madani and M. Adamiak, System-wide protection, IEEE
Power and Energy Magazine, 6(September), 34–42, 2008.
7. S. Gilbert and N. Lynch, Brewer’s conjecture and the feasibility of consistent, available,
partition-tolerant web services, ACM SIGACT News, 33(2), 51–59, 2002.
8. M. Armbrust, A. Fox, R. Grifith, A. D. Joseph, R. Katz, A. Konwinski, G. Lee, et al.,
A view of cloud computing, Communications of the ACM, 53(4), 50–58, 2010.
9. M. Armbrust, A. Fox, R. Grifith, A. D. Joseph, R. Katz, A. Konwinski, G. Lee, et al.,
Above the clouds: A Berkeley view of cloud computing, Department of Electrical
Engineering and Computer Sciences, University of California Berkeley, Report UCB/
EECS 28, 2009.
10. K. Birman, L. Ganesh and R. V. Renesse, Running smart grid control software on
cloud computing architectures, in Proceedings of the Computational Needs for the Next
Generation Electric Grid Workshop, Cornell University, April, 2011, New York.
11. A. Iosup, S. Ostermann, M. Yigitbasi, R. Prodan, T. Fahringer and D. H. J. Epema,
Performance analysis of cloud computing services for many-tasks scientiic computing,
IEEE Transactions on Parallel and Distributed Systems, 22(6), 931–945, 2011.
12. K. Birman, A. Bose, D. Bakken and C. Hauser, GridControl: A software platform to
support the smart grid, A Cornell University and Washington State University Research
Collaboration, 2011. Available at: http://www.cs.cornell.edu/Projects/gridcontrol/index.
html#gridcloud (accessed 3 June 2013).
13. H. Gjermundrod, D. Bakken, C. Hauser and A. Bose, GridStat: A lexible QoS-managed
data dissemination framework for the power grid, IEEE Transactions on Power Delivery,
24(1), 136–143, 2009.
14. C. Hauser, D. Bakken and A. Bose, A failure to communicate: Next generation communication requirements, technologies, and architecture for the electric power grid, IEEE
Power and Energy Magazine, 3(2), 47–55, 2005.
15. K. P. Birman, D. A. Freedman, Q. Huang and P. Dowell, Overcoming CAP with consistent soft-state replication, IEEE Computer, 45(2), 50–58, 2012.
16. K. Birman, Isis2 cloud computing library, 2013. Available at: http://isis2.codeplex.com/.
17. A. Agapi, K. Birman, R. M. Broberg, C. Cotton, T. Kielmann, M. Millnert, R. Payne,
R. Surton and R. van Renesse, Routers for the cloud: Can the Internet achieve 5-nines
availability?, IEEE Internet Computing, 15, 72–77, 2011.
18. D. Anderson, C. Zhao, C. Hauser, V. Venkatasubramanian, D. Bakken and A. Bose,
A virtual smart grid: Real-time simulation for smart grid control and communications
design, IEEE Power and Energy Magazine, 10(1), 49–57, 2012.
19. D. E. Bakken, A. Bose, C. H. Hauser, D. E. Whitehead and G. C. Zweigle, Smart generation and transmission with coherent, real-time data, Proceedings of the IEEE, 99(6),
928–951, 2011.
16
Smart Grids
20. E. Solum, C. Hauser, R. Chakravarthy and D. Bakken, Modular over-the-wire
conigurable security for long-lived critical infrastructure monitoring systems, in
Proceedings of the Third ACM International Conference on Distributed Event-Based
Systems (DEBS), Nashville, TN, 2009.
21. K. Birman, Reliable Distributed Systems: Technologies, Web Services, and Applications,
Springer, New York, 2005.
22. L. Lamport, “Paxos made simple, ACM SIGACT News, 32(4), 18–25, 2001.
23. R. Surton, K. Birman, R. Broberg, T. Marian and R. van Renesse, Man-in-the-Middle
TCP recovery, Cornell University. Available at: http://goo.gl/9S4Jz (accessed 3 June
2013).
24. W. Jiang, V. Vittal and G. Heydt, A distributed state estimator utilizing synchronized
phasor measurements, IEEE Transactions on Power Systems, 22(2), 563–571, 2007.
25. M. Zhao and A. Abur, Multi area state estimation using synchronized phasor measurements, IEEE Transactions on Power Systems, 20(2), 611–617, 2005.
2
Power Application
Possibilities with
Mission-Critical
Cloud Computing
David Bakken, Pranavamoorthy Balasubramanian,
Thoshitha Gamage, Santiago Grijalva,
Kory W. Hedman, Yilu Liu, Vaithianathan
Venkatasubramanian, and Hao Zho
CONTENTS
2.1 Overview......................................................................................................... 18
2.2 Robust Adaptive Topology Control ................................................................ 18
References ................................................................................................................ 19
2.3 Adaptive Real-Time Transient Stability Controls ...........................................20
Reference ................................................................................................................. 21
2.4 Prosumer-Based Power Grid........................................................................... 21
2.4.1 Introduction ........................................................................................ 21
2.4.2 Prosumer-Based Control Architecture ............................................... 22
2.4.3 Computational Challenges .................................................................. 23
2.4.4 Cloud Computing in the Future Electric Grid .................................... 23
2.4.5 Economic Dispatch of Stochastic Energy Resources ......................... 23
2.4.6 Cloud-Based Apps ..............................................................................24
2.4.7 Scenario Analysis and Transmission Planning ..................................24
2.4.8 Model Integration ...............................................................................24
2.4.9 Exploiting and Abstracting Self-Similarity ........................................25
References ................................................................................................................25
2.5 Wide-Area Frequency Monitoring .................................................................26
2.5.1 Introduction ........................................................................................26
2.5.2 FNET Architecture .............................................................................26
2.5.3 FNET Applications ............................................................................. 27
2.5.4 Cloud Computing and FNET.............................................................. 27
2.5.5 Rapidly Elastic Data Concentrators .................................................... 27
2.5.6 Computational Requirement Flexibility .............................................28
2.5.7 Cloud-Based Applications ..................................................................28
References ................................................................................................................ 29
17
18
Smart Grids
2.6 Oscillation Mitigation Strategies .................................................................... 29
References ................................................................................................................ 30
2.7 Automatic Network Partitioning for Steady-State Analysis ........................... 30
References ................................................................................................................ 31
2.1
OVERVIEW
As we have seen in Chapter 1, not only is cloud computing coming to the grid, but
mission-critical implementations such as GridCloud can provide mission-critical
properties. This chapter explores new applications enabled by such technology.
While this chapter is only scratching the surface of what is likely to be routine in a
decade, we hope that it provides a tantalizing glimpse of what is possible.
What, then, is mission-critical cloud computing? To recap, in a nutshell, it
• Keeps the same fast throughput as generic commercial cloud platforms
• Does not deliberately trade off this throughput to allow “inconsistencies,”
for example, when a replica does a state update on a copy of the state but
this update is “forgotten”
• Is much more predictable (and faster) in terms of ramp-up time, central
processing unit (CPU) performance per node, and number of nodes
Therefore, the question for power application developers is how they can use:
• Hundreds of processors in steady state.
• Thousands or tens of thousands of processors when a contingency is reached
or is being approached. Note: often there are many minutes of advanced
warning of this, sometimes an hour or more.
• Data from all participants in a grid that is enabled quickly when a crisis is
approached (though, for market reasons, not necessarily during steady state).
With this in mind, we now present groundbreaking applications that can exploit
such mission-critical cloud platforms.
2.2
ROBUST ADAPTIVE TOPOLOGY CONTROL
Balasubramanian and Hedman
The electric power transmission system is one of the most complex systems available
today. Traditionally, bulk power transmission systems (lines and transformers) are
treated as static assets, even though these resources are controllable. However, it is
known that transmission topology control has been used in the past and is still being used
for corrective-based applications; for example, PJM uses corrective topology control as
a special protection scheme (SPS) [1]. These switching actions are primarily taken on
an ad hoc basis, determined by the system operators based on past historical data rather
than in an automated way based on decision support tools. Past research has demonstrated the ability of topology control to help improve voltage proiles, increase transfer
capacity, improve system reliability, and provide cost beneits to the system [2–8]. Even
Power Application Possibilities with Mission-Critical Cloud Computing
19
though transmission topology control can provide these beneits, harnessing such lexibility from the transmission network in existing operational procedures is limited due
to the computational challenges of optimizing the transmission topology.
More recently, sensitivity-based methods have been proposed as a mechanism to
reduce the computational complexity [9–12]. The robust adaptive topology control
method develops a sensitivity-based heuristic, which reduces the computational time
of the topology control problem. An expression is derived indicating the impact of
changing the state of a transmission line on the objective. This expression is used
to generate a line-ranking system with the potential candidate lines for switching
based on a direct current (dc) optimal power low, which builds on the work of [12].
This approach selects a single feasible switching action per iteration, which provides
an improvement to the system. The advantage of this method is that it solves linear
programs iteratively to come up with a beneicial line-switching solution, which is
computationally simple as compared with other methods employing mixed integer
programming. All the possible switching solutions are lined up in the ranked list, with
the switching action most likely to be beneicial placed at the top of the list. As the
list is formed based on a sensitivity study, the switching action is not guaranteed to
improve the system. Hence, the switching actions need to be checked for alternating
current (ac) feasibility and whether they truly provide an improvement in the objective
before they are implemented. This is done by selecting the irst action from the ranked
list and simulating the switching to ind the improvement in the system. If the switching is not beneicial, the next action in the ranked list is checked for improvement.
This process is continued until a beneicial switching action is found. While such a
procedure is a heuristic, prior work has shown substantial economic savings [9] as
well as strong performance in comparison with global optimization techniques [12].
The processing time taken to come up with a beneicial switching action could
be signiicantly reduced if this process were parallelized so that all the proposed
switching solutions could be checked at once. This opens up enormous opportunities
for the application of cloud computing to transmission-switching applications, which
would drastically reduce the computational time and improve the solution quality,
as the best solution from the ranking list could be identiied very quickly. With prior
research demonstrating cost savings of close to 4% for a $500 billion industry [3],
there is a great opportunity for advanced decision support tools to ill this technological need, in terms of both algorithm sophistication and advanced computing
capabilities, such as cloud computing.
REFERENCES
1. PJM, Manual 3: Transmission Operations, Revision: 40, 2012. Available at: http://www.
pjm.com/~/media/documents/manuals/m03.ashx.
2. W. Shao and V. Vittal, Corrective switching algorithm for relieving overloads and voltage violations, IEEE Transactions on Power Systems, 20(4), 1877–1885, 2005.
3. K. W. Hedman, M. C. Ferris, R. P. O’Neill, E. B. Fisher, and S. S. Oren, Co-optimization
of generation unit commitment and transmission switching with N-1 reliability, IEEE
Transactions on Power Systems, 25(2), 1052–1063, 2010.
4. K. W. Hedman, R. P. O’Neill, E. B. Fisher, and S. S. Oren, Optimal transmission switching
with contingency analysis, IEEE Transactions on Power Systems, 24(3), 1577–1586, 2009.
20
Smart Grids
5. A. Korad and K. W. Hedman, Robust corrective topology control for system reliability,
IEEE Transactions on Power Systems, 28(4), 4042–4051, 2013.
6. K. W. Hedman, R. P. O’Neill, E. B. Fisher, and S. S. Oren, Smart lexible just-in-time transmission and lowgate bidding, IEEE Transactions on Power Systems, 26(1), 93–102, 2011.
7. E. B. Fisher, R. P. O’Neill, and M. C. Ferris, Optimal transmission switching, IEEE
Transactions on Power Systems, 23(3), 1346–1355, 2008.
8. K. W. Hedman, S. S. Oren, and R. P. O’Neill, A review of transmission switching and
network topology optimization, in Proceedings of IEEE Power and Energy Society
General Meeting, July 2011, Detroit, MI.
9. P. A. Ruiz, J. M. Foster, A. Rudkevich, and M. C. Caramanis, On fast transmission
topology control heuristics, in Proceedings of IEEE Power and Energy Society General
Meeting, July 2011, Detroit, MI.
10. J. M. Foster, P. A. Ruiz, A. Rudkevich, and M. C. Caramanis, Economic and corrective
applications of tractable transmission topology control, in Proceedings of 49th Annual
Allerton Conference on Communication, Control, and Computing, pp. 1302–1309,
September 2011, Monticello, IL.
11. P. A. Ruiz, J. M. Foster, A. Rudkevich, and M. C. Caramanis, Tractable transmission
topology control using sensitivity analysis, IEEE Transactions on Power Systems, 27(3),
1550–1559, 2012.
12. J. D. Fuller, R. Ramasra, and A. Cha, Fast heuristics for transmission line switching,
IEEE Transactions on Power Systems, 27(3), 1377–1386, 2012.
2.3
ADAPTIVE REAL-TIME TRANSIENT STABILITY CONTROLS
Venkatasubramanian
The power system is expected to undergo major changes in the next decade, resulting
from rapid growth in system loads (such as electric cars) and from increased dependence on renewable intermittent generation. To face up to these challenges, power
utilities are making major upgrades to wide-area monitoring and control technologies, with impetus from major federal investments in the past few years.
Power system operation is designed to withstand small- and large-scale disturbances. However, when the system is subjected to several large disturbances in a short
span of time, it may become vulnerable to blackouts. Some recent events, such as the
2012 San Diego blackout and the 2003 Northeastern blackout, point to the need for
adaptive real-time transient stability control designs that are speciically designed on
an adaptive premise of making control decisions during the evolution of the event.
In the present-day power system, wide-area transient stability controls such as remedial action schemes (RAS) or SPS are hard-coded control algorithms that are triggered
by a central controller in response to the occurrence of speciic contingencies based on
preset switching logic. When the system is subject to any “unknown” set of contingencies that is not part of the RAS controller logic, the system operation typically switches
to a “safe mode” whereby interarea power transfers are limited to low conservative
settings. The tie-line transfers remain at safe low values until the reliability coordinator
completes a new set of transient stability simulation studies, which results in signiicant
economic losses due to operation at nonoptimal power transfer levels.
Cloud computing emerges as an ideal platform for handling transient stability mitigation issues, both for the present-day power system and for future control
designs. In present-day operation, whenever the system operation is found to be in
Power Application Possibilities with Mission-Critical Cloud Computing
21
one of the “unknown” operating conditions, the reliability coordinator can dial in
a vast amount of cloud-based processing power to carry out the massive number of
new transient stability simulations needed for determining the safe transfer limits.
In the future, we need to rethink the design of transient stability controls such as
RAS or SPS schemes. The massive computational capability offered by cloud computing opens up truly novel futuristic control schemes for mitigating transient stability events, as proposed in [1]. In the present-day power system, simulation studies
are performed off-line for a “guesstimated” list of potential contingencies, and RAS
schemes are implemented for a subset of problematic N−2 or higher-order contingencies whenever needed. Such RAS schemes, then, only work for a limited number of
potential scenarios. Moreover, the respective control actions in these RAS schemes
are also designed to be conservative, being based on off-line studies.
Zweigle and Venkatasubramanian [1] propose to select and implement transient stability controls based on simulations of the system in real time during the evolution of the
events themselves. Wide-area monitoring from an abundance of phasor measurement
units (PMU) in the future will pave the way for real-time monitoring of the state and system topology of the full-size power system. Combining this real-time state information
with real-time simulations will allow us to evaluate which control actions are optimally
suited to the system at the present time, and the decisions are fully adaptive to whatever the system conditions are. Since the controller continues to monitor the system in a
closed-loop fashion, the proposed control schemes are also robust with respect to simulation errors and communication or actuation failures. The formulation is not restrictive to
any subset of contingencies, and can handle low-probability events consisting of multiple
outages, such as those that have served as precursors to large blackouts in the past.
In this proposed formulation, denoted as adaptive real-time transient stability controls, massive processing power is needed to carry out “what if” simulations of many
potential control candidates in parallel before deciding on whether any control action is
needed and which speciic action(s) will be implemented. The system monitoring and
simulations of “what if” scenarios will continue throughout the event until the system
has been stabilized. Once the controller recognizes that the system has returned to its
normal state, the controller returns to dormant system monitoring mode, and cloud
resources can be released. Details of the control algorithms can be seen in [1].
REFERENCE
1. G. Zweigle and V. Venkatasubramanian, Wide-area optimal control of electric power
systems with application to transient stability for higher order contingencies, IEEE
Transactions on Power Systems, 28(3), 2313–2320, 2013.
2.4
PROSUMER-BASED POWER GRID
Gamage and Grijalva
2.4.1
IntroductIon
The electric power grid, in a bid to improve its sustainability, is aggressively exploring ways to integrate distributed renewable energy generation and storage devices
22
Smart Grids
at many levels. The most obvious integration is at the level of generation, where
renewable generation sources such as large wind turbine and solar panel farms will
supplement and eventually (it is hoped) supplant traditional nonrenewable power
generation sources. Another natural integration is at the distribution level, where
relatively smaller-scale renewable energy generation by utilities and other power distribution entities offers cheaper and greener energy options to customers. While not
on the same bulk scale as the generation or distribution level, an emerging trend
in recent years is for end consumers who are typically below the distribution level
(e.g., households, microgrids, and energy buildings) to generate their own power
using renewable sources and become self-sustainable and energy-independent of the
grid.
A fascinating aspect of this changing energy landscape is the drastic changes in
the roles of the players involved. For example, end consumers, in addition to their
typical energy consumption role, are economically motivated to sell excess energy
and provide energy storage services to the grid. Modern utilities also go beyond
their traditional energy distribution role in buying energy from end consumers when
available. Similar role augmentations can be observed at all levels of the modern
electric power grid [1]. As a consequence, what traditionally has been a one-way
energy transfer—from bulk generation, through transmission and distribution, to
end-user consumption—is transforming into a two-way energy exchange.
2.4.2 Prosumer-Based control archItecture
The key entities in this evolving electric power grid are prosumers, economically
motivated power system participants that can consume, produce, store, and transport
energy [1,2,3]. Examples of prosumers include a building equipped with an electric
vehicle parking lot that provides storage services to a utility, a hybrid (generates
its own power but is not completely off-grid) renewable energy microgrid that sells
excess power at peak generation to a neighboring microgrid or utility, and two households with renewable resources exchanging power on demand. In fact, almost all
current power grid players can be conceptualized as prosumers [2]. A key characteristic of prosumers is that they assume different roles depending on the situation and
underlying conditions; a prosumer who is a consumer at one moment can become
a producer at the next. This is analogous to a peer-to-peer model versus a typical client–server model in computer science. Prosumers interact with one another
through the services they offer. For example, a utility prosumer that aggregates heterogeneous home-user prosumers provides consumption and storage services to a
distributed independent service operator (ISO) prosumer. Moreover, at any given
instant, different prosumers may operate under different satisfaction and objective
functions (comfort, cost, eficiency, security, sustainability, reliability, etc.) that will
directly inluence their control decisions. Thus, the control architecture that revolves
around prosumers is radically different from the current centralized hierarchical
control architecture. Instead of waiting for control decisions to trickle down the hierarchy, a prosumer-based control architecture promotes autonomy by taking proactive
and distributed control steps that relect the local, internal, and external state of the
network of prosumers.
Power Application Possibilities with Mission-Critical Cloud Computing
2.4.3
23
comPutatIonal challenges
An important, but often overlooked, trait of the future electric power grid is its
multidimensional, multiscale nature. The existing electricity industry is confronted by a massive invasion of energy-aware ubiquitous household devices that
aim to empower consumers to make intelligent and mindful energy choices. The
industry itself is invested in improving the grid’s visibility and real-time sensing
capabilities by deploying PMUs, smart meters, and other intelligent and highly
accurate electronic devices. An inadvertent effect of this inlux of new devices
is that they produce data with temporal, spatial, and scenario signiicance that
are useful in planning, simulation, operation, and control of the future power
grid [4]. If the next evolution of the power grid is a prosumer-oriented two-way
energy exchange, the true beneit of such massive quantities of multidimensional
data profoundly depends on the ability to extract and render usable, useful, and
mission-critical information at the right time and deliver it to the right destinations. Inarguably, the computational complexity of this daunting task is beyond
the capability of today’s general-purpose computers. Additional major driving
forces, such as federal and industry initiatives to incorporate renewable energy
resources, various energy eficiency programs, and unconventional and novel
system behaviors, coupled with consumer perceptions about the type of services
they expect from the future power grid, add granularity on multiple scales to this
computational challenge.
2.4.4
cloud comPutIng In the Future electrIc grId
There are numerous computational challenges and limitations associated with
existing computational models and analytical tools [4]. A primary objective of
today’s power engineers and researchers alike is to design, develop, and innovate
new models and analytical tools that can support and fully exploit massive multidimensional and multiscale data. Below are a few such examples that have cloud
computing implications in the domain of potential solutions.
2.4.5
economIc dIsPatch oF stochastIc energy resources
The inherent unpredictability and the variable nature of stochastic energy
resources such as wind energy are making it dificult for utilities with large
penetration of renewables to fully exercise real-time economic dispatch with a
iner granularity [5]. The research community has already identiied the need to
replace current economic dispatch software with short-term stochastic scheduling
software that has a granularity of seconds, minutes, or a few hours [6,7]. Such
software has highly elastic computational needs that scale up and down based on
the conditions, the level of penetration, and the availability of information, thus
offering an invitation to exploit the cloud’s rapid elasticity feature [8] to provision
software, platform, and computational infrastructure with different conigurations
on demand, rather than in-house computational stacks that offer little lexibility,
scalability, or availability.
24
2.4.6
Smart Grids
cloud-Based aPPs
Most internal events experienced by prosumers are not visible to external entities.
However, there are circumstances in which knowledge of such events becomes signiicant beyond the immediate control domain. For example, conventional load forecasting by ISOs within a 2% error bound will need to be tightened even further with
the deployment of distributed energy resources and utility-scale storage devices [9].
The enablers needed for such capability are widely available and acceptable new
modeling and simulation methods—single- versus three-phase modeling, nonstandardization of operational and planning models, node/breaker versus bus/branch
modeling, data privacy, operational liability—which can be offered as softwareas-a-service (SaaS) solutions.
2.4.7
scenarIo analysIs and transmIssIon PlannIng
Fundamentally, the power grid has resorted to using an N−1 contingency analysis of bulk transmission system security. The reality, however, is that, to achieve
a higher level of system reliability, the power grid must be capable of evaluating a
large number of plausible contingencies beyond N−1 conditions. This is certainly a
computationally very intensive process. A heuristically feasible option would be to
systematically prioritize searches, reduce scenario spaces, and make dynamically
updated multidimensional information available to a wider audience of power grid
entities. This is still beyond the capability of general-purpose computers, and would
require a suficiently large high-performance computing facility. It is highly unrealistic to assume that every power grid entity will be excited about the proposition
of individually housing such a highly expensive facility, not to mention the highly
skilled workforce required to maintain it.
A more likely solution is to utilize cloud computing infrastructures on the basis of
need and usage, while greatly reducing the capital investment needed and abstracting the underlying maintenance cost and workforce requirements. Cloud computing can commission massive parallel computations to explore the huge permutation
space on demand, and can also share the indings much more quickly with all interested entities. Even the data aggregation, mining, and analysis tasks can be ofloaded
onto cloud-based computational platforms that offer better scalability features under
changing conditions.
2.4.8
model IntegratIon
Power systems at various spatial scales are modeled differently. Information
exchange and seamless modeling integration require signiicant processing capabilities. Often, the data from more detailed models have to be aggregated and transferred “by hand” to other applications. Cloud-based modeling solutions, on the other
hand, can greatly improve this process by using in-the-cloud dynamic and highly
automated aggregation and delivery capabilities. These capabilities can be encapsulated as a library of models with a consistent description of phenomena, and various
scales can be generalized and offered as platform-as-a-service (PaaS) options with
well-deined application programming interfaces (APIs) to system developers who
Power Application Possibilities with Mission-Critical Cloud Computing
25
can custom-build models that suit the individual spatial scaling requirements of different power system entities. Moreover, entities such as utilities or microgrids that
operate individual control systems with homogeneous objective functions (e.g., cost
minimization) can be aggregated for better control stability. Thus, the natural hierarchy of spatially separated power grid entities can be uniied under common protocols
while still abiding by any regulatory or compliance standards.
2.4.9
exPloItIng and aBstractIng selF-sImIlarIty
Self-similarity is the notion of entities at various levels of a system hierarchy exhibiting similar characteristics [10,11]. This is observable in the power grid, where entities
at a lower level of the system are governed by physics similar to that of higher-level
entities, but at a lower scale. What this means for a prosumer-based power grid is
that multiscale prosumers at different levels can beneit from models that are based
on the same reference model, which enables expansion and collapse of lower-level
information as well as seamless integration and aggregation at higher levels. Since
similar physics takes place at various hierarchical levels of the power system, data
at different aggregation levels can be theoretically collapsed or expanded dynamically depending on the required granularity. For example, a combined securityconstrained optimal power low application may start with transmission-level
congestion management, but certain buses (distribution systems) may be dynamically expanded, depending on the solution state, to address the internal constraints or
conditions of the distribution system. This again raises an important computational
issue—the ability to dynamically expand highly available computational capabilities
on demand—which is yet another cloud computing possibility.
REFERENCES
1. S. Grijalva, M. Costley, and N. Ainsworth, Prosumer-based control architecture for
the future electricity grid, in Proceedings of 2011 IEEE International Conference on
Control Applications (CCA), 28–30 September 2011, Denver, CO.
2. S. Grijalva and M. Tariq, Prosumer-based smart grid architecture enables a lat, sustainable electricity industry, in 2011 IEEE PES Innovative Smart Grid Technologies (ISGT),
17–19 January 2011, Anaheim, CA.
3. M. Tariq, S. Grijalva, and M. Wolf, Towards a distributed, service-oriented control infrastructure for smart grid, in Proceedings of the 2011 IEEE/ACM International Conference
on Cyber-Physical Systems (ICCPS), 11–14 April 2011, Chicago, IL.
4. S. Grijalva, Research needs in multi-dimensional, multi-scale modeling and algorithms
for next generation electricity grids, in DOE Conference on Computing Challenges for
the Next-Generation Power Grid, 2011, Ithaca, NY.
5. M. H. Albadi and E. F. El-Saadany, Overview of wind power intermittency impacts on
power systems, Electric Power Systems Research, 80(6), 627–632, 2010.
6. P. A. Ruiz, C. R. Philbrick, and P. W. Sauer, Wind power day-ahead uncertainty management through stochastic unit commitment policies, in IEEE/PES Power Systems
Conference and Exposition (PSCE‘09), 15–18 March 2009, Seattle, WA.
7. W. B. Powell, A. George, H. S. Simão, A. W. Lamont, and J. Stewart, SMART: A stochastic multiscale model for the analysis of energy resources, technology, and policy,
INFORMS Journal on Computing, 24(4), 665–682, 2012.
26
Smart Grids
8. P. Mell and T. Grance, The NIST Deinition of Cloud Computing: Recommendations of
the National Institute of Standards and Technology, The National Institute of Standards
and Technology, 2011, Gaithersburg, MD.
9. S. Fan, L. Chen, and W.-J. Lee, Short-term load forecasting using comprehensive combination based on multimeteorological information, IEEE Transactions on Industry
Applications, 45(4), 1460–1466, 2009.
10. J. Maver, Self-similarity and points of interest, IEEE Transactions on Pattern Analysis
and Machine Intelligence, 32(7), 1211–1226, 2010.
11. C. Vandekerckhove, Macroscopic simulation of multiscale systems within the equationfree framework. PhD Thesis, Katholieke Universiteit Leuven, 2008, Leuven, Belgium.
2.5
WIDE-AREA FREQUENCY MONITORING
Gamage and Liu
2.5.1
IntroductIon
The power system frequency monitoring network (FNET), proposed in 2001 and
established in 2004, is a pioneering wide-area monitoring system in the United
States. Composed of a network of over 200 frequency disturbance recorder (FDR)
devices—a member of the PMU family—that are spread throughout the country, the
FNET serves the entire North American power grid through advanced situational
awareness techniques, such as real-time event alerts, accurate event location estimation, animated event visualization, and postevent analysis [1–3].
An FDR is a single-phase PMU that is capable of measuring voltage phase angle,
amplitude, and frequency from a single-phase voltage source, which can provide
useful information for power system event recognition and status estimation. The
frequency measurement algorithms in FDRs are highly accurate, with virtually zero
error in the 52–70 Hz frequency range, and real hardware accuracy of ±0.0005 Hz,
which outperforms even some of the commercial PMUs.
The power grid operates in a very narrow frequency range, such that frequency
excursions outside this range give clues to the problems. Frequency is a universal
parameter across the entire interconnected power system that provides insight into
generation electromechanical transients, generation-demand dynamics, and system
operations such as load shedding, breaker reclosing, and capacitor bank switching.
This characteristic allows frequency monitoring to be as informative at the distribution level as it is at the transmission level.
2.5.2
Fnet archItecture
The overall FNET architecture, illustrated in Figure 2.1, comprises a widely deployed
network of FDRs that report phasor measurements to either a local processing unit
or a remote data center through Ethernet. The outermost layer of the data center is a
data concentrator for the incoming ield FDR data. The data concentrator performs
certain data preprocessing tasks, such as creating global positioning system (GPS)
time-aligned records from the received data before forwarding them to FNET applications and storage agents in the interior levels of the data center.
Power Application Possibilities with Mission-Critical Cloud Computing
Client
FDR 1
Satellite FDR 2
27
Firewall Data streaming
Real-time applications
and router
Ethernet
WAN
Real-time data sharing
Firewall Data concentrator
and router
FDR n
Firewall and router
Data
storage
Non-real-time
applications
Client
Sensors
Medium
and clients
Data center
FIGURE 2.1 The FNET architecture is composed of widely deployed FDRs that transmit
phasor measurements to either a local client or a remote data center for processing and longterm storage. Data centers host a multilayer agent hierarchy comprising a data concentrator
in the top layer, and a data storage agent and real-time and non-real-time application agents
in subsequent layers.
2.5.3
Fnet aPPlIcatIons
There are two types of FNET application classes—real-time applications and nonreal-time applications [4]. Real-time applications have stringent timing requirements; responses need to be produced within seconds or subseconds after receiving
the data. Examples of real-time FNET applications include frequency monitoring,
event trigger, interarea oscillation trigger, line-trip detection and identiication, and
event visualization. Non-real-time applications, on the other hand, have much more
relaxed timing requirements. Event location, interarea oscillation modal analysis,
data historians, and associated visualizations are some examples of non-real-time
applications.
2.5.4
cloud comPutIng and Fnet
Fundamental to the FNET is the ability to process and analyze measurements from
FDRs in an eficient and effective manner. There is an inherent communications
requirement even before data arrive at a data center for processing. Timely delivery
of data with an adequately high availability is critical for an accurate depiction of
the system state.
2.5.5
raPIdly elastIc data concentrators
The data concentrator layer is pivotal to the underlying FNET applications, since it
acts as the main interface to the data center for the incoming phasor measurements.
High availability and high consistency are two key data concentrator properties
28
Smart Grids
that ultimately dictate the accuracy of the system state representation for the FNET
applications. One way to ensure high availability is by making the data concentrators fault-tolerant through replication. When data concentrators are suficiently
replicated, it also becomes important to make sure that the replicas are consistent
with each other. Inconsistent replicas adversely affect the underlying applications by
forwarding out-of-sync or stale measurement records.
On the measurement side, the granularity of the sensor measurements can rapidly
scale up and down based on speciic grid conditions. Thus, the data concentrators
must be capable of absorbing large quantities of measurement data arriving at high
rates when the grid is stressed as well as when it is operating under normal conditions. This requires an underlying computational infrastructure that can rapidly scale
on demand, which is an inherent feature of the cloud [5].
2.5.6
comPutatIonal requIrement FlexIBIlIty
The multilayer computational data center hierarchy has different computational
needs at different layers. The outermost data concentrator units are mostly preprocessing units (aligning data as GPS–time-synced records) that require more
primary memory than secondary memory. Preprocessed records are streamed
immediately to real-time applications and are only buffered when the number of
records reaches the memory cache limit. Large memory caches result in faster
processing times. Data collection and record padding are not interrupted by the
saving procedure, because the memory cache is designed as a stack; only the oldest
portion of the data is saved.
On the contrary, FNET applications have higher CPU and processing requirements than those of data concentrators. Furthermore, different applications can
have different coniguration requirements (number of CPUs, memory, storage, network, etc.) based on the class of the application (real-time or non-real-time) and
the type of spatial scope of the data that are being handled. Clearly, given the
large capital investment, skilled workforce requirements, and maintenance cost,
supporting such diverse computational requirements using in-house solutions is
not cost effective to any power grid entity. A much more feasible solution would
be to take advantage of cloud computing facilities, which can provide any type of
platform or infrastructure requirement as a service, greatly improving the overall
value proposition.
2.5.7
cloud-Based aPPlIcatIons
The most intriguing aspect of using cloud computing in the FNET is supporting FNET
applications as SaaS [5]. The FNET provides a rich pool of applications for power
grid entities at all levels, both spatially and temporally. Prebuilt applications can be
hosted as SaaS for interested grid entities to use at any scale, based on their speciic
and individual organizational requirements. Furthermore, the cloud can provide a
PaaS model for application developers to develop new applications under a uniied
reference model. The data delivery to these applications is already handled by the data
concentration level and can be offered as a well-deined API. This not only greatly
Power Application Possibilities with Mission-Critical Cloud Computing
29
simpliies the instrumentation aspect of application development, but also helps to
standardize the application development life cycle and data processing pipeline.
REFERENCES
1. B. Qiu, L. Chen, V. Centeno, X. Dong, and Y. Liu, Internet based frequency monitoring
network (FNET), in Proceedings of the IEEE Power Engineering Society Winter
Meeting, vol. 3, pp. 1166–1171, 28 January–1 February 2001, Columbus, OH.
2. Y. Liu, A US-wide power systems frequency monitoring network, in Proceedings
of the IEEE PES Power Systems Conference and Exposition (PSCE‘06), 29
October–1 November 2006, Atlanta, GA.
3. Power Information Technology Lab, FNET Web Display, University of Tennessee 2013.
Available at: http://fnetpublic.utk.edu/sample_events.html.
4. Y. Zhang, P. Markham, T. Xia, L. Chen, Y. Ye, Z. Wu, Z. Yuan, et al., Wide-area frequency monitoring network (FNET) architecture and applications, IEEE Transactions
on Smart Grid, 1(2), 159–167, 2010.
5. P. Mell and T. Grance, The NIST Deinition of Cloud Computing: Recommendations of
the National Institute of Standards and Technology, The National Institute of Standards
and Technology, 2011, Gaithersburg, MD.
2.6
OSCILLATION MITIGATION STRATEGIES
Venkatasubramanian
Real-time oscillation monitoring tools such as the oscillation monitoring system
(OMS), developed at Washington State University based on synchrophasors, are
being implemented in many control centers all over the world. Excellent progress
has been made recently on centralized [1,2] and distributed algorithms [3] for widearea oscillation detection and analysis. Typically, the monitoring tools issue alarms
to the operators whenever the damping ratio of any of the local or interarea modes
goes below a preset threshold, say 2%. Unbounded growth of negatively damped
oscillations can lead to tripping of critical system components and may even lead to
blackouts such as the August 10, 1996, western American blackout. The presence of
poorly damped or sustained oscillations over a period of time can cause damage to
rotor shafts and also impact on consumer power quality, with potentially signiicant
economic consequences. Therefore, it is important to mitigate poorly damped oscillations as quickly as possible. However, the choice of what control actions should be
used to mitigate different types of oscillatory modes is a nontrivial task. As real-time
oscillation monitoring tools are adopted by the industry, there is an urgent need to
develop mitigation strategies whereby operators can respond to oscillation alarms in
the form of speciic operator actions whenever such oscillation alarms occur.
Cloud computation is an ideal platform for developing such mitigation control
strategies. The problem can be handled in two ways:
1. Baseline studies: Massive analysis of historical synchrophasor data can
be carried out using oscillation monitoring engines such as OMS, and the
results can be recorded together with corresponding supervisory control
and data acquisition (energy management system) data. Correlation studies
30
Smart Grids
can be conducted to identify patterns in how different modes are sensitive to different tie lows and generation patterns. Cloud computing enables
analysis of different groups of synchrophasor data in parallel to target different interarea modes, and, possibly, all the observable local modes of relevant generators.
2. Online simulations: Whenever the oscillation engine issues an alarm,
cloud resources can be called on to carry out a parallel simulation of a
large number of feasible candidate operator actions to decide which actions
are the most effective in mitigating the speciic mode-related alarm. For
instance, a critical tie-line interface may be identiied, and the necessary
reduction or increase of tie-line low can be determined based on the simulation studies to improve the damping of a problematic interarea oscillatory mode.
REFERENCES
1. G. Liu, J. Quintero, and V. Venkatasubramanian, Oscillation monitoring system based
on wide area synchrophasors in power systems, in Proceedings of the 2007 IREP
Symposium Bulk Power System Dynamics and Control, VII, 19–24 August 2007,
Charleston, SC.
2. V. Venkatasubramanian and G. Liu, Systems and methods for electromechanical oscillation monitoring, US Patent 8,000,914, 2011.
3. J. Ning, X. Pan, and V. Venkatasubramanian, Oscillation modal analysis from ambient
synchrophasor measurements using distributed frequency domain optimization, IEEE
Transactions on Power Systems, 28(2), 1960–1968, 2013.
2.7 AUTOMATIC NETWORK PARTITIONING
FOR STEADY-STATE ANALYSIS
Zhu
Cloud computing is emerging as a new paradigm for contemporary computational
tasks with explosive demand for computing resources [1]. It offers thousands of computers instant and affordable access to powerful remote computing platforms. Cloud
computing resources can match up to the highest instantaneous demand peaks, while
providing lexibility on a pay-by-use basis when the demand peaks are over. Thanks
to its attractive computational abilities, cloud computing is advocated as a great
opportunity for diverse power system computational tasks, ranging from operations
and control to economics and planning [2].
One key factor to enable cloud computing for the power grid would be an eficient
network partitioning scheme. Typically, the interconnected bulk electric system is
divided into several zones, based on speciic applications. Perhaps the most obvious
example is the division of North America into Regional Transmission Organization
territories for market administration and transmission management purposes [3].
Partitioning the grid into zones with minimal intercoupling serves as a basis for eficiently paralleling computational tasks, since it leads to more succinct information
exchange among the computers. However, with the growing size of the power grid, an
Power Application Possibilities with Mission-Critical Cloud Computing
31
exciting future direction would be to develop an automatic tool to partition any electric network depending on the level of granularity as well as the application domain.
Beneiting from eficient parallelism, cloud computing can be immediately
applied to power system steady-state analysis, which is routinely performed in all
control centers. In this realm, cloud computing with conidentiality is suggested in
[4] for the optimal power low problem—a crucial power system operational task.
Nonetheless, it would be more interesting to perform the security analysis when the
grid is approaching or reaching stressed conditions, leveraging the increase in number of processors from the existing order of dozens to thousands. In fact, the potential
of parallel high-performance computing has been preliminarily investigated in [5]
for massive contingency analysis on the Western Electricity Coordinating Council
power grid.
Nonetheless, it is not possible to apply cloud offerings directly to the power grid
without some further considerations. For example, the GridCloud architecture project (see Chapter 1), developed by Cornell and Washington State University researchers, aims to keep the throughput offered by the commercial cloud offerings while
eliminating a deliberate trade-off of consistencies. This feature may not seem necessary for applications such as web browsing. However, it is of great importance for
power system computational tasks that involve storing replicas of, for example, some
state variables on different computers.
Another issue with the cloud offerings is that they can take many minutes to ramp
up. This motivates the power grid control centers to seek close cooperation with a
cloud provider. As another option, separate cloud installations for power grids may
be more viable in this regard. For example, RTE in France has its own data center,
and multiple ISOs in the eastern US grid are reportedly giving serious thought to
developing their own GridCloud pilots in 2014.
REFERENCES
1. K. Birman, Guide to Reliable Distributed Systems: Building High-Assurance
Applications and Cloud-Hosted Services, 2012, Springer, London.
2. G. Dán, R. B. Bobba, G. Gross, and R. H. Campbell, Cloud computing for the power
grid: From service composition to assured clouds, in Proceedings of the 5th USENIX
Workshop on Hot Topics in Cloud Computing (HotCloud), June 2013, San Jose, CA.
3. North American Electric Reliability Corporation (NERC), 2013. Available at: http://
www.nerc.com/.
4. A. R. Borden, D. K. Molzahn, P. Ramanathan, and B. C. Lesieutre, Conidentialitypreserving optimal power low for cloud computing, in Proceedings of the 50th Annual
Allerton Conference on Communication, Control, and Computing (Allerton), 1–5
October 2012, Monticello, IL.
5. Z. Huang, Y. Chen, and J. Nieplocha, Massive contingency analysis with high performance computing, in Proceedings of the IEEE Power and Energy Society General
Meeting, 26–30 July 2009, Calgary, AB.
3
Emerging Wide-Area
Power Applications with
Mission-Critical Data
Delivery Requirements
Greg Zweigle
CONTENTS
3.1
3.2
3.3
3.4
3.5
Introduction .................................................................................................... 33
System Analysis .............................................................................................. 35
Situational Awareness ..................................................................................... 37
Stability Assessment ....................................................................................... 38
Wide-Area Control ......................................................................................... 39
3.5.1 Automatic............................................................................................ 39
3.5.2 Operator Initiated ...............................................................................40
3.6 Transient Instability Mitigation ...................................................................... 41
3.7 System Protection ...........................................................................................46
3.8 Summary Tables ............................................................................................. 48
References ................................................................................................................ 50
3.1 INTRODUCTION
Our electric power system plays a critical role in present-day society. The many
beneits of electricity include its versatility, reliability, safety, and low cost. In terms
of versatility, when new sources of energy such as wind or solar are considered, the
best method to transmit their power to consumers is by converting it to electricity.
Other transmission options, for example, generating a hydrogen gas intermediate,
are signiicantly more complicated and expensive to implement. In terms of reliability and economics, the interconnected electric power system has demonstrated
exceptional robustness over many decades of operation and with very low cost.
Worldwide, electric power is responsible for delivering approximately one-third of
overall requirements, and the percentage of total power consumption in the form of
electricity is growing due to trends in the electriication of automobile transportation
and increases in the computing and communication infrastructure [1].
Because of its importance to society, there is a continuous need to improve the reliability and economics of electric power. Key to this is maintaining correct operation
33
34
Smart Grids
during challenging events, including equipment failure, electrical faults, and any
weather-induced variability affecting renewable generation sources. Hierarchies of
protection and control, starting at a local distributed level and extending to a wide-area
centralized level, are applied to keep the system operating in acceptable ranges during times of stress. At the local level, devices for equipment protection are deployed
to quickly disconnect devices that develop a fault. This isolates failures and prevents
disturbance spreading. When the fault is temporary, protection devices reclose the line
back into service. This keeps the system strong against future disturbances. Equipment
protection primarily utilizes local signals to make decisions. Recently, there is a move
toward building systems that communicate between neighboring devices in a distributed manner to aid selectivity, security, and dependability. At the top, most centralized
level, human operators monitor and control the system centrally. They receive signals
that are communicated over a wide geographical area. Between the local and top levels
of observation, a variety of distributed control systems are applied for responding to
situations that require wide-area information but must operate with intervals that are
too short for a human to process and make a good decision [2].
Modern electric power systems are designed to connect generators and consumers
of power over large distances. This creates a unique engineering challenge for reliable operation, because of the communication demands that result when state measurements need to be shared with mission-critical applications for proper control and
operation. The electric power system, or a speciied subset, can span hundreds or even
thousands of miles. Measuring, transmitting, and verifying measurements collected
over these long distances require a sophisticated communication infrastructure. The
distances involved are particularly challenging for control of electric power systems.
Most other control systems, such as cell phone receivers or automobile engine regulators, are packaged within relatively small spaces. Communication within these
systems is fast. Long-distance communication networks can result in long latencies,
dificulty in aligning measurements, and unreliability in reception quality. Control
systems are particularly sensitive to these impacts.
Historically, engineers have overcome the challenge of geographic distance for power
system measurements by only measuring quantities that change slowly. Both the voltage
magnitude and injected power in a transmission system vary slightly over an interval
of several seconds. This allows for larger communication delays and simpliies aligning the measurements in time. Limiting measurements to those that change slowly is
acceptable for many applications, but, in general, it is a problem because the state of the
power system consists not only of the magnitude but also the phase angle. Phase angles
change rapidly. Furthermore, depending on the type of control required, the required
measurements can also include topology, synchronous machine rotor angle, and even
the internal conditions of governors, voltage regulators, and other local control devices.
In the past, and in many systems even today, the fast-changing phase angle is estimated at rates as slow as every minute, and is treated on a relative basis. In addition
to providing angle values, estimators calculate missing data from physical locations
where direct measurement is impractical or for cases when communication system
issues lead to data interruption. These estimators also improve the accuracy of measurements. State estimation was previously based on a nonlinear method, driven by
the need to derive angles from power measurements [3]. Because of its nonlinearity,
Wide-Area Power Applications with Mission-Critical Data Delivery
35
the traditional state estimation is computationally expensive and can sometimes fail
to converge. This leads to intervals when critical data are unavailable, not because of
communication failures but because of algorithmic limitations. Convergence problems have become a larger issue recently because renewables result in a violation of
the assumption that the power system is changing slowly. Good estimates are needed
even during times of fast dynamics.
Over the past several decades, a new measurement approach has evolved.
A shared, high-accuracy time source enables measuring devices to precisely timestamp signals before transmitting them over a communication network. One common method of receiving this accurate time is from the global positioning system
(GPS). When measurements are the network voltage and current related states, they
are called synchrophasors [4]. More recently, this measurement approach has also
been applied to rotor angles and other connected machine and device internal states
[5]. The time-stamped measurement technology has unlocked new ways to control
electric power, increase its reliability, and decrease its cost.
Time-synchronized systems now directly measure voltage and current phase angles.
This enables the application of a linear calculation for illing in missing values [6]. The
linear calculation does not have convergence issues. All values, whether directly measured or calculated, are available during times of both slow and fast system dynamics.
Also, with these measurements simple applications become possible, such as a textbased meter command at the measuring devices to help an engineer determine whether
line phases have been swapped [7]. Direct measurements also provide the means for
independent topology checking at the substation level [8] and enable the state estimation to be distributed [9], which results in a more reliable system.
This chapter overviews new or signiicantly improved power system applications that are now possible due to wide-area time-synchronized measurements. The
emphasis is on those applications with mission-critical delivery requirements. The
ability to communicate information over large distances and meet quality of service
constraints is facilitated by the advanced communication systems and distributed
computing architectures covered in other chapters of this book.
Applications are divided into ive categories: system analysis, situational awareness, stability assessment, wide-area control, and system protection. Each area is
surveyed to explain emerging applications and research directions. For wide-area
control, one of the most dynamically advancing ields due to these new measurement
techniques, a speciic application is explained in full detail. Common data delivery
requirements for each category are summarized, and special focus is given to interaction with the communication subsystem. In general, the categories are listed in order
of increasing performance requirements. These requirements then become the basis
for communication infrastructure design and analysis, as described in Chapter 4.
3.2
SYSTEM ANALYSIS
Wide-area time-synchronized measurements provide a new tool for power system
analysis. Figure 3.1 shows an architecture suitable for collecting this information,
acting as a total system data recorder. Buses are labeled 1, 2, …, N, each representing a separate substation. At each substation a phasor data concentrator (PDC)
36
Smart Grids
1
N
2
PDC
PDC
PDC
Communication
network
Data
recorder
FIGURE 3.1 Data collection architecture.
buffers the measurements [10]. The purpose of the PDC is to provide an intermediate storage location at the substation, which is accessed in the case of interrupted
communication to the central data recording device. The PDC also provides an
additional security layer between the substation and the external communication
network. Layered security is an important strategy for achieving overall system
resilience against cyberattacks. In some implementations the PDC time-aligns the
measurements and then sends all measurements for a given time instant together
as a single communicated entity. While phasor data concentrator is a common
term in North America, others might refer to this intermediate communication
and computation device by another name, and additional capabilities and features are possible with more complicated data delivery architectures. However,
the term phasor data concentrator is retained here because of its common usage
in the industry.
One application of the collected data is fault recording. The electric power system is very reliable and major events are fortunately rare. However, when an event
occurs, getting to the root cause is important. Also, although minor events are common and have little impact on power system customers, gaining information from
minor disturbances helps improve the system and avoids major disturbances. Timesynchronized measurements enable event reports to be combined into a single view
of what happened and how the disturbance evolved over a wide area. Measurements
are received from a variety of devices, such as digital fault recorders, protective
relays, and automation controllers, and are then communicated through the PDC to a
centralized location, as shown in Figure 3.1. These measurements might be sampled
at different rates, but with high-accuracy time stamps it is simple to combine them
into a single time base.
Another application of these archived data is model veriication [11,12]. Good
models are the prerequisite for effective power system planning and it is important to
Wide-Area Power Applications with Mission-Critical Data Delivery
37
validate models against measured data to verify their accuracy. Time-synchronized
measurements provide a way to combine data from various parts of the electric
power system and apply this information to test models [13]. Often, these data are
from relatively small events, such as the frequency response of the system after a
generator trips. Also, when load tap-changing transformers make a step change, the
time-stamped measurements help identify load characteristics. Because these new
time-synchronized systems directly measure voltage and current angles, they provide a baseline state when the system is in the normal operating range. This enables
power low models to be veriied.
With respect to the communication system and data delivery requirements, collecting events puts almost no restriction on latency. Receiving the complete set of
events with as much as a minute of delay is acceptable. This means that, if the overall
communication is interrupted, it is possible to retrieve data buffered at the PDC at a
later time, after the communication system has been restored. However, a high sample rate is required and the volume of data is large. Present systems sample as fast as
60 times per second, but in the near future sample rates on the order of 240 samples
per second will become common. Higher sample rates are important for capturing
and analyzing details of the power system’s response to disturbances.
3.3 SITUATIONAL AWARENESS
Operators continuously monitor the electric power system to help maintain reliability. Existing situational awareness is built on the supervisory control and data
acquisition (SCADA) architecture, which is an asynchronously scanned system
with update rates on the order of seconds. Slowly changing voltage magnitude
and power low measurements are adequately represented with this architecture.
However, when the power system changes rapidly, as is becoming more common
with the increase in wind and solar renewable energy sources, then these measurements become suspect during transients. Over the scan interval there are no
guarantees that the measurements are acquired at the same time, and this causes
problems. For example, consider the case of power measurements on parallel lines
and the need to know the total power low. This is necessary for assessing the stability of the Guatemalan power system [14]. During fast dynamics, the power on each
line can transition from nominal levels to levels requiring action in seconds. With
a multisecond SCADA update rate, it is not possible to correctly calculate the total
power at a 1 s interval. A solution through the application of time-synchronized
measurements is described later in this chapter. Also, SCADA displays cannot
effectively show phase angles and fast-changing machine state values because of
the asynchronous scan.
New time-synchronized systems update as frequently as 60 times per second. All
measurements are taken at a common time instant. The angles are directly measured
and displayed. Operators obtain a more detailed understanding of the system state
[15–19]. Additionally, these details improve the ability of alarming and event notiication applications to provide information based on sophisticated assessments [20].
The important beneits of the new information include increasing the conidence
of operators. They know that they are seeing the exact power system conditions in
38
Smart Grids
1
N
2
NI
NI
NI
Communication
network
Operations
FIGURE 3.2 A representative situational awareness architecture.
real time. For example, if a line is taken out of service, the steady state as well as the
transient result on the voltage and current level is immediately visible. If the system
exhibits any undesirable characteristics, such as sudden oscillations, the operator
sees them immediately and can take action. As another beneit, standing line angles
across disconnected lines are directly measured. The lack of these measurements has
posed problems historically. This was a contributor to the Southern California outage in September 2011 [21]. A line was open and the adjusted power lows resulted in
a large angle developing. The operators could not always see this angle and take the
actions to get this line back into service quickly. With time-stamped information the
phase angle across each line is directly measured and displayed.
Figure 3.2 shows an architecture suitable for situational awareness and also
operational control. In contrast to Figure 3.1, communication paths are bidirectional
because of the need for operators to issue control actions. These might include opening a breaker or redispatching generation power. The PDC is replaced with a network
interface. There is no need for a signiicant amount of intermediate data buffering
because data are needed as fast as they are generated. Going back at a later time to
retrieve data after a network outage provides little beneit. Communication requirements for operator displays are driven by the need for near real-time streaming data.
Some amount of latency is tolerated because humans have a limit on their ability to
respond. Latency on the order of seconds is acceptable. However, the update rate
should remain on the order of 30–60 samples per second because this can show
higher frequency oscillations and other detailed information of interest.
3.4 STABILITY ASSESSMENT
Assessing the electrical security of a power system is always an important need.
These assessments relate to several classes of possible instability: voltage, frequency,
Wide-Area Power Applications with Mission-Critical Data Delivery
39
transient [22], and oscillatory. With high-rate, streaming, time-synchronized measurements, the accuracy and prediction power of security assessments are improved
[23–25]. For example, SCADA systems, with asynchronous update rates, do not allow
ixed-rate signal processing applications. With the new time-synchronized systems
and direct state measurement, the assessment algorithms can use well-known iltering and frequency-domain processing algorithms.
For voltage stability assessment, as an example, a simple method is the calculation of stability indices based on reactive power measurements [26]. The voltage stability index is based on the difference between the maximum value that the system
can support and the present value.
QVmargin ( ti ) = Qcollapse ( ti ) − Qmeasured ( ti )
(3.1)
The reactive power Qmeasured is collected at each bus at time instant ti. Then, the
inductive loading at each bus is incremented by a real-time simulation until the system becomes unstable. This is the estimated value Qcollapse. The difference is the
reactive power margin at that time. When QVmargin crosses a deined threshold, the
system is assessed as becoming unable to handle the reactive power demands. It is
important that all measurements are taken at the same time to improve accuracy and
to ensure that measurement artifacts are not causing artiicial convergence issues.
More sophisticated methods utilizing time-synchronized measurements have also
been developed [27,28]. Once a system is assessed as becoming unstable, the information is provided to the operators, who can then adjust reactive power or generation
dispatch to alleviate the situation.
3.5
WIDE-AREA CONTROL
A wide range of diverse applications fall under the category of wide-area control.
They are either distributed and automated or centralized and operator initiated. All
of these control applications are based on a directly measured time-synchronized
system state. This section overviews several of these applications and then examines
in detail an emerging model-based approach suitable to mitigate transient instability.
3.5.1
automatIc
An example of a distributed automatic application is collecting voltage measurements to increase the control coordination optimality of a set of static var compensators (SVC). The SVC is a reactive power device that acts in a fast manner to improve
the voltage proile of the power system. These devices monitor voltages locally and
inject reactive power to keep that local voltage stable and within preset ranges. Using
the streaming nature of time-synchronized measurements, there are systems in service that also receive more distant measurements and make control adjustments that
not only correct for local variations but also hold voltages at remote locations within
proper boundaries [29].
Another wide-area control application is interarea oscillation damping [30–32].
These oscillations are driven by rotating machine controls interacting with each
40
Smart Grids
other through the electrical coupling of the power system. Again, using the streaming nature of time-synchronized measurements results in a system in which remote
oscillations are measured and then corrected at the point of initiation.
Islanding control for distributed generation is a challenging problem that has beneited from direct state measurements. Present standards require distributed generators to stop producing power when their region is disconnected from the main
power system [33]. A key challenge is detecting this disconnected state. Currently,
many systems make this assessment through measurement of breaker status. When
prespeciied breakers indicate that they have opened, a controller determines that
the distributed generation site has islanded and sends disconnection commands. The
problem with this approach is its complexity. Power systems can consist of many possible connection arrangements, so the controller must track all of them in determining the islanded state. Also, when connections are reconigured it is required that the
controller be reconigured to relect this new arrangement. Another approach is for
each distributed generator to continuously push its synthesized frequency against the
bulk power system received frequency. The connected bulk power system is much
stronger than the distributed generator, so when the local system is not islanded the
attempted frequency perturbation is unsuccessful. When the system islands, the frequency rapidly changes and this indicates the islanded condition. The problem with
this approach is coordinating the various frequency forces.
In contrast to these existing methods, the comparison of measurements across a
point of interconnection has both simpliied the problem and resulted in more reliable
controllers [34,35]. Time-stamped measurements allow the direct comparison of the
phase angle at various points of the power system. When the region containing distributed generation goes into an islanded condition, the phase angle differences, along
with their rate of change, provide a strong indication of islanding. The controller then
sends appropriate commands for disconnection of the distributed generators. In the
future, this same scheme will allow control over the real and reactive power output of
distributed generators in response to instantaneous power system conditions. So, as
utilities install the more advanced time-stamped schemes to meet present interconnection requirements, these same systems become ready for more advanced applications.
It is required that communication of data for automatic wide-area control is fast
and reliable. These applications tend to be closed loop, and latency directly affects
the performance and stability of the loop. Latency must be low. Speciic values are
quantiied at the end of this chapter.
3.5.2
oPerator InItIated
The precise time nature of the modern power system enables accurate timing not
only of measurements, but also of control actions. Consider, for example, the scenario of an operator removing a line from service. First, commands are sent to open
the breaker on each side of the line. The result is that the voltage decreases because
of a higher effective impedance. Local tap-changing transformers and SVCs automatically adjust to correct the voltage. Meanwhile, the operator can switch manually operated reactive power devices, such as shunt capacitors, into service. There
is a signiicant time delay between automatic local controls and operator-initiated
Wide-Area Power Applications with Mission-Critical Data Delivery
41
controls. As a result, the system voltage and power lows experience transients. Once
an operator has put shunt capacitors into service, the local control devices react to
return more closely to their original state. The various signal level changes add variability to the power system, which is already experiencing larger signal swings due
to renewable generation. As control devices respond to varying voltage levels, this
adds stress and reduces their life span.
Operators can now generate recipe lists of commands, each with an associated time
stamp [36], and execute all required changes simultaneously. These lists are distributed to the various control actuation points in the system. When a certain operation
is required, for example, removing a line from service, the collection of distributed
recipes all execute in lockstep. This prevents transients and reduces system stress.
Other operator control applications include black-starting generators and monitoring line angles prior to reinserting a line into service. In these applications it is
primarily the measured angles that provide the needed capability. Operator-assisted
controls tend to have less strict communication requirements than the feedback controls. For example, with recipes, it is possible to send the commands seconds, minutes, or even hours ahead of the control execution time. This allows for feedback
testing to ensure that the requested commands are validated, thus increasing the
cybersecurity of the system.
3.6
TRANSIENT INSTABILITY MITIGATION
The electric system interconnects large synchronous machines that provide the bulk
of generated power. During normal operation, the rotors of these machines operate in coordination and maintain slowly varying angle differences. However, during
large contingencies it is possible for power imbalances to disturb the rotor angle relationships beyond the capability of restoring forces. In these cases a set of machines
can swing out of step with each other. This condition is called transient instability. A typical initiating condition is a fault that signiicantly unloads a section of the
system, which causes the affected machines to accelerate past the point of stability.
Modern, fast power system protection has provided signiicant resilience against this
problem [37]. However, there are still cases in which transient instability can develop.
These are often high-order contingencies in which multiple lines are lost, often in
conjunction with the loss of primary protection and the resulting delay of backup
protection causing longer fault duration; they cannot adapt to the present conditions.
System integrity protection schemes, also called remedial action schemes, can
provide a certain level of resiliency against transient instability. However, they are
limited to the speciic conditions studied and must be conigured for these cases in
advance of the disturbance. With time-stamped measurements it is now possible to
build feedback control schemes that measure a wide-area collection of power system
states and respond to a much larger range of contingency cases than is currently possible. A speciic example that demonstrates these ideas in detail is presented next.
Figure 3.3 shows a block diagram of the controller, focusing on three buses out
of the larger power system. Each device controller (DC) takes measurements of the
network state and, when connected to a generator
bus, the machine
state. The net
work state consists of time-stamped voltage V j  k  and current I j  k  measurements
42
Smart Grids
Controller
Communication
network
DC
Machine
controller
Machine
controller
DC
Other sections of
power system
DC
FIGURE 3.3 Architecture for mitigation of transient instability.
at each bus in the power system. Buses range over j = 1, …, N. The machine state
consists of the rotor angle θi[k], rotor ield excitation Efd,i[k], and prime mover power
Pm,i[k]. Machines range over i = 1, …, M. The index k represents the time stamp.
For transient stability control a suitable measurement sample rate is fnom, meaning that the time stamps are incremented at an interval of T = 1/fnom. This is the
rate at which data are transmitted over the communication network to the controller. Although currently not all installed systems are able to acquire such detailed
machine measurements, this technology is becoming more viable. Time-stamped
rotor measurements are included in the appendix of the latest Institute of Electrical
and Electronics Engineers (IEEE) C37.118 standard [38] and are implemented today
in some advanced controllers [39]. Furthermore, time-stamped measurements of the
additional synchronous machine state values are under investigation [40].
The controller monitors the power system dynamics and determines the need for
control actions. For transient instability correction, the control actions selected here
consist of tripping generation, shedding load, or inserting series compensation on
select lines. Other control actions are possible.
Time-stamped received signals are processed and controls are selected through the
methodology of model-driven assessment and performance metric measures [41,42].
The electric power system is nonlinear by nature. This makes controller designs
particularly challenging. Although linearization or model-free approaches are possible, the strong nonlinearities that characterize rotor angle–related instability make
a model-based controller attractive. Figure 3.4 shows the basics of operation. At time
tF the power system experiences a large contingency that, for example, involves faults
on multiple lines. Local protection devices clear these faults at time tT. Subsequently,
it is the task of the controller to determine whether the state trajectories are evolving toward stability and whether control actions are required. If control actions are
required, then it is the task of the controller to select an appropriate control and send
it to the appropriate DC. These tasks are performed by predicting the state trajectories according to a model of the power system. The initial state is from the measured
time-stamped state values. Then, for each control option, the state trajectories are
predicted. A cost measure is applied to the resulting trajectories. The irst control
action for the sequence of control actions that results in a minimal cost is selected
for application. Because the option of not applying any controls is included in the
Wide-Area Power Applications with Mission-Critical Data Delivery
Evolution
43
Modeling
Predicted
trajectory 1
Predicted
trajectory 2
Reference trajectory
State
Predicted trajectory
for no controls
tF
tT
Control 1
Control K
FIGURE 3.4 Basic operation of a model-based controller with prediction.
considered list, the algorithm inherently assesses stability. When the system is not
evolving toward instability the controller takes no action.
The cost measures consist of both the state cost, Cstate, and the control cost, Ccontrol.
Each is calculated individually based on the received state measurements and a prediction of the future state. The total cost is the product of these two individual costs.
J = CstateCcontrol
(3.2)
The state cost Cstate is the summation of the individual cost for three classes of
states. These are the rotor angle Cθ, frequency Cω , and bus voltage magnitude CV.
(3.3)
Cstate = Cθ + Cω + CV
Individual state costs in turn are calculated as follows. Equation 3.4 provides the cost
measure for the rotor angle, following a mean-square formulation. The center of inertia
angle θɶ i [ k ] = θi [ k ] − ΣiM=1 Hi θi [ k ] ΣiM= 0 Hi, where Hi is the inertia for machine i, at time
kT is compared against the average of the center of inertia angle θɶ i over window W. This
difference is normalized by the maximum expected angle, π, and then squared. The
duration of the prediction window is WT seconds. The resulting cost measure for the
rotor angle of machine i is deined as Cθ′ ( i ). A large value of Cθ′ ( i ) indicates that the rotor
angle state for machine i has deviated signiicantly from a possible stable path.
1
Cθ′ ( i ) = 2
W
 θɶ i  k  − θɶ i
  

π
k =1 
W
∑




2
(3.4)
The total cost for all of the rotor angles in the complete power system is the summation of the individual machine rotor angle costs.
M
Cθ =
∑C′ (i )
θ
i =1
(3.5)
44
Smart Grids
For frequency, the rate of change of the rotor angle ωi[k] is compared with the
nominal frequency ωnom and normalized by a frequency limit, ωlimit. This quantity is
then summed over the prediction window.
1
Cω′ ( i ) = 2
W
 ωi  k  − ωnom

ωlimit
k =1 
W
∑



2
(3.6)
The total cost of the frequency for all machines is the summation of the individual
machine frequency costs.
M
Cω =
∑C′ (i )
(3.7)
ω
i =1
The inal state-related cost measure is based on the voltage magnitude, computed
for each of the N network buses. The voltage is compared with the predisturbance
equilibrium voltage for that bus. The difference is divided by the reference voltage,
squared, and then summed over the prediction window.
1
CV′ ( i ) = 2
W
 Vi  k  − Veq,i

Vmax
k =1 
W
∑



2
(3.8)
The total cost of the voltage magnitude is the summation of individual bus voltage
costs, normalized by the ratio between the number of bus voltages and the number
of machines in the system.
M
CV =  
N
2
N
∑C′ (i )
V
(3.9)
i =1
While the state costs are calculated with a mean-square approach, the control
costs, Ccontrol, are taken from tabulated values. The total control cost is then a summation of the individual control actions, taken sequentially over the prediction window.
In Equation 3.10, the number of controls considered over the prediction window is W′.
W′
Ccontrol =
∑C
j
(3.10)
h =1
Table 3.1 provides an example cost allocation for control actions. The table
approach allows the selection of each value based on the individual needs of the application. In Table 3.1, two sets of costs are provided. When the cost (Equation 3.3) for
the case without application of control is below a speciied threshold, then all tested
controls use the values in the second column of Table 3.1. When the cost (Equation
3.3) is above this threshold, then the system is estimated as becoming unstable and the
tested controls in the third column are applied. In either case, the inal selected control
set is based on the sequence that minimizes (Equation 3.2), independently of whether
the cost table selection is based on a stable or an unstable threshold check.
45
Wide-Area Power Applications with Mission-Critical Data Delivery
TABLE 3.1
Control Action Costs
Structural Change Control
Action
No control action
Series capacitance
Load shedding
Generator tripping
Base Case is
Modeled Stable
1
2
4
3
Base Case is
Modeled Unstable
4
1
3
2
The capability of this controller is demonstrated with the IEEE 39-bus system.
Figure 3.5 shows the system response to an example of three lines simultaneously
faulted. The upper left signals are the phase angles of each generator referenced to
the center of inertia angle. The lower left signals are the frequency values, referenced
to the nominal frequency. The right of the igure shows the voltage magnitudes for
all 39 buses.
Assume that the fault is cleared by backup protection with a long clearing time.
The probability of losing three lines is very low. However, it is these low-probability
events that can lead to large cascading outages. An advantage of applying timestamped measurements with wide-area control in a feedback architecture is the ability to respond effectively to both common and rare system events.
Figure 3.6 shows the system response after correction by the controller. The controller receives time-synchronized measurements, estimates the response against a
model for a set of control options, and then selects the control that minimizes that
total cost. The result is that the controller selects tripping a generator 100 ms after the
4
6 × 10
1.5
2
0
–2
1
0
5
10
15
20
15,000
0.5
10,000
mHz
Voltage (p.u.)
Degrees
4
5,000
0
–5,000
0
5
10
Time (s)
15
20
0
0
5
10
Time (s)
FIGURE 3.5 Unstable response of the 39-bus system to a three-line outage.
15
20
46
Smart Grids
1.5
100
Degrees
50
0
–50
1
0
5
10
15
20
800
600
Voltage (p.u.)
–100
0.5
mHz
400
200
0
–200
0
5
10
Time (s)
15
20
0
0
5
10
Time (s)
15
20
FIGURE 3.6 Stable response after control application.
fault is cleared, then inserting series capacitance 200 ms after the fault is cleared, and
inally shedding some load 300 ms after the fault is cleared. The advantage of this
method is that controls are selected based on minimizing a combined cost of state
trajectory deviation and control action costs. This helps to ind a solution that minimizes the negative effect of the controls while simultaneously stabilizing the system.
Automated control for resolving transient instability requires a very reliable communication network. As seen in Figure 3.3, measurements are collected and controls
are sent over a wide area. Latency on the order of less than 10 ms is required for this
application. This allows time for measurement iltering and control selection such
that the irst control action can take place only 100 ms after the initiating event.
3.7
SYSTEM PROTECTION
The most demanding application in terms of communication requirements is system
protection. Time-synchronized measurements can now be found in line-current differential and distributed bus differential protection schemes [43]. For wide-area protection, new approaches to transient stability [44,45] and voltage control [46] have
become available.
A detailed example is the correction of oscillations using time-stamped measurements [14]. In Central America, a 230 kV transmission network connects from
Mexico through Guatemala and south to Panama. The Central American system
capacity is approximately 7 GW, which is signiicantly smaller than the power
system in Mexico. Historically, oscillations in the Central American system have
required disconnection from the Mexican system during certain intervals of the day.
This negatively impacted the reliability and economics of the Guatemalan system.
The wholesale operator of Guatemala, the Administrador del Mercado Mayorista
Wide-Area Power Applications with Mission-Critical Data Delivery
47
(AMM), decided to implement a system that could detect these oscillations and proactively disconnect lines. This eliminates the need to isolate based on time of day
and, instead, acts only when necessary. Thus, a measurement-based control system
is in place instead of predeined table lookup actions.
What has prevented such a scheme in the past is the lack of time-synchronized
measurements. There are two factors involved. First, it is necessary to measure the
power low on two separate lines in order to determine the state of the power low.
While SCADA systems can measure and sum power when slowly varying, as is
the case during normal operation, they cannot provide accurate results during fast
changes. It is precisely during these fast changes that the measurements are required.
Second, a ixed-interval sampling scheme is required for estimating oscillation
modes.
Figure 3.7 shows a block diagram of the system. The time-stamped voltage and
current at each important line are measured with a phasor measurement unit (PMU).
A PMU is a type of device that makes time-synchronized measurements and conforms to the IEEE C37.118 standard. This standard includes requirements on iltering, measurement accuracy, and communication frame layout. In the Guatemalan
system, the PMU is included as part of a protective relay. The advantage of including
a PMU internal to the relay is the availability of the control function necessary to act
based on the oscillation state. Each PMU is connected to a clock. The clock receives
a GPS signal, extracts the time component, and sends it to the PMU. The result is that
all measurements are based on a common, highly accurate time reference.
The voltages and currents are measured at a rate of 60 per second. They are then
streamed over an Ethernet connection to the AMM control center and received by
a synchrophasor vector processor (SVP), an advanced PDC. The lower section of
Figure 3.7 shows the SVP algorithm in detail. Starting with the block on the left, irst
the measurements are time aligned. This is possible because of their precise time
stamp, which is included with each measurement.
Substation 1
Substation N
PMU
Clk
PMU
Clk
Wide-area Ethernet provider
SA
Time
alignment
Calculate
power
SVP
Modal
analysis
FIGURE 3.7 Block diagram of protection system.
Control
selection
48
Smart Grids
The control device includes an IEEE 61131 logic engine. With this capability,
the power on each line is calculated from the received voltages and currents. Each
calculation is at a speciic time instant.
*
P ( tk ) + Q ( tk ) = V ( tk ) ⋅ I ( tk )
(3.11)
The resulting power values for speciied lines are summed in order to determine
the total low in the system. These are compared with a threshold. The power values are also sent to a modal analysis engine that computes the oscillatory mode
frequency, amplitude, and damping. When an oscillation is detected with negative
damping, and it persists in the system for a suficient duration, the control selection
logic determines an isolation command. This is sent back through the network to
open the appropriate breaker.
Local protection that shares time-synchronized measurements for applications
such as line-current differential schemes must act with latencies on the order of
power system cycles, which are tens of milliseconds. For wide-area protection and
control the allowed latencies may be slower because the instability evolves over periods from 100 ms, for transient stability, to seconds, for voltage stability or oscillatory
stability.
3.8
SUMMARY TABLES
Table 3.2 presents communication requirements in a quantitative form with scaling
to indicate the level of dificulty on a range of 1–5 [47]. Included in Table 3.2 are
values of the following data delivery requirements: required latency, required rate,
the criticality level, the quantity of data that is delivered, and the geographical area
over which the data travel. No applications with less than a high level of criticality
are considered in this chapter. Table 3.3 quantiies each of the previously described
applications based on the ranges shown in Table 3.2. These data delivery requirements are referenced in Chapter 4.
For system analysis an input latency on the order of seconds or longer is acceptable because these are off-line applications. The data rate requirement, however, is high because resolution is important when investigating system behavior.
The criticality of the data is moderately important. Of course, it is never good to
lose data, but for off-line applications the impact is mitigated by the possibility of
collecting the data at another time. The quantity of the data can be high and data
are often collected over wide geographical areas. There are no output requirements
for analysis.
Situational awareness requires less latency than system analysis, but the effect of
human operators in the loop limits the beneit of latencies that are below approximately 1 s. The required data rate is moderate. Rates of up to 60 messages per second are useful, but higher rates probably provide detail that an operator is unable
to fully utilize when monitoring the power system in real time. The criticality level
of the information received by operators is high. A fairly large quantity of data is
Wide-Area Power Applications with Mission-Critical Data Delivery
49
TABLE 3.2
Normalized Values of Parameters
Dificulty
(5 hardest)
5
Latency
(ms)
5–20
Rate
(Hz)
240+
Criticality
Ultra
Quantity
Very high
4
3
20–50
50–100
120–240
30–120
Very high
High
High
Medium
2
1
100–1000
>1000
1–30
—
—
—
Low
Very low (serial)
Geography
Across grid or
multiple ISOs
Within an ISO/RTO
Between a few
utilities
Within a single utility
Within a substation
Note: ISO, independent system operator; RTO, regional transmission organization.
TABLE 3.3
Diversity of Data Delivery of Selected Power Applications
Inputs
Outputs
Latency
Rate
Criticality
Quantity
Geography
Latency
Rate
Criticality
Quantity
Geography
System
Analysis
1
3–4
3
5
2–5
—
—
—
—
—
Situational
Awareness
2
2–3
4
3–4
2–5
1
—
4–5
1–2
2–5
Stability
Assessment
3
2–5
4
4
2–5
—
—
—
—
—
Wide-Area
Control
3–5
3–4
5
2–4
1–5
3–5
2–4
5
1–3
1–5
System
Protection
4–5
4–5
5
2–4
1–4
5
—
5
1–3
1–3
received, and over a wide area. The quantity of data is limited by human factors
and above a certain point it becomes dificult to effectively utilize a large amount
of data. Too much information for an operator, especially during times when rapid
response is required, can make assessing and responding to the situation more
dificult.
The output communication requirements for situational awareness are constrained
by the fact that these commands are issued by human operators. Therefore, the tolerated latency is relatively high. The data rate aspect of the output is not relevant.
Operators issue single commands, not streaming instructions. Operator commands
are critical. The data quantity of the output commands is low. The geographical coverage of the commands is the same as for input data.
For stability assessment, requirements are fairly similar to those of the situational awareness application. However, the latency is stricter and the data rate
is higher because often the initial consumer of the data is a computer algorithm,
50
Smart Grids
which is able to process a more detailed information level than is possible for a
human. There are no output requirements because it is assumed that an operator
acts based on the output of the assessment. These output requirements are covered under the situational awareness column. However, stability assessment is also
an intermediate stage of wide-area control and system protection. In these cases,
the assessment requirements are part of the data delivery requirements for those
applications.
It is important that wide-area control latency is low because latency negatively
impacts the stability of feedback control loops. Similarly, the rate is fairly fast. All
control schemes are critical for correct power system operation, and this necessitates
not only a reliable communication network but also a distributed architecture when
possible. The quantity of data is large, but typically not as large as required for system analysis or situational awareness. This is because control schemes often only
require a few signals, such as positive sequence voltage or positive sequence current. The geographical span of the data is dependent on the speciics of each control
application and can range from every location within a substation all the way up to
an entire power system grid. The output data for control also require low latency and
fairly high rates. Like the received measurements, output commands have a critical
reliability need.
System protection places the most demanding requirements on latency, data rate,
and criticality of the application. The quantity of data may be lower because typically only a few key signals are monitored. The geographic scale can cover a wide
area but is also limited by the required latency and reliability. Output signals for protection are single acting, and this is why there is no speciication for rate. However,
a low latency is required.
REFERENCES
1. Annual Energy Outlook 2013, with Projections to 2040, US Energy Information
Administration, 2013.
2. V. Madani, D. Novosel, S. Horowitz, M. Adamiak, J. Amantegui, D. Karlsson, S. Imai,
and A. Apostolov, IEEE PSRC report on global industry experiences with system integrity protection schemes (SIPS), IEEE Transactions on Power Delivery, 25(4), 2143–
2155, 2010.
3. F. C. Schweppe and J. Wildes, Power system static-state estimation, part I: Exact model,
IEEE Transactions on Power Apparatus and Systems, PAS-89, 120–125, 1970.
4. J. S. Thorp, A. G. Phadke, and K. J. Karimi, Real time voltage-phasor measurements
for static state estimation, IEEE Transactions on Power Apparatus and System, PAS104(11), 3098–3106, 1985.
5. G. Zweigle, D. Finney, and R. Moxley, Adding shaft angle measurement to generator
protection and monitoring, in Western Protective Relaying Conference, October 2012,
Spokane, WA.
6. R. Zivanovic and C. Cairns, Implementation of PMU technology in state estimation:
An overview, in Proceedings of the IEEE 4th AFRICON, vol. 2, pp. 1006–1011, 24–27
September 1996, Stellenbosch, South Africa.
7. E. O. Schweitzer III and D. E. Whitehead, Real-world synchrophasor solutions, in
Proceedings of the 35th Annual Western Protective Relay Conference, October 2008,
Spokane, WA.
Wide-Area Power Applications with Mission-Critical Data Delivery
51
8. E. O. Schweitzer III and D. E. Whitehead, Real-time power system control using synchrophasors, in Proceedings of the 34th Annual Western Protective Relay Conference,
October 2007, Spokane, WA.
9. T. Yang, H. Sun, and A. Bose, Two-level PMU-based linear state estimator, in
Proceedings of the IEEE PES Power Systems Conference and Exposition, pp. 1–6, 2009,
Seattle, WA.
10. SEL-3373 Station Phasor Data Concentrator, Instruction Manual, Schweitzer
Engineering Laboratories, Inc., 11 March 2013.
11. G. Zweigle, R. Moxley, B. Flerchinger, and J. Needs, Wide-area measurements to
improve system models and system operation, in Proceedings of the 11th International
Conference on Developments in Power System Protection, DPSP, 23–26 April 2012,
Birmingham, UK.
12. D. Shi, D. Tylavsky, N. Logic, and K. Koellner, Identiication of short transmissionline parameters from synchrophasor measurements, in Proceedings of the 40th North
American Power Symposium, 28–30 September 2008, Calgary, AB.
13. Z. Huang, P. Du, D. Kosterev, and S. Yang, Generator dynamic model validation and
parameter calibration using phasor measurements at the point of connection, IEEE
Transactions on Power Systems, 28(2), 1939–1949, 2013.
14. J. V. Espinoza, A. Guzman, F. Calero, M. Mynam, and E. Palma, Wide-area measurement and control scheme maintains Central America’s power system stability, in Western
Protective Relay Conference, 16–18 October 2012, Spokane, WA.
15. R. Moxley, C. Petras, C. Anderson, and K. Fodero II, Display and analysis of transcontinental synchrophasors, in Proceedings of the 6th Annual Western Power Delivery and
Automation Conference, April 2004, Spokane, WA.
16. D. J. Trudnowski, J. W. Pierre, N. Zhou, J. F. Hauer, and M. Parashar, Performance of
three mode-meter block-processing algorithms for automated dynamic stability assessment, IEEE Transactions on Power Systems, 23(2), 680–690, 2008.
17. J. M. Ordagci, H. C. T. Santos, S. R. Morand, R. Cespedes, R. Mano, and D. Caceres,
ONS—Brasil new control center architecture conceptual design, in Proceedings of the
IEEE/PES Transmission and Distribution Conference and Exposition: Latin America,
13–15 August 2008, Bogota.
18. G. Zhang, S. Lee, R. Carroll, J. Zuo, L. Beard, and Y. Liu, Wide area power system visualization using real-time synchrophasor measurements, in Proceedings of IEEE Power
and Energy Society General Meeting, 25–29 July 2010, Minneapolis, MN.
19. J. Wang, Y. Hu, A. Johnson, H. Tram, and R. Nasri, System requirements of visualization platform for wide area situation awareness system, in 2010 IEEE Power and Energy
Society General Meeting, 25–29 July 2010, Minneapolis, MN.
20. SynchroWAVe Central Software Instruction Manual, Schweitzer Engineering
Laboratories, Inc., 1 April 2013.
21. Arizona-Southern California Outages on September 8, 2011—Causes and
Recommendations, Federal Energy Regulatory Commission and the North American
Electric Reliability Corporation, April 2012.
22. P. Kundur, Power System Stability and Control, McGraw-Hill, New York, 1994.
23. K. Sun, S. Likhate, V. Vittal, V. S. Kolluri, and S. Mandal, An online dynamic security
assessment scheme using phasor measurements and decision trees, IEEE Transactions
on Power Systems, 22(4), 1935, 2007.
24. Z. Wang, A. Giris, V. Aravnthar, and E. Makram, Wide area power system transient stability assessment using catastrophe theory method and synchrophasors, in 2011 IEEE
Power and Energy Society General Meeting, 24–29 July 2011, San Diego, CA.
25. C. Zheng, V. Malbasa, and M. Kezunovic, Regression tree for stability margin prediction using synchrophasor measurements, IEEE Transactions on Power Systems, 28(2),
1978–1987, 2013.
52
Smart Grids
26. M. Donolo, M. Venkatasubramanian, A. Guzman, and F. De Villiers, Mitigating voltage
collapse problems in the Natal region of South Africa, in Proceedings of the IEEE Power
Systems Conference and Exhibition, 15–18 March 2009, Seattle, WA.
27. M. Glavic and T. V. Cutsem, Wide-area detection of voltage instability from synchronized phasor measurements. Part I: Principle, IEEE Transactions on Power Systems,
24(3), 1408–1416, 2009.
28. M. Glavic and T. V. Cutsem, Wide-area detection of voltage instability from synchronized phasor measurements. Part II: Simulation results, IEEE Transactions on Power
Systems, 24(3), 1417–1425, 2009.
29. A. Johnson, R. Tucker, T. Tran, D. Sullivan, C. Anderson, and D. E. Whitehead, Static
var compensation controlled via synchrophasors, in Proceedings of the 34th Annual
Western Protective Relay Conference, October 2007, Spokane, WA.
30. Y. Gong and A. Guzmán, Synchrophasor-based online modal analysis to mitigate
power system inter-area oscillation, in Proceedings of the DistribuTech Conference and
Exhibition, 2009, San Diego, CA.
31. J. Chow, J. Sanchez-Gasca, H. Ren, and S. Wang, Power system damping controller
design using multiple input signals, IEEE Control Systems Magazine, 20(4), 82–90,
2000.
32. M. Mokhtari, F. Aminifar, D. Nazarpour, and S. Golshannavaz, Wide-area power oscillation damping with a fuzzy controller compensating the continuous communication
delays, IEEE Transactions on Power Systems, 28(2), 1997–2005, 2013.
33. IEEE Application Guide for IEEE Std 1547, IEEE Standard for Interconnecting
Distributed Resources with Electric Power Systems, 2009.
34. J. Mulhausen, J. Schaefer, M. Mynam, A. Guzmán, and M. Donolo, Anti-islanding
today, successful islanding in the future, in Proceedings of the 36th Annual Western
Protective Relay Conference, 20–22 October 2009, Spokane, WA.
35. M. Mills-Price, M. Scharf, S. Hummel, M. Ropp, D. Joshi, G. Zweigle, K. Ravikumar,
R. Moxley, and B. Flerchinger, Interconnection control of distributed generation with
time-synchronized phasors, in IEEE Power Systems Conference and Exposition, pp. 1–8,
20–23 March 2011, Phoenix, AZ.
36. E. O. Schweitzer III, D. E. Whitehead, G. C. Zweigle, and K. G. Ravikumar,
Synchrophasor-based power system protection and control applications, in
Proceedings of the 36th Annual Western Protective Relay Conference, October 2009,
Spokane, WA.
37. SEL-421 Relay Protection and Automation System, Instruction Manual, Schweitzer
Engineering Laboratories, Inc., 22 February 2013.
38. IEEE C37.118.1–2011, IEEE Standard for Synchrophasor Measurements for Power
Systems, 2011.
39. Q. Yang, T. Bi, and J. Wu, WAMS implementation in China and the challenges for bulk
power system protection, in IEEE Power Engineering Society General Meeting, 24–28
June 2007, Tampa, FL.
40. P. Wang, W. Deng, and Z. Qian, Method for the measurement of rotor voltage and current of power generator, IET Generation, Transmission, and Distribution, 4(7), 854–
860, 2010.
41. G. Zweigle and V. Venkatasubramanian, Model prediction based transient stability control, in IEEE PES Conference and Exhibition, 7–10 May 2012, Orlando, FL.
42. G. Zweigle and V. Venkatasubramanian, Wide-area optimal control of electric power
systems with application to transient stability for higher order contingencies, IEEE
Transactions on Power Systems, 28(3), 2313–2320, 2013.
43. A. Guzmán, V. Mynam, and G. Zweigle, Backup transmission line protection for ground
faults and power swing detection using synchrophasors, in Proceedings of the 34th
Annual Western Protective Relay Conference, October 2007, Spokane, WA.
Wide-Area Power Applications with Mission-Critical Data Delivery
53
44. D. Hu and V. Venkatasubramanian, New wide-area algorithms for detection and mitigation of angle instability using synchrophasors, in IEEE Power Society General Meeting,
24–28 June 2007, Tampa, FL.
45. E. Martinez, N. Juárez, A. Guzmán, G. Zweigle, and J. León, Using synchronized phasor angle difference for wide-area protection and control, in Proceedings of the 33rd
Annual Western Protective Relay Conference, October 2006, Spokane, WA.
46. C. Taylor, D. Erickson, K. Martin, R. E. Wilson, and V. Venkatasubramanian, WACS—
Wide-area stability and voltage control system: R&D and online demonstration,
Proceedings of the IEEE, 93(5), 892–906, 2005.
47. D. E. Bakken, A. Bose, C. H. Hauser, D. E. Whitehead, and G. C. Zweigle, Smart generation and transmission with coherent, real-time data, Proceedings of the IEEE, 99(6),
928–951, 2011.
4 High Availability, Low
GridStat
Latency, and Adaptive
Sensor Data Delivery
for Smart Generation
and Transmission
David E. Bakken, Harald Gjermundrød,
and Ioanna Dionysiou
CONTENTS
4.1
4.2
4.3
4.4
Introduction .................................................................................................... 56
Requirements and Guidelines for Coherent Real-Time Data Delivery .......... 57
4.2.1 System Model ..................................................................................... 59
4.2.2 Delivery Requirements for a WAMS-DD ..........................................60
4.2.3 Implementation Guidelines for a WAMS-DD ....................................64
4.2.3.1 Context .................................................................................64
4.2.3.2 Implementation Guidelines .................................................. 65
4.2.3.3 Analysis................................................................................ 70
Analysis of Existing Technologies for a WAMS-DD ..................................... 71
4.3.1 Technologies and Standards at the Traditional Network Layers ........ 71
4.3.1.1 Very Inadequate ................................................................... 71
4.3.1.2 Better Yet Incomplete .......................................................... 75
4.3.2 Middleware Technologies and Standards ........................................... 76
4.3.2.1 Deinition of Middleware ..................................................... 76
4.3.2.2 Broker-Based Publish-Subscribe Middleware ..................... 77
4.3.2.3 Peer-to-Peer Publish-Subscribe ........................................... 78
4.3.2.4 Other Middleware ................................................................ 79
4.3.3 Existing Electric Sector Communications-Related Technologies
and Standards .....................................................................................80
4.3.4 Summary ............................................................................................ 82
NASPInet ........................................................................................................ 83
4.4.1 Architecture ........................................................................................ 83
4.4.2 Service Classes ...................................................................................84
55
56
Smart Grids
4.5
Why WAMS-DD Must Include Middleware, Not Solely Network Layers .... 85
4.5.1 Middleware in General ....................................................................... 86
4.5.2 Middleware and QoS+ ........................................................................ 86
4.5.3 Middleware and DRs and IGs ............................................................ 87
4.5.4 Middleware and Legacy Systems ....................................................... 88
4.5.5 Middleware and Avoiding QoS Stovepipe Systems ...........................90
4.6 Security and Trust for WAMS-DD ................................................................. 91
4.7 GridStat...........................................................................................................92
4.7.1 Overview............................................................................................. 93
4.7.1.1 Capabilities .......................................................................... 93
4.7.1.2 More Detail .......................................................................... 93
4.7.1.3 Performance and Scalability ................................................ 95
4.7.2 Rate Filtering for Per-Subscriber QoS+ ............................................. 96
4.7.2.1 Forwarding Algorithm .........................................................96
4.7.2.2 Example of the Forwarding Algorithm ...............................96
4.7.3 Condensation Functions......................................................................97
4.7.3.1 Condensation Function Architecture and Capabilities ........ 98
4.7.3.2 Condensation Function Development ..................................99
4.7.4 Operational Modes .............................................................................99
4.7.4.1 Mode Change Algorithms....................................................99
4.7.4.2 Hierarchical Mode Change Algorithm .............................. 101
4.7.4.3 Flooding Mode Change Algorithm.................................... 102
4.7.4.4 Overview of Performance .................................................. 102
4.7.5 Systematic Adaptation via Data Load Shedding with Modes .......... 102
4.7.6 Remote Procedure Call..................................................................... 103
4.7.6.1 Mechanism Details ............................................................ 104
4.7.6.2 The 2WoPS Protocol .......................................................... 104
4.7.6.3 The Ratatoskr RPC Mechanism ........................................ 105
4.7.7 Security and Trust in GridStat .......................................................... 106
4.8 A New World for Power Application Programmers and Researchers.......... 107
4.9 Conclusions ................................................................................................... 109
Acknowledgments.................................................................................................. 109
Disclaimer .............................................................................................................. 110
References .............................................................................................................. 110
4.1 INTRODUCTION
Electric power grids around the world are becoming increasingly stressed by factors
including inadequate transmission growth, system operators and power engineers
retiring in large numbers, and integration of renewable sources of energy whose
physics is different from well-known sources such as hydro and coal. All these factors, and more, can be mitigated by greatly increased sharing of sensor data between
utilities (and independent system operators [ISOs]/regional transmission organizations [RTOs]) in a grid, something that is extremely limited in power grids today.
This is an important step in helping make grids more resilient, because inadequate
situational awareness resulting from little interutility data sharing has been a major
GridStat
57
contributing factor in virtually all recent major blackouts. In such blackouts, there
are always a few physical or operational root causes that are blamed. However, due
to the poor situational awareness, the grids’ operators are unaware of emerging problems until it is too late to prevent a blackout.
Additionally, closed-loop applications such as distributed control and distributed
protection are being increasingly deployed. However, their data delivery requirements (DRs) are the most extreme of any infrastructure in the world. Thus, it is
crucial to ensure that data delivery for power grids is appropriate for the task at
hand. Conversely, adopting more generic data delivery infrastructures from other
industries, such as air trafic control, which seem supericially similar, can cause
blackouts. Similarly, using network mechanisms such as multiprotocol label switching (MPLS) that provide very weak statistical guarantees and very crude control
over the network (e.g., MPLS has only eight trafic categories) is also unwise.
In this chapter, we describe GridStat, which is in essence an overlay network providing strong guarantees and specialized semantics for wide-area monitoring system
data delivery (WAMS-DD). In conigurations where GridStat’s specialized routers
do not have 100% penetration, they can be augmented by other mechanisms providing very strong network-level quality of service (QoS) guarantees in order to support
closed-loop applications over the wide area. GridStat can also overlay (and help better manage and integrate) other networking subsystems with weaker QoS guarantees
in a critical infrastructure. This is in essence a software-deined network (SDN).
However, it is not a generic one: it is an SDN infused with semantics from both middleware and power grid applications and sensors that allow it to exert ine-grained
control over its trafic. This not only provides very low latency and extremely high
availability, but also allows the SDN to change its delivery patterns in a small fraction of a second to adapt to power anomalies, benign information technology (IT)
failures, and cyberattacks.
The remainder of this chapter is organized as follows. Section 4.2 describes
baseline DRs that any WAMS-DD system must meet. It then derives 20 implementation guidelines (IGs) that those requirements mandate any WAMS-DD
to use. Section 4.3 then presents a detailed analysis of how well existing technologies at the network layers, at the middleware layers, and from the electricity sector meet these requirements and guidelines. Next, Section 4.4 overviews
NASPInet, a WAMS-DD initiative in North America. After this, Section 4.5
explains why WAMS-DD must include the middleware layers. Section 4.6 then
summarizes security and trust issues for WAMS-DD. Section 4.7 describes
GridStat. Section 4.8 summarizes how GridStat enables a new way of thinking and programming for programmers and researchers of power applications.
Finally, Section 4.9 concludes.
4.2 REQUIREMENTS AND GUIDELINES FOR
COHERENT REAL-TIME DATA DELIVERY
Data delivery in the power system today can be improved by reducing the
use of hard-coded protocols, developing more reusable systems, and providing real end-to-end (E2E) performance guarantees. For example, in protection
58
Smart Grids
applications, overprovisioning provides low latencies and high availability in the
steady state but not necessarily in the face of IT failures, bugs in software or hardware that cause spurious trafic, or cyberattacks. As more applications that can exploit
coherent, real-time data delivery emerge, such as those outlined in Chapter 3, using
isolated networks may soon become unsustainable, as will designing a new communications system for each new application or application family.
Fortunately, the state of the art in distributed computing, real-time systems, and
fault-tolerant computing does support provision of strong guarantees with data delivered to many applications. If designed, implemented, and validated correctly, a stateof-the-art data delivery system can greatly lower the barrier to entry (in both time
and money) and enable deployment of new power applications by simplifying the
process of adding new sensors. If designed incorrectly, it will be dificult to maintain
in the future because it will not be able to keep up with increasing demands. Further,
these data delivery systems will have a long life, and no single network-level mechanism (for multicast, security, or QoS) can be assumed to be everywhere.
It is crucial, therefore, that data delivery systems between the mission-critical
peer-to-peer (P2P) automatic protection and control systems, and the power grid’s
operations IT backbone have interoperability between different kinds of network
mechanisms providing the same property, such as delay guarantees [1].
In this section, we examine how a WAMS-DD will be an enabling technology for the new and emerging power system. We irst overview the performance
and reliability requirements that a WAMS-DD must meet. We then present IGs,
based on best practices in other industries and in the ield of distributed computing systems, for achieving these requirements. Next, we compare how existing
technologies meet these DRs and design guidelines when used in isolation without additional overlay networks. These include technologies and standards at the
network layers (and below), at the middleware layer(s), and related ones from the
power industry. We also discuss relevant research and development for wide-area
middleware. After this, we discuss the emerging NASPInet effort and the GridStat
data-delivery middleware. Finally, we conclude this section with a brief discussion
of pertinent cybersecurity issues for next-generation data-delivery services for the
electric power grid.
Note that the following analysis focuses on coherent but asynchronous data
delivery for operations, but the emerging communications infrastructure will
provide the additional beneit of distributing the time signal required for timesynchronized measurements and control. Typically, time is received via a global
positioning system (GPS) and distributed over a separate physical network using
protocols such as the inter-range instrumentation group (IRIG). This results in a
physical cable connection to the measuring devices, such as phasor measurement
units (PMUs). Combining time distribution with the communications network
provides advantages such as simplicity and reliability [2]. Furthermore, for many
applications, such as a control scheme or system protection scheme, that use separate mission-critical P2P communications, operation can proceed even if global
time is lost, as long as they maintain a local coherent time signal. The communications infrastructure can provide this locally common time signal when the primary
GPS signal is unavailable.
59
GridStat
4.2.1 SYSTEM MODEL
Figure 4.1 depicts the architecture of a WAMS-DD. Application programs or irmware that emit a stream of updates are called publishers, which are denoted as Pub1
through PubN in the diagram; Pub1, for example, outputs updates to variables X and Y.
Applications that receive these updates are called subscribers, which are denoted as
Sub1 through SubN. In the diagram, Sub1 subscribes to Y from Pub1 and to W from
PubN.
In the usual case in publish-subscribe (pub-sub) systems, neither publisher nor
subscriber needs to know about the other; they are decoupled such that they only
know about the variable they publish or subscribe to and how to contact the delivery system. In cases where the subscriber requires conirmation that the update
came from its legitimate publisher—which may be common with a WAMS-DD—
data-integrity techniques from the computer security ield can be used by the datadelivery system.
Creating a pub-sub delivery path requires two steps. Publishers register their variables with the delivery system (only once per variable, not once per subscriber),
and later subscribers request a subscription to a given variable. For both publishers
and subscribers, the delivery system returns a handle to a piece of code called a
proxy, which is generated at compile time by the data-delivery middleware. This
proxy contains logic provided by the data-delivery service, which, besides doing
the usual middleware proxy activities such as packaging of the parameters into a
message, is also a place where data-delivery mechanisms may reside. In Figure 4.1,
we denote a publisher-side proxy as Pub-Prx-Mech and the subscriber-side proxy as
Sub-Prx-Mech.
After the variable is registered and subscribed to, updates to variables low from
publishers to subscribers, as shown by the gray arrows in Figure 4.1. To do this, they
traverse what we call the WAMS-DD cloud. This is opaque because, as shown later
in this section, it can be implemented in different ways resulting in different tradeoffs. For the purposes of our system model, the WAMS-DD cloud consists of a graph
in which the edges are network links and the nodes contain forwarding mechanisms
that can forward a message on its way toward a subscriber.
X
Y
Pub1
Sensor app
or firmware
Sub1
WAMS-DD cloud
Pub-Prx-Mech proxy
X, Y updates
Z
W
PubN
Sensor app
or firmware
Pub-Prx-Mech proxy
Z, W updates
FIGURE 4.1 Architecture and system model of a WAMS-DD.
Power app
Y
W
Proxy Pub-Prx-Mech
Y, W updates
SubN
Power app
Z
X
Proxy Pub-Prx-Mech
Z, X updates
60
Smart Grids
We note that this WAMS-DD is in practice a virtual overlay implemented with
a customer’s existing links. These existing links will have inherent trade-offs, so,
considering the capabilities, most utilities/ISOs would conclude that their existing
infrastructure will not support any closed-loop applications, at the very least. Thus,
the existing links would in practice be augmented by adding key additional links
from various sources, such as a Tier 1 iber provider, lighting up utility dark iber,
or buying some new links with technologies providing very strong guarantees (see
Table 4.2). This is discussed further in Section 4.5.
Updates from a publisher of a sensor variable thus traverse one or more paths to
be delivered to a given subscriber. Along a given path, an update may be delayed, so
that its required delivery latency cannot be met, or the update may be dropped due
to failures in a network link or forwarding node or due to a cyberattack. However, the
probability of an update not meeting its DRs can be kept extremely low by carefully
designing the WAMS-DD and by allocating multiple paths for important updates.
That is, a WAMS-DD can be constructed so that the on-time delivery probability
is very high, as long as its design constraints are met. Informally, these include forwarding capacity per node, maximum link trafic, number and kind of benign failures and cyberattacks, and so on.
We now overview the DRs in Section 4.2.2; then, in Section 4.2.3, we describe
IGs that can be used to meet these DRs with extremely high probabilities. These
probabilities offer the potential to practice using dual isolated networks for critical protection applications, while at the same time supporting many more application families with thousands of update lows. However, such delivery technologies
clearly need to be proven in the ield before any migration to them can begin to be
contemplated.
4.2.2
DELIVERY REQUIREMENTS FOR A WAMS-DD
The following DRs must be met by a WAMS-DD [3]; these do not include the details
of cybersecurity-related requirements (which are overviewed in Sections 4.5.5 and
4.7.7). These DRs encompass the requirements of other network layers that the
WAMS-DD layer builds upon. They are summarized in Table 4.1, along with their
IGs, which will be overviewed in the next subsection.
Requirement 1: Hard, E2E guarantees must be provided over an entire grid
because protection and control applications depend on the data delivery. The guarantees must be deterministic: met unless the system’s design criteria have been violated (e.g., trafic amount, number of failures, and severity of cyberattack).
Requirement 2: WAMS-DDs must have a long lifetime and thus must be
designed with future prooing in mind. This is crucial in order to amortize costs over
many projects, utilities, grids, and so on. The goal of NASPInet, for example, is to
last at least 30 years. To achieve this, it is important not to have applications hardcoded directly using the application programming interfaces (APIs) of a single QoS+
mechanism for a given kind of resource (e.g., encryption, network delivery). Rather,
wrappers or, better, more comprehensive middleware should be used. The system’s
libraries can be updated to add new lower-level QoS mechanisms as they become
available without having to recompile the application.
GridStat
61
Requirement 3: One-to-many is the normal mode of communications, not pointto-point. Increasingly, a given sensor value is needed by multiple power applications.
This is best done by pub-sub middleware, but in a worst case (unfortunately common in power grids) it is done by network-level multicast (e.g., Internet protocol [IP]
multicast [IPMC]).
Requirement 4: E2E guarantees must be provided for a wide range of QoS+.
Data delivery for the power system is not “one size its all” [4], as shown in
Chapter 3. For example, to provide very low latencies, very high rates, and very
high criticality/availability to all applications would be prohibitively expensive.
Fortunately, many applications do not require these stringent guarantees, but their
less stringent requirements must nevertheless be met.
Examples of the wide ranges that must be provided follow:
1. Latency and rate: Ten milliseconds or less, up to seconds (or hours or days
for bulk transfer trafic), 001–720 Hz or more.
2. Criticality/availability: IntelliGrid [4] recommends ive levels of availability of data, from medium up to ultra (99.9999%), that is, six nines.
3. Cybersecurity: Support a range of trade-offs of strength of encryption and
authentication compared with delay induced, resources consumed, and the
exact lavor of authentication provided.
Requirement 5: Some merging and future system integrity protection schemes
(SIPS), transient stability, and control applications require ultralow latencies and
one-way delivery on the order of a half or full power cycle (8–16 ms in the United
States) over hundreds of miles [5]. Thus, any forwarding protocols should not add
more than a millisecond or two of latency (through all forwarding hops) on top of the
speed of light in the underlying communications medium.
These latencies must be provided in a way that:
1. Is predictable and guaranteed for each update message.
Each sensor update needs to arrive with an extremely high probability
within its required guaranteed deadline, not a much weaker aggregate guarantee over longer periods of time, applications, and locations such as is
provided by MPLS technology [6].
2. Tolerates nonmalicious failures in the WAMS-DD infrastructure.
No system can tolerate unlimited kinds and numbers of failures. However,
much as the power system must continue in the face of one or more known
contingencies, the IT infrastructure on which it increasingly depends must
still provide these hard, E2E guarantees in the face of failures (up to design
limits).
3. Tolerates malicious cyberattacks.
Power systems are known to be subjects of extensive study and probing by multiple organizations that have signiicant information warfare capabilities, including nation-states, terrorist organizations, and organized crime. A WAMS-DD
must adapt and continue to deliver data despite cyberattacks of a designed severity
62
TABLE 4.1
IGs and the DRs That Mandate Them
DR1: Hard
E2E WAN
Guarantees
DR2:
Future
Prooing
DR3:
DR4:
One-ToWide
Many Range of
Comms QoS+ …
4A:
Latency
and Rate
4B:
Criticality/
Availability
4C:
Cybersecurity
5A:
DR5:
Per-Update
5B:
5C:
Ultralow
and
Tolerating
Tolerating
DR6: High
Latencies … Predictable Failures Cyberattacks Throughput
IGx
Prerequisites
—
—
—
—
—
—
—
X
X
—
—
—
X
—
—
—
X
—
—
—
X
X
X
X
—
—
—
X
—
—
—
—
—
—
—
—
X
—
—
—
X
—
—
—
—
—
—
—
—
X
—
—
—
—
—
—
—
—
—
—
—
—
—
2, 3
X
—
—
—
—
X
—
—
X
—
—
X
—
X
X
—
—
X
X
X
—
—
—
—
—
—
X
—
—
—
—
—
—
—
X
—
—
X
—
X
—
—
—
—
—
—
—
X
X
X
—
8
X
—
—
—
—
—
—
—
X
—
—
—
—
X
—
—
—
—
—
—
—
X
—
—
—
8–10
IG1: Avoid posterror
recovery
mechanisms
IG2: Optimize for
rate-based sensors
IG3: Provide
per-subscriber QoS+
IG4: Provide eficient
multicast
IG5: Provide
synchronized rate
downsampling
IG6: Do not depend
on priority-based
“guarantees”
IG7: Provide
end-to-end
interoperability
across different/new
IT technologies
(multicast, QoS+)
IG8: Exploit a priori
knowledge of trafic
IG9: Have systematic,
quick internal
instrumentation
IG10: Exploit smaller
scale of the
WAMS-DD
IG11: Use static, not
dynamic, routing
Smart Grids
X
Summary of
Implementation
Guideline IGx
—
—
—
—
—
—
—
X
X
X
—
—
IG12: Enforce
complete perimeter
control
X
—
—
—
—
—
—
—
X
X
X
X
12
IG13: Reject
unauthorized
messages quickly
and locally
—
—
—
—
—
—
—
—
—
—
—
X
2, 8
IG14: Provide only
simple subscription
criteria
—
—
—
—
—
—
—
—
X
—
—
X
2
IG15: Support
transient, not
persistent, delivery
—
—
—
—
—
—
—
—
X
—
—
X
—
IG16: Do not
overdesign
consistency and
(re)ordering
—
—
—
—
—
—
—
—
—
—
—
X
2, 8, 14–16
IG17: Minimize
forwarding-time
logic
X
X
—
—
X
X
X
—
—
—
—
—
—
IG18: Support
multiple QoS+
mechanisms for
different operating
conditions
—
—
—
—
—
—
—
—
X
—
—
X
17
X
—
—
—
—
—
—
—
X
—
—
X
—
IG19: Inspect only
message header, not
payload
IG20: Manage
aperiodic trafic
GridStat
X
63
64
Smart Grids
(it should be possible to increase this threshold over the life of the system). Note
that a bug in hardware or software that generates spurious trafic can have an effect
similar to that of a cyberattack.
Requirement 6: Extremely high throughput is required. Today’s synchrophasor
applications are generally limited to 30 or 60 Hz in the United States, in part because
the communications systems they use are not designed to support higher rates. Not
to provide much higher sustainable throughput would greatly limit the number of
new applications that can help the power system’s stability. Indeed, not just synchrophasors but digital fault recorders (DFRs) and intelligent electronic devices in
substations provide a wealth of data. It is quite conceivable and likely that “if you
build it, they will come” and there will be many thousands of synchrophasors, relays,
DFRs, and other sources of sensor updates across a grid. These devices can output
at 720 Hz and sample at 8 kHz, but their full output is not always used remotely due
to communications limitations. If key relay or DFR data could be delivered from a
set of devices across a grid at 720 Hz, many new opportunities would open up for
transient protection without using expensive dedicated networks or “drilling down”
into the root causes of an ongoing power contingency using additional contingencyspeciic data.
We are not aware of any commercial or military market for a wide-area data
delivery infrastructure that has the stringent requirements of a WAMS-DD, including the ability to enforce complete perimeter control, the ability to know the vast
majority of the trafic ahead of time, and other factors incorporated into the IGs
described next in this chapter. The reason is quite simple; electric power is the only
market with such stringent requirements. However, these requirements are achievable using state-of-the-art distributed real-time embedded computing [7,8], as long
as a careful E2E analysis is done [9] and the core data delivery mechanisms are not
saddled with unnecessary features. Much broader reliability has been explored in
the fault-tolerant distributed computing community, from where appropriate lessons,
both good and bad, should be heeded [10,11].
4.2.3 IMPLEMENTATION GUIDELINES FOR A WAMS-DD
We now discuss practical issues that arise when trying to construct a WAMS-DD
that meets the above requirements.
4.2.3.1 Context
The requirements outlined in the previous section were kept to a bare minimum. In
order to achieve them, however, we believe it will be necessary to utilize a number
of IGs, many of which are quite different from what is provided in today’s best-effort
Internet and what has been the conventional wisdom in networking research. In this
section, we enumerate and explain these IGs. However, they are necessary (or at least
highly advisable) in order to meet the stringent DRs.
Some of the IGs below (e.g., IG4 and IG5) are actually deemed requirements
for NASPInet [12], but we describe them here as IGs because it is possible to build
a WAMS-DD without them (although we believe that would be inadvisable for an
interutility backbone such as the proposed NASPInet; see Section 4.4). These IGs
GridStat
65
are drawn from a number of sources, including our knowledge of what the state
of the art in distributed computing has demonstrated is feasible, best practices in
other industries, and decades of experience gained in Defense Advanced Research
Projects Agency wide-area application and middleware projects, both our own and
those of others.
We note that these guidelines refer to best practices of how to build a WAMS-DD.
Other guidelines (beyond the scope of this chapter) will apply on how to use one,
and will need to be developed as best practices. For example, in our experience,
many power engineers assume that, with synchrophasors, they should have phasor
data concentrators (PDCs) inside their utility at many levels. However, this is a very
bad idea for updates that need to be delivered with ultralow latencies (DR5). A PDC
aggregates many PMU signals, performs error correction and angle computation,
then outputs the collection of this information for a given PMU time slot (this is
called time alignment). Such a PDC may have dozens of PMU signals coming into
it, so performing time alignment means that the output has to wait until the slowest
PMU update arrives. In this case, the updated sensor values will have suffered signiicant delays even before they leave the utility to be transported by a WAMS-DD
such as NASPInet. Thus, for those updates that require ultralow delivery latency,
any PDC or other time alignment should be placed as close to the subscribers as possible (ideally in their local area network), even at the cost of a small amount of either
wasted bandwidth or duplication of PDCs. Similarly, data that are required with
extremely low latency should not have a database in their path: they can be entered
into a database after they are sent out, but the database must not slow down the fast
delivery path.
We also note that the scope of these IGs involves only the data-delivery system
for WAMS-DD. It does not include the supporting services that will be required for
coniguration, security, path allocation, resource management, and so on. It will be
important for the WAMS-DD that the use of these tools avoids hard-coding choices,
but rather allows them to be speciied in a high-level policy language (or at least a
database) [13–15]. For an example of a hierarchical version of such services (a “management plane”), see [16,17].
4.2.3.2 Implementation Guidelines
Table 4.1 provides an overview of the IGs and the DRs that require the given IG. We
now explain each of the IGs in turn.
Guideline 1: Avoid posterror recovery mechanisms. Traditional protocols for
the Internet in general, and reliable multicast protocols from the fault-tolerant computing research community, use posterror recovery. In these protocols the receiver
sends either a positive acknowledgment (ACK) when it receives a message, or a negative acknowledgment (NACK) when it concludes that the message will not arrive.
However, both add considerable latency when a message* is dropped: three one-way
latencies are required, plus a timeout that is much greater than the average one-way
message latency.
* We use the term message rather than packet, because in many cases we are describing middlewarelayer mechanisms above the network and transport layers.
66
Smart Grids
The better alternative is to send sensor updates (messages) proactively over multiple disjoint paths, each of which meets the latency and rate requirements [18,19].
Indeed, if multiple independent messages, each going over a QoS-managed path,
cannot meet the delivery deadline, then sending ACKs or NACKs is very unlikely to
help, and, indeed, will almost certainly only make things much worse.
Guideline 2: Optimize for rate-based sensors. WAMS-DD can be made with
higher throughput and robustness if they are not overengineered. General-purpose
pub-sub systems offer a wide range of trafic types, because they are designed to
support a wide range of applications. However, in a WAMS-DD, the vast majority of
the trafic will be rate based. Design accordingly.
Guideline 3: Provide per-subscriber QoS+. It is crucial that different subscribers
to the same sensor variable are able to have different guarantees in terms of latency,
rate, and criticality/availability. If not, then a lot of bandwidth will be wasted: all
subscribers will have to be delivered that sensor’s updates at the most stringent QoS+
that any of its subscribers requires.
Guideline 4: Provide rate-eficient multicast. In order to achieve the highest
throughput possible, it is imperative to avoid unnecessary network trafic. Thus,
never send an update over a link more than once. Also, as a sensor update is being
forwarded through the network, if it is not needed downstream in the multicast tree
(e.g., those subscribers require it at a lower rate than other subscribers), the update
message should be dropped, which can best be implemented using an in-network
rate down-sampling mechanism, as is done in GridStat [18,19].
These irst four guidelines add up to a need for multicast routing heuristics that
provide multiple, disjoint paths to each subscriber, with each path meeting the subscriber’s latency requirement. A family of heuristics developed for this multicast
routing problem [20,21] conirms the feasibility of the approach at the anticipated
scale (see IG10) if routing decisions are made statically (IG11).
Guideline 5: Provide synchronized rate downsampling. In providing rate
downsampling, it is important not to downsample in a way that destroys the
usefulness of some data. For example, synchrophasors are used to take a direct
state measurement at a given microsecond. If some subscribers require only a
small fraction of the updates for a set of synchrophasor sensors, it is important
that the updates that reach the subscriber at each interval carry the same time
stamp. For example, if a subscriber only requires one-tenth of the updates from
two different variables, then it would not be meaningful to get updates {#1, #11,
#21, …} from one synchrophasor and updates {#2, #12, #22} from another synchrophasor, because the given measurements (e.g., #1 vs. #2) do not correspond
to the same time (they are not the same snapshot), which is the main point of
synchrophasors.
Guideline 6: Do not depend on priority-based “guarantees.” Pub-sub delivery systems typically offer a way to specify a priority, so if the trafic gets too
heavy less important trafic can be dropped. However, this does not provide a hard
E2E guarantee to subscribing applications, and even applications that are not of the
highest criticality still need their DRs to be met. Instead of priorities, mechanisms
must be used that exploit the characteristics of WAMS-DD (as outlined in these
guidelines) to provide each subscriber with irm assurances that its guarantees will
GridStat
67
be met as long as there are not more than the agreed-upon number of failures or
severity of cyberattack.
Guideline 7: Provide E2E interoperability across different/new IT technologies
(providing multicast, latency, rate, etc.). A grid-wide WAMS-DD will ipso facto have
to span many utility and network organizations. It is unlikely that the same mechanisms will be present across all these organizations. And, even if they are today, if the
WAMS-DD gets locked into the lower-level APIs and semantics of a given multicast
or QoS mechanism, it will be dificult to “ride the technology curve” and utilize new
and better mechanisms that will inevitably become available over the long lifetime of
the WAMS-DD. This is a stated goal of the GridWise community, for example [22].
Fortunately, it is possible to use middleware to span these different underlying technologies in order to provide guarantees that span this underlying diversity.
Guideline 8: Exploit a priori knowledge of predictable trafic. Internet routers
cannot make assumptions or optimizations based on the characteristics of the trafic
that they will be subjected to, because they are intended to be general purpose and
support a wide range of trafic types. WAMS-DDs, however, have trafic that is not
just rate based, but is almost all known months ahead of time (e.g., when an engineering survey is carried out on a new power application). This common case can be
optimized, as described in later IGs below.
Guideline 9: Have systematic, quick internal instrumentation. In order to provide
E2E guarantees across a wide area despite failures and cyberattacks, IG8 must be
exploited to provide systematic and fast instrumentation of the WAMS-DD. This allows
much quicker adaptation to anomalous trafic, whether accidental or malicious in origin.
Guideline 10: Exploit smaller scale of the WAMS-DD. This is crucial if the challenging DRs are to be met over a wide area with reasonable cost. However, this
requires rethinking the conventional wisdom in networking research and commercial middleware products.
A WAMS-DD for even the largest electric grid will be orders of magnitude
smaller in scale than the Internet at large,* so it is feasible for the entire coniguration to be stored in one location for the purposes of (mostly off-line) route selection.
Additionally, academic computer science researchers historically consider something that is O(N2) for path calculation with N routers or forwarding engines (FEs) to
be infeasible; see, for example, [17]. However, this assumption ignores two key factors for WAMS-DD. First, N is not in the neighborhood of 108, as in the Internet, but
rather is more likely to be ~103, at least for the next 5–10 years; that is, even if O(N2)
algorithms are feasible at this scale. Second, as a rule, power engineers do not decide
that they need a given sensor’s values seconds before they really need them, due
in part to the fact that today’s data-delivery infrastructure requires them to recode
hard-coded socket programs and then recompile. Rather, power engineers plan their
power contingencies (and what data they will need in them) months ahead of time
with detailed engineering studies, and similarly for their monitoring, protection,
* For example, in the entire United States there are approximately 3500 companies that participate
in the grid. We thus believe that the number of router-like FEs that would be required for a NnDB
backbone (at least in the case of broker-based pub-sub; deined later) is at most 10 4 and likely only
around 103.
68
Smart Grids
control, and visualization needs. Thus, the routing/forwarding decisions involved in
path selection can be done off-line well ahead of time, while still allowing for handling a modest number of subscription requests at runtime.
It is also feasible for router-like FEs to store state for each low. Having a router
keep per-low state has long been considered a bane to networking researchers,
because it is considered to be prohibitively unscalable. However, with the much
smaller scale and the much more limited type of applications for a WAMS-DD,
storing per-low state is not only feasible but is also a requirement for providing
IG3 (per-subscriber QoS+) with IG4 (eficient multicast); this is something that our
GridStat project has been advocating for many years [16]. However, recently networking researchers have realized the necessity of storing per-low state to provide
any reasonable kind of QoS [23].
Guideline 11: Use static, not dynamic, routing and naming. Much stronger
latency guarantees can be provided when using complete knowledge of topology
coupled with static routing. Complete topology knowledge is a reasonable assumption in a managed NnDB, given that it will be a carefully managed critical infrastructure with complete admission control. Also, almost all of the sensors and power
applications will be known well ahead of time, so optimizations for static (or slowly
changing) naming can potentially be useful and can be done while still providing
more lexible and dynamic discovery services at a much lower volume. We note
that networking and security researchers generally assume that the membership of
multicast groups (or a set of subscribers) may change rapidly; see, for example, [17].
However, as noted above, that is not the case with WAMS-DD.
Guideline 12: Enforce complete perimeter control. All trafic put onto a
WAMS-DD must pass admission control criteria (permissions based on both security and resource management) via a management system: the publisher registering a
sensor variable (at a given rate) and the subscribers asking for a subscription with a
given rate and E2E latency. This is essential to provide guarantees at a per-message
granularity. It also enables quicker adaptations. This should ideally be done in the
publishing application’s address space, via a middleware proxy, so that spurious trafic does not consume any network or router capacity.
Guideline 13: Reject unauthorized messages quickly and locally. Messages that
have gone around the admission control perimeter should be rejected as soon as possible, ideally at the next WAMS-DD FE, rather than going most or all the way across
the WAMS-DD consuming resources along the way. Detection of such unauthorized
packets is an indicator of anomalous trafic and evidence of a failure or cyberattack that needs to be reported to the management infrastructure. When suficient
evidence is collected over suficient time, an appropriate adaptation can occur, but
such rejection would be in practice reported to a management infrastructure, given
that it is an anomaly.
Guideline 14: Provide only simple subscription criteria. This is exactly the
opposite of what is usually done with general-purpose pub-sub in either academic
research or commercial products: both tend to favor complex subscription criteria,
which are expensive to evaluate as each update is forwarded through the system
(think of complex “topics”) [24]. For example, in GridStat, the subscription criteria
are latency, rate, and number of paths, and, as noted below, the forwarding decision
GridStat
69
is made completely based on rate, with static routing. Note also that the lower-level
ID of a sensor variable could still be looked up through a complicated discovery
service; this guideline is concerned with avoiding complex forwarding logic.
Guideline 15: Support only transient delivery, not persistent delivery. Most
pub-sub systems offer persistent delivery, whereby if an event cannot be immediately forwarded it is stored for some time and then the delivery is retried. This
harms throughput, however, as well as potentially the per-packet predictability
(because it requires storing the data). In our experience it is completely unnecessary for real-time visualization, control, and protection, due to the temporal
redundancy inherent in rate-based update streams: the next update will be arriving
very soon anyway, so the usefulness of a given update fades very quickly. Thus,
it is inadvisable to complicate the critical paths of delivery mechanisms to support persistent delivery (though it can be provided “on the side” by other mechanisms). Furthermore, in the power grid, historian databases are already required
for archiving data for regulatory reasons, so there is no reason to complicate the
design or otherwise bog down the fastest and highest-availability mechanisms of
WAMS-DD to deliver historical data.*
Guideline 16: Do not overdesign for consistency and (re)ordering. Research in
fault-tolerant multicast tends to provide different levels of ordering between updates
from the same publisher, or between different clients of the same server, as well as
consistency levels between different replicas or caches of a server. There is no need
for anything like this in a WAMS-DD. Present data-delivery software provides no
kind of consistency at all, so power applications assume nothing in terms of consistency and ordering. The only requirement for such consistency that we have found
is relected in IG5 for synchrophasors, and the only ordering of any kind is where a
PDC combines updates from different PMUs into one message to pass onward. With
devices such as synchrophasors that have accurate GPS clocks, the order of events
can be directly known and no delivery ordering mechanism is required other than
that which is done by a PDC.
Guideline 17: Minimize forwarding-time logic. In order to provide the highest
throughput, the forwarding logic that decides how a packet or update is to be forwarded on should be kept as simple as possible. In the GridStat project, forwarding
decisions are made based solely on the subscription rate of subscribers downstream
in the multicast tree [16,19]. Given that the trafic is rate based (IG2) and known
ahead of time (IG8), that subscription criteria are kept simple (IG14), that only
transient delivery is supported (IG15), and that there are no consistency semantics
(IG16), much logic can be pushed off to subscription set-up time or even off-line.
This reduces the logic necessary when an update arrives at an FE (or P2P middleware mechanisms at an edge) and hence greatly increases throughput and decreases
latency.
Guideline 18: Support multiple QoS+ mechanisms for different runtime conditions, beyond just network resources, in a uniied, coherent way. A given mechanism
* We note that such postevent historical data can be delivered by the same physical network links as the
fast trafic with trafic isolation mechanisms; indeed, this is one of the main trafic categories for the
emerging NASPInet.
70
Smart Grids
that provides guarantees of latency and security, for example, will not be appropriate
for all the runtime operating conditions in which a long-lived WAMS-DD may have
to operate. This is because different implementations of a given QoS+ mechanism
can require very different amounts of lower-level resources such as a central processing unit (CPU), memory, and bandwidth [25]. Also, in a very big WAMS-DD, many
QoS+ mechanisms will not be present everywhere, or close, so spanning them in
a coherent manner and such that applications do not have to be aware of the above
limitations is important. This is discussed further in Section 4.5.
Guideline 19: Inspect only packet header, not payload. In order to provide the
highest throughput and lowest latency, ensure that subscription criteria and consistency semantics allow a forwarding decision to be based solely on a packet header.
This is not possible for pub-sub middleware that has complicated subscription topics,
as is typical of commercial and research systems. For them, data ields in the payload
also have to be inspected.
Guideline 20: Manage aperiodic trafic. Any trafic that is aperiodic (i.e., not
based on rate but on a condition) must be isolated from rate-based periodic trafic
and managed accordingly. This can be done deterministically, for example, with
(OSI Layer 1) optical wave division multiplexing (OWDM) hardware. Further, aperiodic trafic should be aggregated intelligently—for example, based on updateable
policies rather than hard-coded settings—instead of sending all alarms/alerts to the
next level up for processing.
4.2.3.3 Analysis
It is important to recognize that you cannot have the highest level of all the properties described in the DRs for every sensor variable. As noted in [1]:
1. Different properties inherently must be traded off against others.
2. Different mechanisms for a given property are appropriate for only some of
the runtime operating conditions that an application may encounter (especially a long-lived one).
3. Different mechanisms for the same nonfunctional property can have different trade-offs of lower-level resources (CPU, bandwidth, storage).
4. Mechanisms most often cannot be combined in arbitrary ways.
Even if you somehow could have them all at once, it would be prohibitively
expensive. Given these realities, and the fact that application programmers are rarely
experts in dealing with the above issues, middleware with QoS+ properties supported in a comprehensive and coherent way is a means to package up the handling
of these issues and allow reuse across application families, organizations, and even
industries.
Finally, because of length constraints it is not possible in this chapter to fully discuss
the cybersecurity issues that arise in a WAMS-DD. Clearly, a WAMS-DD providing universal connectivity creates cybersecurity challenges beyond those arising in a
conventional, single-utility supervisory control and data acquisition (SCADA) system.
Cybersecurity also interacts with DRs and IGs: for example, techniques used for message conidentiality and authentication must not impose too much additional latency,
GridStat
71
yet the multicast requirement appears to limit use of symmetric-key cryptography
for authentication. See Sections 4.5.5 and 4.7.7 for more information, and, of course,
the references cited therein.
4.3
ANALYSIS OF EXISTING TECHNOLOGIES FOR A WAMS-DD
We now analyze how existing technologies and standards meet the above DRs and
IGs. We review, in turn, traditional network layer technologies, middleware technologies, and electric sector technologies. This section ends with Table 4.2, which
very clearly depicts how almost all technologies are very inadequate for WAMS-DD,
especially when closed-loop applications must be supported (or at least not designed
out, either deliberately or inadvertently).
4.3.1 TECHNOLOGIES AND STANDARDS AT THE TRADITIONAL NETWORK LAYERS
Almost all WAMS-DD products and research to date have been drawn from the
“network” layers. We analyze representative technologies and categorize them
under subsections 4.3.1.1 and 4.3.1.2. Their limitations are clearly summarized in
Table 4.2.
It is important to note that a number of these IGs have the appropriate networklevel mechanism to implement the given IG, but they cannot do it without help from
higher layers: the middleware (or the application if middleware is not used) and also
sometimes application management. This is to help parameterize the mechanisms
with higher-level information outside the scope of the network level. This is discussed further in Section 4.5.
4.3.1.1 Very Inadequate
Traditional network protocols, including the open systems interconnection (OSI)-2
“data link” layer (e.g., Ethernet), OSI-3 “network” layer (e.g., IP), and the OSI-4
“transport” layer (e.g., transmission control protocol [TCP], user datagram protocol
[UDP], and stream control transmission protocol [SCTP]), do not provide E2E QoS+
guarantees or multicast [26,27]. This is because they are at lower networking layers
and E2E functionality is not their intended use. All of these lower-layer protocols
can be part of the E2E solution above which WAMS-DD sits. Nevertheless, some
systems do apply them in ways that are nearly E2E in scope, and therefore we now
examine these protocols and extensions to them to see how they would meet the
requirements and guidelines if they were implemented as the complete E2E solution. We note that a major limitation of network-level solutions is that, based on our
long investigations in the ield, there is no other application domain that has DRs
anywhere near as stringent as those of the bulk power grid.
IPv6 low labels [28] associate each “reservation” with an application-to-application
network socket connection, which contains many different sensor-update streams
with a wide range of required QoS+. Packets are processed in a low-speciic manner by the nodes that have been set up with a low-speciic state. The nature of the
speciic treatment and the methods for the low state establishment are out of scope
of the speciication.
72
TABLE 4.2
Coverage of DRs and IG by Existing Technologies
Network Level
Very Inadequate
Better Yet Incomplete
Middleware
Electric Sector
TCP,
IPv6
VLANs
BB
P2P
UDP, Flow
IP
and
Open
Int-Serv
Net
COTS
COTS Streaming Military
SCTP Labels Multicast MPLS VPNs ATM Flow Guara. Svcs Anagran Insight SeDAX Pub-Sub Pub-Sub SQL/CEP RT Apps
SOA/Web
Services
IEC
61850,
OPC
UA
DNP3,
MMS
Grid
Stat
?
—
—
—
Y
?
Y
—
Y
—
D
D
—
D
—
—
—
Y
—
—
S
?
—
Y
—
Y
—
?
—
—
S
Y
—
—
—
Y
S
Y
S
Y
Y
Y
Y
Y
D
D
Y
Y
Y
S
—
—
—
—
Y
Y
—
—
S
—
—
—
—
—
S
Y
—
S
?
—
?
Y
—
—
S
D
φ
—
φ
φ
D
—
S
Y
Y
S
—
—
Y
Y
Y
Y
Y
Y
Y
—
—
—
Y
Y
Y
—
—
S
—
—
—
—
—
—
Y
Y
Y
—
—
—
—
—
Y
—
—
—
Y
—
D
D
—
D
—
—
—
Y
φ
φ
Y
—
φ
φ
Y
φ
φ
φ
Y
—
φ
φ
Y
—
φ
φ
?
—
?
—
Y
Y
D
D
Y
Y
—
—
Y
Y
?
?
Y
Y
S
Y
Y
Y
Y
?
Y
—
D
?
Y
?
—
?
Y
—
—
D
Y
D
—
S
Y
?
S
—
—
—
—
—
—
—
—
—
—
—
Y
S
Y
Y
—
—
—
—
—
—
—
—
—
S
—
?
?
D
?
—
—
—
Y
φ
—
—
D
—
—
S
—
—
S
—
?
?
?
?
D
—
—
Y
φ
φ
S
?
—
—
S
S
S
S
—
?
D
D
?
—
—
—
Y
—
—
—
—
—
—
—
—
—
—
—
—
—
D
—
—
—
—
Y
DR1: Hard E2E WAN
guarantees
DR2: Future prooing
DR3: One-to-many
(pub-sub or multicast)
DR4: Wide range of QoS+:
4A: Latency and rate
4B: Criticality/availability
4C: Cybersecurity
DR5: Ultralow latencies:
5A: Per-message and
predictable
5B: Tolerating failures
5C: Tolerating cyberattacks
DR6: High throughput
IG1: Avoid posterror
recovery mechanisms
IG2: Optimize for
rate-based sensors
IG3: Provide per-subscriber
QoS+
IG4: Provide rate-eficient
multicast
IG5: Provide synchronizedrate downsampling
Smart Grids
—
Delivery Requirement or
Implementation
Guideline
φ
—
—
—
—
—
—
Y
φ
?
?
D
?
—
—
—
Y
—
—
—
—
—
—
S
φ
φ
S
S
Y
Y
S
Y
Y
—
—
Y
—
?
—
?
—
—
—
—
—
S
—
?
?
L
?
—
—
—
Y
—
φ
—
?
—
—
—
—
?
—
?
?
?
D
D
—
—
—
S
—
—
—
—
—
—
—
—
—
—
—
?
?
D
?
—
—
—
Y
—
—
—
—
—
—
S
S
S
Y
—
?
?
D
?
—
—
—
Y
—
—
—
—
—
φ
S
φ
φ
—
—
—
D
—
—
—
—
Y
—
?
—
—
—
S
S
S
S
—
?
?
D
?
S
—
—
Y
—
Y
—
—
—
—
φ
φ
φ
φ
—
D
D
—
D
—
—
—
Y
—
Y
Y
—
—
—
Y
Y
Y
Y
—
S
S
D
D
—
—
—
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
?
D
?
S
—
—
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
—
?
?
—
?
—
—
—
Y
—
—
—
—
—
—
S
—
—
—
—
?
?
—
?
—
—
—
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
—
—
—
—
—
—
—
Y
—
—
—
—
—
—
—
—
—
—
—
D
D
—
D
—
—
—
F
IG6: Do not count on
priority “guarantees”
IG7: Provide E2E
interoperability across
different/ new IT
technologies (multicast,
QoS+)
IG8: Exploit a priori
knowledge of trafic
IG9: Have systematic,
quick internal instrument
IG10: Exploit smaller scale
of the WAMS-DD
IG11: Use static, not
dynamic, routing
IG12: Enforce complete
perimeter control
IG13: Reject unauthorized
packets quickly and
locally
IG14: Provide only simple
subscription criteria
IG15: Support transient, not
persistent, delivery
IG16: Do not provide
unnecessary consistency
IG17: Minimize
forwarding-time logic
IG18: Support multiple
QoS+ mechanisms for
different operating
conditions
IG19: Inspect only packet
header, not payload
IG20: Manage aperiodic
trafic
GridStat
—
φ
Note: ‘—’ means the requirement is not met or the guideline not followed. Also note how few ‘Y’ entries there are in almost all columns.
73
74
Smart Grids
IPMC provides eficient multicast for a single, nonreplicated low. However, if
multiple IPMC addresses are used as a replication mechanism to attempt to provide
redundant path delivery, there is no guarantee that the corresponding multicast trees
will be disjoint, which is important not only for eficient multicast (IG4) but also
for providing low latencies in the face of failures (DR5B). Further, IPMC addresses
must be used carefully in even a mid-sized infrastructure; for example, due to the
instabilities IPMC can cause, it has been banned from many cloud computing centers
[29–31]. Finally, IPMC also does not, by itself, have other E2E capabilities that are
necessary for a WAMS-DD, as clearly shown in Table 4.2.
MPLS is designed to give Internet service providers (ISPs) a set of management
tools for bandwidth provisioning, not to provide ine-grained (per-update) QoS [32].
Its guarantees are weak compared with the needs of a critical infrastructure. For
example, it gives aggregate economic guarantees over user, location, and protocol,
not hard guarantees (DR1) for each update (DR5A). As noted in [33], “DiffServ
does not guarantee that lows classiied with a higher priority will really observe a
better quality of service than lower priority ones.” Given that MPLS is a variant of
DiffServ, this is directly applicable. It is worth noting that in the relatively tranquil
steady state MPLS may give extremely good latency and availability. However, in
power contingencies, IT failures, cyberattacks, or other situations that place stress on
the WAMS-DD, the guarantees are nonexistent.
Further, different ISPs can implement MPLS in different ways. There are no facilities for combining lows across different ISPs, as would be required in a WAMS-DD,
or for predicting the E2E delays.
MPLS also has severe limitations with respect to the granularity of its “guarantees” and adaptation. The MPLS header only has three bits for the class ield, so each
low is assigned one class of only eight. This is a far cry from providing guarantees,
management, adaptation, and so forth on a per-subscriber-per-publisher-each-update
basis with hard guarantees. This is why MPLS meets few of the DRs and IGs, as
shown in Table 4.2.
MPLS has some fault-tolerance mechanisms, such as a fast reroute feature, detour
merging, and E2E path protection. However, these mechanisms presently provide a
minimum latency of about 50 ms, which is too long for the emerging SIPS and other
applications described in Chapter 3. Finally, MPLS is a closed system; OpenFlow
can match its performance while having an open and extensible system (though, as
shown below, one at present far from adequate for closed-loop applications) [34].
Virtual local area networks (VLANs) and virtual private networks (VPNs) do not
meet the DRs listed above because their purposes are orthogonal to the DRs. A VPN
or VLAN could be part of a WAMS-DD, but VPN and VLAN technologies alone do
not meet the requirements and can add to latency and decrease throughput.
Pragmatic general multicast (PGM) is a transport-layer multicast protocol [35].
Implementation by Microsoft is known as reliably delivered messages (RDM). PGM
runs over a datagram multicast protocol such as IPMC, to provide basic reliable
delivery by use of NACKs. PGM uses a rate-based transmission strategy to constrain the bandwidth consumed. However, it does not provide real-time guarantees.
For brevity, it is not summarized in Table 4.2, but its coverage of the DRs and IGs
should be clear from the technologies that it depends upon, which are covered there.
GridStat
75
Spread can be considered a high-level multicast protocol that provides a range
of ordering strengths across a wide area network (WAN) [36]. It supports ordered
delivery and the resulting consistency, even in the face of network partitions, and it
is used largely for replicating databases. It has no real-time mechanisms.
4.3.1.2 Better Yet Incomplete
Asynchronous transfer mode (ATM) and synchronous optical networking (SONET)
are networking technologies sometimes employed in WANs. They offer strong latency
guarantees on a per-message basis. ATM does not support multicast (DR3) or multiple disjoint paths (DR4B). Given ATM’s strong latency guarantees, however, at the
right granularity, the ATM protocol can be part of a WAMS-DD that overlays ATM
and other protocols. Of course, such an overlay network would by deinition be middleware, which is needed for WAMS-DD for other reasons discussed in Section 4.5.
OpenFlow is a network communications protocol that gives remote access to the
forwarding plane of a network switch or router [37,38]. It does this by abstracting
common features of routers and switches to provide constructs such as low tables
and an instruction set that allows remote access to those constructs. Many vendors
have implemented OpenFlow on their products with varying degrees of completeness and consistency [39]. Finally, OpenFlow can implement all the features of
MPLS (a closed system) and with similar performance in power settings [33].
One fundamental limitation of OpenFlow is that it is ipso facto a lowest-commondenominator (LCD) approach, having to be mapped to a wide range of devices. Very
few of these have any mission-critical requirements, let alone anything close to their
LCD. Worse, OpenFlow has no notion of global time. While it is possible to provide
some degree of predictable performance by very carefully managing lows, providing strong guarantees with tight deadlines and low jitter under dificult conditions is
problematic at best. Finally, OpenFlow has no primitives that can do anything like
rate downsampling even on a single low (i.e., IG3 and IG4), let alone synchronized
with another low (IG5).
The standard Internet integrated services (int-serv) has a guaranteed service (GS)
that strives to provide strong delivery guarantees. It does provide hard E2E guarantees (DR1). However, it guarantees a maximum latency, not an average, so it has to
be very conservative in what it promises. Further, it is not designed to provide low
jitter, and its delay can be quite signiicant compared with closed-loop requirements
(it can easily be far longer than these requirements). Still, while it does not meet most
of the DRs and IGs, using int-serv GS is signiicantly better than using MPLS.
The signaling mechanism of int-serv, resource reservation protocol (RSVP), does
perform admission control. However, it is done in the router, not in the application’s
address space (as is simple to do with middleware proxies, which are in the application’s address space), so spurious trafic can still slow down routers. Also, using this
admission control with a ine enough granularity, that is, to ensure that each publication from a given publisher applications does not go over rate, would be cumbersome
with RSVP and take a large number of ports, potentially exhausting the 16-bit space
of UDP ports.
Anagran technology [23], which is based on earlier Caspian technology, has lowbased routers that keep per-low state to provide better QoS via low management.
76
Smart Grids
It has not historically released much information on its router. The survey [34] notes
that Anagran technology provides only weak statistical guarantees, not strong ones as
required for closed-loop applications. It also has no per-sensor, per-subscriber granularity, or anything close: its lows use a 5-tuple in IPv4 (source and destination address and
port, protocol) or an IPv6 low label (see above) if available. This is the opposite of what
is needed in critical infrastructures: to be unfair to provide strong guarantees! Further,
Anagran technology automatically classiies lows into a number of classes, rather than
being provided with the QoS+ requirements for each publisher and subscriber.
Anagran’s corporate life cycle may well be a lesson for the realities of inding
QoS+ mechanisms from other industries that are appropriate for closed-loop applications: in the broader market, QoS+ mechanisms stronger than MPLS but weaker
than what was needed for closed-loop applications did not ind a market big enough
to sustain Anagran. It is worth noting that the company had a lot of deep technical
expertise and credibility, given that it was founded by Larry Roberts, the person who
envisioned and commissioned the Advanced Research Projects Agency Network
(ARPANET) in the late 1960s, and it was very well funded by investors.
Net Insight’s Nimbra platform [40] comes much closer than any other network-level
technology to meeting the DRs and IGs that closed-loop applications mandate, and is
actually the one counterexample we are aware of regarding the negative lessons from
Anagran’s demise. Net Insight’s primary market has been broadcast television backhaul and distribution networks, but its technology is very applicable to closed-loop
applications: Net Insight uses tight time synchronization (without requiring GPS) to
provide both very strong latency guarantees and stronger security, even with multicast
delivery. Nimbra also has wrappers for some network technologies such as MPLS,
SONET, and so on, and presumably it also will for future networking technologies.
4.3.2 MIDDLEWARE TECHNOLOGIES AND STANDARDS
There is a wide range of commercial, off-the-shelf (COTS) middleware frameworks
providing different kinds of services with some relevance for a WAMS-DD. We irst
consider middleware supporting the pub-sub paradigm. There are two distinct architectures for pub-sub middleware, each with advantages and disadvantages.
4.3.2.1 Deinition of Middleware
Middleware is a layer of software above the network but below the application that
provides a common programming abstraction across a distributed system (computers connected by networks and communicating only by message passing, or constructs built on top of the system) [41,42]. Middleware exists to help manage the
complexity and heterogeneous mix of resources that are inherent in distributed
systems. Middleware provides higher-level building blocks (these abstractions),
which, compared with socket programming, can make code more portable, make
the programmers more productive, and have fewer errors due to handling the lowerlevel communications details. Indeed, comparing programming with middleware
with socket programming is much like comparing programming with a high-level
language such as C# or Java with assembler programming: relatively few people do
it any longer, and for very similar reasons.
GridStat
77
Middleware masks the kinds of heterogeneity that are inherent in any wide-area
distributed computing system in order to enhance interoperability and make it much
simpler to program. All middleware masks heterogeneity in network technology and
in CPU architecture. Almost all middleware masks heterogeneity across operating systems and programming languages. A notable exception to the latter is Java’s remote
method invocation (RMI), which is the remote invocation system for Java (note that the
Java language is not middleware; RMI is). Java RMI does this so it can exploit features
in the Java programming language such as serialization and relection. A programmer can still use other middleware with Java, for example, the common object request
broker architecture (CORBA) object-oriented middleware or the pub-sub data distribution service (DDS), both from the Object Management Group (OMG). Finally, some
middleware does a very good job of masking heterogeneity across implementations of
the same standard from different vendors; the OMG in particular does this well.
Middleware can also shield programmers from many (but not all) of the dificulties inherent in programming a distributed computing system. For example, it can
largely hide from the programmer the exact location of an object, the fact that an
object may be replicated or may relocate, and so on.
It is for these reasons, and more, that middleware has been considered “best practice”
for 15–20 years in virtually any other industry that has applications spread across a
network. It is our hope that it will become much more prevalent in the electricity sector.
4.3.2.2 Broker-Based Publish-Subscribe Middleware
Broker-based (BB) pub-sub middleware systems rely on an infrastructure of broker
nodes to forward messages toward subscribers. The data delivery plane (DDP) for
a WAMS-DD, though not necessarily in many commercial systems, is a managed
WAN because it implements IG12 (complete perimeter control).
A BB pub-sub WAMS-DD is depicted in Figure 4.2. A node in the DDP is called an
FE and is a device specialized for the particular BB pub-sub framework. We depict the
mechanisms that a BB WAMS-DD system can exploit in gray; these consist of the FEs
and the proxies in the same process as the applications (publishers and subscribers).
In BB WAMS-DD systems intended for mission-critical applications, there is often
a separate “plane” for managing the system* and providing services. This data management plane (DMP) is depicted by the dashed gray lines in Figure 4.2; it is shown
here as a single entity, but it is often distributed. Publishers provide the DMP with basic
QoS metadata (QMD) about their publications, for example, the rate at which they
will output updates. Subscribers provide QoS requirements (QR), including rate and
latency. The DMP then exerts control over the DDP (gray lines) in order to provide the
delivery guarantees, for example, by updating a forwarding table for an FE.
BB pub-sub systems require a broker/server infrastructure to be installed; you
cannot just buy an IP router from Cisco or others. This can be a disadvantage, which
often, for small and medium scales, cannot be amortized over enough applications to
be justiied. BB pub-sub systems have an advantage, however, in that they place intelligence inside the network, not just at the edges. This enables, for example, eficient
* In telecommunications parlance, this is often called the “control plane,” hence our use of the term
plane. Telecommunications nomenclature also refers to WAMS-DD as the “data plane.”
78
Smart Grids
Data management plane and services
QoS metadata (QMD)
X
Y
QoS requirements (QR)
Data-delivery plane
(managed WAN)
Pub1
Sensor app
or firmware
Pub-Prx-Mech proxy
X, Y updates
Sub1
Power app
FE
FE
Y
W
Proxy Pub-Prx-Mech
Y, W updates
FE
Z W
PubN
Sensor app
or firmware
FE
FE
Pub-Prx-Mech proxy
Z, W updates
SubN
FE
Power app
Z
X
Proxy Pub-Prx-Mech
Z, X updates
FE
FIGURE 4.2 BB WAMS-DD.
multicast (IG4) and rate downsampling throughout the data delivery system, not just
at the edges. It also creates the potential to reject unauthorized packets at their next
“hop” through the system (IG13).
Additionally, BB systems can exploit mechanisms in the graph of FEs in order to
meet more of the IGs. For example, such an FE can be used to provide per-subscriber
QoS+ (IG3), provide synchronized rate downsampling (IG5), exploit a priori knowledge of trafic (IG8), and exploit the smaller scale of the WAMS-DD (IG10); that is,
it can contain per-subscriber state, a forwarding table entry for every subscription for
which it forwards updates. An example of a BB WAMS-DD is GridStat.
4.3.2.3 Peer-to-Peer Publish-Subscribe
P2P pub-sub systems place mechanisms for reliability and iltering only at the edges
of an infrastructure. A canonical architecture for a P2P pub-sub coniguration of
a WAMS-DD is given in Figure 4.3. For the DDP, P2P systems typically rely on a
Data-delivery plane
(LAN/Intranet/WAN)
Pub1
X
Y
Pub-Prx-Mech proxy
X, Y updates
Z
W
Sub1
Sensor app
or firmware
PubN
Sensor app
or firmware
Pub-Prx-Mech proxy
Z, W updates
FIGURE 4.3 P2P WAMS-DD.
Power app
IP
IP
Y
W
Proxy Pub-Prx-Mech
Y, W updates
IP
IP
IP
SubN
IP
IP
Power app
Z
X
Proxy Pub-Prx-Mech
Z, X updates
GridStat
79
combination of IPMC and Ethernet broadcast to be as eficient as possible. Note that
in Figure 4.3 we omit the DMP, which is often not present as a separate core entity
in P2P systems; the edge mechanisms collectively implement it.
One other thing to note in Figure 4.3 is that the controllable mechanisms for affecting trafic lie at the edges, in the proxies. Certainly a P2P WAMS-DD will exploit
IPMC as much as possible, but this has its limits, as described previously. Because its
control mechanisms are at the edges, both QMD and QR are communicated to other
proxies that collectively provide the delivery guarantees. Similarly, the only WAMSDD-speciic mechanisms are in the proxies, so control messages also must go there.
In Figure 4.3, to help aid understanding, the dashed gray and the gray trafic lines are
omitted. In practice, such trafic lines would be delivered via the DDP using IP routers.
P2P pub-sub systems have an advantage in smaller- and medium-sized deployments, but, for larger scales, the lack of mechanisms in the backbone core for rate
downsampling and quicker fault tolerance, adaptation, and instrumentation limit
their abilities to achieve extremely low latencies in the presence of failures.
A federated combination of P2P and BB pub-sub systems has the potential to
offer much of the best of both worlds. Here, P2P pub-sub systems are employed near
the edges, that is, within a single utility or sometimes within an ISO. Between utilities or ISOs, BB pub-sub systems are used in order to support higher throughputs
and the lowest possible latencies over distance. A federated amalgamation of P2P
systems would feature a globally unique namespace for variables and utilities, and
could seamlessly pass messages with standardized wire and message formats [1].
4.3.2.4 Other Middleware
SEcure Data-centric Application eXtensible (SeDAX) is low-level middleware that
provides both pub-sub and querying capabilities [43]. It is implemented with an overlay network on top of TCP/IP that is organized into a Delaunay Triangle network.
SeDAX mainly targets the distribution side of the power grid, and as such aims to
handle many more hosts than are present in a mission-critical WAMS-DD. Similarly,
its availability requirements for lows are only 99.5%, which is far from the six nines
required by DR4b (which comes from the Electric Power Research Institute in [4]).
Finally, it does not provide subsecond latency, let alone strong latency guarantees. It
is thus not appropriate for WAMS-DD.
Another category is called streaming queries (also known as complex event processing). It consists of a network of computer nodes that manipulate data streams
through continuous queries in order to selectively propagate data, merge streams
with existing data, or store data in a distributed database. Such systems are not
designed to provide hard E2E WAN guarantees (DR1) with per-message granularity
(DR5A) while tolerating failures (DR5B). Given their intended application domain,
they also do not follow most of the IGs.
Finally, over the last decade or so, crude middleware based on web technologies
such as hypertext transfer protocol (http), extensible markup language (XML), and
“web services” has become popular in other industries, in large part because it is
easy to get around irewalls with it. We note that scalability and throughput of such
systems are highly questionable due to the many integration layers they typically add
to make it possible to glue just about any application to another [44]. Ken Birman,
80
Smart Grids
a leading expert in reliable distributed computing whose software has been widely
ielded, notes in [45] (emphasis ours):
It doesn’t take an oracle to see that the mania for web services, combined with such
rampant online threats, contains the seeds of a future debacle. We’re poised to put airtrafic control, banking, military command and control, electronic medical records,
and other vital systems into the hands of a profoundly insecure, untrustworthy platform cobbled together from complex legacy software components.
Unfortunately, one can add the smart grid to this list: a number of utilities and
organizations see web services as a key enabling technology for the smart grid (e.g.,
[12,46]). This is unfortunate because it inadvertently rules out building an extremely
low-latency, extremely high-availability, extremely high-throughput WAMS-DD
suitable for the more challenging closed-loop applications.
This is generally done because power engineers, or IT staff at utilities (which tend
to be understaffed), buy into the hype. But is it really wise to have the operations of
increasingly stressed power grids depend on a technology whose raison d’être is to
circumvent irewalls? We think not. Caveat emptor!
4.3.3 EXISTING ELECTRIC SECTOR COMMUNICATIONSRELATED TECHNOLOGIES AND STANDARDS
Middleware is rarely used in today’s electric power systems, despite being considered best
practice in many other industries for 15–20 years [1]. Further, the extreme conditions for
closed-loop applications are far more stringent than in any other industry. Additionally,
it is an unfortunate fact that almost all of the communications standards in the electric
sector have been developed by power engineers or others who are not expert at building
communications protocols, are unaware of the state of the art and practice, and so on.
It is not surprising, then, that there seem to be no networking technologies developed
for the power grid that meet all of the DRs above. Part of this limitation is because
commonly used power technologies are intended for a substation scope, with the QoS+
“mechanism” being overprovisioning of bandwidth. When moving from a LAN to a
WAN environment, there are implicit design decisions that cannot be solved by layering a
new “WAN-appropriate” API over existing LAN-based protocols [25]. We now overview
some of the more common power protocols and standards related to communications.
Object linking and embedding for process control uniied architecture (OPC UA)
[47] was designed for substations. It uses TCP, which was not designed for predictable latency and does not support multicast. Subscribers and publishers “ping” each
other to verify whether the other is up, which not only does not scale but also ignores
best practices for pub-sub systems.
Institute of Electrical and Electronics Engineers (IEEE) C37.118 is a standard for
synchrophasors that includes standard message formats. C37.118 is being revised
to allow different data delivery mechanisms to be used. If successful, then C37.118
synchrophasor updates should easily be deliverable by any WAMS-DD transport.
The IEC 61850 communications standard was also designed primarily for applications associated with a substation automation system (SAS). It was conceived and
GridStat
81
created by protection providers in order to move data and information to, from,
and among intelligent protection, control, and monitoring devices instead of legacy
SCADA and remote terminal unit (RTU) methods.
IEC 61850’s companion common information model (CIM), IEC 61970 Part 3,
can potentially be of use in a WAMS-DD, especially when the harmonization with
C37.118 is completed, in particular in helping automate default QoS+ settings and
perhaps adaptation strategies for a wide variety of sensors and applications that use
them. This would be signiicant value added. Further, if IEC 61850 manufacturing message speciication (MMS) and generic object-oriented substation events
(GOOSE) APIs are successfully extended across the WAN, then IEC 61850 may well
be able to successfully use a WAMS-DD transport. This would be of huge beneit to
the many utilities that have a lot invested in IEC 61850 applications.
However, this will only be true if the WAMS-DD transport is carefully designed
to meet the DRs in Section 4.2.2. Unfortunately, this seems doubtful for a number of
reasons, including the following:
• Presently, IEC 61850-90-5 discusses delays of 50–500 ms, which completely rules out its use with closed-loop apps outside a single substation
and also SIPS/remedial action schemes (RAS)/special protection systems
(SPS), including those from Chapter 3.
• Overall, the standard seems to many observers with a computer science or
software engineering background far more complex than it has to be, given
the problem it is tackling. Indeed, to the authors it seems more like something that a mechanical engineer in 1975 would have written, not a software
engineer from 1995 (let alone today).
• Vendors do not, perhaps without exception, implement the full standard.
That makes interoperability between vendors very problematic.
• There is no reference implementation and intervendor test suite, which
makes true cross-vendor interoperability almost impossible because vendors have to test implementations pairwise, which is in turn problematic
because, as noted above, few if any implement the full standard.
• There are no tools for coniguring and otherwise using the standard that work
well (or even fairly completely) across different vendors’ implementations.
This, of course, is totally contrary to the goals (and hype) of the standard.
• It takes twice the bandwidth to deliver synchrophasor data as C37.118 does,
with no extra useful information.
• Subscriber applications have to be able to detect missing and duplicate data.
It would be much better if middleware did this, so that application programmers (many of whom are power engineers, or at least are not very familiar
with distributed computing) did not have to do it themselves.
• GOOSE authentication originally mandated the use of RSA signatures.
However, they are too slow for closed-loop applications even with a highend PC, let alone a more modest embedded device in a substation (and one
that cannot devote all its time to RSA calculations!) [48,49]. There is no
silver bullet here: approaches that use a shared key for each publisher and its
subscribers are vulnerable to a subscriber spooing a publisher if they do not
82
Smart Grids
have other help from other mechanisms. The one solution we know of that
works fast enough is described in [50]. It only adds 10 ms of latency, and has
been integrated into GridStat, and thus is suitable for use with closed-loop
applications such as SIPS. This is discussed in Section 4.7.7.
• With 61850 in a substation, developers and deployers have to be very careful
that the multicast (which is done there via Ethernet multicast) does not overload small embedded substation devices (which do not even need to receive
most or almost all of the trafic). Over a WAN, that kind of spamming is
much more likely to cause bandwidth problems than CPU problems.
• It was designed by a committee with a “kitchen sink” approach before anyone
had done close to a full implementation, which is very problematic. Indeed,
Internet pioneer David Clark famously stated about the Internet’s development
and the Internet Engineering Task Force (IETF) that deines its protocols [51]:
We reject: kings, presidents, and voting.
We believe in: rough consensus and running code.
MMS also does not have data-delivery mechanisms. It can map onto the OSI protocol stack (which was not adopted in practice) and TCP/IP; see [26,27].
An information architecture for the power grid is proposed in [52], which contains
an analysis of 162 disturbances between 1979 and 1995. This paper demonstrates that
information systems have an impact on power grid reliability and points out major
deiciencies in the current communications scheme. The paper contains proposals for
different ways to structure interactions between control centers and substations, and
it also contains reliability analyses of different schemes. However, it does not propose
communication mechanisms and relies on off-the-shelf network technologies, which
do not meet many of the DRs and IGs. Sadly, this is common, almost universal, practice in communications mechanisms and analysis in grids today.
4.3.4
SUMMARY
The coverage of the DRs and IGs by various commercial and research technology is given
in Table 4.2. The columns of this table are existing networking and middleware technologies, while the rows are the DRs and IGs outlined previously in this section. The table cells
denote how well the given technology meets the given IG or DR. These have the following
values: Y, yes; —, no; S, some; L, likely (but not conirmed); ?, unknown (documentation
is not suficient); D, doubtful (but not conirmed, because the product has insuficient
documentation or it is a broad product category but not mission critical); F, future plans
(architected for this); and φ, not applicable (and thus, by deinition, does not provide).
In some cases, it is not possible to discern the exact coverage of a DR or an IG
due to lack of detail in a research paper, product documentation, and so on. However,
rather than leaving them unspeciied, if we believe we understand the technology’s
goals, environment, and so on enough to speculate, we grade them as “L” or “D”
rather than simply “?”. This is possible because it is highly unlikely that the developers of the mechanisms inadvertently overdesigned them to meet WAMS-DD requirements that the designers were not even aware of!
83
GridStat
4.4
NASPINET
The North American Synchrophasor Initiative (NASPI) is a government–industry
consortium dedicated to effective deployment of synchrophasors in the United
States. In this section, we describe its WAMS-DD initiative. Unfortunately, for a
number of reasons, including budgetary, regulatory oversight, and technical culture,
as of late 2013 there has been no deployment of anything resembling a full-features
NASPInet, that is, one implementing all the DRs (and ideally most or all IGs) in
Table 4.2. Nevertheless, it is an example worth studying.
4.4.1
ARCHITECTURE
The NASPI network (NASPInet) concept is the only effort worldwide we are aware
of that is dealing with E2E WAMS-DD issues at a more than supericial level.
To support the use of synchrophasors, NASPI has been developing the notion of
NASPInet, which has two main components: the data bus (NnDB) and the phasor
gateway (NnPG).
The NnDB is the electricity version of what is sometimes called an enterprise service bus (ESB), which provides communications services for business-to-business
exchanges. NnDB satisies the DRs described earlier in this paper if it is implemented with appropriate technologies. The NnPG is the edge component of Nn,
interfacing the utility or ISO to the NnDB. This is depicted in Figure 4.4.
Note that, while NASPInet was originally conceived as being only between utilities (and ISOs/RTOs, of course), it is not limited to this. The same technologies
would work in a large utility that wanted to have much more control over its network
Monitoring center X
Utility A
Archive
Apps
PMU
Phasor
gateway
Apps
Utility C
PMU
PMU
Phasor
gateway
PMU
Historian
NASPlnet data bus
Utility B
Apps
Phasor
gateway
Historian
Monitoring
center Y
Phasor
gateway
Apps
PMU
PMU
PMU
PMU
FIGURE 4.4 NASPInet conceptual architecture. (Courtesy of NASPI.)
Archive
84
Smart Grids
and applications to provide WAMS-DD properties. This would be particularly
important for closed-loop applications.
4.4.2
SERVICE CLASSES
Five initial service classes have been identiied for the NnDB in recognition of the
fact that different kinds of trafic with different DRs must be carried:
1.
2.
3.
4.
5.
Feedback control (e.g., small signal stability)
Feedforward control (e.g., enhancing state estimators with synchrophasors)
Postevent (postmortem event analysis)
Visualization (for operator visibility)
Research (testing or R&D)
Each class has associated qualitative requirements for such properties as low
latency, availability, accuracy, time alignment, high message rate, and path redundancy. This is shown in Table 4.3.
Distinguishing the classes in this way is an important irst step for a WAMS-DD
system. The performance and availability requirements in Chapter 3, and the DRs
and IGs in Section 4.2, can be considered a signiicant reinement of these NASPInet
trafic classes. However, we note that the NnDB classes also consider the lowest required latency to be 100 ms, which is demonstrably insuficient for some
applications.
One issue with the class deinitions is that a customer, such as a utility, an ISO,
an RTO, or the North American Electric Reliability Council (NERC), cannot specify only what it wants from a telecom provider, for example, a “Class A” network,
and be done with it. This will not necessarily result in a WAMS-DD that meets
TABLE 4.3
NASPInet Trafic Classes and Their Attributes
Real-Time Streaming Data
Historical Data
NASPInet Trafic
Attribute
Class A
Feedback
Control
Class B
Feedforward
Control
Class C
Visualization
Class D
Postevent
Low latency
Availability
Accuracy
Time alignment
High message rate
Path redundancy
4
4
4
4
4
4
3
2
2
4
2
4
2
1
1
2
2
2
1
3
4
1
4
1
Class E
Research
1
1
1
1
1
1
Source: Hu, Y. et al., Data bus technical speciications for North American synchrophasor initiative network (NASPInet), May 2009. https://www.naspi.org/File.aspx?ileID=587.
Note: 4, Critically important; 3, important; 2, somewhat important; 1, not very important.
GridStat
85
the requirements across multiple trafic classes. For example, if too much trafic
of “easier” classes is on the network, you will not get Class A guarantees. Rather,
to provide the DRs identiied above, one needs to perform resource management
within the data-delivery service. Application management and network management
components must account for all trafic associated with each subscription using a
given level of QoS+. This is part of enforcing a number of IGs, including IG8 (exploit
trafic knowledge), IG9 (systematic, quick, internal instrumentation), IG12 (complete
perimeter control), IG13 (reject unauthorized packets quickly and locally), and IG20
(manage aperiodic trafic). Further, implementing this requires middleware, as we
discuss next.
Finally, we note that, despite the US government spending more than $4 billion of
federal iscal stimulus money on the power grid over the last 5 years, there has been no
signiicant progress in deploying a full-featured NASPInet. There are a number of reasons for this. First of all, advanced communications are something that organizations
are not bureaucratically set up to deal with. Second, power engineers and researchers almost always assume that communications are good enough. Third, through the
imperfect and power-centric bureaucratic process, through no fault of its own the
company hired to write a NASPInet speciication, while excellent at power issues, had
zero experience with anything but the old hard-coded grid communications; nobody
working for them has even a BS in computer science, or even an undergraduate minor,
despite the fact that WAMS-DD is very advanced computer science.
Further, even the most ambitious communications upgrade as part of this investment is not only not using pub-sub middleware, it is not even using IPMC. This was
because the developers had so many problems getting vendor software running with
200 PMUs, rather than the 10 PMUs it was used to, that they had no time to signiicantly modify the communications. They know that with just point-to-point communications they will not scale to many more PMUs. Further, they are only using
MPLS to provide QoS. As noted above, it is very deicient for WAMS-DD, and they
are de facto ruling out any closed-loop applications.
As demonstrated in this chapter, a full-featured WAMS-DD is very possible, but
it has to be done by computer scientists leveraging the state of the art in real-time
distributed computing, not by power researchers, or, for that matter, by software vendors repackaging an existing pub-sub infrastructure from other domains. However,
the pressure is building: grids are becoming more stressed each year for a host of
reasons that can be mitigated by a proper WAMS-DD.
4.5 WHY WAMS-DD MUST INCLUDE MIDDLEWARE,
NOT SOLELY NETWORK LAYERS
In this section, we explain how implementing a WAMS-DD requires the use of
middleware. In general, according to the stated goals of the smart grid community,
middleware is required [1] (this paper won a best paper award for demonstrating this
proposition). Further, it is simply not possible to implement a WAMS-DD without
information provided by middleware, sometimes augmented by application monitoring. In other cases, although it is possible to build some features and properties
without middleware, with the help of middleware the lower-level mechanisms can
86
Smart Grids
do a much better job. In general, the network level only knows about packets and IP
addresses, not middleware-layer application variables (e.g., published sensor variable updates). It also knows nothing about the power applications that subscribe to
these updates, and their importance in different situations. For example, GridStat’s
systematic adaptation described in Section 4.7.5 cannot possibly be implemented
without help above the network layers.
In this section, we consider why middleware in general is required for WAMS-DD.
We next describe how it is needed when QoS+ is required, as, of course, it is in any
WAMS-DD. We next discuss how middleware is needed to properly implement some
of the DRs and IGs. Finally, we discuss how middleware is very helpful in integrating legacy systems.
4.5.1
MIDDLEWARE IN GENERAL
Middleware is required to accrue the beneits overviewed in Section 4.3.2.
Additionally, in practice, in any system of moderate size and complexity there will
be products from different vendors. While they may implement the same protocols,
their coniguration and management may well be different. For example, even router
vendors will admit that the way you conigure IPMC is different across vendors.
Thus, if for no other reason, most such systems will have at least a thin layer above
the network for applications to use.
A utility, therefore, has three choices regarding this:
1. Have each application programmer recreate its own thin layer (these are
sometimes called “application-level protocols” by network researchers)
2. “Roll your own” middleware in-house
3. Use commercial middleware
Application programmers already have enough to deal with, so making them also
handle Option 1 is highly inadvisable (but common practice in the electricity sector,
regrettably). Rolling your own middleware is certainly better than Option 1 because it
allows the implementation of the thin layer to be shared across application programs.
However, utilities may not have sophisticated network programmers to do this, and
if they do, most likely their time could be better spent on other tasks. This leaves us
with Option 3: commercial middleware. Here experts in networking, middleware,
QoS+ mechanisms, interoperability, and so on create middleware that is reused over
many companies and projects. And, as noted in Section 4.3.2, programmers then are
more productive and the code they write is more portable and manageable.
4.5.2
MIDDLEWARE AND QOS+
When QoS+ properties are required, as of course they are in a WAMS-DD, it is
even more important to use middleware. Without it, application programmers have
to combine lower-level QoS+ mechanisms in arbitrary ways. For example, how is an
application programmer going to know how to combine network QoS from MPLS or
int-serv GSs with Internet protocol security and nonrepudiation? These mechanisms
GridStat
87
were designed independently of each other and interactions between them can be
very subtle. It is thus a really bad idea to assume that application programmers can
combine them in reasonable ways, knowing any inherent trade-offs or limitations of
their combination.
By far the better way to provide QoS+ properties in a WAMS-DD is for a middleware developer to combine multiple mechanisms, evaluate and measure their combination, document any trade-offs and limitations the application programmer should
be aware of, and so on.
In addition to providing QoS+ properties in the steady state, there are additional
reasons why WAMS-DD requires middleware: coordinated adaptation driven from
the top down. In the face of failures, cyberattacks, or other runtime IT problems, it
is extremely undesirable to have different mechanisms adapting independently of
each other and without any higher-level guidance. For example, having ACK and
NACK messages clog the WAMS-DD is a really bad idea. It is much better to have
coordinated, E2E adaptation driven by knowledge of the applications via knowledge
captured in applications’ middleware.
4.5.3 MIDDLEWARE AND DRS AND IGS
As mentioned previously, a number of the DRs and IGs cannot be implemented properly, or even at all, without the help of middleware. These include the following:
DR2 (future prooing): Applications that use middleware can exploit new and
better QoS+ mechanisms as they become available, because the applications are not locked into lower-level speciic QoS+ mechanisms. The applications do not need to be reprogrammed or recompiled, just to have a new
version of their middleware library linked in.
DR5B (ultralow latency tolerating benign failures): Providing this high level
of fault tolerance without jeopardizing low-latency guarantees requires
knowledge of the applications using the WAMS-DD, not just socket-level
information on packets.
IG2 (optimize for rate-based sensors): In order to optimize for rate-based
sensor data delivery, it is necessary to know the rates that the different
applications require. QoS-enabled middleware APIs capture this, while
network-level mechanisms are generally unaware of application rates.
IG3 (provide per-subscriber QoS): Similar reasons to IG2 above.
IG4 (provide eficient multicast): Similar reasons to IG2 above.
IG7 (provide E2E interoperability across different or new IT technologies—
multicast, encryption, authentication, network-level QoS, etc.): Network-level
mechanisms do not cover the entire space here, and, with few exceptions,
cannot generally be composed with other kinds of network mechanisms providing the same property (e.g., latency) without a higher-level layer above
them both, which is, of course, by deinition middleware.
IG12 (enforce complete perimeter control): Network QoS mechanisms generally do not know much about the trafic delivered on them, including their
rates. Further, if overrate trafic comes to a router that knows the rate, it is
88
Smart Grids
still consuming resources on that router rejecting it. Middleware with QoS
APIs can, of course, capture rate and other QoS parameters. Further, middleware generally uses a proxy object that is returned when the application
connects to the middleware. The proxy object is then what the application
(publisher or subscriber) uses to send and receive sensor updates, respectively. This proxy object can perform rate-based policing to ensure that no
more than the guaranteed rate escapes the application and taxes the irst
hop across the network.
IG18 (support multiple QoS+ mechanisms for different operating conditions and conigurations): In a big WAMS-DD, no single mechanism for
providing a QoS+ property such as latency will be present everywhere
in the WAMS-DD. Sometimes the path from publisher to subscriber has
to traverse multiple QoS+ mechanisms, for example, MPLS and Nimbra.
Similarly to IG7, a layer is needed above the network layer to sensibly connect these different mechanisms providing the same property, and to know
the limits and trade-offs inherent in their combination.
IG20 (manage aperiodic trafic): Aperiodic trafic, also called event triggered
or condition based, must obviously be isolated from rate-based trafic (also
called time triggered or periodic). This can be done at the network layer.
However, it is also necessary to manage that trafic with knowledge of the
data and applications using the data. This is because in many circumstances
there may be more data than can it in the bandwidth allocated to them.
Fortunately, there is often a lot of redundancy in this trafic (e.g., a large
number of alerts/alarms sounding from a single root cause, or fairly unimportant messages that can be logged and ignored, at least for some time,
in some circumstances, or both). Carrying out this management requires
middleware to capture the semantics of the data and application. It also
requires higher-level management to intelligently aggregate the data using
policies or other high-level mechanisms, but this requires the information
that middleware provides.
4.5.4
MIDDLEWARE AND LEGACY SYSTEMS
Middleware is often used not to write complete systems from scratch, but rather to
integrate legacy systems that were not designed to interoperate with different kinds
of systems. It does this by adding a middleware wrapper, which surrounds the legacy
subsystem and maps down from a high-level middleware interface to the interface
speciic to a given subsystem. When middleware wraps a legacy system in such a
way, the APIs and also the management infrastructure can be written at a high level
consistently across the different legacy subsystems, because the middleware shields
them from the lower-level details of the individual legacy systems. This allows the
subsystems to interact in a very high LCD (the middleware layer and its high-level
abstractions), with strong type checking, and allowing the use of exception handling
mechanisms across different programming languages. This is of great beneit.
As an example, client–server-distributed object middleware such as CORBA is
often used to wrap legacy servers. In this way a client written in Java or Python can
89
GridStat
invoke a server object on a server written in C#, C, or FORTRAN, even though the
last two languages are not object oriented.
The previous example involved client–server interactions, but the technique of
wrapping legacy subsystems with middleware also applies to pub-sub infrastructures such as WAMS-DD. It can be highly beneicial to do so, because communications infrastructures for utilities can be a jumbled mess of communications
subsystems that were not written to work well together (not only due to the relative
backwardness of the industry on IT matters, but also because many of the bigger
vendors believe that intervendor interoperability can provide a path for them to lose
their lucrative vendor lock-in). Such wrapping is a necessity, not just to be able to
create a coherent WAMS-DD out of these legacy subsystems, but also because in the
future no one mechanism for a given QoS+ property (such as latency) is ever going to
“take over the world” and completely displace all others. Given the large investment
in the legacy systems, and the fact that (at least if utilized correctly) they can still
provide signiicant value, it will always be a case of trying to seamlessly interoperate
across multiple subsystems.
We give an example of this in Figure 4.5. Here an existing utility communications infrastructure includes a large number of legacy protocols such as MPLS,
microwave links, and others (the “others” could include broadband over power
lines/power line communications, satellite links, and even 4G cellular, for example, to remote substations where it is impractical to run iber or copper). This
existing infrastructure has been strengthened by programming some of the existing routers using OpenFlow and by adding stronger guarantees with subsystems
of GridStat FEs and Nimbra devices. This could in turn be managed as a coherent
whole by GridStat’s management plane. In this scenario, non-GridStat subsystems look to GridStat as a single link between two FEs. However, the management plane can exhibit control over them to strengthen QoS via a control wrapper
that has a high-level middleware control API, and the wrapper then maps down to
the subsystem’s speciic control APIs (the control wrappers and communications
Management plane
MPLS
Publisher 1
Publisher 2
C
o
m
m
o
n
M
W
Publisher N
A
P
I
Nimbra
MPLS
Subscriber 1
OpenFlow
Legacy microwave
Subscriber M
GridStat forwarding engines
= Cntl wrapper
Nimbra
= Data wrapper
FIGURE 4.5 Middleware integrating legacy and new communications subsystems.
90
Smart Grids
are shown in light gray). Finally, the data wrappers at the boundaries of the subsystem not only provide a common, high-level middleware API but could also be
used to carry out rate-based policing, similarly as described above, using middleware proxies.
Given such a high-level coherent WAMS-DD and a management plane such as
GridStat’s, the different subsystems can be treated as a uniied whole, and route
across different subsystems, taking into account not only the performance characteristics of the subsystem but also its reliability, and, of course, considering aggregate
bandwidth. Further, the subsystems in Figure 4.5 may have fairly different failure
characteristics and security vulnerabilities, which can greatly help the overall E2E
data availability from publisher to subscriber. The reader may be curious why old
microwave links, which have much lower bandwidth and often higher drop rates,
would even be included. However, they have very different failure characteristics
from most of the other subsystems, and this can add a lot of value, as long as their
maintenance is not becoming cost-prohibitive. These links do have much less capacity than the other technologies, so only the most important trafic must be sent over
them, perhaps at a lower rate. Of course, this requires middleware to capture these
semantics for the management system.
Indeed, it is crucial to note that, even if a WAMS-DD has no stronger QoS
mechanisms than MPLS or even simple IPv4, it will still need the management and
policing, wrappers, and so on that are shown on the edges below. If it does not have
these, things may work well in the calm steady state, but when things get bad the
predictability of the latency and other properties is extremely suspect. In short, when
you need it the most you can count on it the least.
Sadly, this is the situation at almost all utilities today. The only real QoS
“mechanism” is massive overprovisioning of bandwidth. This works quite well
most of the time, but when there are benign IT failures, let alone cyberattacks, the
WAMS-DD is likely to perform quite unacceptably. This in itself could be a major
vulnerability: terrorists can not only try to change actuator and relay settings to
cause problems in the grid, but they may also attack the WAMS-DD so that it is
much more dificult for utilities and ISOs/RTOs to understand what is going wrong
and ix it before a large blackout is caused. Indeed, such attackers would, of course,
attack when the grid is already severely stressed. The conditions under which a grid
will be stressed, and what power assets can cause most problems if changed, are
relatively easy for a determined adversary with a long-term view (e.g., Al Qaeda) to
ascertain long ahead of time.
4.5.5
MIDDLEWARE AND AVOIDING QOS STOVEPIPE SYSTEMS
Finally, we close this section with a system-level admonition that is crucial for the
long-term beneits that WAMS-DD can provide, if done right. A stovepipe system is
a legacy system that is comprised of different parts that are very dificult to combine
or refactor. They were identiied as major problems in big systems 20–30+ years
ago, and much software engineering and system architecting has been devoted to
avoiding them on many newer systems (at least in other industries that are not as far
behind in best practices as the electricity sector is).
GridStat
91
But, given that WAMS-DD has to provide QoS+, and the mission-critical nature
of its applications, stovepipe systems have another dimension. Bakken et al. [1] provide the following deinition:
QoS stovepipe system (QSS): A system of systems whose subsystems are locked into
low-level mechanisms for QoS and security such that
1. It cannot be deployed in many reasonable conigurations, or
2. Some programs cannot be combined because they use different lowerlevel QoS mechanisms for the same property (e.g., latency) that cannot be
directly composed, or
3. It cannot be upgraded to “ride the technology” curve as better low-level
QoS and security mechanisms become available
Of course, it is crucial that WAMS-DD installations avoid QSS, and middleware,
while no “silver bullet,” is a key enabling technology for this. It is for this reason
(and others from this section) that many military programs in the United States and
elsewhere require the use of QoS-enabled middleware; examples can be found in [1].
4.6 SECURITY AND TRUST FOR WAMS-DD
One of the United States Department of Homeland Security’s strategic plan objectives for the iscal years 2008–2013 has been to protect and strengthen the resilience
of the nation’s critical infrastructure and key resources. WAMS-DD, being a middleware that could be deployed in a critical infrastructure such as the power grid, must
provide support for key security services (authentication, conidentiality, integrity,
and availability) not only at the system level, but at the data level as well. There are
many points of interaction among a variety of participants, and a local change can
have immediate impact everywhere. Thus, to cope with the uncertainties associated
with security, there must be provision for a trust service that assesses the trustworthiness of the received data that come through nontrivial chains of processing. Trust,
when used properly, has the potential to enable the entities that operate within critical
infrastructures to share conidential, proprietary, and business-sensitive information
with reliable partners.
Even though WAMS is becoming one of the popular technologies for upgrading the
electric power grid, still there is no consensus on a security standard or even a set of
comprehensive security guidelines that will guarantee the secure generation, distribution, and consumption of data. It is a nontrivial and rather challenging task to provide
trustworthy and secure interactions in such operating environments, where private,
public, and national entities must collaborate in such a way that information is not compromised due to either malicious or accidental attacks. However, research has been
done on the security aspects of an envisioned power grid communication infrastructure that could be used as a basis to form the security requirements for WAMS-DD. It
is beyond the scope of this chapter to map the security spectrum for WAMS-DD.
One of the irst research efforts to deine the security, trust, and QR in a datadelivery middleware tailored for the power grid is described in [53]. Starting with
the application requirements for a variety of interactions (e.g., communication
92
Smart Grids
between substations and control centers, and communication between control centers), the authors proceed to specify the implications of the requirements for security,
trust, and QoS.
The work in [54] complements the aforementioned speciications by presenting a
conceptual lexible trust framework that models an entity’s trust as a relation whose
state is updated as relevant conditions that affect trust change. Trust is expressed
as a set of security, behavioral, and QoS properties. The novelty of the model is its
ability to integrate trust reasoning and monitoring into a single model by extending
the traditional concept of trust conditions into more expressive expectations, which
include not only expected values for particular properties but also covering, aggregating, and triggering mechanisms that manipulate the observed value. In addition,
and unlike other approaches, the model derives E2E trust assessments without using
transitive indirect trust explicitly in the derivations. A recent work [55] presents a
theoretical model based on Bayesian decision theory that automatically assesses the
trustworthiness of entities in the electric power grid.
One of the few published works on WAMS security is [56]. The authors interviewed the owners of two WAMS networks in order to identify the primary implementations of WAMS, and thus address security challenges in the light of each
implementation. The three WAMS implementations are an isolated WAMS network,
an isolated WAMS that interfaces with the control center network, and, inally, a
WAMS fully integrated into the control center network infrastructure. An analysis
of WAMS security concerns is presented, accompanied by a number of threat scenarios. The vulnerabilities identiied by the authors are addressed in terms of these
three implementation approaches because their impact differs in each case.
4.7
GRIDSTAT
GridStat is a data delivery service designed to support the DRs discussed in this
chapter. Its research results have inluenced the shape of NASPInet. GridStat
research started in 1999 by looking at the QoS+ requirements of innovative power
applications being developed by power researchers and analyzing closely what the
state of the art in applied distributed computing systems could support. After signiicant gaps were identiied, the detailed design and then development of the GridStat
framework began in 2001.
GridStat is a BB pub-sub system that meets all of the DRs from this chapter
except 5C: Tolerating Cyberattacks, which it does to a modest degree (by redundant
paths); covering this more completely is near-term future research. GridStat also
implements all but three of the IGs, which have similarly been planned for and are
also near-term future research.
In this section, we irst provide an overview of GridStat. Next, we describe
GridStat’s novel mechanisms that allow different subscribers to be delivered different QoS+, which is crucial for a WAMS-DD. We then describe condensation functions, a mechanism that allows arbitrary computations to be placed in a WAMS-DD
coniguration. After this, we describe GridStat’s mechanisms for rapidly adapting
its subscription base. Finally, we describe GridStat’s remote procedure call (RPC)
mechanism and then overview its security and trust infrastructure.
93
GridStat
4.7.1
OVERVIEW
4.7.1.1 Capabilities
The architecture of GridStat is given in Figure 4.6. The architecture is modular and
it consists of two modules, the data plane and the management plane. Each of the
modules is responsible for part of the overall functionality of the framework.
GridStat’s data plane is responsible for delivering the sensor data. A single sensor
datum is called a status variable, and can be subscribed to by multiple subscribers.
The data plane consists of publishers, which generate the sensor data, subscribers,
which consume the data, and a graph of FEs, which are specialized middleware-layer
routers (“brokers” in pub-sub parlance) that deliver the status variables.
GridStat’s management plane is responsible for managing the data plane, including making admission control decisions. It is organized into a hierarchy of QoS brokers (QB) that is designed to map directly onto the geographically based hierarchies
into which the power grid and other critical infrastructures are normally organized.
A leaf QB is the lowest level in this hierarchy and is responsible for, and directly connected to, a collection of FEs that are organized into a single administrative domain
called a cloud. Brokers can host policies for resource allowance (how much bandwidth an organization or app can receive), security permissions (potentially default
ones overridden in some cases), adaptation strategies, data aggregation, and specifying other coniguration and operational issues.
4.7.1.2 More Detail
GridStat’s data-delivery semantics are novel in three broad ways in order to better support WAMS-DD. First, they are a specialization of the pub-sub paradigm
Management plane
QoS broker
QoS broker
QoS broker
QoS broker
QoS broker
Data plane
FE
FE
FE
FE
FE
Subscriber
FE
FE
FE
FE
FE
FE
FE
FE
Publisher
FIGURE 4.6 GridStat architecture.
FE
FE
FE
FE
Publisher
Subscriber
Publisher
94
Smart Grids
tailored to leverage the semantics of periodic streams of data updates. For example,
in a normal pub-sub system, an event that is being forwarded from a publisher to
its subscribers cannot be arbitrarily dropped, because it is generally impossible to
know how it will affect the application programs that subscribe to it. However, in
GridStat the subscribers’ required rate, latency, and number of redundant (disjoint)
paths are part of the API to subscribe to a given status variable. With this knowledge,
the management plane conigures an FE such that it can discard updates, an operation we call rate iltering, when downstream subscribers have received an update
recently enough to satisfy their latency and rate requirements. This can potentially
save a large amount of bandwidth, because many sensors in the power grid produce
updates with a hard-coded rate that is deliberately conservative, that is, at a rate
above what its engineers believed any application would require. The absence of rate
iltering means that each subscriber has to subscribe to a status variable at the highest rate that any subscriber has to subscribe at, which is, for example, what IPMC
requires. With many subscribers, and only a few needing high-rate updates, this can
be extremely wasteful in a WAMS-DD.
The second novelty is that GridStat implements synchronized rate downsampling
(IG5), a requirement that GridStat researchers irst described and that is necessary when performing rate iltering on synchrophasor data. The third way in which
GridStat’s data delivery is novel is that different subscribers to a given status variable
can be provided with different rate, latency, and number of redundant paths. We call
this QoS+ heterogeneity. This allows a power device, such as a relay, that is physically very close to the publisher to be given a high rate and low latency, while remote
subscribers that only need to loosely track the variable can be given a much lower
rate and higher latency. If this lexibility were not supported, the remote subscribers
would have to receive the same high rate of delivery as the device in the same substation, thus consuming unnecessary resources. This is what IPMC requires, because
there is no rate iltering of any kind.
These operations are depicted in Figure 4.7, where the management plane is not
shown as a hierarchy for simplicity. When updates of any publication, such as Y,
arrive at an FE, it knows Y’s publication rate and the rate at which each of its outgoing links needs Y. Consider the case where Y is being published by Pub1 at 120 Hz
and where Sub1 requires it at 120 Hz but Sub2 requires it at 30 Hz. At the FE (in this
example this could be FE4), where updates to Y split off on different links toward
Sub1 and, Sub2, the former link will pass on all updates while the latter will drop
three-quarters of them. Also, if Sub1 receives Y over two different paths, an update
will traverse a different set of FEs to reach it. In the example above, the irst path
would be FE0 → FE1 → FE2 and the second path would be FE0 → FE3 → FE4 → FE2.
In a proxy (piece of middleware code beneath the application), duplicate updates for
Y will be dropped.
We are not aware of any system that has such broad coverage, or close to it, of
the DRs and IGs, as should be clear from Table 4.2. Systems that provide either
application-driven rate iltering or QoS+ heterogeneity are uncommon, and we know
of no system that provides both. Yet, as is shown above, in Chapter 3, and in [57],
leading-edge and emerging power applications can greatly beneit from having such
capabilities.
95
GridStat
Data management plane
QoS metadata
QoS requirements
Control
Pub1
Sensor
app
Sub1
X
Y
GridStat data-delivery plane
(managed WAN)
Sensor
app
X
Y
Proxy
Proxy
X, Y updates
FE1
FE0
FE2
X, Y updates
FE3
FE4
PubN
Sensor
app
SubM
FE5
FE6
Z
FE7
Proxy
Z updates
FE8
Sensor
app
Y
Z
Proxy
FE9
Y, Z updates
FIGURE 4.7 GridStat FE operation.
4.7.1.3 Performance and Scalability
On a 2010-era PC in Java, GridStat adds less than 0.1 ms per overlay hop. Because,
in any real deployment of GridStat, this would be dominated by the speed of light,
it is essentially free, adding no more than 1 ms over the speed of the underlying
networks, because there would be no more than about 10 hops. This PC version can
handle ~500 K forwards/s at each FE. This could be further improved if we made
the assumption that rates that can be subscribed to are power-of-two multiples of a
base rate; for example, 15, 30, 60, …, Hz. Further, given how simple the rate-based
forwarding logic is and how ineficient Java is at high-performance input/output,
a C version (which is planned for the very near term) should achieve 10–20 times
the Java version, or roughly 5–10 M forwards/s. This is likely more than adequate
for the backbone needs of major grids for at least 5 years, given the slow growth in
interutility sharing.
On 2003-era network processor hardware, it adds ~0.01 ms/hop and scales to a
few million forwards/s [58]. Using 2013-era network processing hardware (which is
at least 10 times more scalable than 2003 versions) and a small cluster, these results
should easily be extendable to achieve 50–100 M forwards/s, while rejecting unauthenticated messages, monitoring trafic patterns, and checking for (and reporting)
evidence of intrusions and cyberattacks. Custom hardware implementations in ieldprogrammable gate array or application-speciic integrated circuit would likely be
highly parallelizable as well as “unrolling” the logic in the software FE’s two-deep
loops. It is thus likely to support signiicantly higher than 100 M forwards/s, and
likely up to line speeds on a 2.5 Gbps or faster iber link. This is possible because
the logic in a GridStat FE is simpler than a general-purpose IP router: it is doing less
deliberately!
96
Smart Grids
4.7.2 RATE FILTERING FOR PER-SUBSCRIBER QOS+
In order to get an operational view of a wide-area system, its monitoring and control
system must be able to deliver a snapshot of the state of the entire system (or at least
a snapshot of a set of interesting measurements). There are two requirements in order
to achieve this. First, the measurements must be taken at the same time. Second, as
described above, if any rate iltering of the measurements is performed by the communication system, the temporal relationship between the events must be preserved.
GridStat’s forwarding logic provides deterministic multicast rate iltering of event
streams for temporally related status variables. This feature is provided by the underlying data structure and the routing algorithm that the FEs are using; hence it is
transparent to the end points and always activated for all events. In addition, there is
no need for any coordination between the FEs or intervention from the management
plane. The multicast feature is a side effect of the data structure used in the FE, and
as a result no extra computational resources are needed. The computational resource
usage by the iltering algorithm is further analyzed in [59].
4.7.2.1 Forwarding Algorithm
The forwarding algorithm is performed for each event that is received by an FE. The message format includes a header that stores the variable ID and the time stamp when the
speciic measurement was taken. For each subscription that is established over a speciic
FE, the following information is included during the subscription set-up command: variable ID, published interval, subscribed interval, and outgoing link. With this information
the data structure for the forwarding algorithm is built (for more information see [59]).
The three steps that are taken during the event forwarding are the following:
1. The irst step is to add to the event’s time stamp half of the publishing interval, that is, shift the event’s time stamp by half the publishing interval to the
right.
2. From the result of Step 1, take the mod of a subscription interval. Note
that this is done for all of the unique subscription intervals for this speciic
variable.
3. If the result of Step 2 is less than the publisher interval, the event is
forwarded.
The three steps can be illustrated with the following pseudocode:
if ((event.TS + ceil(pubInt/2)% subInt[0..i]) < pubInt)
forward event
else
drop event
4.7.2.2 Example of the Forwarding Algorithm
The example in Figure 4.8 illustrates how events from four variables (p1, p2, p3, p4)
that publish events every 50 ms will be forwarded through an FE for a subscriber
downstream that subscribed to receive the set every 100 ms. Due to various reasons,
97
GridStat
Original event streams:
50
ts = 80
100
ts = 95
p1
150
ts = 105 ts = 120
p3
p2
p4
ts = 130
p1
Events in 100 ms
Step 1:
50
ts = 145
p2
p3
150
ts = 105
p1
ts = 120
ts = 130
p2
p3
ts = 145
p1
p2
ts = 45
ts = 55
p3
p4
p1
p1
ts = 70
p2
ts = 80
ts = 95
p3
p4
50
ts = 20
p2
ts = 30
ts = 45
p3
p4
Events in 100 ms
p3
p4
Events in 150 ms
Step 3:
0
ts = 5
p2
ts = 180 ts = 195
100
ts = 30
Events in 100 ms
ts = 170
Events in 150 ms
50
ts = 20
200
ts = 155
p1
p4
Events in 100 ms
ts = 5
p4
Events in 150 ms
100
Step 2:
0
200
ts = 155 ts = 170
100
Events in 150 ms
FIGURE 4.8 Illustration of the iltering steps.
the events from the different publishers are published with slightly different time
stamps.
According to the algorithm, the following will occur in the FE:
1. The addition of 25 ms (publication interval divided by 2) to the time stamp
of each of the events. After the addition, the time stamps that belong to the
100 ms group will have a time stamp of 100–149 ms.
2. The subscription interval is 100 ms, and that means the events in the group
with time stamps 100, 200, 300, and so on should be forwarded. Taking the
mod of the result of Step 1 results in events with time stamps 0–99.
3. After Step 2, there are two phases with events from the publishers, because
subscription divided by publication interval equals 2. The framework will
always pick the irst phase, so the result of Step 2 is checked to test whether
it is less than the publication interval (the irst phase). If this is the case, the
event is forwarded; if not, it is dropped. Events with original time-stamp
values between 75 and 124 will be forwarded, while the others (time-stamp
values between 125 and 174) will be dropped.
4.7.3
CONDENSATION FUNCTIONS
In an environment where multiple entities must cooperate in order to maintain the
stability of the system, there will be some common patterns that the various entities will be interested in. Without support from the control and monitoring system,
98
Smart Grids
these patterns will have to be determined at the end points. If the determination for
these patterns can be provided as a service by the monitoring and control application,
a number of beneits can be observed. The most obvious beneit is the reuse of application logic, but also reduction of computational and network resource, as the computation will be performed only once, close to the source, and shared with all the
interested parties.
GridStat provides the condensation function mechanism, which allows an end
user to specify a transformation of multiple raw event streams into a new stream
during transmission. This transformation allows the end user to migrate application
logic into the middleware layer, with great lexibility.
4.7.3.1 Condensation Function Architecture and Capabilities
A condensation function consists of four modules, as shown in Figure 4.9: input
ilter, input trigger, calculator, and output ilter, with input ilter and output ilter
modules being optional.
The input ilter module ilters status events from the input event streams. The
iltering process compares the status event value with the upper or lower predeined
thresholds, or both, and decides whether to drop the event or not.
The trigger module is responsible for receiving events from the input status
variables and initiating the calculation function. For each of the input variables,
there is an entry in the data structure where the trigger module stores the latest
value from each of the variables. The calculator module is initiated by one of the
three built-in triggering mechanisms (others can be added). The built-in triggering
mechanisms are: time triggered, number of received events, and number of alert
events received.
The calculator module must be provided by the end user. The user will have to
overwrite two pure virtual methods, namely, the Initializer method (initializes the
object) and the Calculator (transforms the input to an output).
Lastly, the output ilter is similar to the input ilter, with the difference that it ilters the result produced by the calculator. The trigger module implements the same
interface as an outgoing link. What this entails is that, during the routing process, an
event routed to a condensation function has the same interface as if it were routed to
any of the outgoing links. As a result, the condensation function does not interfere
with the low of the routing process.
Input
filter
Input
trigger
Calculator
Output
filter
Filter the
input data
to either a
lower or an
upper
threshold
When
should the
calculation
start, delay
or number
of input
Applies a
user-defined
function to
the
arguments
Filter the
result to
either a
lower or an
upper
threshold
FIGURE 4.9 Condensation function modules.
GridStat
99
4.7.3.2 Condensation Function Development
The GridStat framework provides a graphical user interface (GUI) tool that assists
the end user in deining a condensation function. If the input stream or the produced
output stream uses data types other than those built in, an interface deinition language (IDL) compiler will be used to unmarshal and marshal the events for iltering
and calculation.
The placement of the condensation function within the data plane is the responsibility of the management plane. The GUI tool sends the speciication for the condensation function to the leaf QB, through an FE. The leaf QB will then ind an
appropriate FE and place the condensation there. In addition, it will set up all the
subscription paths that are the input for the condensation function as well as registering the resulting event stream as a irst-class variable, just like all the other
publications.
More detailed information on the condensation function can be found in [60].
4.7.4 OPERATIONAL MODES
Routing algorithms that disjoint path routing with additive constraints (latency) are
computationally very complex. It would in practice be very bad if there were thousands of new subscription requests in a crisis: that would likely be an inadvertent
but effective denial of service attack on the subscription infrastructure. Fortunately,
power grid operators do not choose random sensor variables to start observing during a power contingency. Rather, they have (paper) checklists for each contingency,
which were created months or years ago by engineering studies.
GridStat’s mode mechanism exploits this fact to enable its subscription base to
react very rapidly. It allows multiple forwarding tables to be preloaded in its FEs,
then rapidly switched when appropriate [61]. This action is called a mode change.
Depending on the GridStat deployment, FEs can utilize several forwarding tables
corresponding to the operating modes, while inactive forwarding tables lie dormant
until their corresponding mode is activated.
GridStat also supports a hierarchy of modes. QBs also inherit mode sets from
their ancestors in the hierarchy. A QB will have one set of forwarding tables for each
mode set, and will have exactly one active mode in each mode set.
4.7.4.1 Mode Change Algorithms
A mode deinition consists of an ID, a name, and a set of data plane subscriptions (a
forwarding table), and is owned by a single QB in the management hierarchy. Modes
deined and owned by a QB constitute a mode set, and exactly one of the modes in a
mode set is active at any time; that is, the operating mode. This means that every QB
always has exactly one of its modes active, which could be a default and unnamed mode
if no modes are deined. A QB that operates in a mode assures that all +subscriptions
(and parameterizations thereof) contained in the subscription set of that mode are active
in the data plane by using the forwarding table corresponding to the active mode.
A QB can quickly switch forwarding tables in the data plane through modes.
Figure 4.10 shows how operating modes are used in both the management plane and
the data plane. In this example, QB-A is the parent of both QB-B and QB-C.
100
Smart Grids
Management plane
QoS broker A
Mode sets B: (Stable, unstable)
Active modes: Stable
QoS broker B
Mode sets A: (Green, yellow, red)
Active modes: Green
QoS broker C
Mode sets C: (Normal, waming, critical)
Active modes: Normal
Data plane
B1
B2
Publisher
B3
Publisher
C4
B4
Subscriber
C2
C1
Mode sets: A, B
Active modes: (A.Green, B.Stable)
C3
Mode sets: A, C
Active modes: (A.Green, C.Normal)
Subscriber
FIGURE 4.10 Mode example.
QB-A deines a mode set that contains {Green,Yellow,Red}. However, it has
no cloud directly under its control, so this mode set is used to inluence forwarding
in its children, B and C. The active mode at this time for QB-A is Green. If it later
changes its mode, FEs belonging to its descendants will change forwarding tables.
QB-B deines the mode set with labels {Stable,Unstable}, and its active
mode presently is Stable. Thus, QB-B will utilize forwarding tables for both
A.Green and B.Stable.
QB-C deines the mode set with labels {Normal,Warning,Critical}.
Its active mode is Normal. Thus, QB-C will utilize forwarding tables for both
A.Green and C.Normal.
A mode change operation that switches the forwarding tables in all of the involved
FEs at the expected time is called a consistent mode change operation. Otherwise,
the operation is called an inconsistent mode change operation, and additional recovery mechanisms must be utilized in order to restore the operating modes on the
FEs that are considered inconsistent. An inconsistent mode change operation will
most likely result in some subscribers not receiving all the events that they have
subscribed to, or receiving a few updates to a new mode while still receiving updates
that should only have come in the old mode. However, these glitches will only happen for the duration of the mode change, which is very small.
The recovery mechanism is provided in order to tolerate some degree of network
failures and to eventually ensure consistent mode change operations. The recovery
mechanism is triggered by the QB, which detects missing mode change acknowledgments, and attempts to resolve these situations when possible.
Two mode change algorithms have been implemented in GridStat: the hierarchical mode change and the looding mode change. The hierarchical mode change algorithm uses the management hierarchy to disseminate mode change operations and
gather acknowledgments from FEs and QBs. The hierarchical mode change algorithm enables all subscriptions registered to operate in the coordinator’s current and
GridStat
101
new mode to low during the entire mode change. Thus, subscribers with subscriptions registered in both the current and new modes continue to receive status events
during hierarchical mode change operations that switch between those modes.
The looding mode change algorithm disseminates mode change operations
directly out on the data plane by using the limited looding mechanism in GridStat.
When an FE receives the operation, it forwards the operation on all outgoing event
channels, except the event channel from which it received the operation. As FEs may
receive multiple copies of the same operation, redundant copies are discarded. FEs
are informed to change from the current mode to the new mode at some future time
stamp.
The two mode change algorithms provide different trade-offs. The hierarchical
mode change algorithm is a resource-intensive algorithm that is split into ive message phases in order to enable transferred subscriptions present in both modes in a
mode change, for example, from Green to Yellow, to low. A message phase indicates
that the coordinator has to initiate and propagate a mode change phase (message)
down to all FEs in its hierarchical scope, and the next phase cannot be initiated
until the previous phase has completed. The coordinator must receive aggregated
acknowledgments from all FEs and QBs. The looding mode change algorithm, on
the other hand, is a best-effort algorithm and disseminates mode change operations
directly out on the data plane, where FEs are informed to switch at a predetermined
future time. The looding mode change algorithm is eficient, in terms of resource
usage and performance, but does not guarantee any subscriptions to low during
the mode change operation. The looding mode change algorithm relies on the FEs’
ability to switch modes at the exact same time and therefore requires all FEs to be
time synchronized.
We now provide more detail on both of the mode change algorithms.
4.7.4.2 Hierarchical Mode Change Algorithm
As mentioned earlier, the hierarchical mode change algorithm is divided into ive
distinct phases that enable transferred subscriptions to be present in both modes of
a mode change operation. Furthermore, the ive phases eliminate any FE overload
scenarios during a hierarchical mode change operation.
The following steps show how the hierarchical algorithm affects the FEs in a
mode change from Green to Yellow:
1. The inform phase: The FEs that have subscribers attached to them inform
the subscribers about the upcoming mode change.
2. The prepare phase: The FEs that have publishers connected to them switch
to the temporary forwarding table Green ∩ Yellow. The highest subscription interval (lowest rate) of transferred subscriptions is issued in this phase
in order to reduce the load on downstream FEs. This phase ensures that
subscription trafic that belongs in both modes (Green and Yellow) is forwarded through the FE network. This step only eliminates subscriptions,
and hence cloud resources cannot become overloaded.
3. The internal change phase: FEs that have neither publishers nor subscribers connected to them switch to Yellow’s forwarding table. Since all other
102
Smart Grids
FEs operate in a temporary forwarding table and only forward a smaller set
of subscriptions (in mode Green and Yellow), the FEs can safely switch to
mode Yellow without overloading any FEs downstream.
4. The edge change phase: The FEs that are currently in the temporary forwarding table Green ∩ Yellow will switch to Yellow’s forwarding table.
5. The commit phase: The FEs inform their subscribers about the completed
mode change.
4.7.4.3 Flooding Mode Change Algorithm
The looding mode change algorithm disseminates mode change operations directly
out on the data plane by using the limited looding mechanism in GridStat [62]. The
leaf QB that is responsible for the speciic cloud initiates the looding by using its
embedded publisher. When an FE receives the operation it forwards the operation
on all outgoing event channels, except the event channel from which it received the
operation. As FEs may receive multiple copies of the same operation, redundant
copies are discarded. FEs are informed to change from the current mode to the new
mode at some future time stamp. Upon receiving the mode change command, an FE
immediately responds with an acknowledgment to its leaf QB. It will then activate
the new mode at the destined future time stamp.
The looding mode change algorithm offers better statistical delivery guarantees
to the data plane than the hierarchical mode change algorithm. This is because of
the number of redundant paths in the data plane, and it is thus more resilient to network failures than the hierarchical mode change algorithm. While the hierarchical
mode change algorithm attempts to preserve the subscriptions registered in the two
involved modes, the looding mode change algorithm switches directly to the new
mode, even though some of the FEs have not acknowledged that they will perform
the switch.
4.7.4.4 Overview of Performance
The two algorithms presented above are quite different with respect to their complexity. This is also relected in their performance. In an experimental setting with
three layers of QBs and ive clouds with ive FEs in each cloud, with the packet loss
varied from 0% to 8%, and the minimum consecutive packet loss varied from 0 to 3
and the maximum consecutive packet loss varied from 0 to 5, the following observations were made. The looded mode change algorithm reached all FEs in approximately 45 ms, while the hierarchical algorithm required 1200–3200 ms, depending
on the link loss and burstiness setting. This was observed when the link latency was
set to a very conservative 8 ms for each of the links in order to emulate a wide-area
power grid (the speed of light in iber or copper is roughly 125 mi/ms); this includes
the links between the QBs as well as the links between the FEs. More detailed information about the performance of the mode change algorithms can be found in [61].
4.7.5
SYSTEMATIC ADAPTATION VIA DATA LOAD SHEDDING WITH MODES
The mode change algorithms described above, coupled with other information
that can easily be captured by middleware and an application management system,
GridStat
103
provide powerful capabilities in helping a WAMS-DD not only to adapt to IT problems, but also to optimize its beneit for the power problems being dealt with at any
given time, not just the (usually calm) steady state.
Electric utilities perform load shedding, whereby they cut off power to some
customers as a last resort to try to avoid a blackout. We call this power load
shedding.
There is a communications analog for WAMS-DD: data load shedding. For example, GridStat’s QoS+ APIs are for rate, latency, and number of paths. However, they
require not only the desired amount (which would be provided in the steady state)
but also the worst case that the application can tolerate. This implicitly gives permission to throttle back some lows from their desired QoS+ toward the worst case if
necessary, by reducing the rate or the number of paths, or both. This can help the
WAMS-DD adapt to fewer functioning resources being available due to benign IT
failures or malicious cyberattacks. But it can also help free up capacity in order to
add new subscriptions speciic to a given contingency. For example, dozens of new
synchrophasor sensors could be subscribed to, the rates of some existing subscriptions could be increased, and a few feeds from high-rate DFRs could be added (at
720 Hz or more). This could help operators have a much better understanding of
what is happening in the grid at that moment. This could be done very rapidly: in a
fraction of a second, not only are the data feeds changed, but also visualization windows pop up to alert operators and give them quick insight into the ongoing power
problems.
Given GridStat’s semantics, such data load shedding can easily be done in a
systematic way that dynamically optimizes (or at least greatly improves) the value
of the WAMS-DD to the utility. A very simple scheme would be one in which different subscriber applications are tagged with their importance (think of a number
from 0 to 10, for example), not only for the steady state but for that application’s
importance in different contingencies. These data can then be used to decide which
subscriptions are throttled back toward their worst case in any of a number of ways.
This adaptation could be further optimized using more complex beneit functions.*
With beneit functions, an application (or system integrator) would specify how
much beneit a given level of QoS+ provided to the application; again, for simplicity, think of a number from 0 to 10 here. Examples of a beneit function speciication, where individual list items are (rate:beneit), might be {(<30:0), (30:3), (60:7),
(≥120:10)}. This means that there is no beneit to the application from a rate less
than 30 Hz and no added beneit from going above 120 Hz, and the values between
have the beneits listed.
4.7.6 REMOTE PROCEDURE CALL
GridStat’s one-way delivery mechanisms are suficient for delivering updated values from remote sensors. However, they are inadequate for a round-trip RPC, for
example, to an actuator in the power grid or between control centers. To overcome
* Beneit functions are more commonly called utility functions, but we use the former term in the
electricity sector to avoid confusion.
104
Smart Grids
this limitation of the one-way delivery mechanism, the Ratatoskr round-trip RPC
mechanism is built over GridStat [63]. It is a highly tunable RPC mechanism tailored
to the needs of the power grid. The RPC mechanism implemented is a simple invocation of remote subroutines without any object-oriented semantics, which provides
a familiar programming environment and lends itself well to serial programming.
More importantly, the mechanism proves the feasibility of a two-way control system
on top of the GridStat framework.
4.7.6.1 Mechanism Details
Ratatoskr is a two-level protocol stack that is built on top of the GridStat network
(see Figure 4.11). The bottom level is the 2 Way over Pub-Sub (2WoPS) protocol that
provides for two-way communication, while the top level provides an RPC mechanism (Ratatoskr RPC [RRPC]). The 2WoPS protocol utilizes the QoS semantics
provided by the GridStat network (spatial redundancy), and it introduces two additional redundancy techniques: temporally redundant sends and ACK/retries. RRPC
provides extensive customization of the redundancy levels for each call, which provides for various trade-offs.
Further, pre- and postconditions built into the call semantics provide additional
safety mechanisms for application designers. These are described after we irst
explain the details of the RPC protocol.
4.7.6.2 The 2WoPS Protocol
To achieve a two-way communication style, needed for RPC signaling, Ratatoskr
uses two separate GridStat pub-sub connections. Each RPC host utilizes both a publisher and a subscriber—the publisher for sending data as published status variables
and the subscriber for receiving data. Currently, the 2WoPS protocol supports only
communication between two hosts, but the GridStat framework has support for multicast of status events, so future versions could add one-to-many and group communication styles.
Control center
Substation
Control center system
Actuator
Ratatoskr
RPC
Legacy
control
2WoPS
GridStat
publisher
Sensor
state
monitor
GridStat
subscriber
Substation
sensors
Ratatoskr
RPC
2WoPS
GridStat publisher
GridStat data plane
Network: UDP-IP, ATM, network processors, . . .
FIGURE 4.11 RPC module stack.
Legacy
control
Protection
scheme
GridStat subscriber
GridStat
105
The protocol aims to provide a high degree of fault tolerance while retaining
predictable delivery times. To this end, three redundancy techniques are employed
to strengthen delivery guarantees. These are:
1. Spatial redundancy: GridStat provides QoS that a subscription stream should
be forwarded through multiple disjoint network paths. The 2WoPS protocol
uses this feature to provide spatial redundancy for its communication.
2. Temporal redundancy: While spatial redundancy provides protection from
a range of network failures, the network costs and delay properties of a large
number of paths could prove prohibitive in cases where a very high degree
of fault tolerance is required or where the network topology is unsuited for
spatial redundancy. Ratatoskr offers temporal redundancy by allowing each
data unit sent by the 2WoPS protocol to be repeatedly published without
waiting for a round trip. A data unit published n times has a temporal redundancy level of n and tolerates n−1 losses.
3. ACK/resend: While spatial and temporal redundancy provide a high degree
of fault tolerance, they do not provide any feedback to the application about
successful delivery, and do not provide protection from failures that affect
the whole network, such as periods of heavy congestion. Ratatoskr provides a third technique for redundancy by having message receivers ACK
successfully received 2WoPS messages, and allowing the user to specify
that messages that do not produce an ACK should be resent up to n times.
A message that is set to be resent up to n times when ACKs are missing has
an ACK/retry level of n.
4.7.6.3 The Ratatoskr RPC Mechanism
RRPC is an RPC protocol for power grid control communication built on top of
the 2WoPS protocol. It draws extensively on the features provided by the 2WoPS
protocol. Delivery guarantees for calls are achieved using GridStat’s QoS-enabled
network communications.
In addition to increased safety through fault tolerance, pre- and postconditions
on calls are built into the RPC semantics. Preconditions are conditional expressions
over GridStat status variables that are evaluated at the server (a power grid actuator) before execution of an RRPC call, and, if the expression is false, the call to the
physical actuator is not executed. Postconditions are similar expressions evaluated
after the call has executed, and negative results are reported back to the application
via exception. This can provide a physical conirmation that the actuator succeeded
even if all of the redundant reply messages are lost. This helps circumvent a wellknown limitation in distributed computing—that in an asynchronous system, even
with only one failure, a client cannot know whether its request was processed by the
server or not [64]—by using a physical feedback loop. It also handles the case of a
failed actuator, when the irmware returns a successful return code but the physical
device is stuck open or closed.
An example of precondition in power grid operation could be the control of an isolator. Isolators are actuators that connect and disconnect de-energized power circuits.
106
Smart Grids
A precondition could be to verify that a line is de-energized before attempting isolation. An example of a postcondition could be the control of a load tap changer.
Load tap changers are components of certain transformers that allow adjustment of
output voltage. A postcondition could be to verify the new voltage level after load
tap changer operation, or even to generate a status report from all connected devices
and send it back to the client.
4.7.7 SECURITY AND TRUST IN GRIDSTAT
As discussed in Section 4.6, security and trust are important for the operation
of any information infrastructure supporting the power grid, and there are multiple dimensions upon which they must be addressed. GridStat research has led to
implementation within the framework of mechanisms for securing and authenticating published messages [65] as well as mutual authentication between components
of the GridStat system [66]. Both mechanisms are built using an architecture for
composable and replaceable security modules under the control of security managers operating in cooperation with QBs. This architecture was motivated by two
principal facts. First, the long lifetime of power grid IT infrastructure strongly suggests that security implementations developed today will not remain secure over
the decades that the system will be deployed. Second, no single existing protocol,
particularly for message authentication, simultaneously provides both adequate
security and adequate latency to meet the needs of all potential applications for the
GridStat system: it is necessary to make trade-offs between the level of security
provided and the performance, and these trade-offs will be different for different
applications and data streams [67]. The security module architecture addresses
both needs. A number of different encryption, authentication, and obfuscation
modules have been built for GridStat. It is relatively easy to build modules to
explore new cryptographic protocols having different performance trade-offs. For
example, recent work on protocols based on one-time public-key signature protocols has been incorporated as a new module choice offering a favorable latency/
security trade-off for some situations [50,68]. The module replacement mechanism
is designed to allow the modules in use for a particular stream or node authentication purpose to be securely replaced during operation, even if the algorithm
used by the replaced module has been compromised. The scheme for this relies
on sharing a collection of symmetric-key material between a security manager
node and the nodes that it manages at the time that the nodes are commissioned.
This preshared key material is slowly consumed over the life of the system when
module replacements occur.
Another aspect concerns securing the nodes of the security management system
themselves. Compromising the security management node associated with a particular QB potentially compromises communications of the management subtree that the
QB anchors. To address this, a distributed security service has been designed using
threshold cryptographic techniques to ensure that compromising even several nodes
does not compromise operation of the system [69].
As yet, the GridStat framework does not contain any speciic support for trust
decision making, though the work in [55] is designed to support GridStat or any
GridStat
107
other critical-infrastructure pub-sub system. A utility operator’s decisions rely on
data received and inferred. In an adversarial world, it must be possible for operators to assess how likely it is that the data they are provided truly relect the state of
the system or whether they have been manipulated or corrupted. Security mechanisms such as those described above are extremely important for reducing the
likelihood of tampering, but decisions should not be taken on the blind assumption
that the security mechanisms are completely effective in all cases (e.g., data could
be manipulated by tampering with sensors, ahead of the point where a digital signature is applied).
Research undertaken so far addresses what such mechanisms should look like
[70] and how they can be used to support utilities’ operational decision making
[55,71].
4.8 A NEW WORLD FOR POWER APPLICATION
PROGRAMMERS AND RESEARCHERS
The capabilities provided by GridStat, or any WAMS-DD that meets the DRs and
IGs and implements security appropriate for WAMs-DD as outlined above, are
a quantum leap in data delivery for electric power grids and other critical infrastructures. However, they also open up new ways of thinking for power researchers and engineers in terms of their power applications. We summarize these in
this section.
As a baseline, power applications can be written assuming that the WAMS-DD
will be:
• Extremely fast: less than 1 ms or so added over the speed of light in the
underlying network layers
• Extremely available: and providing the performance guarantees even in the
face of failures
• Appropriately cybersecure: security mechanisms appropriate for longlived embedded devices and as lightweight as possible so as not to preclude
closed-loop applications
• Tightly managed: on a per-subscriber-per-publication basis to provide the
above strong guarantees
• Adaptive: can change its subscription base very quickly
In order to exploit this much-improved baseline, power researchers and engineers
need to ask themselves a number of questions, including:
• What {rate, latency, redundancy} does my application really need for its
remote data?
• Why fundamentally does it need this level of QoS+?
• How sensitive is my application to occasional longer delays, period drops
(perhaps a few in a row), or data blackouts for longer periods of time?
• How can I formulate and test hypotheses for the above?
108
Smart Grids
Further, note that the above is only for the steady state. However, the questions
should also be asked for different power contingencies. For example:
• How important is my application in a given contingency (e.g., simply assign
a number 0–10)?
• How much lower {rate, latency, redundancy} can I live with in a contingency (here, even at a lower rate, note that these are still strong guarantees)?
• Can I program my application to run at different rates (assuming, if necessary, that antialiasing is done appropriately for that rate), or is there a fundamental reason why my application has to run at a given rate?
• What extra data feeds (or existing ones at higher rates) could I use in a given
contingency, assuming that they could be available within a second of the
contingency being identiied? How can I best use these for either operator
situational awareness or closed-loop responses?
From the above, it may not be clear that power researchers need to work closely
with computer scientists in order to help the power grid be more resilient. The status quo today is that power researchers tend to ignore communications issues and
computer scientists do not have nearly enough useful information about the power
grid’s applications, as well as its pragmatic IT constraints, in order to offer significant help. However, if they work together, much can be achieved. For example,
power researchers can assume much “better” communications (strong guarantees,
systematic adaptation, etc.) while computer scientists can come up with even better
WAMS-DD mechanisms if they know not only emerging “killer apps” for power but
also the trade-offs that power applications can live with.
Finally, note that the above only involves communications. The computational
resources available to make power grids more stable should not be assumed to be static,
in particular thanks to cloud computing [72]. Indeed, the GridCloud project is combining GridStat researchers with Cornell researchers who are developing advanced cloud
computing mechanisms to provide a comprehensive, E2E solution for mission-critical
cloud computing. Such a system will keep the same high throughput, as generic, commercial cloud computing mechanisms improve consistency and predictability. The
consistency is improved by removing the deliberate (and, as it turns out, unnecessary)
trade-off to allow occasional inconsistencies in cloud data to preserve high throughput.
In terms of performance, GridCloud is also implementing ways to make the aggregate
performance and ramp-up time of cloud computations much better.
Given such an improved computational infrastructure, then, questions for power
researchers include how they could use
• Hundreds of processors in steady state
• Thousands or tens of thousands of processors when the grid is approaching
a blackout
• Data from all participants in a grid enabled quickly when a blackout is
being approached
Such improved computational support will likely have a huge impact on power
grid operations.
GridStat
109
The questions in this section represent a completely new way of thinking for
power researchers. However, they have the potential to make the power grid much
more resilient than simply using today’s applications (or ones planned for tomorrow)
and assuming that communications and computations are limited and nonadaptive.
4.9
CONCLUSIONS
This chapter has described the requirements for building WAMS-DD, and IGs that
the authors believe are necessary for a proper WAMS-DD, or at least highly advisable. It then analyzed how completely these requirements and guidelines are met by
network-level technologies, middleware-level technologies, and technologies developed in the electricity sector. Following this, the chapter overviewed the NASPInet
initiative, and explained how NASPInet and other WAMS-DDs must employ middleware (as well as, of course, a network-level mechanism with middleware layers
on top). Finally, we described GridStat in detail, and overviewed the completely new
way of thinking of new power application programs that it enables.
ACKNOWLEDGMENTS
The GridStat project has been going on since 1999, so there are many people to acknowledge for their support and guidance besides the authors, and we apologize in advance
for any omissions. GridStat co-principal investigator Professor Carl Hauser has been an
invaluable partner in the development of GridStat. Professor Anjan Bose has been an
invaluable collaborator from the power grid side. Dr. Thoshitha Gamage has contributed heavily to GridStat in the last few years. Graduate students Erlend Viddal, Stian
Abelsen, Venkata Irava, Erik Solum, Ryan Johnston, Joel Helkey, Kim Swenson, Sunil
Muthuswamy, Supreeth Sheshadri, Ping Jiang, Rasika Chakravarthy, Michael Carter,
and Punit Agrawal have all contributed heavily. Power researchers who have helped
inluence GridStat’s applications and requirements include graduate student Tao Yang
and power researchers Dr. Chuanlin Zhao and Professor Mani Venkatasubramanian.
Lead GridStat staff programmer David Anderson has contributed immensely
to GridStat’s implementation as a full-time programmer for the last 4 years.
Undergraduate staff programmers include Andrew Lytle, Raeann Marks, Loren
Hoffman, Nathan Schubkegel, Alex Schuldberg, Ryan Hoffman, Kyle Wilcox, Travis
Gomez, Chad Selph, Constance Bell, Simon Ngyuen, Nick Newsome, and DeAndre
Blackwell. Suggestions, questions, and advice from a large number of people have
helped inluence GridStat. These are too numerous to list, but include Jeff Dagle,
Greg Zweigle, Phil Overholt, Patricia Hoffman, Dave Bertagnolli, Rich Stephenson,
Jeff Gooding, Mahendra Patel, Ritchie Carroll, Ken Martin, Floyd Galvan, Lisa
Beard, Neeraj Suri, Deb Frincke, Jim McNierney, Dan Brancaccio, Dick Wilson,
Dan Lutter, and Don Kopczynski.
Many funding agencies have provided inancial support for GridStat. These
include US Department of Energy grant DE-OE0000097 (TCIPG), US Department
of Energy grant DE-OE0000032 (GridSim), ARPA-E grant DE-AR0000230
(GridCloud), grants CNS 05-24695 (CT-CS: Trustworthy Cyber Infrastructure for the
Power Grid (TCIP)) and CCR-0326006 from the US National Science Foundation;
110
Smart Grids
and by grant #60NANB1D0016 (Critical Infrastructure Protection Program) from
the National Institute of Standards and Technology, in a subcontract to Schweitzer
Engineering Labs Inc. We thank the US Department of Energy and the Department
of Homeland Security for their part in funding the National Science Foundation
TCIP center. We thank Schweitzer Engineering Labs for their donation of equipment
and Paciic Northwest National Laboratory for their loan of equipment.
DISCLAIMER
This chapter was prepared as an account of work sponsored by an agency of the US
government. Neither the US government nor any agency thereof, nor any of their
employees, makes any warranty, express or implied, or assumes any legal liability
or responsibility for the accuracy, completeness, or usefulness of any information,
apparatus, product, or process disclosed, or represents that its use would not infringe
privately owned rights. Reference herein to any speciic commercial product, process,
or service by trade name, trademark, manufacturer, or otherwise does not necessarily
constitute or imply its endorsement, recommendation, or favoring by the US government or any agency thereof. The views and opinions of authors expressed herein do not
necessarily state or relect those of the US government or any agency thereof.
REFERENCES
1. D. E. Bakken, R. E. Schantz, and R. D. Tucker, Smart grid communications: QoS
stovepipes or QoS interoperability, in Proceedings of the 2009 Grid-Interop, GridWise
Architecture Council, Denver, CO, 2009. Available at: http://www.gridstat.net/
publications/TR-GS-013.pdf.
2. K. Fodero, C. Huntley, D. E. Whitehead, and B. Kasztenny, A novel scheme for widearea time synchronization, in Proceedings of the 10th IET International Conference on
Developments in Power System Protection, Managing the Change, pp. 1–5, 29 March–1
April, Manchester, 2010.
3. D. E. Bakken, C. H. Hauser, H. Gjermundrød, and A. Bose, Towards more lexible and
robust data delivery for monitoring and control of the electric power grid, Washington
State University, Pullman, WA, Technical Report EECS-GS-009, May 2007.
4. Electric Power Research Institute (EPRI), The integrated energy and communication
systems architecture, vol. IV, Technical Analysis, 2004.
5. S. Horowitz, D. Novosel, V. Madani, and M. Adamiak, System-wide protection, IEEE
Power and Energy Magazine, 6(5), 34–42, 2008.
6. E. Rosen, A. Vishanathan, and R. Callon, RFC-3031: Multiprotocol label switching
architecture, The Internet Society, 2001.
7. Y. Krishnamurthy, V. Kachroo, D. A. Karr, C. Rodrigues, J. P. Loyall, R. E. Schantz, and
D. C. Schmidt, Integration of QoS-enabled distributed object computing middleware for
developing next-generation distributed applications, in Proceedings of the 2001 ACM
SIGPLAN Workshop on Optimization of Middleware and Distributed Systems, vol. 36,
no. 8, pp. 230–237, 18 June, Snowbird, UT, 2001.
8. R. E. Schantz and D. C. Schmidt, Research advances in middleware for distributed systems: State of the art, in Proceedings of the 2002 IFIP World Computer Congress, 25–30
August, Montreal, QC, 2002.
GridStat
111
9. J. Saltzer, D. Reed, and D. Clark, End-to-end arguments in system design,
Transactions on Computer Systems, Association of Computing Machinery, 2(4),
277–288, 1984.
10. G. V. Chockler, I. Keidar, and R. Vitenberg, Group communication speciications:
A comprehensive study, ACM Computing Surveys, 33(4), 1–43, 2001.
11. X. Défago, A. Schiper, and P. Urbán, Totally ordered broadcast and multicast algorithms: Taxonomy and survey, ACM Computing Surveys, 36(4), 372–421, 2004.
12. Y. Hu, Data bus technical speciications for North American synchro-phasor initiative
network (NASPInet), US Department of Energy, May 2009. https://www.naspi.org/File.
aspx?ileID=587.
13. R. E. Schantz, J. P. Loyall, M. Atighetch, and P. P. Pal, Packaging quality of service control behaviors for reuse, in Proceedings of the 2002 5th IEEE International Symposium
on Object-Oriented Real-Time Distributed Computing, 29 April–1 May, Washington,
DC, 2002.
14. P. P. Pal, J. P. Loyall, R. E. Schantz, J. A. Zinky, R. Shapiro, and J. Megquier, Using QDL
to specify QoS aware distributed (QuO) application coniguration, in Proceedings of
the 2000 3rd IEEE International Symposium on Object-Oriented Real-Time Distributed
Computing, pp. 310–319, 15–17 March, Newport, CA, 2000.
15. D. E. Bakken, Quality of service design considerations for NASPInet, Presentation to the
North American Synchrophasor Initiative (NASPI) Work Group Meeting, 4 February,
2009. Available at: http://www.naspi.org.
16. K. H. Gjermundrød, I. Dionysiou, C. H. Hauser, D. E. Bakken, and A. Bose, Flexible
and robust status dissemination middleware for the electronic power grid, School of
Electrical Engineering and Computer Science, Washington State University, Pullman,
WA, Technical Report EECS-GS-003, September 2003.
17. R. Bobba, E. Heine, H. Khurana, and T. Yardley, Exploring a tiered architecture for
NASPInet, in Proceedings of the 2010 IEEE Conference on Innovative Smart Grid
Technologies, pp. 1–8, 19–21 January, Gaithersburg, MD, 2010.
18. K. Tomsovic, D. E. Bakken, M. Venkatasubramanian, and A. Bose, Designing the next
generation of real-time control, communication and computations for large power systems, Proceedings of the IEEE Special Issue on Energy Infrastructure Systems, 93(5),
965–979, 2005.
19. K. H. Gjermundrød, D. E. Bakken, C. H. Hauser, and A. Bose, GridStat: A lexible
QoS-managed data dissemination framework for the power grid, IEEE Transactions on
Power Delivery, 4(1), 136–143, 2009.
20. V. S. Irava and C. H. Hauser, Survivable low-cost low-delay multicast trees, in
Proceedings of the IEEE Global Telecommunications Conference, vol. 1, pp. 6–115, 2
December, St Louis, MO, 2005.
21. V. S. Irava, Low-cost delay-constrained multicast routing heuristics and their evaluation,
PhD Dissertation, Washington State University, Pullman, WA, August 2006.
22. GridWise Architecture Council, Interoperability Constitution Whitepaper (v1.1),
December 2006.
23. L. Roberts, A radical new router, Proceedings of the IEEE Spectrum, 46(7), 35–39,
2009.
24. P. T. Eugster, P. A. Felber, R. Guerraoui, and A. M. Kermarrec, The many faces of publish-subscribe, ACM Computing Surveys, 35(2), 114–131, 2003.
25. J. Zinky, D. E. Bakken, and R. Schantz, Architectural support for quality of service for
CORBA objects, Theory and Practice of Object Systems, 3(1), 55–73, 1997.
26. K. P. Birman, J. Chen, K. M. Hopkinson, R. J. Thomas, J. S. Thorp, R. van Renesse,
and W. Vogels, Overcoming communications challenges in software for monitoring and
controlling power systems, Proceedings of the IEEE, 93(5), 1028–1041, 2005.
112
Smart Grids
27. K. Hopkinson, G. Roberts, X. Wang, and J. Thorp, Quality of service considerations in
utility communication networks, IEEE Transactions on Power Delivery, 24(3), 1465–
1474, 2009.
28. J. Rajahalme, A. Conta, B. Carpenter, and S. Deering, RFC3697: IPv6 low label speciication, The Internet Society, 2004. Available at: http://www.faqs.org/rfcs/rfc3697.html.
29. Y. Vigfusson, H. Abu-Libdeh, M. Balakrishnan, K. Birman, R. Burgess, H. Li,
G. Chockler, and Y. Tock, Dr. Multicast: Rx for data center communication scalability, in Proceedings of the 5th European Conference on Computer Systems Eurosys, pp.
349–362, April, ACM SIGOPS, Paris, France, 2010.
30. D. Basin, K. Birman, I. Keidar, and Y. Vigfusson, Sources of instability in data center
multicast, in Proceedings of the 4th International Workshop on Large Scale Distributed
Systems and Middleware (LADIS 2010), August, ACM Principles of Distributed
Computing (PODC), 25–28 July, Zurich, 2010.
31. K. Birman, Guide to Reliable Distributed Systems: Building High-Assurance
Applications and Cloud-Hosted Services, Springer, London, 2012.
32. A. Farrel (ed.), Network Quality of Service Know It All, Elsevier, Boston, 2009.
33. R. Wojcik and A. Jajszczyk, Flow oriented approaches to QoS assurance, ACM
Computing Surveys, 44(1), 1–37, 2012.
34. A. Sydney, J. Nutaro, C. Scoglio, D. Gruenbacher, and N. Schulz, Simulative comparison of multiprotocol label switching and openFlow network technologies for transmission operations, IEEE Transactions on Smart Grid, 4(2), 763–770, 2013.
35. PGM Reliable Transport Protocol Speciication, RFC 3208, IETF, 2001.
36. Y. Amir, C. Danilov, and J. Stanton, A low latency, loss tolerant architecture and protocol for wide area group communication, in Proceedings of the 2000 IEEE First
International Conference on Dependable Systems and Networking, pp. 327–336, 25–28
June, New York, 2000.
37. N. McKeown, T. Anderson, H. Balakrishnan, G. Parulkar, L. Peterson, J. Rexford,
S. Shenker, and J. Turner, OpenFlow: Enabling innovation in campus networks, ACM
SIGCOMM Computer Communication Review, 38(2), 69–74, 2007.
38. Open Network Foundation, OpenFlow Switch Speciication 1.3.2, April 25, 2013.
39. S. McGillicuddy, Not all OpenFlow hardware is created equal: Understanding the
options, Open Network Foundation, 25 September 2013. Available at: http://www.
opennetworking.org.
40. Net Insight, The Nimbra platform, 2013. http://www.netinsight.net/Global/Documents/
Products/Brochures/Nimbra_Brochure.pdf.
41. K. Geihs, Middleware challenges ahead, IEEE Computer, 34(6), 24–31, 2001.
42. D. Bakken, Middleware, unpublished article, 2001. http://www.eecs.wsu.edu/~bakken/
middleware.pdf.
43. Y.-J. Kim, J. Lee, G. Atkinson, H. Kim, and M. Thottan. SeDAX: A scalable, resilient,
and secure platform for smart grid communications, IEEE Journal on Selected Areas in
Communications, 30(6), 1119–1136, 2012.
44. K. Birman, Like it or not, web services are distributed objects! Communications of the
ACM, 47(12), 60–62, 2004.
45. K. Birman, The untrustworthy web services revolution, IEEE Computer, 39(2), 98–100,
2006.
46. PJM, PJM 2007 Strategic Report, 2 April, 2007. http://www.pjm.com/~/media/Images/
ctc-display/modules/timeline/2007-10-yr-pdf.ashx.
47. W. Mahnke, S. Leitner, and M. Damm, OPC Uniied Architecture, Springer, Berlin,
2009.
48. J. Zhang and C. A. Gunter, Application-aware secure multicast for power grid communications, in Proceedings of the IEEE International Conference on Smart Grid Communications
(SmartGridComm), pp. 339–344, 4–6 October, Gaithersburg, MD, 2010.
GridStat
113
49. C. Hauser, T. Manivannan, and D. Bakken, Evaluating multicast message authentication
protocols for use in wide area power grid data delivery services, in Proceedings of the
45th Hawaii International Conference on System Sciences (HICSS), pp. 2151–2158,
4–7 January, IEEE, Maui, Hawaii, 2012.
50. K. Cairns, C. Hauser, and T. Gamage, Flexible data authentication evaluated for the smart
grid, in Proceedings of the 4th Annual IEEE SmartGrid Communications Conference,
21–24 October, Vancouver, BC, IEEE. Forthcoming.
51. D. Clark, A cloudy crystal ball—Visions of the future, in Proceedings of the TwentyFourth Meeting of the Internet Engineering Task Force (IETF), pp. 539–543, July,
Cambridge, MA, 1992.
52. Z. Xie, M. Govindarasu, V. Vittal, A. Phadke, and V. Centeno, An information architecture for future power systems and its reliability analysis, IEEE Transactions on Power
Systems, 17(3), 857–863, 2002.
53. C. H. Hauser, D. E. Bakken, I. Dionysiou, K. H. Gjermundrød, V. Irava, J. Helkey, and
A. Bose, Security, trust and QoS in next-generation control and communication for large
power systems, International Journal of Critical Infrastructures, 4(1–2), 3–16, 2008.
54. I. Dionysiou, D. Frincke, C. Hauser, and D. Bakken, An approach to trust management challenges for critical infrastructures, in Proceedings of the 2nd International
Workshop on Critical Information Infrastructures Security (CRITIS07), Lecture Notes
in Computer Science Series volume 5141, pp. 173–184, 2–5 October, Springer Berlin,
Malaga, Spain, 2007.
55. Y. Wang and C. Hauser, An evidence-based trust assessment framework for critical
infrastructure decision making, in Proceedings of the 5th IFIP Working Group 11.10
International Conference on Critical Infrastructure Protection (ICCIP), pp. 125–135,
23–25 March, Hanover, NH, 2011.
56. M. D. Hadley, J. B. McBride, T. W. Edgar, L. R. O’Neil, and J. D. Johnson, Securing
wide area measurement systems, Paciic Northwest National Laboratory, Technical
Report PNNL-17116, 2007.
57. D. E. Bakken, A. Bose, C. H. Hauser, D. E. Whitehead, and G. C. Zweigle, Smart generation and transmission with coherent, real-time data, Proceedings of the IEEE, 99(6),
928–951, 2011.
58. K. Swenson, Exploiting network processors for low latency, high throughput, rate-based
sensor updated delivery, MS Thesis, Washington State University, Pullman, WA, 2009.
59. H. Gjermundrod, C. Hauser, and D. Bakken, Scalable wide-area multicast with temporal
rate iltering distribution framework, in Proceedings of the 2011 IEEE 11th International
Conference on Computer and Information Technology (CIT), pp.1–8, 31 August–2
September, Washington, DC, 2011.
60. H. Gjermundrød, D. E. Bakken, and C. H. Hauser, Integrating an event pattern mechanism
in a status dissemination middleware, in Proceedings of 1st International Conference on
Pervasive Patterns and Applications (PATTERN’09), pp. 259–264, 15–20 November,
IEEE, Athens, Greece, 2009.
61. S. F. Abelsen, H. Gjermundrød, D. E. Bakken, and C. H. Hauser, Adaptive data stream
mechanism for control and monitoring applications, in Proceedings of 1st International
Conference on Adaptive and Self-Adaptive Systems and Applications (ADAPTIVE’09),
pp. 86–91, 15–20 November, IEEE, Athens, Greece, 2009.
62. K. H. Gjermundrød, Flexible QoS-managed status dissemination middleware framework
for the electric power grid, PhD Dissertation, Washington State University, Pullman,
WA, August, 2006.
63. E. S. Viddal, D. E. Bakken, H. Gjermundrød, and C. H. Hauser, Wide-area actuator
RPC over GridStat with timeliness, redundancy, and safety, in Proceedings of the
International Conference on Complex, Intelligent and Software Intensive Systems
(CISIS’2010), pp. 17–24, 15–18 February, Krakow, Poland, 2010.
114
Smart Grids
64. M. J. Fischer, N. A. Lynch, and M. S. Patterson, Impossibility of distributed consensus
with one faulty process, Journal of the ACM, 32, 374–382, 1985.
65. E. Solum, C. H. Hauser, and R. Chakravarthy, Modular over-the-wire conigurable security for long-lived critical infrastructure monitoring systems, in A. Gokhale and D. C.
Schmidt (eds), Proceedings of the 3rd ACM International Conference on Distributed
Event-Based Systems, pp. 1–9, ACM, New York, 2009.
66. R. Chakravarthy, C. H. Hauser, and D. E. Bakken, Long-lived authentication protocols
for critical infrastructure process control systems, in Proceedings of the Fourth IFIP WG
11.10 International Conference on Critical Infrastructure Protection, March, 2010.
67. C. H. Hauser, T. Manivannan, and D. E. Bakken, Evaluating multicast message authentication protocols for use in wide area power grid data delivery services, in Proceedings of
the 45th Hawaii International Conference on System Science (HICSS), pp. 2151–2158,
4–7 January, Maui, HI, 2012.
68. K. Cairns, T. Gamage, and C. Hauser, Eficient targeted key subset retrieval in fractal hash sequences, in Proceedings of the 20th ACM Conference on Computer and
Communications Security, pp. 1273–1284, 4–8 November, Berlin, Germany, 2013.
69. P. Agrawal, A fault-tolerant key storage service with proactive recovery. MS Thesis,
School of Electrical Engineering and Computer Science, Washington State University,
August, 2012. Available at: http://www.dissertations.wsu.edu/Thesis/Summer2012/p_
agrawal_091412.pdf.
70. I. Dionysiou, Dynamic and composable trust for indirect interactions, PhD Dissertation,
Washington State University, Pullman, WA, August, 2006.
71. C. H. Hauser, Trust research to address uncertainty in security for the smart grid, in
Proceedings of the IEEE PES Innovative Smart Grid Technologies (ISGT), pp. 1–2,
16–20 January, Washington, DC, 2012.
72. K. P. Birman, L. Ganesh, and R. van Renesse, Running smart grid control software
on cloud computing architectures, in Workshop on Computational Needs for the Next
Generation Electric Grid, 19–20 April, Cornell University, Ithaca, NY, 2011.
5
A Distributed Framework
for Smart Grid Modeling,
Monitoring, and Control
Alfredo Vaccaro and Eugenio Zimeo
CONTENTS
5.1
5.2
5.3
Introduction .................................................................................................. 115
Related Works ............................................................................................... 118
Web Services–Based Framework for SG Management................................ 120
5.3.1 GISinterfaceWS ................................................................................ 122
5.3.2 DataAcquisitionWS........................................................................... 122
5.3.3 eAssessmentWS ................................................................................ 122
5.3.4 ComputationalWS............................................................................. 123
5.3.5 DataStorageWS ................................................................................. 124
5.3.6 Worklow Enactor ............................................................................. 125
5.4 Experimental Setting .................................................................................... 126
5.4.1 Data Acquisition Web Service .......................................................... 126
5.4.2 Job Submission Web Service ............................................................ 127
5.4.3 Computational Web Service ............................................................. 127
5.4.4 Data Storage Web Service ................................................................ 129
5.4.5 Worklow Enactor ............................................................................. 129
5.5 Conclusions ................................................................................................... 130
References .............................................................................................................. 131
5.1 INTRODUCTION
Electrical power systems were traditionally characterized by the presence of numerous utilities, heterogeneous standards, overlapping territories, and a general lack
of integration. Recently, under pressure from deregulated electricity markets and
new environmental policies, the main limitations intrinsic to this environment have
become clear and the structures of modern power systems have evolved to accommodate large changes induced by these new energy policy trends.
In this emerging scenario, the large-scale deployment of the smart grid (SG) paradigm could play a strategic role in supporting the evolution of conventional electrical
grids toward active, lexible, and self-healing web energy networks composed of
distributed and cooperative energy resources.
115
116
Smart Grids
From a conceptual point of view, the SG is the convergence of information and
operational technologies applied to the electric grid, providing sustainable options to
customers and improved security [1].
SG technologies include advanced sensing systems, two-way high-speed communications, monitoring and enterprise analysis software, and related services to get
location-speciic and real-time actionable data in order to provide enhanced services
for both system operators (i.e., distribution automation, asset management, advanced
metering infrastructure) and end users (i.e., demand-side management, demand
response) [2,3].
Advances in research on SG could increase the eficiency of modern electrical power systems by (i) supporting large-scale penetration of small-scale distributed energy resources (DERs); (ii) facilitating the integration of renewable energy
sources; (iii) reducing system losses and greenhouse gas emissions; and (iv) increasing the reliability of electricity supply to customers.
From this perspective, a crucial issue is how to increase the penetration of DERs
into the SG by properly coordinating their operation and by restricting their negative
impact on grid operations and control.
It is well known that the integration of DERs into running distribution systems
affects the power lows and voltage conditions of customer and utility equipment by
inducing a number of side effects (e.g., bidirectional power lows, increased fault
current levels, and a need for new concepts of network protection [4]). This increases
the complexity of the control, protection, and maintenance of distribution systems,
which are typically designed to operate radially without any generation at the distribution line or customer side [5]. This requires extensive analysis aimed at (i) identiication of the optimal generation schedule that minimizes production costs and
balances the demand and supply that come from both micro sources and the distribution feeder; (ii) online assessment of the SG security and reliability levels; and (iii)
identiication of proper preventive and corrective measures that mitigate the effects
of critical contingencies. The main limitations that should be addressed in order to
overcome these issues originate mostly from the following intrinsic features characterizing the SG operation:
• The rising level of DERs, which makes the distribution systems more vulnerable with respect to dynamic perturbations
• The increasing number of smaller geographically dispersed generators,
which could signiicantly increase the number of power transactions
• The dificulties arising in predicting and modeling market operators’
behavior, governed mainly by unpredictable economic dynamics, which
introduce considerable uncertainty into short-term SG operation
• The need for more detailed security studies assessing the impact of multiple
contingencies on the distribution system (i.e., N-2 security criteria)
• The high penetration of generation units powered by renewable energy
sources, which induces considerable uncertainty in SG operation
• The low level of upgrade and interoperability of existing distribution management systems that are typically based on low-scalable and proprietary
A Distributed Framework for Smart Grid Modeling, Monitoring, and Control 117
software platforms characterized by different information technologies and
standards for the data exchange
• The need for standard interfaces between the distributed tools and components characterizing the power system’s operation environment; these
interfaces must be standardized so as to work in a “plug-and-play” concept
similar to the hardware used in substation automation
In this context the adoption of web services, a new paradigm for enterprise application integration built on the foundation of open standard and common infrastructure could contribute to overcoming some of these limiting aspects.
A web service is an interface that describes a collection of modular, self-contained
applications that are network-accessible through standardized extensible markup
language (XML) messaging. The interface hides the implementation details of the
service, allowing it to be used independently of the hardware or software platform
on which it is implemented and also independently of the programming language
in which it is written [6,7]. The use of standard XML protocols makes web services
technology platform-, language-, and vendor-independent, and they are thus an ideal
candidate for supporting the deployment of the SG paradigm [8].
In agreement with these arguments, this chapter outlines the design of a prototype
web services–based framework for integrated SG modeling, monitoring, and control.
The architecture of the proposed framework is based on the following web
services:
1. GISinterfaceWS acquires georeferential information on the current network
topology and the control devices asset by interacting with a geographical
information system (GIS).
2. DataAcquisitionWS acquires the network ield data interacting with the
entire set of supervisory control and data acquisition (SCADA) systems and
ield electrical measurements (FEM) by standardized XML messaging.
3. eAssessmentWS develops the distributed generation/storage unit capability
assessment by interacting with a network of intelligent units.
4. ComputationalWS employs a parallel solution engine, based on a computational grid, to develop the calculations needed to support the entire set of
SG services in terms of state estimation, network analysis, online security
assessment, and so on.
5. DataStorageWS is used to permanently store historical data and alarm conditions that can be analyzed off-line using a web browser. The service can
be remotely accessed by using data access middleware.
In order to prove the effectiveness of the proposed framework, the results of ield
activities developed on an experimental test bed are presented and discussed.
The experimental results obtained show that the web service technologies in the
SG are an effective tool to integrate monolithic and hard-to-customize power system
control and monitoring software tools, allowing system operators to easily manage
proprietary hardware instrumentation and multiple software modules.
118
5.2
Smart Grids
RELATED WORKS
The control and operational issues of SGs are presented and discussed in [9,10]. The
analysis in this chapter shows that the required control and operational strategies
of an SG can be observably and conceptually different from those of conventional
power systems. These differences will depend on the type and depth of penetration
of DER units, load characteristics, power quality constraints, and market participation strategies.
Consequently, novel control strategies and ad hoc energy management systems
should be deployed for this new application. These facilities would support such
applications as DER and demand response dispatch, distribution automation and
substation automation, adaptive relaying, energy management, market pricing, grid
modeling, operator displays, and advanced visualization systems [9].
SGs could play an important role in deregulated energy markets, as outlined in [11].
In this context, SGs could be operated in order to serve the total load demand, using
local production as much as possible, without exporting power to the upstream transmission grid. This strategy represents a beneit for the transmission system operator,
since the SG relieves possible network congestion during peak demand by partly or
fully supplying the system’s energy needs.
As outlined in [12], operational control and energy management systems for the
SG should be implemented through the cooperation of various controllers, located
at all these levels, on the basis of the communication and collection of information
about distributed energy systems and control commands. This could be deployed
according to a centralized or decentralized control paradigm.
In a centralized control paradigm, the SG controllers optimize the power
exchanged between the distributed grid system (DGS) and the main grid, thus maximizing local production depending on market prices and security constraints. This
is achieved by issuing control set points to DERs and controllable loads in order to
optimize local energy production and power exchanges with the main distribution
grid [12].
The decentralized control paradigm is intended to provide maximum autonomy
for the DER units and loads. This implies that local controllers are intelligent and
can communicate with each other to form a larger intelligent entity. In a decentralized control framework, the main task of each controller is not to maximize the
revenue of the corresponding unit but to improve the overall performance of the SG.
Thus, the architecture must be able to include economic functions, environmental
factors, and technical requirements, for example, black start. These features indicate
that a multiagent system (MAS) can be a prime candidate for decentralized SG control [13–15]. Armed with such a vision, [14,15] present a general framework for the
control of DERs organized in SGs. The proposed architecture is based on the agent
technology and aims at integrating several functionalities, as well as being adaptable
to the complexity and size of the SG.
Another important issue to address for effective SG operational control is dealing
with the problem of security assessment in both steady-state and dynamic scenarios [10]. The main steady-state security concerns for the operation of SGs are satisfying voltage constraints and maintaining power lows within thermal limits. Dynamic
A Distributed Framework for Smart Grid Modeling, Monitoring, and Control 119
security concerns the SG operation under a number of contingencies both within and
above it (i.e., during a transition between the grid-connected and islanded modes of
operation). Effective SG security assessment requires detailed analysis of phenomena that can compromise SG operation. In this connection, it is expected that online
time-domain-based analysis such as voltage stability and transient angular stability
can be carried out in real time [16].
All of these functions require the design of reliable, resilient, secure, and manageable standards-based open communication systems [12]. These infrastructures represent a key issue in SG deployment, as they would provide the fundamental backbone
for connecting the SG elements, the data providers, and the decision-making entities
in an open and interoperable framework.
They should support ubiquitous connectivity between decision-making points
and dispersed and heterogeneous data sources characterized by varying degrees of
transport, security, and reliability requirements [12,17].
They must ensure the capability of SG control systems to transmit data to and
from distributed intelligent controllers; moreover, the prospect of the development
of innovative energy market policies imposes the need for the realization of eficient
and reliable widespread bidirectional communication links between SG operators
and customer premises.
In the light of these needs, IEC 61850 provides a standardized framework for
substation automation integration that speciies the following: the communication
requirements, the functional characteristics, the data structure within devices, how
applications interact with and control other devices, and how conformity to the standard should be tested [18,19]. In the IEC 61850 standard, part 7-420, the information
exchange between DER units and monitoring and control devices [20,21] is considered. The standard also deines some measures for the success of an integrated
mobile gateway (MG) communication network design [22]:
• Flexibility to adjust and grow the system topology as requirements change
• Performance, especially quality of service (QoS), to enable effective prioritization among competing applications and to meet critical requirements of
the most important protection and control functions
• Reliability, for critical protection systems, but also because so many different systems are relying on the same infrastructure
The deployment of a low-cost and high-performance communication infrastructure allowing the cooperation of distributed controllers on the basis of
communication and collection of information about distributed energy systems and control commands is a challenge that must be faced by the emerging
technologies [12].
The literature analysis discussed above reveals that the present power system
operation environment is primarily composed of many distributed tools and components, each addressing only a speciic functionality (i.e., energy management, voltage
control, network modeling, security assessment, data acquisition, market interface).
They do not deal with the deinition of an integrated platform capable of easily executing very complex applications, built by composing required functionalities in a
120
Smart Grids
standardized, easy-to-use, and well-deined way. To address this problem, serviceoriented architectures (SOAs), in particular those based on the standard web service
technologies, seem particularly suitable for SG control and monitoring due to their
ease of application programming, portability, maintenance, and integration with
legacy services [6].
5.3
WEB SERVICES–BASED FRAMEWORK FOR SG MANAGEMENT
The proposed framework for SG control, modelling, and monitoring is based on
well-deined interfaces and contracts between services, according to the SOA principles. The interface is deined in a neutral manner that should be independent of the
hardware platform, the operating system, and the programming language in which
the service is implemented. This allows services built on a variety of such systems
to interact with each other in a uniform and universal manner. The main beneits of
an SOA system are the improvement of interoperability, the integration of new and
legacy applications, the ability to survive evolutionary changes in structure, and the
implementation of the internals of each service.
Even though many technologies can be adopted to implement an SOA, for example, the common object request broker architecture (CORBA) and message-oriented
middleware systems, web services technologies are emerging as particularly suited
for ensuring a high degree of integration among existing or newly created services
available on the web. Web services, in fact, extend the advantages of software components, making it possible to employ an existing low-level middleware infrastructure
based on web servers and hypertext transfer protocol (http). From a programming
point of view, a web service is a set of operations that can be easily accessed independently of the service deployment details.
The World Wide Web Consortium (W3C) is standardizing many protocols and
languages for supporting web services technologies. At low level, simple object
access protocol (SOAP) represents the exchange format based on XML for data and
speciic data types in a speciic web service call. This protocol is used over the http
to issue a call to a remote service whose interface is described through web service description language (WSDL), an XML-based language that is more dynamic
and lexible than interface description language (IDL), which characterizes remote
method invocation (RMI)-based middleware such as CORBA.
Figure 5.1 shows the overall distributed meta-architecture for SG monitoring and
control. The core component of the meta-architecture is the SG engine, which is
responsible for the execution of SG control, modeling, and monitoring functions in
a geographically distributed scenario. It includes high-level components, mainly the
submission service, the operation service, and the notiication service.
The submission service is responsible for the handling of user submission
requests and is designed to simplify the submission phase performed by a nonexpert network operator. The SG operation service is responsible for the execution of
SG control and monitoring. Finally, the notiication service is responsible for the
asynchronous management of output data of an application able to notify speciic
events. A basic and fundamental component of the SG engine is the worklow enactor, which manages the execution of the SG service and adds some functionalities
A Distributed Framework for Smart Grid Modeling, Monitoring, and Control 121
Intelligent electronic device 1
Field energy meter n
Intelligent electronic device k
SCADA 1
Field energy
meter 1
SCADA m
.....
.....
.....
WAN
Utility j
Utility 1
GISinterfaceWS
DataAcquisitionWS
DataStorageWS
System
operator
http/IP interactions
Field data acquisition
State estimation
eAssessmentWS
RBDMS
Security analysis
Data storage
Workflow engine
Advanced dispatching
XML-based
database
Monitoring
Computational engine
Smart grid engine
RBDMS
ComputationalWS
Computational grid
FIGURE 5.1 Overall web services–based architecture.
related to the speciic monitoring application. Here the worklow enactor is traditionally described using its business logic description, and it is written in a certain
worklow language.
The WSDL interfaces of the three kinds of web services deined above are the
following:
1. GISinterfaceWS: for georeferential information acquisition
2. DataAcquisitionWS: for real-time data acquisition
3. eAssessmentWS: for distributed generation/storage unit capability
assessment
4. ComputationalWS: to process the mathematical computations required
by the SG control and monitoring functions
5. DataStorageWS: related to the data storage functionalities
In Figure 5.1, GISInterfaceWS and DataAcquisitionWS are abstracted by the
generic DataExchangeWS interface, whereas eAssessmentWS can be bound to the
eAssessment module shown in the igure.
122
Smart Grids
5.3.1
gIsInterFaceWs
GISinterfaceWS was deined for the acquisition of georeferentiated information
on the SG. It delivers the following services:
• GetNetworkdata: acquires SG structural data concerning the geographical map, the network topology, and the sources of power supply
• GetControlDevicesAsset: type, location, and characteristic parameters of the available control and monitoring devices
• GetElectricalNetworkDetails: line parameters, power transformer rating and numbers, impedance values, bus bar scheme, and circuit
breaker type and installation
• GetOperationalParameters: substation equipment status, feeder
breakdowns, failure of distribution transformers, tripping on feeders/lines,
consumer outages
5.3.2
dataacquIsItIonWs
DataAcquisitionWS was deined for the real-time acquisition of the entire set of
measurable SG ield data.
This comprises, in particular, the SG topology, the available measurements of
the electrical nodes parameters, and the actual state of the generator/storage units
computed locally by intelligent electronic devices (IEDs). All possible kinds of measurable parameters are well deined; for each kind of parameter, a code and a basic
measurement unit are deined according to IEC 61850-7-420 [20,21]. This standard
deines the communication and control interfaces for all SG devices, and develops
DER object models. The main beneits of using this standard are:
•
•
•
•
Consistent data models
Easy maintenance of data models
Sharing interoperability of communication based on IEC 61850
Seamless integration into the station automation and the power control
system
Each measurement value returned by the DataAcquisitionWS is correlated to
a time stamp, which indicates the time at which the value was acquired.
The returned type by the described portType is a string that corresponds to the
uniform resource locator (URL) of the ile containing the required information,
which can be downloaded at a convenient time.
5.3.3
eassessmentWs
A reliable assessment of the actual and future generator/storage unit capabilities
appears to be crucial for reliable SG operation.
This demands the design of mathematical models able to predict—given the actual
unit state and the forecasted environmental conditions—the evolution of the supply
A Distributed Framework for Smart Grid Modeling, Monitoring, and Control 123
capabilities [23]. These predictive models should also exhibit adaptive features, to
deal with the time-varying phenomena affecting the generation/storage unit performances (i.e., aging, parameter drifts) and low computational requirements in order
to ensure effective hardware implementation.
To address this problem, the proposed framework integrates a network of distributed IEDs to dynamically assess the supply capability of the main SG components.
These IEDs are fully managed by a dedicated web service.
This web service delivers the getCapabilitydata portType, which acquires
the georeferentiated input data for the capability calculations by interfacing with
dedicated web-based public services (forecasted environmental variables), and the
submitCapabilitydata portType, which submits the right input data to the
distributed IEDs for the capability calculations.
5.3.4
comPutatIonalWs
To support the SG’s monitoring and control functions, intensive online network computations are required. These comprise system state estimation, optimal power low
studies (i.e., voltage control), security assessment (i.e., contingency analysis), and
economic analysis (i.e., the generators must be dispatched based on an economic
assessment of fuel cost, electric power cost, weather conditions, and anticipated process operation).
The online solution of these tasks is computer-intensive activity, especially in
the presence of a large number of DGSs, the extension of electrical networks, and
the depth of analysis. So, to address these computationally demanding analyses in
useful execution times, parallel algorithms to be executed in a distributed environment have to be adopted [8,24]. In this context, the computational engine has to
ensure high reliability, lexibility, and scalability. A practical solution is to exploit
the interesting features offered by cloud computing providers. A cloud system is able
to provide several services at different abstraction levels: infrastructure as a service
(IaaS), platform as a service (PaaS), and software as a service (SaaS). By exploiting
these services, customers can satisfy every kind of information and communications
technology (ICT) need by simply exploiting a high-bandwidth network connection.
In particular, the availability of a large number of resources in cloud systems and
sophisticated software infrastructures for handling the available hardware makes it
possible to ensure high levels of scalability, reliability, and performance. Therefore,
unlike the approaches used in the past for solving similar problems, such as middleware for clusters of computers or geographically dispersed computers networked
through the Internet, by using cloud systems, computational resources can be handled by exploiting the features of the PaaS level of the cloud stack, where speciic
middleware can beneit from the security, scalability, reliability, and performance of
the overall infrastructure. This kind of middleware is abstracted in our approach by
the speciic implementation of the JobSubmissionWS interface, can be adopted
by the SG engine to execute the computations supporting the monitoring services,
and so can be easily integrated into the overall architecture.
The JobSubmissionWS interface exports a set of brokering functionalities
that are related to the interaction with a computational engine and are necessary for
124
Smart Grids
the submission of an application and the management of execution results in both
asynchronous and synchronous ways. In particular, following the grid broker pattern
proposed in [25], the following portTypes are deined:
1. setJob. This is used to deine all the necessary information for the application execution (which includes information on the execution code and on
the input data ile), information on the application structure and its execution
requirements in relation to the resource coniguration (such as the operating
system, a particular run-time environment, library, or transport communication layer, etc.), to QoS parameters, such as the maximum execution time,
and, in the case of an economics-driven grid system, the maximum budget
that can be spent. A call to the setJob portType returns an application ID,
which can be used successively to perform actions on it as required. For
generalization purposes, we considered the application setup made through
a compressed ile, structured in a well-deined way, and understandable by
the speciic underlying grid-computing middleware. This ile will typically
contain source and executable iles or both, input data iles application, and
QoS requirement description iles.
2. submitJob. This is used to verify that the computational engine will
be effectively able to execute the application, respecting all the necessary
hardware, software, and QoS requirements.
3. discardJob. When the submitter does not judge the execution offer of
the computational engine to be suficient, he or she can remove it from the
list of submitted applications.
4. executeJob. This is used to request the execution start of the application.
5. statusMonitor. This is used to continuously verify online the status
of the application execution, such as information on the state of the overall
execution (ready, running, waiting, ended, suspended, etc.).
6. collectResults. This is used for synchronous waiting for the normal
execution termination, in order to receive contextual information on the
output data.
7. terminateJob. This is used to abruptly terminate the application
execution.
5.3.5
datastorageWs
Storage systems are required to permanently store alarm conditions that can be
analyzed off-line, for example, by using a web browser or a dedicated application.
Since they can be adopted as a reference in network planning or as a knowledge
base for expert systems based on decision support systems, these historical data are
extremely useful in power system analysis. In order to be compatible with the rest
of the framework, these storage systems have to export a DataStorageWS WSDL
interface. So, a standardized and convenient method of accessing and managing distributed and heterogeneous data storage resources, such as a remotely accessed realtime database management system (RDBMS) or XML repository, must be deined.
A Distributed Framework for Smart Grid Modeling, Monitoring, and Control 125
5.3.6
WorkFloW enactor
The worklow enactor composes and integrates web services that deliver different
functionalities, related mainly to (1) real-time electrical data acquisition from different sources; (2) supply of the high computational power to develop SG monitoring and control functions; and (3) storage functionalities to handle huge amounts
of data.
The worklow enactor handles the execution request of the monitoring/control
service described in a certain worklow language, which describes the temporal
composition of the web services delivering the basic functionalities, submitted
by means of the submission service. Figure 5.2 shows the action sequence performed by the worklow, referring to a single execution of the monitoring/control
application [8].
Typically, the monitoring/control application has to be executed continuously,
requiring for each execution the acquisition of current data on the SG’s state, particularly the following:
• Phase 1. getNetworkTopology and 2. getControlAsset are performed to obtain the necessary SG description and the characteristic data
of the available control and monitoring devices.
• Phase 3. getCapabilitydata and 4. submitCapabilitydata are
performed to acquire the input data for the generator/storage supply capability calculations, to arrange these data for each IED as a function of its
geographical position, and to submit the corresponding information to the
distributed IEDs.
• Phase 5. getNodesParameters is performed to obtain the real-time
data of each node and is made to interact with the service implementing
the DataAcquisitionWS interface, typically delivered by the electrical
network for analysis.
DataAcquisitionWS
Operator
Submission
Workflow
engine
Monitoring
Computational engine
3. setJob
4. Submit
5. collectResults
JobSubmissionWS
FIGURE 5.2 Main actions of the worklow enactor.
6. Update
DataStorageWS
Sensors data acquisition
Notification
Data
storage
Submit
Internet-http/IP interactions
1. getNetworkTopology
2. getNodesParameters
126
Smart Grids
• Phase 6. setJob, 7. submit, and 8. collectResults are performed to
request the execution of the control and monitoring computations by a computational engine, and, in particular, are made to interact with a service
implementing the JobSubmissionWS interface exported by a grid computing system. These computations comprise: (1) the SG’s state estimation,
computed by processing the network data acquired and solving the system
state equations; (2) the checking of the distribution network operational
constraints; (3) the assessment of the system’s security and reliability levels; and (4) the control and regulation functions of the local SG (i.e., power
factor, frequency and voltage control, and load management).
Finally, Phase 9, update, is executed to save the output data of the analysis in a data
storage system interfaced through a service implementing the DataStorageWS
interface.
5.4
EXPERIMENTAL SETTING
In the following section, the implementation details of the basic web services and the
worklow enactor for SG modeling and control are presented.
5.4.1
data acquIsItIon WeB servIce
Field data can be acquired by means of geographically dispersed resources (i.e.,
SCADA systems, IED, and ield energy meters [FEM]). A prototype web service
implementing the DataAcquisitionWS interface considering the FEM units
located on each SG unit (load/generation/storage), and the direct interaction with
them for real-time acquisition of the electrical parameters, was developed. In particular, a network of distributed FEMs based on ION 7330-7600 units was considered. They can measure demand on any instantaneous value and record peak
(maximum) and minimum demand with a date and time stamp per second. The units
are equipped with an onboard web server, which supports the XML format and the
protocols ION, Modbus transmission control protocol (TCP), and Telnet, for their
full remote control. The FEMs were physically connected to a fast ethernet local
area network (LAN), conigured with an Internet protocol (IP) address. The authors
prototyped the interaction through the http protocol to access, in particular, an XML
ile, called realtime.xml, which contains the most recent measured ield data.
To assess the supply capability curves of the SG components, a prototype version
of an IED based on the ZWORLD-BL2100 unit (an advanced single-board computer
that incorporates the Rabbit 2000 microprocessor) was adopted. These IEDs are
installed directly on the SG components. They allow us to continuously monitor
their state and to dynamically predict the corresponding supply capability curves as
a function of the forecasted climatic variable proiles. These functionalities are fully
available via an IP-based connection [23].
The authors are currently developing another implementation of the
DataAcquisitionWS interface, considering the interaction with proprietary
SCADAs. This new functionality will allow the integration of all the information
A Distributed Framework for Smart Grid Modeling, Monitoring, and Control 127
coming from existing distributed measurement systems, and so decrease the response
time of the service.
5.4.2
JoB suBmIssIon WeB servIce
The irst implementation of the JobSubmissionWS was based on the hierarchical metacomputer middleware (HiMM), version 1.1, developed at the University of
Sannio [26]. The main features of HiMM that are useful in this context are (1) the
virtual hierarchical topology, which, allowing applications to exploit dedicated networks or clusters that are not directly accessible by the Internet, makes the system
highly scalable; and (2) the support for the object-oriented programming paradigm
and, in particular, for the Java language, which simpliies application development
and easy integration with web technologies.
In addition, HiMM supports a grid broker service, which allows an application
to exploit computing resources in a simpliied way, hiding the complexity of the
underlying system. The grid broker delivers a speciic service to easily submit a
master/slave application by using its nearly sequential version and some information necessary for deployment. It simpliies distributed programming, since it supports the separation of concerns related to functional (algorithmic issues, such as
the creation of high-level domain-dependent abstractions in the form of objects and
classes) and system (lower-level tasks such as object distribution, mapping, and load
balancing) aspects of programming. The grid broker is responsible in particular for
resource discovery and resource selection on the basis of QoS parameters speciied
by a user (such as the desired deadline and the maximum budget), task mapping on
the selected resources, and task scheduling according to the hierarchical master/
slave parallel model. In order to submit an application, the following information is
required: (1) application code; (2) input data; and (3) application description and QoS
requirements. For (3), descriptors based on XML called job description format (JDF)
and user requirements description format (URDF) are to be deined.
5.4.3
comPutatIonal WeB servIce
The computational grid used for performing the online computations was deployed
on a test bed organized according to a logical hierarchical and heterogeneous topology of two levels, as shown in Figure 5.3 [27]. The irst level consists of a set of
directly accessible heterogeneous resources composed of eight computers, and two
front-end computers that were used as masters to coordinate the computations of
two pools of computers located in two different buildings and representing the second level of the network. The irst pool consists of a cluster of workstations (COW)
composed of eight homogeneous computers interconnected by a fast ethernet LAN
through a hub 3Com TP16C. The second pool consists of a network of eight heterogeneous workstations (NOW), interconnected by a fast ethernet LAN through
a Switch HP Procurve 2524. The irst-level machines are interconnected by a fast
ethernet LAN.
The mathematical solution engine adopted to support the online computations is
based on a JavaRuntime Environment equipped with speciic power systems analysis
128
Smart Grids
Servlet container
HiMM broker (R2)
Client
Level 1
Fast ethernet LAN
COW front end
(R2)
NOW front end
(R2)
COW
NOW
Node (R6)
Node (R2)
Node (R1)
Node (R1)
Node (R6)
Node (R3)
Node (R1)
Node (R1)
Node (R6)
Node (R2)
Node (R6)
Node (R3)
Node (R8)
Node (R4)
Node (R1)
Node (R7)
Node (R1)
Node (R4)
Node (R8)
Node (R5)
Node (R1)
Node (R1)
Node (R4)
Node (R8)
Level 2
FIGURE 5.3 Computational grid adopted for online MG monitoring and control.
features. It allows system operators to investigate steady-state and dynamic issues in
electrical networks by performing load low analysis, time domain simulations, and
optimal economic dispatch.
Various control and monitoring functions could be executed in parallel by the
computational web service. It adopts a brokering service, based on an economydriven model, to satisfy the QoS constraints speciied by the user (i.e., a time deadline to solve a speciic computation). By using an economy-driven model, the broker
is able to (1) identify the optimal asset of the computational resources needed to satisfy the user requirements (i.e., resource discovery and resource selection process);
(2) transparently split a sequential object-oriented task (i.e., a set of contingencies)
into subtasks (i.e., a subset of contingencies) according to a hierarchical master/slave
computing model; and (3) automatically distribute subtasks to the set of selected
computational resources.
In order to evaluate the parallel processing performances of the computational
web service, many experimental tests were executed by varying the number of
machines used for the computational engine in order to evaluate its speedup factor.
The computation time and the speedup obtained by increasing the slaves in powers
of two (two, four, and eight slaves) were evaluated. Each computation was performed
several times in order to use the average. The speedup obtained by using a two-level
architecture is nearly linearly related to the number of slave machines and is very
close to the ideal case.
A Distributed Framework for Smart Grid Modeling, Monitoring, and Control 129
5.4.4
data storage WeB servIce
In this irst phase, the authors used a relation database, installed on a Linux workstation and based on the Postgres RDBMS.
5.4.5
WorkFloW enactor
The main implementation issue related to the worklow enactor is the choice of a
worklow language and its related technologies, which can be usefully adopted to
describe the wireless asset monitoring (WAM) application in the context of a web
services–based approach. Many proposals have been made in the literature, especially by the major web services software providers, and many of them are continuously evolving. In these proposals the application worklow is usually described with
a textual scripting language (typically the XML format) that is used to provide the
“glue” that links involved services together.
Among the possible solutions, we chose business process execution language
(BPEL). BPEL is a programming abstraction that allows developers to compose
multiple discrete web services into an end-to-end process low. At the present time
it could be considered the most complete, and has risen in popularity since many
vendors are developing their toolkits, many of which are also becoming commercial.
Among the available BPEL engines, we chose to integrate in our framework the
worklow enactor delivered by BPWS4J, since it is free and suficiently stable.
The BPWS4J toolkit includes a worklow execution platform, a tool that validates
BPEL4J documents, and an Eclipse plug-in that provides a simple editor for creating and modifying BPEL4J iles. In particular, the worklow enactor of our platform
is based on the worklow execution of BPEL4J integrated in Apache Tomcat, version 5.0.24, and is specialized to process the online power systems security analysis
(OPSSA) worklow.
For each process, the BPWS4J engine takes in a BPEL document that describes
the process worklow to be executed, a WSDL document (without binding information) that describes the interface that the process will present to clients, and WSDL
documents that describe the services that the process may or will invoke during its
execution. From this information, the process is made available as a web service
with a SOAP interface. The SG worklow describes the composition of the web
services involved, where each service is deined as a <partner>. Service composition in the SG worklow is deined using a sequence tag, which deines how the
partners will be sequentially executed. The speciication of a sequence includes
deinitions of the input message (the receive tag), of the service invocation (the
invoke tag), and of the output message (the reply tag). The assign tag establishes the relationship between the output message of a service and the input message of the service in the worklow.
Similarly to the individual web services, which are described through the
corresponding WSDL ile, the SG web service is also described in a WSDL ile
(Figure 5.4). Messages, portTypes, and operations are also deined, and there is an
additional section (serviceLinkType tag) to deine the role of each service
during their interaction.
130
Smart Grids
<definitions ….
<message name="StringMessageType">
<part name="echoString" type="xsd:string"/>
</message>
<wsdl:message name="OPSSARequest">
<wsdl:part name="in0" type="xsd:string"/>
</wsdl:message>
<wsdl:message name="OPSSAResponse">
<wsdl:part name="setJobReturn" type="xsd:int"/>
</wsdl:message>
<portType name="OPSSAService">
<operation name="execOPSSA" parameterOrder="in0">
<input name="OPSSARequest"
message="tns:StringMessageType"/>
<output name="OPSSAResponse”
message="tns:StringMessageType"/>
</operation>
</portType>
.…..
</definitions>
FIGURE 5.4 WAM WSDL ile.
The web services implemented, in particular the data acquisition and the job submission web services, were written using the Java language and the Apache Axis
toolkit. The Axis toolkit essentially included an SOAP engine, that is, a framework
for constructing SOAP processors such as clients and servers, a simple stand-alone
server, a server that plugs into servlet engines such as Apache Tomcat, and extensive
support for WSDL.
5.5
CONCLUSIONS
The chapter proposed an advanced web services–based framework for integrated
SG control, modelling, and monitoring. The core component of the proposed framework is the SG engine, which is responsible for the execution of the SG management
functions in a geographically distributed scenario. It includes a network of remotely
controlled units distributed in the most critical SG sections for ield data acquisition
and advanced protective functionalities; a GIS interface for the acquisition of georeferentiated information on the SG; a grid computing–based solution engine for the
online computations; and a web-based interface for graphical synoptic and reporting
development.
The web services–based framework has been deployed on a computational grid.
The irst experimental results have emphasized the important role of web services
technology in SG control, modelling, and monitoring.
The application of this technology is very promising, since it makes the proposed
framework platform-, language-, and vendor-independent, and thus an ideal candidate for an effective integration in existing energy and distribution management
systems (EMS/DMS).
The future directions of our research are oriented toward the conceptualization of
a fully decentralized nonhierarchical architecture for SG monitoring and control that
A Distributed Framework for Smart Grid Modeling, Monitoring, and Control 131
will move away from the older centralized paradigm to a system distributed in the
ield with an increasing pervasion of intelligence devices (smart sensors). The adoption of smart sensors is expected to lead to a more eficient task distribution among
the monitoring system resources and, consequently, to signiicantly less demand on
centralized computing resources.
REFERENCES
1. S. M. Amin and B. F. Wollenberg, Toward a smart grid, IEEE Power and Energy
Magazine, 3(5), 34–38, 2005.
2. A. Schroder, I. Laresgoiti, K. Werlen, B. Schowe-von der Brelie and A. Schnettler,
Intelligent self-describing power grids, in Proceedings of the 20th International
Conference and Exhibition on Electricity Distribution (CIRED), 8–11 June 2009,
Prague, Czech Republic.
3. R. L. King, Information services for smart grids, in Proceedings of 2008 IEEE Power
and Energy Society General Meeting—Conversion and Delivery of Electrical Energy in
the 21st Century, 20–24 July 2008, Pittsburgh, PA.
4. N. Hadjsaid, J.-F. Canard and F. Dumas, Dispersed generation impact on distribution
networks, IEEE Computer Applications in Power, 12(2), 22–28, 1999.
5. R. Comfort, A. Mansoor and A. Sundaram, Power quality impact of distributed generation: Effect on steady state voltage regulation. Presented at the PQA 2001 North
America Conference, 2001, Pittsburgh, PA.
6. J. Zhu, Web services provide the power to integrate, IEEE Power and Energy Magazine,
1(6), 40–49, 2003.
7. Q. Liu, J. Wen and H. Sun, Implementing power system management via semantic web
services composition, in Lecture Notes in Computer Science, Springer, Berlin, 2009.
8. Q. Morante, N. Ranaldo and E. Zimeo, Web services worklow for power system security assessment, in Proceedings of IEEE International Conference on e-Technology,
e-Commerce, e-Service, IEEE Computer Society Press, 29 March–1 April 2005, Hong
Kong.
9. J. R. Roncero, Integration is key to smart grid management, in IET-CIRED Seminar on
SmartGrids for Distribution, 23–24 June 2008, Frankfurt.
10. H. Johal, W. Ren, Y. Pan and M. Krok, An integrated approach for controlling and optimizing the operation of a power distribution system, in 2010 IEEE PES Innovative
Smart Grid Technologies Conference Europe (ISGT Europe), Institute of Electrical and
Electronics Engineers, 11–13 October 2010, Gothenburg.
11. B. Ramachandran, S. K. Srivastava, D. A. Cartes and C. S. Edrington, Distributed
energy resource management in a smart grid by risk based auction strategy for proit
maximization, in Power and Energy Society General Meeting, IEEE, 25–29 July 2010,
Minneapolis, MN.
12. A. Mahsoom, O. Olusola and L. Jonathan, Exploring control systems to implement
SMART GRID solutions, in Proceedings of 10th IET International Conference on
Developments in Power System Protection (DPSP 2010), 29 March–1 April 2010,
Manchester, UK.
13. T. M. Overman and R. W. Sackman, High assurance smart grid: Smart grid control
systems communications architecture, in 2010 First IEEE International Conference
on Smart Grid Communications (SmartGridComm), 4–6 October 2010, Gaithersburg,
MD.
14. M. Pipattanasomporn, H. Feroze and S. Rahman, Multi-agent systems in a distributed
smart grid: Design and implementation, in IEEE/PES Power Systems Conference and
Exposition, PSCE ‘09, 15–18 March 2009, Seattle, WA.
132
Smart Grids
15. V. M. Catterson, E. M. Davidson and S. D. J. McArthur, Agents for active network management and condition monitoring in the smart grid, in First International Workshop on
Agent Technologies for Energy Systems, 11 May 2010, Toronto, ON.
16. F. Li, P. Zhang and N. Bhatt, Next generation monitoring and control functions for
future control centers, in IEEE PES General Meeting, 26–30 July 2009, Calgary, AB.
17. J. G. Cupp and M. E. Beehler, Implementing smart grid communications, TECHBriefs,
No. 4, 5–8, 2008.
18. D. Baigent, M. Adamiak and R. Mackiewicz, IEC 61850 Communication Networks and
Systems in Substations: An Overview for Users, SIPSEP, Monterrey, Mexico, 2004.
19. A. Apostolov, C. Brunner and K. Clinard, Use of IEC 61850 object models for power
systems quality/security data exchange, in Quality and Security of Electric Power
Delivery Systems. CIGRE/IEEE PES International Symposium, pp. 155–164, 8–10
October, Montreal, Canada, 2003.
20. M. C. Janssen, C. A. Andersen, C. Brunner and J. Dall, Modelling of DER schedules using IEC 61850, in Proceedings of 20th International Conference on Electricity
Distribution, 8–11 June 2009, Prague, Czech Republic.
21. C. Brunner, IEC 61850 for power system communication, in Proceedings of IEEE/PES
Transmission and Distribution Conference and Exposition, 21–24 April 2008, Chicago,
IL.
22. P. Lloret, J. L. Velasquez, L. Molas-Balada, R. Villafaila, A. Sumper and S. GalceranArellano, IEC 61850 as a lexible tool for electrical systems monitoring, in Proceedings
of the 9th International Conference on Electrical Power Quality and Utilisation, EPQU,
9–11 October 2007, Barcelona, Spain.
23. A. Vaccaro and D. Villacci, Prototyping of a distributed JAVA based client/server architecture for e-Assessment of power equipments loading capability, in Proceedings of
IEEE-Powertech, 22–26 June 2003, Bologna, Italy.
24. M. Di Santo, N. Ranaldo, A. Vaccaro and E. Zimeo, Java-based distributed architectures
for intensive computations in electrical grids, in Proceedings of International Workshop
on Java for Parallel and Distributed Computing, April 2004, Santa Fe, NM.
25. M. Irving, G. Taylor and P. Hobson, Plug in to grid computing, IEEE Power and Energy
Magazine, 2(2), 40–44, 2004.
26. M. Di Santo, N. Ranaldo and E. Zimeo, A broker architecture for object-oriented master/slave computing in a hierarchical grid system, in Parallel Computing, September
2003, Dresden, Germany.
27. Q. Morante, N. Ranaldo, A. Vaccaro and E. Zimeo, Pervasive grid for large-scale power
systems contingency analysis, IEEE Transactions on Industrial Informatics, 2(3),
165–175, 2006.
6
Role of PLC Technology
in Smart Grid
Communication
Networks
Angeliki M. Sarai, Artemis C. Voulkidis,
Spiros Livieratos, and Panayotis G. Cottis
CONTENTS
6.1
6.2
Introduction .................................................................................................. 133
Overview of Smart Grid Networks ............................................................... 135
6.2.1 Energy-Related Applications/Services ............................................. 136
6.2.2 Applications/Services Related to Smart Grid Communications ...... 136
6.2.3 IT Applications/Services .................................................................. 138
6.3 Role of PLC in the Smart Grid Communications Infrastructure ................. 139
6.3.1 PLC in CPNs..................................................................................... 140
6.3.2 PLC in NANs.................................................................................... 141
6.3.3 PLC in FANs .................................................................................... 142
6.3.4 PLC in SNs ....................................................................................... 143
6.4 Aggregation and Power Control in SGC Networks Employing PLC ........... 143
6.4.1 Aggregation of PLC Trafic .............................................................. 143
6.4.2 Power Control at PLC Information Generators ................................ 144
6.5 Design Guidelines for Smart Grid Implementations .................................... 145
6.6 Conclusions ................................................................................................... 146
References .............................................................................................................. 147
6.1 INTRODUCTION
Although the demand for electric power is increasingly changing, the dimensioning and speciications of electric power systems are greatly inluenced by statistical
studies and historical proiles of energy demand. The biggest part of the power grid
was engineered several decades ago based on the operational conditions and the
technological options of that time. Since then, decisive changes in both the size and
proile of electricity demand have taken place that necessitate the upgrading of major
operational modules or parts of the equipment and, also, changes in the switching
functionalities of the power grid with regard to load redistribution and balancing.
133
134
Smart Grids
These important issues can be addressed either by applying proper strategies to manage power demand or by appropriately incorporating distributed energy resources
(DERs). Furthermore, the mandatory use of smart energy meters combined with
lexible pricing tools can mitigate the cost of congestion by limiting the peak load
where the electric energy is delivered. At the same time, demand control from the
consumption side should be enabled. According to the European Union’s 20-20-20
smart grid (SG) target toward the low-carbon economy, a 20% reduction in CO2
emissions, an increase of the share of renewable sources to 20%, and 20% lower
energy consumption are the targets for 2020 [1].
Thus, a major challenge for utilities is to provide reliable and eficient SG services that will enable seamless integration of DERs and support of smart metering applications, while enhancing power quality through sophisticated distribution
automation (DA) services. The SG constitutes the modern approach to monitoring
and controlling the power grid. As such, the SG can be regarded as an information and communication technology (ICT)-enhanced electric power network,
which can intelligently integrate/incorporate the actions of all related stakeholders, namely, the generators, distributors, and consumers of energy. As the main
target of the SG is to deliver electrical power in a sustainable, economical, and
secure way, the development of reliable and eficient data transmission schemes is
imperative.
Recent developments in SG communications (SGCs) technologies address the
crucial objectives of reliability, state awareness, and inancial viability of the SG
stakeholders, producers/operators/customers, and society as a whole. These technologies employ wireless and wired options that are usually integrated and operate interchangeably to form hybrid communications platforms that serve the SG.
The SGC networks (SGNs) support the two-way transfer of SG data employing a
wide range of telecommunication technologies. Indicative examples encountered in
SG implementations are cellular technologies [2], power line communication (PLC)
options [3], Wi-Fi [4], and optical technology [5]. The resulting hybrid communications platform, usually consisting of multiple, sometimes superposed, networks,
should have suficient capacity to accommodate the quality of service (QoS)-enabled
SG trafic and guarantee an all-Internet protocol (IP), secure, two-way, end-to-end
transfer of the SG data.
A critical question concerning SGCs is which transmission medium should be
employed each time. As the SG is composed of a variety of constituent networks,
namely, high-speed enterprise networks, ield area networks (FANs), substation networks (SNs), and customer premises networks (CPNs), the answer is not unique.
The present chapter explores the role of the PLC transmission medium in the hybrid
communications platform employed by SGNs and proposes relevant techniques. The
recently standardized broadband over power lines (BB-PLC) networks [6,7] aim
at effectively serving QoS-enabled trafic. The ubiquitous presence of the power
grid, along with the easy and scalable deployment of the PLC SGNs over the power
grid [8], renders BB-PLC the only broadband access technology readily available to
remote or rural areas [9]. The above considerations, together with the independence
of utilities from a third-party communications provider, render PLC technology a
very promising SG option, especially in the medium-voltage (MV) grid.
135
Role of PLC Technology in Smart Grid Communication Networks
The rest of the chapter is organized as follows. In Section 6.2, SG networks are
outlined, considering the relevant energy, communications, and information technology (IT) aspects. In Section 6.3, the perspectives of SGCs are analyzed and the role
of the PLC is examined in the framework of the various networking options arising
in SGNs. In Section 6.4, PLC transmission techniques are examined in their attempt
to handle QoS-enabled SG data. The relevant design guidelines are given in Section
6.5. Finally, conclusions are drawn in Section 6.6.
6.2 OVERVIEW OF SMART GRID NETWORKS
There have been several attempts in the literature to present the fundamental elements
of SGs from various aspects. The authors in [10] present the systems constituting
the SG, namely, the smart infrastructure system, the smart management system, and
the smart protection system. The authors in [11–13] follow the National Institute of
Standards and Technology (NIST) conceptual model [14], depicted in Figure 6.1. Also,
the authors in [15] consider SGs from the IT perspective, presenting reference models related to the application layer. Finally, a layered approach is proposed in [16,17].
This section adopts the Institute of Electrical and Electronics Engineers (IEEE) 2030
[18] interoperability reference model to classify the SGs under three perspectives,
Conceptual model
Operations
Markets
Service
provider
Bulk
generation
Transmission
Distribution
Customer
Secure communication interface
Electrical interface
Domain
FIGURE 6.1 NIST conceptual model of the smart grid. (National Institute of Standards
and Technology, NIST framework and roadmap for smart grid interoperability standards,
release 1.0, January 2010. Available at: http://www.nist.gov/public_affairs/releases/upload/
smartgrid_interoperability_inal.pdf [accessed March 2013].)
136
Smart Grids
namely, (i) the energy perspective, (ii) the communications perspective, and (iii) the IT
perspective, and presents the related fundamental applications/services.
6.2.1
energy-related aPPlIcatIons/servIces
As depicted in Figure 6.1, systems interconnected over the power grid include bulk
generation, the transmission and distribution networks, and the customer premises.
The seamless integration of such entities requires a reliable, two-way low of both
electric power and information. The following challenges should be addressed:
• Bulk generation: A successful power grid operation depends strongly on
the equilibrium between energy production and energy consumption. The
main challenge on the power generation side is to incorporate intermittent
DERs to increase microgrid penetration and enhance the sustainability of
the power grid. However, the generation of DERs exhibits severe luctuations that prohibit their seamless integration into the power grid [19] and
disrupt the equilibrium between energy generation and consumption. It is
expected that the formulation of appropriate prediction models will signiicantly increase the reliability of DERs and, consequently, their use in
generation.
• Transmission and DA: To increase power quality and redundancy, the smart
infrastructure should enable DA services employing sensors and actuators throughout the grid to monitor power production and energy demand
and detect faults and component failures. Also, though the substations are
already supported by supervisory control and data acquisition (SCADA)
automation services, there is still the need to incorporate digital devices that
monitor and control the operation of the power grid in an autonomous way.
• Customer premises: The end users can interact with the SG in various
ways. First, real-time monitoring of power consumption is enabled through
automatic meter reading/advanced metering infrastructure (AMR/AMI)
systems. The operator of the power grid provides demand-response (DR)
applications for both direct and indirect load control, where demand-side
management (DSM) applications notify the customers about excessive
energy demand and provide incentives to reduce consumption in an attempt
to reduce peak load. Moreover, SG customers are potential prosumers; that
is, they not only consume but may also produce energy. In addition to DERs,
new load paradigms such as plug-in hybrid electric vehicles (PHEVs) [20]
are present in the SG, as PHEVs consume energy while charging and return
energy to the power grid when they are fully charged [10].
6.2.2
aPPlIcatIons/servIces related to smart grId communIcatIons
From the SGC point of view, a variety of access technologies and protocols should
be engaged to support a two-way information low to/from the entities appearing in
Figure 6.1. In Figure 6.2, the reference communications model for the SG deined in
the IEEE 2030 standard [18] shows how SGNs act as an intermediate layer between
137
Role of PLC Technology in Smart Grid Communication Networks
End-to-end smart grid communications model
Communications network security layer
Communications network management layer
Utility local
area network
(LAN)
Public Internet
Regional/
metropolitan
area networks
(MAN)
Backbone/core
network
Substation
network
Wide area networks (WAN)
Bulk
generation
Transmission
Backhaul
Customer
premises
Last mile
Distribution
substation network
HAN: Home area
“hot spots”
network
NAN/AMI: Neighborhood
BAN: Business/building
area network/advanced
area network
metering infrastructure
IAN: Industrial area
EAN/AMI: Extended area
Backhaul
network
network/advanced
metering infrastructure
FAN: Field area network Smart
Distribution
meter
Customer
HAN
Nonrenewable
BAN
Substation
Renewable
IAN
Distributed generation/microgrids
Renewable/
microgrids
Smart grid power and electric system layer
FIGURE 6.2 The IEEE 2030-2011 smart grid communication layered model. (IEEE guide
for smart grid interoperability of energy technology and information technology operation
with the electric power system [EPS], end-use applications, and loads, IEEE Std 2030-2011,
pp. 1–126, 10 September 2011.)
the power grid and the network management layer. As shown in Figure 6.2, multiple
networks are employed to support the various QoS requirements of SG services.
Speciically, the SG services are associated with the following types of networks:
• CPNs: CPNs vary according to their size and to the number of serving
devices and can be classiied into home area networks (HANs), building
area networks (BANs), and industrial area networks (IANs). Regardless of
the CPN type, local area network (LAN) technologies are used to provide
connectivity within CPNs that are employed to support applications such as
AMR, remote load control, monitoring and control of DERs, energy consumption management through DSM, and charging of PHEVs.
• Neighborhood area networks (NANs): NANs are employed to collect information generated by CPNs. Usually, a NAN comprises the AMI installed
to collect the AMR data. NANs constitute the last mile of the SGNs, as
they are intermediate between the customer and the distribution grid. The
signiicant components of NANs are their distribution access points (DAPs)
operating as gateways for the trafic collected via the AMI. The DAPs
decide on data aggregation and routing, based on the type and signiicance
of the transferred SG trafic.
138
Smart Grids
• FANs: FANs are deployed over the distribution grid to monitor and control
various ield devices, such as insulators, feeders, and transformers. These
devices are dispersed over the grid to enable DA services and services
related to the seamless integration and management of DERs. As FANs
cover the whole distribution grid, they may also serve for utilities communications, offering communication and monitoring services to the technical
personnel responsible for grid maintenance and day-to-day operation.
• SNs: SNs are highly reliable enterprise LANs used for protection,
monitoring, and automation services. Among them, SG protection and
SCADA services require a maximum delay of 10 ms [21]. Hence, to meet
the QoS requirements of teleprotection services, SNs must usually have a
one-hop point-to-point link to the utility operations and control center.
Table 6.1 provides an overview of the available transmission media along with
their advantages and disadvantages. The options tabulated in Table 6.1 may be
employed in different segments of SGNs.
6.2.3 It aPPlIcatIons/servIces
The IT applications involved in the SG primarily support the management and
control of various procedures and devices dispersed over the grid. The data related
to SG operation come from a plethora of sources either internal or external to the
grid. Internal sources include: (i) devices for substation monitoring, (ii) smart units
attached to various grid nodes such as AMR devices and smart sensors, and (iii)
databases employed by the utilities as well as information originating from human
intervention. Next, the applications and services supported by various segments of
the grid are presented:
• Bulk generation: Applications related to the management of DER generation are supported. Also, generation scheduling is enabled and relevant data
are sent to the various stakeholders of the electric energy market.
• Transmission and distribution grids: Data acquisition and collection from
devices installed both in the ield and in substations are supported all over
the SG. An overview of various QoS requirements for SCADA automation
applications can be found in [21–23] and in the IEEE standard for substation automation, IEEE 1646 [24]. Also, maintenance and day-to-day
activities that process data collected by geographic information services
are performed to guarantee power availability. Regarding the distribution grid, applications related to DER management and integration are
offered. Finally, parts of the grid that connect end users should support data
aggregation for better collection and management of billing data.
• Customer applications/services: Applications related to the seamless integration of DERs must be provided. Also, various energy management systems
and home automation applications may offer real-time monitoring of energy
consumption. Finally, applications accessed through portals may provide data
and services related to energy usage, billing, and customer authorizations.
Role of PLC Technology in Smart Grid Communication Networks
139
TABLE 6.1
Smart Grid Communications Media Overview
Network Access
Technology
Cellular networks
(GPRS/3G/4G)
Wi-Fi (IEEE
802.11)
Ethernet
Data Rates
(Maximum Values)
GPRS
56–115 Kbps
3G
2 Mbps
4G
50–100 Mbps
~100 Mbps
Advantages
Widely adopted for
certain types of smart
grid applications
Easy installation
Reduced cost
100 Mbps–10 Gbps
Easy installation
Reduced cost
WiMAX (IEEE
WiMAX (8062.16 m) Wide coverage
802.16) and LTE 300 Mbps
Scalability
LTE
100 Mbps
Fiber optics
40–1600 Gbps (when High capacity
(SONET/SDH
WDM is employed) Reliability
E/GPON)
Security
DSL
20 Mbps
Easy installation
Scalability
PLC (NB/BB/
BB-PLC
The transmission medium
BPL)
500 Mbps on CPNs
already exists
Third-party independence
WSN/WPAN
ZigBee
Reliability
(IEEE 802.15.4 250 Kbps
Low delay
ZigBee,
DASH7
DASH7)
200 Kbps
6.3
Disadvantages
Offers only periodic
service monitoring
Dependence on third
parties
Limited scalability and
reliability
Limited scalability
Cost related to leasing
spectrum
High cost
Not available in rural
areas
Public access network
Nonuniversal legislation
Adverse transmission
medium
Short coverage
ROLE OF PLC IN THE SMART GRID
COMMUNICATIONS INFRASTRUCTURE
Although various communication options are available to support SGC, PLC is preferable on many occasions for the following reasons:
• The deployment of the PLC segments of hybrid SGC solutions is immediate,
simple, and scalable.
• As far as the PLC segments of SGNs are concerned, data security oficers
(DSOs) do not have to procure telecommunication services from third parties.
• The privacy and independent processing of SG data are guaranteed, as the
DSOs are the owners of the PLC part of SGNs.
• As PLC technology employs the power lines themselves as the physical
transmission medium, it is the only SGC option that offers monitoring
140
Smart Grids
of the operational status of the electric grid by appropriately processing
inherent noise patterns and spurious signals collected by the PLC nodes.
• Alternative technologies to PLC can serve only part of the desired applications, especially in the MV SG. In any case, the PLC nodes may usually
embed any related technology in order to (i) increase their functional and
communication capabilities and (ii) provide compatibility with alternative
communication options or existing or future automation technologies.
• Direct and selective scalable integration of all application requirements of
the SG can be implemented at the level of the grid nodes.
The rest of the section explores the role of the PLC in SGNs of the SG.
6.3.1
Plc In cPns
The CPNs constitute a multiprotocol, multivendor environment [18], in which many
technologies are involved, with PLC and Wi-Fi being the most mature. Other solutions intended for the SG employ Global System for Mobile Communications/
General Packet Radio Service (GSM/GPRS) and ZigBee technologies. An overview
of the communication technologies used in CPNs, along with their respective level
of maturity, is presented in Table 6.2 [25].
In most cases, BB-PLC seems preferable to narrowband (NB)-PLC as it may provide a variety of broadband applications requiring fast Internet access. NB-PLC may
be the solution for transferring low-rate data over large distances along the MV grid
in remote areas. The main efforts to standardize BB-PLC access resulted in the
International Telecommunication Union Telecommunication Standardization Sector
(ITU-T) G.hn [7] and IEEE 1901 [6] protocols summarized in Table 6.3. The operational principles of the two standards are presented in [26–30]. The large amount
of trafic moving through an SGN, along with the necessity to prioritize data with
regard to their importance and/or bandwidth requirements, renders the time division
TABLE 6.2
Communication Technologies Employed in CPNs
Communication
Technologies
PLC
ZigBee
Wi-Fi
WiMAX
GSM/GPRS
DASH7
HAN
IAN
AMR
PHEV
√
√
√
~
√
Δ
√
√
√
~
√
Δ
√
√
√
Δ
Δ
Δ
~
√
Δ
Δ
√
Δ
Source: Usman, A. and Shami, S.D., Renewable and Sustainable Energy Reviews, 19, 191–199, 2013.
Notes: √, in use, some mature solutions available; Δ, not currently in use, solutions can be developed;
~, ongoing research, some solutions available but under testing.
Role of PLC Technology in Smart Grid Communication Networks
141
TABLE 6.3
BB-PLC Technologies Employed in CPNs
IEEE 1901
Frequency band
Maximum supported
data rate
OFDM
FEC
ITU-T G.hn
FFT-PHY
Wavelet-PHY
2–25 MHz
2–50 MHz
2–100 MHz
1 Gbps
2–30 MHz
2–48 MHz
2–60 MHz
545 Mbps
2–28 MHz
2–60 MHz
Windowed
2048, 4096 carriers
FFT
3072, 6144
carriers
Turbo
convolutional
code
BPSK, QPSK,
8-, 16-, 64-,
256-, 1024-,
4096-QAM
CSMA/CA (4
priority levels)
TDMA
Wavelet
512, 1024
carriers
RS, RS-CC,
LDPC
QC-LDPC
Modulation
BPSK, QPSK, 8-, 16-,
64-, 256-, 1024-,
4096-QAM
Contention-based
MAC scheme
Contention-free
MAC scheme
CSMA/CA (4 priority
levels)
TDMA
544 Mbps
BPSK, 4-, 8-,
16-, 32-PAM
CSMA/CA (8
priority levels)
TDMA
Source: Rahman, M.M., Hong, C.S., Lee, S., Lee, J., Razzaque, M.A., and Kim, J.H.,
IEEE Communications Magazine, 49, 183–191, 2011.
multiple access (TDMA) contention-free medium access control (MAC) option preferable to the carrier sense multiple access (CSMA) contention-based option when
reliable SG services are required. On the other hand, noncritical broadband services
employ contention-based access.
6.3.2
Plc In nans
NANs incorporate the AMI components to collect the trafic generated in CPNs.
Figure 6.3 shows an indicative network architecture, where the NANs employ PLC
links. The trafic generated in the CPNs is routed to the network gateways, namely,
the NAN DAPs, which may also serve for data aggregation. Speciically, critical data
are usually routed to the utility premises without being aggregated, whereas data
aggregation is usually applied to noncritical data. Even though only the PLC option
is employed to implement the NANs in Figure 6.3, strict QoS constraints may necessitate a redundant SGC infrastructure in order to support time-critical applications.
The PLC links encountered in CPNs are short, so that BB-PLC may be effectively
used. This is not the case in NANs, where longer links are encountered. Therefore,
due to the lower attenuation in lower frequencies and the low-rate requirements in
142
Smart Grids
Utility operation and
control
Substation
network
Distribution network
Neighborhood area network (NAN)
Field area network (FAN)
Customer premises
network
AMR
DA
AMR
DER
AMR
DA
DA
High-speed wireless backbone
DER
DAP
PLC link
Wireless link
FIGURE 6.3 Indicative heterogeneous SGN architecture.
NANs, NB-PLC is preferable for NANs, especially to support DR applications, as
argued in [31], where the PLC options available are compared with regard to their
applicability in NANs. As to the standardization of PLC access networks, there is
an ongoing process by the IEEE P1901.2 Task Group to standardize NB-PLC [32].
With regard to BB-PLC, although many PLC options are currently available for
CPNs, this is not the case for NANs, since the only broadband option supporting
access networks is IEEE 1901. The BB-PLC NANs may effectively support scalability and redundancy, and, when combined with node clustering approaches [33,34],
they may successfully meet the various QoS constraints of SG applications. Also, the
use of repeaters and the proper design of BB-PLC networks may render the BB-PLC
option capable of supporting the trafic moving through the NANs [9].
In addition to PLC, other options may be employed to support the NAN infrastructure, provided that they guarantee reliable and secure data transmission. As
to private wide area network (WAN) solutions, there is active research on licensed
wireless solutions in an attempt to serve delay-sensitive applications [35–37]. Other
mature solutions include Wi-Fi and GPRS/GSM infrastructures [38].
6.3.3
Plc In Fans
FANs are usually incorporated into NANs, since they are both SGNs deployed
over the MV grid. However, while NANs usually interface with metering devices
dispersed over CPNs, FANs interface with wireless mesh options, that is, IEEE
802.15.4 protocol, necessitating a different communications platform. The authors
in [39] classify FANs as typical cases of low-power and lossy networks (LLNs),
since they exhibit connection losses and unpredictable errors caused by the severe
voltage luctuations and noise caused by the operation of ield devices. Thus, FANs
usually require different QoS handling, employing network layer protocols such as
IPv6 over wireless personal area networks [40,41] to address the issues of end-to-end
connectivity and reliability.
Role of PLC Technology in Smart Grid Communication Networks
143
The low-rate applications supported by FANs render NB-PLC appropriate for
FAN transmission, as claimed by the upcoming IEEE P1901.2 standard. Also,
interoperability with the wireless IEEE 802.15.4 parts is imperative. Relevant key
applications that require sophisticated network approaches to handle the heterogeneous nature of FANs are related to DA [31]. Also, PLC may effectively be used to
support the detection and localization of faults by monitoring the signal-to-noise
ratio (SNR) of the PLC signals [42].
6.3.4
Plc In sns
The SNs are usually the irst SGNs to implement, since they support the most critical
applications realized in the SG. Substation automation services include the remote
monitoring and control of SCADA systems, phase measurement units (PMUs),
remote terminal units (RTUs), and intelligent electronic devices (IEDs). The common characteristic of all these devices is that they generate SG information that needs
to be collected via the backhaul networks, that is, the SN, in real-time synchronized
mode of operation. Since synchronization is a critical issue, global positioning system (GPS) enabling is usually proposed in the literature [24]. As to PLC transmission, both BB-PLC and NB-PLC seem appropriate to serve SN communications,
taking into account that the former option supports low whereas the latter supports
high transmission rates. Other options may include wireless and optical media [43].
6.4
AGGREGATION AND POWER CONTROL
IN SGC NETWORKS EMPLOYING PLC
The extended ICT infrastructure that enhances the traditional power grid with artiicial intelligence gives rise to the generation of SG information. That is, as power
generators generate power, SG entities, either AMR or other SG devices at the customer premises, act as generators of SG information, which, in the framework of
PLC-enabled SGNs, is transferred via the power lines. The huge amount of data produced by the SG information generators necessitates proper management to optimize
their exploitation. Moreover, as the number of SG information generators scales up,
so does the need for proper handling of the resulting data trafic; the pursued facilitation of power grid monitoring comes at the cost of the growing complexity of the
underlying communications architecture. The extreme heterogeneity of SG information generators and the necessity to support a variety of SG information services
increase the need to optimize the SGN operation.
6.4.1
aggregatIon oF Plc traFFIc
Depending on the type of trafic produced by the smart devices and on the associated services and business-to-business (B2B) or business-to-customer (B2C) service
level agreements (SLAs), the SG data trafic may be categorized into several classes.
Certain applications, such as substation automation, require extremely low transmission delay and are classiied as time-critical applications. Other applications,
such as the transmission of AMR data generated in CPNs, which do not require an
144
Smart Grids
immediate response fall into the class of delay-tolerant services. From the data accuracy point of view, applications such as AMR or voice over IP (VoIP) may tolerate
a certain degree of data loss, allowing the application of lossy aggregation techniques, whereas applications such as substation automation require highly accurate
data transmission from the SG data source to the utility’s back ofice, and vice versa.
The various QoS constraints explicitly imposed by the requirements of the
services and applications supported may be properly traded off in the attempt to
increase the eficiency of an SGN and, in certain cases, to increase its service capacity to accommodate more services or applications. In any case, imposing QoS constraints for satisfactory data services provisioning critically affects the operational
characteristics of NANs, FANs, and SNs.
Data aggregation has been considered in the recent literature as a mechanism to
manage the trade-off between communications eficiency and data accuracy [44,45].
Moreover, several recent studies have highlighted data aggregation as a way to preserve anonymity, without reducing the contextual information of the data reported
to the utilities [46]. Since security and privacy are of primary importance for SG
applications, data aggregation seems appealing for application in SGNs.
Although originally employed in the framework of wireless sensor networks
(WSNs), the principles of data aggregation are also valid in the case of SGNs [47]. As
the topology of the distribution grid may be modeled as a tree structure, and WSNs
usually form tree-like structures for the routing of information, many data aggregation protocols destined for WSNs can be adapted to serve SGNs as well.
The design targets determining the application of data aggregation to QoSconstrained SGNs are primarily related to determining (i) which data lows can
or should be aggregated, (ii) where the aggregation points should be located, and
(iii) which aggregation procedure guarantees conformity with the delay-related
QoS constraints of certain SG service classes. Notably, as delay reduction comes
at the cost of bit error rate (BER) deterioration, the above design targets are not
easy to achieve. Indicatively, determining the optimal aggregation tree in WSNs
and SGNs may be reduced to the solution of the well-known minimum spanning tree (MST) problem, which is known to be NP-complete [48]. Also, proper
scheduling of data transmission reporting by the SG devices, which is related to
serving time-critical applications, has been widely studied and has been proven
to be NP-hard [49]. Taking into account the above considerations, determining
the appropriate parameters for SGN operation is a dificult task, relecting the
computational complexity that comes as a side effect of the enhanced monitoring
and automation functionality offered by the SG.
6.4.2
PoWer control at Plc InFormatIon generators
In line with the need to optimize the transmission schedules of the information generators, the recent IEEE P1901 standard proposes two approaches for proper MAC,
one based on the CSMA family of protocols and one implementing TDMA [6]. The
latter establishes the necessity for collision-free transmission periods to (i) alleviate
the negative effects of restoring information from collided packets received from the
SG devices side and (ii) guarantee QoS provisioning for time-critical applications.
Role of PLC Technology in Smart Grid Communication Networks
145
Although general directions toward exploiting collision-free periods are given, exact
scheduling algorithms have not yet been proposed.
To cope with the lack of a standard MAC scheme, several approaches have been
recently proposed [50,51]. These approaches apply techniques originally destined
for wireless transmission. However, in all the schemes proposed, any similarities
of PLC transmission with wireless transmission were not fully exploited; multipath
propagation, severe signal attenuation, and background noise drastically aggravate
the eficiency of PLC transmission and should be taken into account. To this end,
proper control of the injected power spectral density (IPSD) by the SG devices may
offer advantages in enabling frequency reuse through transmission-slot reuse. By
controlling the IPSD of the SG information generators, the distribution grid may
be segmented into independent transmission areas, where simultaneous transmissions are allowed, increasing SGN throughput while reducing the transmission delay.
The merits expected from the employment of power control combined with data
aggregation in the context of tree-like PLC SG structures are presented in [49–52].
Interestingly, apart from the documented communications-related beneits, power
control may also yield signiicant energy savings, which, considering the large numbers of SG information generators employed in SGNs, may prove nontrivial.
In conclusion, the introduction of ICT technologies enables full monitoring and
control of the distribution grid at the cost of increased complexity and reduced eficiency of the necessary communication schemes. However, data aggregation alone
or combined with proper IPSD control of the SG information generators may alleviate the negative effects of increased data trafic, simultaneously guaranteeing proper
QoS provisioning and preserving end-user information anonymity, a feature of great
importance to DSOs.
6.5
DESIGN GUIDELINES FOR SMART GRID IMPLEMENTATIONS
This section presents several basic guidelines that should be followed when designing SG networks. The design steps that should be followed are:
• Proper deinition of the SG requirements: The irst step when implementing
SGs is to deine the relevant requirements according to the DSO speciications. Many DSOs have deployed backbone iber communication networks
to monitor their crucial operations, especially in the high-voltage grid [53];
hence, in many cases, SNs are already deployed. On the other hand, DSOs
have not yet deployed last-mile communication networks to exchange information with the various nodes of the transmission and distribution grids
as well as with the end users. In these cases, the deployment of NANs and
FANs is imperative. Therefore, the current stage of SG penetration in each
case will determine the priorities in SGN deployment.
Despite the speciic needs of DSOs, all SGNs should enable end users’ connectivity with the utilities’ premises through a reliable, secure, highly available, cost-effective, and resilient infrastructure. The need for reliability and
security may lead to the selection of wired options for SGN implementation,
whereas the necessity for optimal management of the cost–resilience trade-off
146
Smart Grids
may lead to the installation of redundant infrastructures or sophisticated PLC
transmission schemes and optimal network planning, as indicated in [54].
• SG services: After the requirements imposed by the DSOs have been speciied and the SGN type has been selected, the SG services must be deined.
There are many reviews on QoS requirements of various SG services
deployed over NANs, FANs, or SNs. As previously discussed in this chapter, SNs incorporate delay-sensitive applications. However, most of these
networks have already been deployed and are properly managed. As far as
DA is considered, the service requirements might vary signiicantly, since
protection services impose strict delay constraints, whereas, on the other
hand, metering services might be more tolerant to delays. In the case of
NANs, the need for security, resilience, and anonymity calls for a robust
and scalable SGN rather than a high-capacity one.
• Communications network support: Directly related to the SG services is
the selection of the SGN transmission medium. To support mission-critical
applications, the implementation of one-hop communication technologies
is imperative; hence, wireless cognitive approaches are preferable. On the
other hand, PLC seems preferable for delay-tolerant applications. The services QoS speciied in the previous step usually leads to the deployment of
hybrid networks. Techniques that offer interoperability or handle the performance trade-offs of the various transmission options are necessary. The
SG interoperability standard [18] deines the interfaces between various SG
entities, but does not handle issues related to data aggregation techniques or
to the interoperability of various communications schemes.
• Design of IT applications: Last but not least is the design of IT applications
that constitute the interface of the SG nodes with the SG. As far as utilities
are concerned, these applications should address the issue of big data; that
is, they should effectively handle the data generated by the SG devices. In
this context, many application-based platforms employing semantic repositories [55] or cloud-based approaches [56] are considered in the literature.
Regardless of the application platform employed, the introduction of novel
aggregation schemes leads to a signiicant reduction of unnecessary network trafic, enhancing the SGN performance.
6.6 CONCLUSIONS
The transition toward smart power grids that support two-way lows of energy and
information is imperative. In this framework, new services are introduced related to
energy systems, communications platforms, and IT applications. The PLC transmission option is a strong candidate to support the various types of SGNs that should
interoperate to offer reliable, end-to-end SG services:
1. BB-PLC gives rise to a high-capacity infrastructure capable of eficiently
handling the trafic generated by CPNs.
2. Both NB-PLC and BB-PLC options may collect the trafic generated by
CPNs and route this to the utilities’ back ofices, thus supporting NANs’
Role of PLC Technology in Smart Grid Communication Networks
147
communication. When combined with data aggregation techniques, PLC
may satisfy an essential prerequisite of the SG, namely, preserving information anonymity. Also, when combined with injected power control, slot
reuse in TDMA-based PLC transmission enhances network performance to
meet the QoS constraints of time-critical applications.
3. NB-PLC seems appropriate to handle the low-rate trafic generated in heterogeneous FANs, especially when transmission of SG information from
remote SGC nodes on the MV grid is required. When combined with the
use of eficient network layer protocols, FANs may support the seamless
integration of wireless and wired segments, increasing network availability.
In conclusion, there is no speciic approach to designing SGNs. SG applications
of variable QoS responding to different DSOs’ requirements may lead to different
SG implementations. The need for resilience, security, reliability, and scalability,
combined with the need for a cost-effective design, will determine how the various
communications networks supporting the SG will be implemented.
REFERENCES
1. European Commission, The EU climate and energy package 2020. Available at: http://
ec.europa.eu/clima/policies/package/index_en.htm.
2. Hochgraf, C., Tripathi, R., Herzberg, S., Smart grid charger for electric vehicles
using existing cellular networks and SMS text messages, in Proceedings of the First
IEEE International Conference on Smart Grid Communications (SmartGridComm),
pp. 167–172, 4–6 October 2010, Gaithersburg, MD.
3. Bumiller, G., Lampe, L., Hrasnica, H., Power line communication networks for
large-scale control and automation systems, IEEE Communications Magazine, 48(4),
106–113, 2010.
4. Raiei, M., Eftekhari, S.M., A practical smart metering using combination of power
line communication (PLC) and WiFi protocols, in Proceedings of 17th Conference on
Electrical Power Distribution Networks (EPDC), pp. 1–5, 2–3 May 2012, Tehran.
5. Aggarwal, A., Kunta, S., Verma, P.K., A proposed communications infrastructure for
the smart grid, in Proceedings of Innovative Smart Grid Technologies (ISGT), pp. 1–5,
19–21 January 2010, Gaithersburg, MD.
6. IEEE P1901, IEEE standard for broadband over power line networks: Medium access
control and physical layer speciications, IEEE Std 1901-2010, pp. 1–1586, December
2010.
7. ITU-T Recommendation G.9960, Uniied high-speed wire-line based home networking
transceivers—Foundation, January 2010.
8. Patel, A., Aparicio, J., Tas, N., Loiacono, M., Rosca, J., Assessing communications technology options for smart grid applications, in Proceedings of IEEE Conference on Smart
Grid Communications, pp. 126–131, 17–20 October 2011, Brussels.
9. Sarai, A., Tsiropoulos, G., Cottis, P., Hybrid wireless-broadband over power lines: A
promising broadband solution in rural areas, IEEE Communications Magazine, 47(11),
140–147, 2009.
10. Fang, X., Misra, S., Xue, G., Yang, D., Smart grid—The new and improved power grid:
A survey, IEEE Communications Surveys and Tutorials, 14(4), 944–980, 2012.
11. Lopes, A.J., Lezama, R., Pineda, R., Model-based systems engineering for smart grids
as systems of systems, Procedia Computer Science, 6, 441–450, 2011.
148
Smart Grids
12. Metke, A.R., Ekl, R.L., Security technology for smart grid networks, IEEE Transactions
on Smart Grid, 1(1), 99–107, 2010.
13. Sadeghi, S., Yaghmaee Moghddam, M.H., Bahekmat, M., Heydari Yazdi, A.S., Modeling
of smart grid trafics using non-preemptive priority queues, in Proceedings of 2012 2nd
Iranian Conference on Smart Grids (ICSG), pp. 1–4, 24–25 May 2012, Tehran, Iran.
14. National Institute of Standards and Technology, NIST framework and roadmap for smart
grid interoperability standards, release 1.0, January 2010. Available at: http://www.
nist.gov/public_affairs/releases/upload/smartgrid_interoperability_inal.pdf (accessed
March 2013).
15. Hwang, T., Choi, M., Kang, S., Lee, I., Design of application-level reference models for
micro energy grid in IT perspective, in Proceedings of 8th International Conference on
Computing and Networking Technology (ICCNT), pp. 180–183, 27–29 August 2012,
Gyeongju.
16. Lo, C., Ansari, N., The progressive smart grid system from both power and communications aspects, IEEE Communications Surveys and Tutorials, 14(3), 799–821, 2012.
17. Leeds, D.J., The smart grid in 2010: Market segments, applications and industry players,
GTM Research, 2009. Available at: http://www.greentechmedia.com/research/report/
smart-grid-in-2010 (accessed March 2013).
18. IEEE Std 2030-2011. IEEE guide for smart grid interoperability of energy technology
and information technology operation with the electric power system (EPS), end-use
applications, and loads, IEEE Std 2030-2011, pp. 1–126, 10 September 2011.
19. Molderink, A., Bakker, V., Bosman, M.G.C., Hurink, J.L., Smit, G.J.M., Management
and control of domestic smart grid technology, IEEE Transactions on Smart Grid, 1(2),
109–119, 2010.
20. IEC, Road vehicles—Vehicle to grid communication interface—Part 1: General information and use-case deinition, ISO/IEC 15118-1, 2010.
21. Deshpande, J.G., Kim, E., Thottan, M., Differentiated services QoS in smart grid communication networks, Bell Labs Technical Journal Wiley Subscription Services, 16(3),
61–81, 2011.
22. Khan, R.H., Khan, J.Y., A comprehensive review of the application characteristics and
trafic requirements of a smart grid communications network, Computer Networks,
57(3), 825–845, 2013.
23. Kansal, P., Bose, A., Bandwidth and latency requirements for smart transmission grid
applications, IEEE Transactions on Smart Grid, 3(3), 1344–1352, 2012.
24. IEEE Std 1646-2004. IEEE standard communication delivery time performance requirements for electric power substation automation, IEEE Std 1646-2004, pp. 0_1–24, 2005.
25. Usman, A., Shami, S.D., Evolution of communication technologies for smart grid applications, Renewable and Sustainable Energy Reviews, 19, 191–199, 2013.
26. Galli, S., Logvinov, O., Recent developments in the standardization of power line communications within the IEEE, IEEE Communications Magazine, 46(7), 64–71, 2008.
27. Rahman, M.M., Hong, C.S., Lee, S., Lee, J., Razzaque, M.A., Kim, J.H., Medium access
control for power line communications: An overview of the IEEE 1901 and ITU-T G.hn
standards, IEEE Communications Magazine, 49(6), 183–191, 2011.
28. Galli, S., Scaglione, A., Wang, Z., For the grid and through the grid: The role of power
line communications in the smart grid, Proceedings of the IEEE, 99(6), 998–1027, 2011.
29. Galli, S., Kurobe, A., Ohura, M., The inter-PHY protocol (IPP): A simple coexistence
protocol for shared media, in Proceedings of IEEE International Symposium on Power
Line Communications and Its Applications, ISPLC, pp. 194–200, 29 March–1 April
2009, Dresden, Germany.
30. Oksman, V., Galli, S., G.hn: The new ITU-T home networking standard, IEEE
Communications Magazine, 47(10), 138–145, 2009.
Role of PLC Technology in Smart Grid Communication Networks
149
31. Galli, S., Scaglione, A., Wang, Z., Power line communications and the smart grid, in
Proceedings of First IEEE International Conference on Smart Grid Communications
(SmartGridComm), pp. 303–308, 4–6 October 2010, Gaithersburg, MD.
32. IEEE, Low frequency narrow-band power line communications. Available at: http://
grouper.ieee.org/groups/1901/2/ (accessed January 2013).
33. Wang, Z., Wang, Y., Wang, J., Overlapping clustering routing algorithm based on L-PLC
meter reading system, in Proceedings of IEEE International Conference on Automation
and Logistics, ICAL ‘09, pp. 1350–1355, 5–7 August 2009, Shenyang, China.
34. Liu, X., Wang, W., Zheng, J., Hai, T., Zhang, L., Liu, B., Tangential connection clustering
routing algorithm for L-PLC based AMR systems, in Proceedings of 7th International
Power Electronics and Motion Control Conference (IPEMC), vol. 4, pp. 2932–2936,
2–5 June 2012, Harbin, China.
35. Yu, R., Zhang, Y., Gjessing, S., Yuen, C., Xie, S., Guizani, M., Cognitive radio based
hierarchical communications infrastructure for smart grid, IEEE Network, 25(5), 6–14,
2011.
36. Vineeta, Thathagar, J.K., Cognitive radio communication architecture in smart
grid reconigurability, in Proceedings of 1st International Conference on Emerging
Technology Trends in Electronics, Communication and Networking (ET2ECN), pp. 1–6,
19–21 December 2012, Surat, Gujarat, India.
37. Yu, R., Zhang, Y., Chen, Y., Hybrid spectrum access in cognitive neighborhood area networks in the smart grid, in IEEE Wireless Communications and Networking Conference
(WCNC), pp. 1478–1483, 1–4 April 2012, Shanghai.
38. Lopez, G., Moura, P.S., Custodio, V., Moreno, J.I., Modeling the neighborhood area
networks of the smart grid, in Proceedings of IEEE International Conference on
Communications (ICC), pp. 3357–3361, 10–15 June 2012, Ottawa, ON.
39. Chauvenet, C., Tourancheau, B., Genon-Catalot, D., Goudet, P.-E., Pouillot, M., A communication stack over PLC for multi physical layer IPv6 networking, in Proceedings of
First IEEE International Conference on Smart Grid Communications (SmartGridComm),
pp. 250–255, 4–6 October 2010, Gaithersburg, MD.
40. Dawson-Haggerty, S., Tavakoli, A., Culler, D., Hydro: A hybrid routing protocol for
low-power and lossy networks, in Proceedings of First IEEE International Conference
on Smart Grid Communications (SmartGridComm), pp. 268–273, 4–6 October 2010,
Gaithersburg, MD.
41. Ko, J., Terzis, A., Dawson-Haggerty, S., Culler, D.E., Hui, J.W., Levis, P., Connecting
low-power and lossy networks to the internet, IEEE Communications Magazine, 49(4),
96–101, 2011.
42. Kapareliotis, E.S., Drakakis, K.E., Dimitriadis, H.P.K., Capsalis, C.N., Fault recognition on power networks via SNR analysis, IEEE Transactions on Power Delivery, 24(4),
2428–2433, 2009.
43. Qureshi, M., Raza, A., Kumar, D., Kim, S.S., Song, U.S., Park, M.W., Jang, H.S., Yang,
H.S., Park, B.S., A survey of communication network paradigms for substation automation, in Proceedings of IEEE International Symposium on Power Line Communications
and Its Applications, ISPLC, pp. 310–315, 2–4 April 2008, Jeju, Jeju Island.
44. Krishnamachari, B., Estrin, D., Wicker, S., The impact of data aggregation in wireless
sensor networks, in Proceedings of the 22nd International Conference on Distributed
Computing Systems Workshops, pp. 575–578, July 2002, Vienna, Austria.
45. Solis, I., Obraczka, K., In-network aggregation trade-offs for data collection in wireless
sensor networks, International Journal of Sensor Networks, 1(3–4), 200–212, 2006.
46. Lu, C., Liang, X., Li, X., Lin, X., Shen, X., EPPA: An eficient and privacy-preserving
aggregation scheme for secure smart grid communications, IEEE Transactions on
Parallel and Distributed Systems, 23(9), 1621–1631, 2012.
150
Smart Grids
47. Gungor, V.C., Lu, B., Hancke, G.P., Opportunities and challenges of wireless sensor networks in smart grid, IEEE Transactions on Industrial Electronics, 57(10), 3557–3564,
2010.
48. Arein, A.S., Kashem Mia, M.A., NP-completeness of the minimum edge-ranking
spanning tree problem on series-parallel graphs, in Proceedings of 10th International
Conference on Computer and Information Technology, pp. 1–4, 27–29 December 2007,
Dhaka, Bangladesh.
49. Incel, O.D., Ghosh, A., Krishnamachari, B., Chintalapudi, K., Fast data collection in
tree-based wireless sensor networks, IEEE Transactions on Mobile Computing, 11(1),
86–99, 2012.
50. Kim, M.S., Kim, J., Kim, J., Yoo, Y., Design and implementation of MAC protocol for
SmartGrid HAN environment, in Proceedings of IEEE 11th International Conference
on Computer and Information Technology (CIT), pp. 212–217, 31 August–2 September
2011, Pafos.
51. Bu, S., Yu, F.R., Liu, P.X., Zhang, P., Distributed scheduling in smart grid communications with dynamic power demands and intermittent renewable energy resources, in
Proceedings of IEEE International Conference on Communications Workshops (ICC),
pp. 1–5, June 2011, Kyoto, Japan.
52. Moscibroda, T., The worst-case capacity of wireless sensor networks, in Proceedings
of 6th International Symposium on Information Processing in Sensor Networks, IPSN,
pp. 1–10, 25–27 April 2007, Cambridge, MA.
53. Livieratos, S., Vogiatzaki, E., Cottis, P., A generic framework for the evaluation of the
beneits expected from the smart grid, Energies, 6, 988–1008, 2013.
54. Canale, S., Di Giorgio, A., Lanna, A., Mercurio, A., Panili, M., Pietrabissa, A., Optimal
planning and routing in medium voltage power line communications networks, IEEE
Transactions on Smart Grid, (99), 1–9, 2012.
55. Pena, A., Penya, Y.K., Distributed semantic repositories in smart grids, in Proceedings
of 9th IEEE International Conference on Industrial Informatics (INDIN), pp. 721–726,
26–29 July 2011, Lisbon, Portugal.
56. Simmhan, Y., Prasanna, V., Aman, S., Kumbhare, A., Liu, R., Stevens, S., Zhao, Q.,
Cloud-based software platform for big data analytics in smart grids, Computing in
Science and Engineering, 15(4), 38–47, 2013.
7
Power Grid Network
Analysis for Smart
Grid Applications
Zhifang Wang, Anna Scaglione,
and Robert J. Thomas
CONTENTS
7.1
7.2
7.3
Introduction .................................................................................................. 152
System Model ............................................................................................... 153
Topology Measures of Power Grids.............................................................. 154
7.3.1 Topological Metrics .......................................................................... 156
7.3.2 Small-World Wiring ......................................................................... 156
7.3.3 The Nodal Degree Distribution ........................................................ 158
7.3.4 The Graph Spectrum and the Scaling of Connectivity .................... 162
7.4 Line Impedance Distribution ........................................................................ 165
7.5 A Model to Generate RT Test Networks ...................................................... 168
7.5.1 clusterSmallWorld Subnetwork ........................................................ 169
7.5.2 Lattice Connections .......................................................................... 170
7.5.3 Assignment of Line Impedances ...................................................... 170
7.6 Topological Analysis for the Distribution Networks .................................... 171
7.6.1 Structure of the Distribution Network .............................................. 171
7.6.2 Graph Theoretic Analysis of a Sample MV Distribution Network .....172
7.6.3 The PMF of the Node Degrees in the Sample 396-Node MV
Distribution Network ........................................................................ 174
7.6.4 LV Distribution Network .................................................................. 176
7.7 Summary ...................................................................................................... 176
References .............................................................................................................. 176
Smart grid is an umbrella term for the modernization of electricity delivery systems with enhanced monitoring, analysis, control, and communication capabilities in order to improve the eficiency, reliability, economics, and sustainability
of electricity services. The modernization happens at both the high-voltage (HV)
transmission and the medium- or low-voltage (MV/LV) distribution grids and has
a diverse set of goals, including facilitating greater competition between providers,
encouraging greater use of alternative energy sources, implementing the automation
151
152
Smart Grids
and monitoring capabilities needed for bulk transmission, and enabling the use of
market forces to drive energy conservation.
Many of the existing technologies already used by electric utilities will be taking
full advantage of the smart grid, but additional communication and control capabilities will also be established to optimize the operation of the entire electrical grid.
The smart grid is also positioned to adopt new technologies, such as smart metering, synchronous phasor measurement units (PMUs), plug-in hybrid electric vehicles
(PHEVs), various forms of distributed generation, solar and wind energy, electric
load management systems, distribution automation, and many more.
In order to design an eficient communication scheme and examine the eficiency
of any networked control architecture in smart grid applications, we need to statistically characterize its information source, namely, the power grid itself. Investigating
the statistical properties of power grids has the immediate beneit of providing a
natural simulation platform, producing a large number of power grid test cases with
realistic topologies, with scalable network size, and with realistic electrical parameter settings. The second beneit is that one can start analyzing the performance of
decentralized control algorithms over information networks whose topology matches
that of the underlying power network, and use network scientiic approaches to determine analytically whether these architectures would scale well.
7.1
INTRODUCTION
This chapter provides a comprehensive study on the topological and electrical characteristics of a power grid transmission network based on a number of synthetic and
real-world power systems. Besides some basic graph theoretic metrics, it especially
considers the following characteristics:
• Small-World Properties: The transmission network of a power grid is
sparsely connected and manifests salient small-world properties that are
characterized by the larger clustering coeficient and much smaller average
shortest path compared with a random graph network with the same network
size [1]. However, it is found that the power grid assumes a different kind of
small-world topology than that of the Watts–Strogatz small-world model.
• Nodal Degree Distribution: This metric relects how the nodes in a power
grid network connect with neighbors and closely relates to the network
topology robustness under node removal [2,3]. Although most literature
assumes that power grids have an exponential (or geometric) nodal degree
distribution [4,5], it is found that the empirical distribution estimated from
real-world power systems clearly deviates from an exponential distribution,
especially in the range of small node degrees. The probability generation
function is applied to analyzing the node degree distribution and estimating
the parameters for it.
• Graph Spectrum and Connectivity Scaling Property: These two metrics
associate with the eigenvalues of the adjacency matrix and the Laplacian
matrix, respectively, and they carry important topology characteristics of
a network.
Power Grid Network Analysis for Smart Grid Applications
153
• Distribution of Line Impedance: The line impedance dominates the electrical characteristics of a power grid network; hence, a study on its distribution is important and necessary. It is found that the distribution of line
impedance is heavy-tailed and can be captured quite accurately by a clipped
double Pareto lognormal (DPLN) distribution.
This chapter introduces an algorithm that is able to generate random topology
power grids featuring the same topology and electrical characteristics as those found
in the real data. Finally, it also reports recent studies on the topological and electrical
characteristics of a sample MV power distribution network.
7.2
SYSTEM MODEL
The power network dynamics are coupled by the network equation
YU = I ,
(7.1)
where U and I represent the complex phasor vectors of bus voltages and injected currents; and Y is the network admittance matrix, which is determined by not only the
connecting topology but also its electrical parameters. Given a network with n nodes
and m links (which may also be referred to as buses and branches or lines in power
grid analysis, or vertices and edges in graph theory and network analysis), each link
l = (s, t) between nodes s and t has a line impedance zpr(l) = r(l) + jx(l), where r(l) is
the resistance and x(l) the reactance. Usually for an HV transmission network the
reactance dominates, that is, x(l) ≫ r(l). The line admittance is obtained from the
inverse of its impedance, that is, ypr(l) = g(l) + jb(l) = 1/zpr(l).
Assume that a unit current lows along the link l = (s, t) from node s to t; then the
resulting voltage difference between the ends of the link equals Δu = U(s)−U(t) = zpr(l),
or equivalently Δu = 1/ypr(l). Therefore, zpr(l) can be interpreted as the electrical distance between nodes s and t, and ypr(l) relects the coupling strength between the two
end nodes.
Deine the line-node incidence matrix A: = (Al,k)m × n as: Al,s = 1, Al,t = −1 if the lth
link is between nodes s and t with arbitrary directions; and Al,k = 0 otherwise. Then
the network admittance matrix Y: = (Ys,t)n × n can be obtained as
Y = AT diag ( y pr ) A,
(7.2)
with entries as follows:
Y ( s, t ) = − y pr ( s, t ) ,

y pr ( s, t ) ,
Y ( s, s ) =
t ≠s

Y ( s, t ) = 0,
if link s − t exists, for t ≠ s
∑
(7.3)
otherwise.
154
Smart Grids
The network Laplacian L: = (Ls,t)n × n can then be obtained as L = ATA, with the entries
 L ( s, t ) = −1,

 L ( s, s ) = k,
 L ( s, t ) = 0,

if link s − t exists, for t ≠ s
with k = deg ( s ) = Σ t ≠ s − L ( s, t )
otherwise.
(7.4)
A close comparison of the matrix structures of L and Y uncovers some interesting analogies. The Laplacian matrix L fully describes the topology of a network,
while the network admittance matrix Y contains information not only about the
system topology, but also about its electrical coupling. The off-diagonal entries of
Y, Ys,t equal the line admittance of the link between node s and t (with a minus
sign), whose magnitude relects the coupling strength between the two nodes. The
diagonal entries of the Laplacian L represent the total number of links connecting
each node with the rest of the network, while a diagonal entry of Y represents the
total coupling capability of one node with the rest of the network. Therefore, the
network admittance matrix Y can be viewed as a complex-weighted Laplacian; and
the Laplacian L can be equivalent to a “lat” network admittance matrix, which
assumes all the links in the network have the same line impedance (with a common
proportional factor).
7.3
TOPOLOGY MEASURES OF POWER GRIDS
First, we look at some basic topological metrics for a power grid network: the network size n, the total number of links m, the average nodal degree ⟨k⟩, the average
shortest path length counted in hops ⟨l⟩, and the Pearson coeficient. All these metrics can be derived from the Laplacian matrix:
•
•
•
•
The total number of links is m = 1 2 Σi L (i, i ).
The average nodal degree is ⟨ k ⟩ = 1 n Σi L (i, i ).
The network adjacency matrix can be obtained as Madj = −L + diag{L}.
The average shortest path ⟨l⟩ measured in hops can be calculated with
Dijkstra’s algorithm [6] based on Madj.
Deine the nodal degree vector as k = [k1, k2, …, kn] = diag(L), and k is the average
degree of a node seen at the end of a randomly selected link (i,j), that is,
k = ( 2m )
−1
∑ ( k + k ) = ( 2m ) ∑ k
−1
i
2
i
j
(i, j )
i
=
⟨k 2 ⟩
.
⟨k⟩
(7.5)
Thus, we can compute the Pearson coeficient of node degrees of the network as
ρ=
∑
∑
(k − k )(k − k )
(k − k ) (k − k )
i
j
(i, j )
(i, j )
2
i
j
2
,
(7.6)
155
Power Grid Network Analysis for Smart Grid Applications
which is a measure of the correlation of node degrees in the network [7]. It has
been observed that the Pearson coeficient for some kinds of networks is consistently positive, while for other kinds it is negative [8]. Therefore, some researchers proposed using the Pearson coeficient to differentiate technological networks
from social networks [9]. However, it is found that the Pearson coeficient of power
grids does not have restrictive characteristics, but ranges over a wide interval from
negative to positive, which can be veriied by the metric evaluation results listed in
Table 7.1.
Power grids have been found to have the salient features of small-world graphs
(see the work by Watts and Strogatz [1]). That is, while the vast majority of links are
similar to those of a regular lattice, with limited near-neighbor connectivity, a few
links connect across the network. These bridging links signiicantly shorten the path
length that connects every two nodes and critically increase the connectivity of the
network. At the same time, their scarcity puts the connectivity at risk in the case of
link failure for one of these critical bridges. The characterizing measure to distinguish a small-world network is called the clustering coeficient, which assesses the
degree to which nodes tend to cluster together. A small-world network usually has a
clustering coeficient signiicantly higher than that of a random graph network, given
the comparable network size and total number of edges. The random graph network
mentioned here refers to the network model deined by Erdös and Rényi, with n
labeled nodes connected by m edges that are chosen uniformly randomly from the
n(n − 1)/2 possible edges [10].
The clustering coeficient is deined as the average of the clustering coeficients
for all nodes [1]:
C (G ) =
1
n
n
∑C ,
(7.7)
i
i =1
with Ci being the node clustering coeficient as Ci = λG (i)/τG (i), where λG (i) is the
number of edges between the neighbors of node i and τG (i) the total number of edges
that could possibly exist among the neighbors of node i. For undirected graphs,
obviously τG (i) = ki(ki − 1)/2, given ki is the node degree. As pointed out in [11], the
TABLE 7.1
Topological Metrics of the IEEE and Real-World Power
Grid Networks
IEEE-30
IEEE-57
IEEE-118
IEEE-300
NYISO
WSCC
(n, m)
⟨I⟩
⟨k⟩
ρ
C(G)
C(R)
(30, 41)
(57, 78)
(118, 179)
(300, 409)
(2935, 6567)
(4941, 6594)
3.31
4.95
6.31
9.94
16.43
18.70
2.73
2.74
3.03
2.73
4.47
2.67
−0.0868
0.2432
−0.1526
−0.2206
0.4593
0.0035
0.2348
0.1222
0.1651
0.0856
0.2134
0.0801
0.094253
0.048872
0.025931
0.009119
0.001525
0.000540
156
Smart Grids
clustering coeficient for a random graph network theoretically equals the probability of randomly selecting links from all possible links. That is
C ( R) =
7.3.1
m
⟨k⟩
.
=
n ( n − 1) 2 n − 1
(7.8)
toPologIcal metrIcs
The topology metrics evaluated for the Institute of Electrical and Electronics
Engineers (IEEE) model systems, the New York Independent System Operator
(NYISO), and the Western System Coordinating Council (WSCC) systems are
shown in Table 7.1. Especially, we show the clustering coeficients of the IEEE,
NYISO, and WSCC systems compared with the random graph networks with the
same network size and the same total number of links. The former is denoted as
C(G), and the latter as C(R). For the reader’s reference, the IEEE 30, 57, and 118
bus systems represent different parts of the American Electric Power System in the
midwestern United States; the IEEE 300 bus system is synthesized from the New
England power system. More information can be obtained from [12]. The WSCC
grid is the interconnected system of transmission lines spanning the western United
States plus parts of Canada and Mexico, and the NYISO system represents the New
York state bulk electricity grid.
From Table 7.1, one can observe two interesting properties pertaining to the topology of the grid: (a) The connectivity is very sparse, since the average nodal degree
does not scale up as the network size increases. Instead, the average nodal degree
falls into a very restricted range, which is related to the particular geographical area
to which the network belongs, rather than the network size. For example, power grids
in the western United States have an average nodal degree between 2.5 and 3, while
in the northeastern United States the power grids are a little denser, with ⟨k⟩ around
4.5. (b) It is not accurate to attribute “small-world” features to power grids of any
size. It is true that the samples we studied have a clustering coeficient signiicantly
larger than that of a random graph network of comparable size. However, as we will
discuss next, while being similarly sparse, power grids have better connectivity scaling laws than the small-world graphs proposed in [1].
7.3.2
small-World WIrIng
The Watts–Strogatz small-world model is generated starting from a regular ring
lattice, then, using a small probability, rewiring some local links to an arbitrary
node chosen uniformly at random from the entire network (to make it a smallworld topology). A tool to visually highlight small-world topologies is the Kirk
graph, which was proposed by Kirk (2007) [13]. It is a simple but effective way to
show a network topology. First, node numbers are assigned according to physical
nodal adjacency, that is, physically closely located nodes are given close numbers; then all the nodes are sequentially and evenly spread around a circle, and
links between nodes are drawn as straight lines inside the circle. Figure 7.1 shows
157
Power Grid Network Analysis for Smart Grid Applications
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
(a)
19 18
17 16
15 14 13
12 11
10
9
8
7
6
5
4
3
2
1
40 41
42 43
44 45 46
57
56
55
54
53
52
51
50
49
47 48
18
2019
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
3940
41
(c)
17 16
19 18
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40 41
(b)
15 14 13
1211
10
9
8
17 16
15 14 13
12 11
10
9
8
7
6
5
4
3
2
1
42 43
7
44 45 46
57
56
55
54
53
52
51
50
49
47 48
6
5
4
3
2
1
42 43
44 45 46
57
56
55
54
53
52
51
50
49
4748
FIGURE 7.1 (a–c) Topology comparison using Kirk plot.
three representative network topologies, using Kirk graphs, of an Erdös–Rényi
rand-graph network, of a Watts–Strogatz small-world network, and of a real power
grid—the IEEE-57 network. The three networks have the same network size
and almost the same total number of links. The topology difference between the
Erdös–Rényi network and the IEEE-57 network can be easily noticed, in contrast
to a good degree of similarity between the Watts–Strogatz small-world model and
the test power network.
However, the rewiring of the test power grid is not independent, as it is in the
Watts–Strogatz small-world model; instead, long hauls appear over clusters of
nodes. This property is visually noticeable in the IEEE 57 network and differentiates it from the small-world graph, in which these links are chosen independently.
This fact has an intuitive explanation: long hauls require having a right of way
to deploy a long connection, and it is highly likely that the long wires will reuse
part of this space. These physical and economic constraints inevitably affect the
structure of the topology. Besides, there is more than meets the eye. The Watts–
Strogatz small-world model has a scaling property that cannot be validated by
158
Smart Grids
power grid topologies, precisely because the average nodal degree of a power
network is almost invariant with the size of the network. Given a network size
with its speciied average nodal degree, the model fails to produce a connected
grid topology for reasonably large network sizes with realistic power grid degree
distributions. The main reason for the poor scaling lies in the fact that, in order to
produce a sparse but connected topology, the Watts–Strogatz small-world model
requires [1,11]
1 ≪ ln ( n ) ≪ ⟨ k ⟩ ≪ n ,
(7.9)
or, equivalently, ⟨k⟩ = n = e⟨k⟩. On the other hand, real-world power grids have a very
low average nodal degree, with ⟨k⟩ = 2 – 5, regardless of the network size. This will
limit the network size to much smaller than 150 in order to produce a connected topology using the model. However, in the real world, large power grids are connected even
if they are much sparser than is required by the Watts–Strogatz small-world model.
7.3.3
the nodal degree dIstrIButIon
Now we examine the empirical distribution of nodal degrees in the available realworld power grids. Figure 7.3a shows the histogram probability mass function
(PMF) in log scale for the nodal degrees of the NYISO system. If the PMF curve
approximates a straight line in the semilogarithmic plot (i.e., shown as log(p(k))
vs. k), it implies an exponential tail analogous to that of the geometric distribution. However, it is also noticed that for the range of small node degrees, that is,
when k ≤ 3, the empirical PMF curve clearly deviates from that of a geometric
distribution.
The probability generation function (PGF) can be utilized to analyze the node
degree distribution in power grids. The PGF of a random variable X is deined as
GX ( z) = Σ k Pr( x = k ) z k. Given a sample dataset of X with size N, its PGF can also be
estimated from the mean of z X, because
( )
E zX =
=
=
1
n
∑z
1
n
∑n
≈
x
( x =k )
zk
k
∑
k
x
n( x = k ) k
z
n
∑ Pr
( x =k )
zk ,
(7.10)
k
where n(x = k) denotes the total number of the data items equal to k. Due to
lim n →∞ n( x = k ) n = Pr( x = k ), we can have E(z X) ≈ GX(z) with a large enough data size.
159
Power Grid Network Analysis for Smart Grid Applications
If a random variable can be expressed as the sum of two independent random
variables, its PMF is the convolution of the PMFs of the component variables, and its
PGF is the product of those of the component variables. That is,
X = X1 + X 2
Pr( X = k ) = Pr( X1 = k ) ⊗ Pr( X2 = k )
(7.11)
GX ( z ) = GX1 ( z ) GX2 ( z ) .
It is found that the node degree distribution in power grids can be very well
approximated by the sum of two independent random variables, that is,
K = G + D,
(7.12)
where G is a truncated geometric with the threshold of kmax:
(1 − p ) p
Pr(G = k ) =
k
i
∑ i=0 (1 − p ) p
k
max
(1 − p ) p ,
k
+1
1 − (1 − p )
k
=
k = 0,1, 2, … , kmax,
max
(7.13)
with the PGF as
kmax
(1 − p ) pz k
k
+1
1 − (1 − p )
∑
G (z) =
G
=
k
k =0
max
(
(
p 1 − (1 − p ) z
(
1 − (1 − p )
kmax +1
)
kmax +1
)
(7.14)
) (1 − (1 − p ) z )
and D is an irregular discrete { p1, p2 , … , pkt }, with Pr( D= k ) = pk , k = 1, 2, … , kt
whose PGF is GD ( z ) = p1z + p2 z 2 + p3 z 3 + ⋯ + pkt z kt . Therefore, the PMF of K is
Pr(K=k) = Pr(G=k) ⊗ Pr(D=k), and the PGF of K can be written as
GK ( z ) =
(
(
p 1 − (1 − p ) z
(
1 − (1 − p )
)
kmax +1
kmax +1
)∑
kt
i =1
pi z i
) (1 − (1 − p ) z )
.
(7.15)
Equation 7.15 indicates that the PGF G K(z) has kmax zeros evenly distributed around
a circle of radius of 1/(1 − p) that are introduced by the truncation of the geometric G
160
Smart Grids
1.5
1.5
1.0
1.0
0.5
0.5
0
0
Y
Y
(because the zero at 1/(1 − p) has been neutralized by the denominator (1 − (1 − p)z))
and has kt zeros introduced by the irregular discrete D with { p1, p2 , … , pkt}.
Figure 7.2 shows the contour plots of PGF of node degrees for different groups
of buses in the NYISO system. Three interesting discoveries are worthy of note: (a)
clearly each plot contains evenly distributed zeros around a circle, which indicates
a truncated geometric component; (b) besides the zeros around the circle, most contour plots also have a small number of off-circle zeros, which come from an embedding irregular discrete component; (c) the contour plot for each group of nodes has
zeros with a similar pattern but different positions. This implies that each group
of node degrees has similar distribution functions, but with different coeficients.
Therefore, it is necessary and reasonable to characterize the node degree distribution
according to the node types. Otherwise, if the node degrees were to be aggregated
into one single group, as in Figure 7.2a, some important characteristics of a subgroup
of node degrees would be concealed (e.g., comparing a and b–d in Figure 7.2).
−0.5
−0.5
−1.0
−1.0
−1.5
−1
(a)
0
X
1
−1.5
1.5
1.0
1.0
0.5
0.5
0
0
−0.5
−0.5
−1.0
−1.0
Y
1.5
−1.5
(c)
−1
0
X
−1
(b)
1
−1.5
−1.5
(d)
−1.0 −0.5
0
X
0
X
1
0.5
1.0
1.5
FIGURE 7.2 Contour plot of E(z X) of node degrees for different groups of buses in the
NYISO system: (a) all buses; (b) generation buses; (c) load buses; (d) connection buses. The
zeros are marked by plus signs.
161
Power Grid Network Analysis for Smart Grid Applications
Based on the contour plots, one can easily locate the zeros in PGF, and further
determine the coeficients of the corresponding distribution functions. The estimated coeficients for each group of nodes in the NYISO and all the nodes in the
WSCC system are listed in Table 7.2. Because the WSCC system data we have
only contains unweighted raw data without distinguishing node types, its node
degree distribution is only analyzed for one aggregate group. Figure 7.3 compares
the PMF with estimated coeficients and the empirical PMF of the NYISO system,
and shows that the former matches the latter with quite good approximation. The
TABLE 7.2
Estimated Coeficients of the Node Degree Probability
Density Functions of the NYISO and WSCC Systems
Node Groups
All
Generation
Load
Connection
All WSCC
kmax
Max (k)
kt
ρ
34
36
26
18
16
37
37
29
21
19
3
1
3
3
3
0.2269
0.1863
0.2423
0.4006
0.4084
, pkt}
0.4875, 0.2700, 0.2425
1.000
0.0455, 0.4675, 0.4870
0.0393, 0.4442, 0.5165
0.3545, 0.4499, 0.1956
100
100
Empirical PMF
Fitting PMF
Empirical PMF
Fitting PMF
10−1
10
10−2
log (PMF)
log (PMF)
{ p1, p2 ,
10−3
10−4
−1
10−2
10−3
10−5
10−6
0
10
(a)
20
k-node degree
30
10−4
40
Empirical PMF
Fitting PMF
20
k-node degree
30
40
Empirical PMF
Fitting PMF
10−1
10−2
log (PMF)
log (PMF)
10−1
10−3
10−2
10−3
10−4
10−4
(c)
10
100
100
10−5
0
(b)
0
5
10
15
20
k-node degree
25
30
10−5
(d)
0
5
10
15
k-node degree
20
25
FIGURE 7.3 Comparing the empirical and itted PMF of node degrees in the NYISO system: (a) all buses; (b) generation buses; (c) load buses; (d) connection buses.
162
Smart Grids
results from both systems have validated our assumption of node degree distribution in power grids: it can be expressed as the sum of a truncated geometric random
variable and an irregular discrete random variable. The results also demonstrated
the effectiveness of the proposed method of analyzing node degree distribution by
using the PGF.
7.3.4
the graPh sPectrum and the scalIng oF connectIvIty
The eigenvalues of the adjacency matrix Madj and the Laplacian matrix L carry
important topological features of a network. The two matrices are in fact exchangeable, and either of them fully describes a network topology. The set of eigenvalues
of its adjacency matrix Madj is called the spectrum of a graph. Graph spectral density
is deined as
ρ (λ) =
1
n
n
∑ δ (λ − λ ) ,
j
(7.16)
j =1
with {λj, j = 1, 2, …, n} forming the graph spectrum of the network. The kth moment
of ρ(λ) represents the average number of k-hop paths returning to the same node
in the graph [11]. However, please note that these paths can contain nodes which
are already visited. The graph spectral density of network topology from different
categories has distinctively different patterns [11]. An Erdös–Rényi random graph
network G(n;p), given a large network size n and a nontrivial link selection probability p(n) = cn−z, with z < 1 and c as a constant regardless of network size, has its
normalized spectral density converging to a semicircular distribution as n → ∞. This
is known as Wigner’s semicircle law [14,15]. That is
2

 4 − ( λ λ 0 )
ρ (λ) = 
2πλ 0

0
if | λ |< 2λ 0
(7.17)
otherwise
with λ 0 = np(1 − p) and p = m n(n − 1) 2 = ⟨ k ⟩ n − 1, where m is the total number of
links and ⟨k⟩ is the average node degree. And ρɶ = 2πλ 0ρ(λ) = 4 − λɶ 2 with λɶ = λ / λ 0
depicts a semicircle around the origin with a radius of 2.
Next, we refer to spectrum as the set of eigenvalues and to spectral density as its
density of values. Figure 7.4 shows the spectral density of a ring lattice (a), of two
real-world power grids (b) and (c), and of a power grid generated from the proposed
random topology (RT)-nested-Smallworld model (d), compared with the semicircle
law corresponding to random graphs with the same network size and number of
links. The plots demonstrate that power grid graphs have a very distinctive spectral
density, far from that of regular or completely random networks; it also shows that
the proposed model RT-nested-Smallworld (see Section 7.5) is in good agreement
with the real data in terms of graph spectral density.
163
Power Grid Network Analysis for Smart Grid Applications
20
4
20
18
16
14
12
10
8
6
4
2
0
−2
(b) −4
4
20
18
16
14
12
10
8
6
4
2
0
−2
(d) −4
Lattice−1-D: 2935
Random graph
15
10
5
0
(a) −4
0
2
NYISO: 2935
Random graph network
−2
0
Normalized graph spectral density
20
18
16
14
12
10
8
6
4
2
0
−2
(c) −4
−2
(e)
2
RT-nestedSW: 2988
Random graph network
−2
0
2
4
WSCC: 4941
Random graph network
−2
0
2
4
9
8
7
6
5
4
3
2
1
0
−4
−3
−2
−1
0
~
λ
1
2
3
4
FIGURE 7.4 Normalized graph spectral density of different networks, ρɶ (λ ) versus λɶ : the dotted
line of semicircle represents the graph spectral density of random graph networks; (a) a ring lattice; (b) RT-nested-Smallworld; (c) NYISO; (d) WSCC; (e) a 396-node MV distribution network.
Another important measure is the second smallest eigenvalue of the Laplacian
matrix, λ2(L), called algebraic connectivity. This measure is sometimes termed
the Fiedler eigenvalue, because it was irst introduced by Fiedler (1989) [16]. How
well a network is connected and how fast information data can be shared across the
network is relected by λ2(L). The smallest eigenvalue of the Laplacian is always
zero, that is, λ1(L) ≡ 0, and the number of times 0 appears as an eigenvalue in the
164
Smart Grids
Laplacian is the total number of connected components in the network. The eigenvalue λ2(L) is greater than 0 if and only if the network is a connected graph. If the
algebraic connectivity λ2(L) is close to zero, the network is close to being disconnected. Otherwise, if (1/n)λ2(L) tends to be 1, with n as the network size, the network
tends to be fully connected.
Table 7.3 shows the algebraic connectivity of IEEE model systems and the NYISO
system. Figure 7.5 plots the connectivity scaling curve of power grid versus network
size and compares it with that of one-dimensional (1-D) and two-dimensional (2-D)
lattices. The 1-D lattice is a ring-structured topology, with nodes connected to most
TABLE 7.3
Algebraic Connectivity of
the IEEE and Real-World
Power Networks
IEEE-30
IEEE-57
IEEE-118
IEEE-300
NYISO
WSCC
n
λ2(L)
30
57
118
300
2935
4941
0.21213
0.088223
0.027132
0.0093838
0.0014215
0.00075921
101
100
λ2
10−1
10−2
n−0.976
n−1.376
n−0.933
10−3
n−1.0604
n−2
10−4
n−2
−5
10
101
102
103
n (network size)
104
FIGURE 7.5 Connectivity scaling curve versus network size: 2-D lattice with k = 4 (x-dotted
line); 2-D lattice with k = 2 (Δ-dotted line); 1-D lattice with k = 4 (▫-dotted line); 1-D lattice
with k = 2 (∇-dotted line); nested-Smallworld RT with subnetwork size = 30 (small ▫); nestedSmallworld RT with subnetwork-size = 300 (⬨); power grids (star).
Power Grid Network Analysis for Smart Grid Applications
165
adjacent neighbors on both sides. The 2-D lattice is a regular 2-D meshed grid with
each boundary side merging with the other side and each node connected to the most
adjacent neighbors around it. For a 1-D lattice, its connectivity scales as λ2(L) ∝ n−2;
for a 2-D lattice, its connectivity grows as λ2(L) ∝ n−1; interestingly, for power grids,
connectivity grows as λ2(L) ∝ n−1.376~−1.0604, lying between those of the 1-D lattice
and the 2-D lattice.
In trying to it the random wiring of a power network, we postulate a new possible
model, RT-nested-Smallworld, in Section 7.5. The model results from nesting several “small-world” subnetworks, whose size is likely to produce a connected topology with a realistic degree distribution, into a regular lattice again. More details
can be found in Section 7.5. The intuition guiding this modeling is that the network
would have produced connected topologies with a connectivity that was intermediate between the 1-D and 2-D lattices. Figure 7.5 shows an excellent match between
the algebraic connectivity of this type of model and the real data.
7.4
LINE IMPEDANCE DISTRIBUTION
This chapter not only examines the network topology of a power grid, but also
attempts to reproduce its electrical characteristics accurately, because the key element of a power grid network is its network admittance matrix Y, which associates
both with the connecting topology and with its line impedances. The transmission
line impedance in a power grid is, in fact, a complex number.
When it comes to the statistical analysis, we study the distribution of the magnitude of Zpr. The line impedance can be represented as Zpr = R + jX, where R is the resistance and X is the reactance. Usually X is the dominant component, whereas R only
takes a trivial value, which in many cases can even be neglected. Therefore, one can
easily reconstruct the complex value of Zpr given its magnitude. The empirical data on
line impedances of power grids are taken from IEEE model systems and the NYISO
system. The irst clear observation from the empirical histogram probability density
distribution (PDF) is that the distribution of the line impedances is heavy-tailed. The
candidate distribution functions include gamma, generalized Pareto (GP), lognormal,
DPLN, and two new distributions that we call lognormal-clip and DPLN-clip, which
will be explained later. In searching among the heavy-tailed distributions for a it, the
parameters of the candidate distribution can be estimated via the maximum likelihood
(ML) criterion from the data, and the appropriately modiied Kolmogorov–Smirnov
(K-S) test can be used to check whether the hypothesized distribution is a good it.
The DPLN distribution was introduced by Reed and Jorgensen (2004) [17], and has
proved to be very useful to model the size distributions of various phenomena, such
as incomes and earnings, human settlement sizes, and so on. We choose the DPLN
distribution as a candidate to it the line impedance data in a power grid because the
latter implicitly relate to human settlement size in the network. Generally speaking,
large-scale generation is usually located near water or the fuel source, but remotely
from densely populated areas with large-scale human settlements. Therefore, longdistance transmission lines have been constructed in order to provide the interconnection and serve the large-scale settlements. And the long-distance lines usually
exhibit high line impedance.
166
Smart Grids
The lognormal-clip and DPLN-clip are especially suited for itting the NYISO
data because the line impedances in this system appear to approximate a lognormal
or DPLN distribution very well except for having an interrupted tail, which is captured by the clipping. Therefore, it can be assumed that the NYISO data resulted from
some original impedance data being “clipped” by an exponential cutoff tailing coeficient. That is, given the original impedance data Y following a speciic distribution,
−Y Z
Y ∼ f Y (y), the clipped impedance data are X = Z max 1 − e ( max ) , with Z max being the
cutoff threshold, so that Y = − Z max log(1 − X / Z max ). And the resulting clip distribution turns out to be
(
)
X ~ f X ( x ) = Y(′x ) fY (Y( x ) )
=
(
)
Z max
fY − Z max log (1 − x Z max ) .
Z max − x
(7.18)
It is reasonable to introduce the clipping mechanism into the line impedance
distribution because in real-world power grids transmission lines are limited in
length, and so, correspondingly, is the line impedance, which is proportional to the
length. As shown next, the line impedances in real-world power grids have heavytailed distributions. Hence, the candidate distribution functions listed below are
selected from the class of heavy-tailed distributions:
Gamma
Γ ( x | a, b ) =
1
x a −1e x / b ;
baΓ ( a )
(7.19)
GP
( x − θ) 
 1 
gp ( x | k, σ, θ ) =    1 + k


σ 
 σ 
−1− (1 k )
;
(7.20)
Lognormal
logn ( x | µ, σ ) =
1
xσ 2 π
2
e −(log x −u )
2 σ2
;
(7.21)
DPLN
dPlN ( x | α, β, µ, σ ) =
2
αβ 
− α −1  log x − µ − α σ 
 A ( α, µ, σ ) x Φ 

α + β 
σ


 log x − µ + β σ2  
+ A ( −β, µ, σ ) x β −1Φ c 
 ,
σ
 

(7.22)
Power Grid Network Analysis for Smart Grid Applications
where A ( θ, µ, σ ) = e( θµ + θ σ
Lognormal-clip
2 2
167
/ 2)
;
logn clip ( x | µ, σ, Z max )
=



Z max
X 
µ, σ  ;
logn  − Z max log  1 −

Z max − x
 Z max 


(7.23)
DPLN-clip
dPlN clip ( x | α, β, µ, σ, Z max )
=



Z max
X 
dPlN  − Z max log  1 −
α, β, µ, σ  .

Z max − x
 Z max 


(7.24)
For each candidate distribution function, one can estimate its parameters using
the ML method, and then run the K-S test in order to pick the distribution that gives
the best K-S test result. Table 7.4 shows the best-itting distribution functions with the
corresponding ML parameter estimates. Figure 7.6 compares the empirical PDFs and
TABLE 7.4
PDF Fitting for the Line Impedances in the IEEE and
NYISO Power Grids
System
Fitting Distribution
IEEE-30
Γ(x|a, b)
IEEE-118
Γ(x|a, b)
IEEE-57
gp(x|k, σ, θ)
IEEE-300
gp(x|k, σ, θ)
NYISO-2935
lognclip(x|μ, σ, Zmax)
dPlNclip(x|α, β, μ, σ, Zmax)
ML Parameter Estimates
(α = 0.05)
a = 2.14687
b = 0.10191
a = 1.88734
b = 0.05856
k = 0.33941
σ = 0.16963
θ = 0.16963
k = 0.45019
σ = 0.07486
θ = 0.00046
μ = −2.37419
σ = 2.08285
Zmax = 1.9977
α = 44.25000
β = 44.30000
μ = −2.37420
σ = 2.082600
Zmax = 1.9977
168
Smart Grids
NYISO: Empirical PDF of Zpr and DPLN-clip fitting PDF
Density
104
(a)
NYISO Zpr
DPLN-clip
102
100
10−2
10−4
1
10−3
10−2
10−1
100
101
NYISO: Empirical CDF of Zpr and DPLN-clip fitting CDF
F (Zpr)
0.8
0.6
0.4
NYISO Zpr
DPLN-clip
0.2
0
(b)
0
0.2
0.4
0.6
0.8
1
1.2
Zpr (p.u.)
1.4
1.6
1.8
2
FIGURE 7.6 Comparison between empirical PDF/CDF distributions and itted distributions: (a) IEEE 300; (b) NYISO 2935.
cumulative distribution functions (CDFs) (on logarithmic scales) with those from the
best-itting distribution function for the NYISO system. The studies show that the
gamma distribution is the best it for the line impedances in the IEEE 30 and 118
systems and the GP distribution for those of the IEEE 57 and 300 systems, while for
the NYISO system lognormal-clip and DPLN-clip best it the data. This is also reasonable, because, even if it is not one to one, the line impedance usually grows with the
geographic distance between the buses it connects.
In addition, it is well known that the distribution of human settlements tends to be
clustered; thus, it is reasonable that long tails appear between these clusters of smallersize buses, giving rise to these heavy tails. This observation seems to agree with the
topological properties found in the power grid graphs, and it is reasonable to consider
these heavy-weight impedances as good candidates for line impedances of the links
that are rewired in the small-world islands, as well as the ones connecting difference
islands in our nested-Smallworld model, which is introduced in the next section.
7.5
A MODEL TO GENERATE RT TEST NETWORKS
There are a number of works in the power network literature that propose models
to generate scalable-size power grid test cases, such as: the use of ring-structured
grids to study the propagation pattern of disturbances [18]; a tree-structured power
grid model for detecting critical transitions in the transmission network that cause
cascading failure blackouts [19]; the Watts–Strogatz small-world network [1]; or the
formation of a power grid topology starting from a uniform or a Poisson distribution
for the nodal location (emulating what is often done in modeling wireless networks)
[20]. While these proposed models provide some good perspectives on power grid
Power Grid Network Analysis for Smart Grid Applications
169
characteristics, the generated topology fails to correctly or fully relect that of a
realistic power system.
This section presents a random topology power grid model, RT-nested-Smallworld,
which constructs a large-scale power grid in a hierarchical way: irst forming connected subnetworks with size limited by the connectivity requirement; then connecting the subnetworks through lattice connections; and, inally, generating the
line impedances from some speciic distribution and assigning them to the links in
the topology network. The hierarchy in the model arises from observation of realworld power grids: usually a large-scale system consists of a number of smaller-size
subsystems (e.g., control zones), which are interconnected by sparse and important
tie-lines. The model contains three main components: (a) clusterSmallWorld subnetwork, (b) lattice connections, and (c) generation and assignment of line impedances.
These will be described in detail in the following subsections.
7.5.1
clustersmallWorld
suBnetWork
Electrical power grid topology has “small-world” characteristics; it is sparsely connected with a low average nodal degree that does not scale with the network size. On
the other hand, in order for a small-world model to generate a connected topology,
the network size has to be limited. In this proposed model, different mechanisms
from those of the Watts–Strogatz small-world model have been adopted to form a
power grid subnetwork in order to improve its resulting connectivity, as shown in
the following paragraphs. Consequently, the connectivity limitation on the network
size can be expanded from what is indicated by Equation 7.9. The experiments have
shown that for ⟨k⟩ = 2 – 3, the network size should be limited to no greater than 30,
and for ⟨k⟩ = 4 – 5, no greater than 300.
Therefore, the irst step of this new model is to select the size of subnetworks
according to the connectivity limitation. Then a topology is built up through a modiied small-world model, called clusterSmallWorld. This model is different from the
Watts–Strogatz small-world model in two aspects: the link selection and the link
rewires. That is, instead of selecting links to connect most immediate ⟨k⟩/2 neighbors
to form a regular lattice, it selects a number k of links at random from a local neighborhood N d0 with the distance threshold of d0, where k comes from a geometric distribution. The local neighborhood is deined as the group of close-by nodes with mutual
node number difference less than the threshold d0, that is, N d(i0) = { j; | j − i |< d0} for
node i. It is worth noting that the clusterSmallWorld model adopts a geometric distribution with the expectation of ⟨k⟩ for the initial node degree settings (i.e., for the link
selection). Experiments have shown that the process of link selection and link rewiring that follows will transform the initial node degree distribution and inally result in
a distribution with good matching approximation to the observed distribution.
For the link rewiring, the Watts–Strogatz small-world model selects a small portion
of the links to rewire to an arbitrary node chosen at random in the entire network to
make it a small-world topology. The clusterSmallWorld model uses a Markov chain with
transition probabilities of (α, β), as shown in Figure 7.7, to select clusters of nodes and
therefore groups of links to be rewired. This mechanism is introduced in order to produce a correlation among the rewired links. After running the above Markov transition
170
Smart Grids
α
0
1–α
1–β
1
β
FIGURE 7.7 Markov chain for selecting cluster of rewiring nodes (0: not rewire; 1: rewire).
n times (i.e., one for each node in the network), we get clusters of nodes with “1”s alternating with clusters of nodes with “0”s, where “1” means to rewire and “0” means not.
Then, by a speciic rewiring probability qrw, some links are selected to rewire from all
the links originating from each 1-cluster of nodes, and the corresponding local links are
rewired to outside 1-clusters. The average cluster size for rewiring nodes is Kclst = 1/β,
and the steady-state probabilities are p0 = β/(α + β), p1 = α/(α + β). Experiments are performed on the available real-world power grid data to estimate the parameters. The
average rewiring cluster size Kclst, the rewiring probability of links qrw, and the ratio of
nodes having rewire links p1 can be directly obtained from statistical estimates. Then
the transition probability can be computed as β = 1/Kclst and α = βp1/(1 − p1).
7.5.2
lattIce connectIons
The lattice connections are selected at random from neighboring subnetworks to
form a whole large-scale power grid network. The number of lattice connections
between neighboring subnetworks is randomly chosen to be an integer with the mean
of ⟨k⟩.
7.5.3
assIgnment oF lIne ImPedances
At this step m line impedances are generated from a speciied heavy-tailed distribution, and then sorted by magnitude and grouped into local links, rewire links, and
lattice-connection links according to their corresponding proportions, as shown in
Figure 7.8. Finally, line impedances in each group are assigned at random to the corresponding group of links in the topology.
Impedances (p.u.)
Lattice-conn
Local
1
FIGURE 7.8 Grouping line impedances.
Rewires
m
Link number
Power Grid Network Analysis for Smart Grid Applications
7.6
171
TOPOLOGICAL ANALYSIS FOR THE
DISTRIBUTION NETWORKS
Usually the transmission network of a power grid is an HV network with voltage
levels above 50 kV. The studies and analysis in this chapter have so far focused
on the HV transmission network of a power grid. Now we extend the topological
analysis to the distribution side of the power grid. A distribution network carries
electricity from the transmission system and delivers it to end users. Typically,
the network would include MV (<50 kV) lines, electrical substations and polemounted transformers, LV (<1 kV) distribution wiring, and, sometimes, electricity
meters. The results presented are based on a sample 396-node MV distribution
network, which comes from a real-world US distribution utility mainly located in
a rural area.
7.6.1
structure oF the dIstrIButIon netWork
In the LV and MV sections of the grid, the physical layout is often restricted by what
land is available and its geology. The logical topology can vary depending on budget
constraints, system reliability requirements, and load and generation characteristics.
Generally speaking, there are a few typical kinds of topology in the distribution network: ring, radial, and interconnected.
A radial network is the cheapest and simplest topology for a distribution grid, and
the one most often encountered. This network has a tree shape, in which power from
a large supply radiates out into progressively lower-voltage lines until the destination homes and businesses are reached. It is typical of long rural lines with isolated
load areas. Today’s grid is radially operated with respect to the current transmission
system, but this topology will no longer work when distributed energy resources
(DER) are integrated into the grid. Unfortunately, this topology is the worst in terms
of maximum communication delay, because the number of hops between its nodes
tends to grow in the order of the size of the network.
An interconnected or ring network is generally found in more urban areas and
will have multiple connections to other points of supply. These points of connection
are normally open, but allow the operating utility to set up various conigurations by
closing and opening switches. Operation of these switches may be by remote control
from a control center or by a lineman. The beneit of the interconnected model is
that, in the event of a fault or required maintenance, a small area of the network can
be isolated and the remainder kept on supply.
Most areas provide three-phase industrial service. A ground is normally provided, connected to conductive cases and other safety equipment, to keep current
away from equipment and people. Distribution voltages vary depending on customer
need, equipment, and availability. Within these networks there may be a mix of overhead line construction utilizing traditional utility poles and wires and, increasingly,
underground construction with cables and indoor or cabinet substations. However,
an underground distribution network is signiicantly more expensive than lines
deployed through an overhead construction. Distribution feeders emanating from
a substation are generally controlled by a circuit breaker, which will open when a
172
Smart Grids
fault is detected. Automatic circuit reclosers may be installed to further segregate
the feeder, thus minimizing the impact of faults. It is important to remark that long
feeders experience voltage drop, and thus require the installation of capacitors or
voltage regulators.
7.6.2
graPh theoretIc analysIs oF a samPle mv dIstrIButIon netWork
The logical topology of the sample 396-node MV distribution network analyzed
here is shown in Figure 7.9. The power supply comes from the 115 to 34.5 kV stepdown substation. Most nodes or buses in the network are 12.47 kV, and only a small
number of them are 34.5 or 4.8 kV. As shown in Figure 7.9, an MV network usually
comprises different voltage levels, separated by transformers.
In the following topology analysis of the sample MV network, it is assumed that
wireless or wired couplers have been implemented at the locations of transformers
and switches, so that the network connectivity will not be affected by transformer
types or switch status. On the other hand, if couplers are missing, the network
will be segmented into several sections either by the transformers or by the open
switches. For the sample MV network analyzed here, most buses (>95%) in the
network are at the same voltage level of 12.47 kV. Therefore, the topology analysis
result of the separated 12.47 kV subnetwork is in fact very close to that of the whole
connected graph.
FIGURE 7.9 A 396-node MV distribution network in a rural area of the United States.
Components: bus (circle), line branches (line ending with dots), switches (line ending with
“+”s), transformers (lines ending with “x”s), open or out-of-service component (light gray);
the node color represents its voltage level: 115 kV (gray X), 34.5 kV (gray O), 12.47 kV
(black), 4.80 kV (gray).
173
Power Grid Network Analysis for Smart Grid Applications
The topology metrics we evaluated include the following:
• (n,m): the total number of nodes and branches, which represents the network size well.
• ⟨k⟩: the average node degree, which represents the average number of
branches a node connects to.
• ⟨l⟩: the average shortest path length in hops between any pair of nodes.
• ρ: the Pearson correlation coeficient, which evaluates the correlation of
node degrees in the network. This measure relects whether a node adjacent
to a highly connected node also has a large node degree.
• λ2(L): the algebraic connectivity, which is the second smallest eigenvalue
of the Laplacian matrix and is an index of how well a network is connected
and how quickly information data can be shared across the network.
• C(G): the clustering coeficient, which assesses the ratio of nodes tending
to cluster together.
The result of the analysis is shown in Table 7.5 in comparison with two other
transmission networks: the IEEE-300 system represents a synthesized network from
the New England power system and has a network size comparable to the 396-node
MV distribution network we analyzed; the WSCC grid is the interconnected system
of transmission lines spanning the western United States plus parts of Canada and
Mexico, and contains 4941 nodes and 6594 transmission lines. It is well known that
transmission and distribution topologies differ; nevertheless, to comment on these
differences in a quantitative manner, as in this exercise, is useful for several reasons.
For example, it allows us to better understand the characteristics of the transmission
and distribution networks as information sources; it allows us to optimize the design
of the distribution PMU-based wide-area measurement systems rather than attempting to duplicate the existing transmission, which is tailored to a network with very
different topological characteristics; and it can tell us how the distribution topology
can be “modiied” to achieve some advantageous characteristics of the transmission
network, that is, shorter path lengths between nodes, better algebraic connectivity,
and so on.
From Table 7.5, we can see that the 396-node MV distribution network has an
average node degree of ⟨k⟩ = 2.12, which is comparable to, although a little bit lower
than, those of the other two transmission networks, the IEEE-300 system and the
WSCC system. That means its average connecting sparsity is at about the same level
TABLE 7.5
Topological Characteristics of the Transmission Networks and the
MV Distribution Network
IEEE-300
WSCC
MV distribution
n
⟨k⟩
⟨l⟩
ρ
λ2(L)
C(G)
C(R)
300
4941
396
2.73
2.67
2.12
9.94
18.70
21.10
−0.2206
0.0035
−0.2257
0.0094
0.00076
0.00030
0.0856
0.0801
0.0000
0.009119
0.000540
0.005367
174
Smart Grids
as the transmission networks it is compared with. However, the sample MV distribution network has a much longer average path length, ⟨l⟩ = 21.10 in hops, than the
IEEE-300 system, and, interestingly, it is even longer than that of the much larger
4941-node WSCC system. More speciically, any node in this MV distribution network is about 16.50 hops away from node-1 or node-2, which are 115 kV buses at the
HV side of the two step-down supply transformers and may likely serve as the trafic
sinks in the power line communication (PLC)-based network.
Looking at the algebraic connectivity λ2(L), the 396-node MV distribution network has a much weaker overall connectivity compared with the transmission networks, that is, λ2(L) = 0.00030 versus 0.0094 (IEEE-300) and 0.00076 (WSCC). This
result shows that this topology is highly prone to becoming a disconnected graph
under node failure (islanding). Finally, the most distinctive difference we found lies
in the fact that the 396-node MV distribution network has a clustering coeficient
equal to zero, compared with the clustering coeficient of 0.0856 for the IEEE-300
system and 0.0801 for the WSCC system. This means that no node in the sample
MV distribution network is the vertex of a complete subgraph (triangle). MV distribution networks not located in rural areas are generally less prone to becoming
disconnected graphs, as in urban areas it is not unusual for utilities to provide link
redundancy by, for example, adding rings. If the distribution network becomes a
disconnected graph, data connectivity obviously suffers if PLC is used. This vulnerability of the distribution network can be alleviated by judiciously adding wireless
links to complement the PLC-based network with the goal of improving network
connectivity as well as shortest path length characteristics.
7.6.3
the PmF oF the node degrees In the samPle
396-node mv dIstrIButIon netWork
The average node degree of a power grid transmission network tends to be quite low
and does not scale as the network size increases. The topology of a transmission network has salient small-world properties, since it features a much shorter average path
length (in hops) and a much higher clustering coeficient than that of Erdös–Rényi random graphs with the same network size and sparsity. While small-world features have
been recently conirmed for the HV transmission network, the sample MV network
used here implies that a power grid distribution network has a very different kind of
topology from that of an HV network, and obviously it is not a small-world topology.
The node degree distribution of the 396-node MV distribution network, shown in
Figure 7.10, shows that the maximum node degree in the network equals 4, which is
much smaller than that found in the transmission side of the grid, where maximum
nodal degrees of 20 or 30 can be found; and about 16% of the nodes connect to only one
branch, 60% to two branches, 22% to three branches, and only 2% to four branches.
Figure 7.4e depicts the network’s spectral density, which is a normalized spectral
distribution of the eigenvalues of its adjacency matrix. The spectrum of an Erdös–
Rényi random graph network, which has uncorrelated node degrees, converges to
a semicircular distribution (see the semicircle dotted line on the background in
Figure 7.4e). The spectra of real-world networks have speciic features depending
on the details of the corresponding models. In particular, scale-free graphs develop
175
Power Grid Network Analysis for Smart Grid Applications
0.7
0.6
PMF (k)
0.5
0.4
0.3
0.2
0.1
0
1
2
3
k-node degree
4
5
FIGURE 7.10 PMF of the node degrees in the sample 396-node MV distribution network.
100
0.35
Probability mass function
0.30
0.25
10−1
0.20
0.15
10−2
0.10
0.05
0
10−3
0
1
2
Line length (103 m)
3
0
1
2
Line length (103 m)
3
FIGURE 7.11 PMF of the line length in the sample 396-node MV distribution network:
(left) probability versus length; (right) log-probability versus length, where the existence of
an exponential trend in the tail is clearly visible.
a triangle-like spectral density with a power-law tail, whereas a small-world network has a complex spectral density consisting of several sharp peaks. The plot in
Figure 7.4e indicates that the sample MV distribution network is neither a scale-free
network nor a small-world network.
The branch lengths in the MV distribution network are also analyzed. The corresponding PMF is shown in Figure 7.11. It indicates that most of the branches are
shorter than 1067 m (3500 ft) and the branch length distribution has an exponential
tail, with only a very small number of extremely long branches.
176
7.6.4
Smart Grids
lv dIstrIButIon netWork
It is dificult to obtain example data about LV distribution network topologies.
Generally speaking, an LV distribution network is radial, and has a similar network
topology to an MV distribution network, except that it may have more nodes with
shorter branch length.
7.7
SUMMARY
This chapter presents a comprehensive study on both the topological and electrical
characteristics of electric power grids. It inds that the HV transmission network of
the power grid is sparsely connected, with a low average node degree that does not
scale with the network size; the power grid topology manifests obvious small-world
properties, but its small-world rewires are not independent as in the Watts–Strogatz
small-world model; instead, long hauls appear among clusters of nodes. This chapter
provides evidence that the nodal degrees follow a mixed distribution that comes
from the sum of a truncated geometric random variable and an irregular discrete
random variable. It also proposes a method to estimate the distribution parameters
by analyzing the poles and zeros of the average PGF. The chapter studies the graph
spectral density and shows that the power grid network takes on a distinctive pattern
for its graph spectral density. Study of the algebraic connectivity of power networks
has shown that power networks are exceptionally well connected given their sparsity, featuring a better scaling law than the Watts–Strogatz small-world graphs. It is
shown that the distribution of line impedances is heavy-tailed and can be captured
quite accurately by a clipped DPLN distribution. The chapter introduces a novel
random topology power grid model, RT-nested-Smallworld, intending to capture the
above observed characteristics.
Finally, the chapter extends the same topological analysis to the distribution side
of the power grid and reports the results based on a sample 396-node MV distribution network that comes from a real-world US distribution utility mainly located in
a rural area.
REFERENCES
1. D. J. Watts and S. H. Strogatz, Collective dynamics of “small-world” networks, Nature,
393, 393–440, 1998.
2. R. Solé, M. Rosas-Casals, B. Corominas-Murtra, and S. Valverde, Robustness of the
European power grids under international attack, Physical Review E, 026102, 2008.
3. Z. Wang, A. Scaglione, and R. J. Thomas, The node degree distribution in power grid and
its topology robustness under random and selective node removals, in Proceedings of the
1st IEEE International Workshop on Smart Grid Communications (ICC’10SGComm),
23–27 May, Cape Town, South Africa, 2010.
4. R. Albert, I. Albert, and G. L. Nakarado, Structural vulnerability of the North American
power grid, Physical Review E, 69, 1–4, 2004.
5. P. Crucittia, V. Latorab, and M. Marchioric, A topological analysis of the Italian electric
power grid, Physica A, 338, 92–97, 2004.
6. E. W. Dijkstra, A note on two problems in connexion with graphs, Numerische
Mathematik, 1, 269–271, 1959.
Power Grid Network Analysis for Smart Grid Applications
177
7. J. L. Rodgers and W. A. Nicewander, Thirteen ways to look at the correlation coeficient,
The American Statistician, 42(1), 59–66, 1988.
8. M. Newman, The structure and function of complex networks, SIAM Review, 45,
167–256, 2003.
9. D. E. Whitney and D. Alderson, Are technological and social networks really different, in Proceedings of the 6th International Conference on Complex Systems (ICCS06),
Boston, MA, 2006.
10. P. Erdös and A. Rényi, On random graphs. I, Publicationes Mathematicae, 6, 290–297,
1959.
11. R. Albert and A. Barabási, Statistical mechanics of complex networks, Reviews of
Modern Physics, 74(1), 47–97, 2002.
12. University of Washington, Electrical Engineering. Power systems test case archive,
http://www.ee.washington.edu/research/pstca/.
13. E. J. Kirk, Count loops in a graph, Matlab Central, The Mathworks Inc., http://www.
mathworks.com/matlabcentral/ileexchange/10722, 2007.
14. E. P. Wigner, Characteristic vectors of bordered matrices with ininite dimensions, The
Annals of Math, 62, 548–564, 1955.
15. E. P. Wigner, On the distribution of the roots of certain symmetric matrices, The Annals
of Math, 67, 325–328, 1958.
16. M. Fiedler, Laplacian of graphs and algebraic connectivity, Combinatorics and Graph
Theory, 25, 57–70, 1989.
17. W. J. Reed and M. Jorgensen, The double pareto-lognormal distribution: A new parametric model for size distributions, Communications in Statistics, Theory and Methods,
33(8), 1733–1753, 2004.
18. M. Parashar and J. S. Thorp, Continuum modeling of electromechanical dynamics in large-scale power systems, IEEE Transactions on Circuits and Systems, 51(9),
1848–1858, 2004.
19. B. A. Carreras, V. E. Lynch, I. Dobson, and D. E. Newman, Critical points and transitions in an electric power transmission model for cascading failure blackouts, Chaos,
12(4), 985–994, 2002.
20. Z. Wang, R. J. Thomas, and A. Scaglione, Generating random topology power grids, in
Proceedings of the 41st Annual Hawaii International Conference on System Sciences
(HICSS-41), vol. 41, 7–10 January, Waikoloa, HI, 2008.
8
Open Source Software,
an Enabling Technology
for Smart Grid Evolution
Russell Robertson, Fred Elmendorf,
and Shawn Williams
CONTENTS
8.1
8.2
Synopsis ........................................................................................................ 179
Introduction and Brief History ..................................................................... 180
8.2.1 Introduction ...................................................................................... 180
8.2.2 Brief History ..................................................................................... 180
8.3 Beneits and Challenges................................................................................ 181
8.3.1 Beneits ............................................................................................. 181
8.3.2 Challenges......................................................................................... 182
8.3.3 Licensing Considerations .................................................................. 182
8.4 Present State and Examples .......................................................................... 183
8.4.1 Present State ..................................................................................... 183
8.4.2 Examples........................................................................................... 183
8.4.3 Examples: Grid Protection Alliance (GPA) ...................................... 184
8.5 Future of the Electric Utility Industry .......................................................... 189
References .............................................................................................................. 190
8.1
SYNOPSIS
The infrastructure that makes up the electric power grid has always been in a
state of evolution. Design strategies, physical hardware, and materials continue
to improve over time, and the connected nature of the grid makes it impossible
for these changes to take place in isolation. Similarly, with the introduction of
computers and intelligent electronic devices, sources of data continuously multiply and new tools and strategies are developed for better understanding and management of the power grid. A unique interdependency among the companies that
make up the electric power grid creates a need for interactions on common ground.
Open source software (OSS) addresses the need for common ground in application
development.
The purpose of this chapter is to introduce the concepts of OSS, provide a brief
look at OSS history, cover some of its beneits and challenges, show how it meets
179
180
Smart Grids
power system needs through examples of existing tools, and present a future vision
of OSS in the electric utility industry.
8.2
8.2.1
INTRODUCTION AND BRIEF HISTORY
IntroductIon
OSS is a critical part of today’s computing fabric and provides the backbone for
communications, the Internet, and many back-end services. OSS is not the same as
“Freeware,” which is free to obtain but does not include the source code, so the user
assumes the risk that it does what it is supposed to and nothing else. While OSS carries no licensing fees, by deinition it must include the source code. The exposure of
the source code allows everyone to review it and determine that it does what it is supposed to do in the proper way. OSS is maintained in a trusted repository, and code
changes are veriied before distribution. A diagram of the OSS development model
as presented by David Wheeler1 is included in Figure 8.1.
8.2.2 BrIeF hIstory
In the early days of computing, developers viewed writing software as a form of
personal expression. Programming was a functional art form that developers were
happy to share with others in the community. The open sharing of concepts and
techniques facilitated collaboration and allowed rapid advances in software tools.
Computer companies generated most of their income from the hardware, not the
software it supported. However, by the mid- to late 1970s a shift in focus occurred,
and computer manufacturers began to introduce proprietary software2 to increase
differentiation, protect their personal interests, and develop a new income stream.
In the electric utility community, as computing resources became available for
power system engineers and operators, virtually all of their tools were developed
collaboratively and user groups were formed to coordinate activities. Some of the
user-supported activities and early applications have survived through the years,
Source
code
Trusted
repository
Distributors
Problem reports
New feature requests
Trusted
developers
Problem reports
Developers
Executables
Users
FIGURE 8.1 OSS development model.
Open Source Software, an Enabling Technology for Smart Grid Evolution
181
but over time a large majority of applications have become proprietary. An early
response to this transition was the emergence of the term “free software.” However,
this term could be confusing, since to some it meant “free—as in freedom of speech
and expression”3 while to others it meant “free—as in free beer.” In 1998, The Open
Source Initiative (OSI)4 coined the term “open source” as part of a marketing initiative that it hoped would remove ambiguity and convey a more positive message.
In spite of the fact that the majority of applications migrated to proprietary software, a huge portion of the overall computing infrastructure has remained open
source. Some well-known examples are servers that support the Internet and communications, and, more recently, “big data” repositories.
Driven by many converging forces, such as the need for agility, eficiency,
and security, a new emphasis has been placed on developing high-quality open
source tools. A 2003 study showed that OSS was already being widely used in the
Department of Defense, and since that time a number of policy statements have been
made requiring that OSS be considered equally or even preferentially to equivalent
proprietary solutions.
8.3
BENEFITS AND CHALLENGES
OSS provides many beneits compared with similar proprietary applications, but can
be subject to many of the same issues. The quality and reliability of any software are
dependent on the skill of the developer or development team, and are not dictated
by the licensing under which it is distributed. From a user perspective, OSS has the
potential for providing value and lexibility that are not available under a proprietary
license. The following list of beneits is not comprehensive, but presents a variety of
positive qualities that are inherently part of OSS.
8.3.1
BeneFIts
Security: The code visibility provided by OSS provides the opportunity for wide review,
and the lexible nature of OSS allows identiied changes to be deployed rapidly. Jim
Stogdill describes this as maneuverability that can be used in cybersecurity combat.5
No Vendor Lock-in: The user is free to move from one application to another, or
from one source of support to another, and has access to the source code, which can
be maintained or enhanced in-house if desirable.
Low Cost: Since there are no initial or annual licensing fees, there can be substantial savings in the acquisition and deployment of OSS applications. When enhancements are needed, development costs are limited to the task at hand, and do not carry
the burden of supporting a larger organization. Bug ixes and some broadly appealing
enhancements are often donated, which further reduces the total cost of ownership.
High Quality: OSS applications must be of reasonably good quality to be successful in attracting community involvement and engaging users. Once established,
the community provides continuous review by “many eyes,” and there is no way
to “hide” a bug. Additionally, the development capabilities are not limited to one
company’s resources. The code base is maintained by a steward, who veriies any
changes before they are distributed.
182
Smart Grids
State of the Art: OSS is not bound by previous commercial investment, and has
the lexibility to use the latest techniques. It is often developed with intentional
design for extension and enhancement, and the continuous community review avoids
stagnation, which can occur within the developers of a single company.
Perfect Platform for Collaboration: The absence of inancial concerns around
licensing fees and the freedom to share code and ideas with a large community make
OSS an ideal vehicle for bringing together many participants from different entities
such as research, academia, government, utility, and vendors.
Facilitates Innovation: The collaborative aspect of OSS described above eliminates the “not invented here” syndrome, and provides developers with more freedom
to experiment. Community sharing of ideas and reviewing each other’s work foster
creativity.
Decreases Time to Deployment: The community environment of OSS allows
direct application of the concept “many hands make light work,” and decreased
restrictions combined with increased innovation and creativity result in decreased
development time. There are no contract negotiations and no inancial incentive to
stall a release to recoup the investment in the previous version, which also facilitates
more timely delivery of initial deployment and enhancements.
8.3.2
challenges
The biggest challenges for OSS are misconceptions and a lack of education.
Nonproprietary can be misinterpreted to mean noncommercial, but OSS is typically
very high-quality, well-managed, commercial-grade code. Another misconception is
that OSS is developed by inexperienced college students; typically it is developed by
skilled developers or teams. No licensing fee can be misunderstood to mean no cost,
but training, support, transition, and enhancements are usually fee based. Uneducated
assumptions regarding OSS could include the notions that it is less reliable and has
more bugs than proprietary software, but major studies have shown that, in comparisons of OSS with proprietary operating systems and web server software, OSS scores
much more favorably than proprietary counterparts. David Wheeler includes a good
discussion of OSS myths and reality in his 2012 presentation titled “Open Source
Software (OSS or FLOSS) and the US Department of Defense (DoD).”
Another type of challenge for OSS is status quo and momentum. Electric utility
companies are necessarily risk averse, and often ind comfort in continuing on the
current course of action. However, as cost savings become ever more important, and
the increased lexibility and agility of OSS are better understood, the resistance to
change is decreasing.
At the most basic level, OSS faces the same challenges and constraints as any proprietary development. It has to be the right solution for the problem, it has to be developed and maintained professionally, and it has to provide the desired user experience.
8.3.3
lIcensIng consIderatIons
Users of OSS have less concern than developers about speciic licensing issues,
as long as they are not involved in new development, and have identiied adequate
Open Source Software, an Enabling Technology for Smart Grid Evolution
183
resources for maintenance and support. Developers of OSS need to carefully consider their goals regarding further development, marketing options, and integration
with other software. OSI, a nonproit organization, was formed in 1998 to be the
steward of the Open Source Deinition, build a community, educate, and promote the
awareness and importance of nonproprietary software. It maintains a comprehensive
library of open source licenses and is a valuable resource when exploring new licensing options.
8.4
PRESENT STATE AND EXAMPLES
8.4.1
Present state
OSS is a vehicle to effectively share software platforms and technologies throughout the electric industry. This promotes a common ground for communication, data
sharing, analysis, planning and management of the grid. It removes the common
hindrances of the traditional proprietary software model. Methods and processes are
developed by consensus of all parties with a reduced chance of miscommunications
and misunderstandings.
As noted on the OSI web page, OSS is continuing to gain acceptance. It is demonstrating the ability to deliver on the promise of better quality, higher reliability,
more lexibility, lower cost, and no vendor lock-in. Fueled by the MITRE study for
the US DoD in 2003,6 OSS has been continually gaining ground in US government agencies. In addition to the DoD, the US Department of Energy (DoE) is also
encouraging OSS and using it as the vehicle of choice for technology transfer in
applications developed through the department’s research projects. With the everincreasing recognition that OSS is a viable contender as a source for applications,
electric utilities are taking advantage of this opportunity to reduce cost while maintaining quality and increasing lexibility.
8.4.2
examPles
Two of the best-recognized and most successful OSS projects are the Linux7 operating system and the Apache web server. These are foundations of our modern
computing world, but most electric utility users are not directly involved in those
communities. Two examples of very successful OSS projects for power system applications are UWPFLOW8 and TEFTS9 from The University of Waterloo, Ontario,
Canada.
Listed below are additional examples of OSS in the power grid:
IEEE Task Force on Open Source Software for Power Systems: http://ewh.
ieee.org/cmte/psace/CAMS_taskforce/index.htm
This Task Force explores the potential for open source software (OSS) in the
Power Engineering Society (PES). The mission of the Task Force is twofold:
(i) to diffuse the philosophy of OSS in the power systems community; and
(ii) promote OSS for the beneit of the PES ranging all the way from simple
pedagogical OSS to commercial-grade OSS.
184
Smart Grids
OpenPMU: https://sites.google.com/site/openpmu/
The OpenPMU project is a suite of tools to help researchers create new power
system technologies. The project was originally founded at Queen’s University
Belfast, UK, and subsequently joined by the SmarTS Lab at KTH Stockholm,
Sweden. For more information on the origins of the project, see here.
GridLAB-D: http://www.gridlabd.org/
GridLAB-D™ is a new power distribution system simulation and analysis
tool that provides valuable information to users who design and operate
distribution systems, and to utilities that wish to take advantage of the latest energy technologies. It incorporates the most advanced modeling techniques, with high-performance algorithms to deliver the best in end-use
modeling. GridLAB-D™ is coupled with distribution automation models
and software integration tools for users of many power system analysis
tools.
GridLAB-D™ was developed by the U.S. Department of Energy (DOE)
at Paciic Northwest National Laboratory (PNNL) under funding for Ofice
of Electricity in collaboration with industry and academia.
Total Grid Community: http://www.totalgrid.org/
An open source community dedicated to enabling a smarter energy grid.
Green Energy Corp is pleased to sponsor this Community and looks forward
to continued partnership with open source developers world-wide.
OpenDSS: http://electricdss.sourceforge.net/
The OpenDSS is an electric power Distribution System Simulator (DSS) for
supporting distributed resource integration and grid modernization efforts.
OpenETran: http://sourceforge.net/projects/openetran/
OpenETran can presently simulate multi-conductor power lines, insulators,
surge arresters, nonlinear grounds, and lightning strokes. It eficiently calculates energy and charge duty on surge arresters, and iterates to ind the
critical lightning current causing lashover on one or more phases. It is also
suitable for use in substation insulation coordination. Capacitor switching,
TRV, and other applications may be added.
8.4.3
examPles: grId ProtectIon allIance (gPa)
One example of a software development company dedicated to the development of
OSS tools is GPA. GPA is a not-for-proit corporation formed to facilitate and support the development and deployment of comprehensive electric energy system-wide
solutions and enhancements to electric sector security through integrated, adaptable,
collaborative and rapidly evolving programs of threat identiication and analysis, and
infrastructure research, development, demonstration, and technology transfer. GPA
focuses on improvements to electric energy resiliency and maintenance of reliability
and availability. The efforts will accommodate the evolving technology of electric
energy systems, including grid modernization, advanced generation sources, and
enhanced systems integration.
Open Source Software, an Enabling Technology for Smart Grid Evolution
185
The Grid Solutions Division of GPA is located in Chattanooga, Tennessee, and
develops OSS applications to meet the speciic demands of real-time data collection
and storage to support electric grid operations and planning, and an extensible platform for the automation of disturbance analysis.
GPA products are hosted in the CodePlex open source project community provided
by Microsoft. GPA maintains product documentation and code on the CodePlex project
web sites. For those who wish to contribute code, GPA has published coding guidelines.
Many of GPA’s current products focus on synchrophasor applications.
Synchrophasors are global positioning system (GPS) time-stamped measurements
taken rapidly (at least 30 times per second) from different points on the power grid.
These synchronized measurements are transmitted in real time to operations centers,
where they provide information on the state and health of the power system.
Synchrophasor information can be used to monitor power lows on the electric
grid and to identify areas or conditions where problems exist. More information
about synchrophasors can be found in the October 2010 North American Electric
Reliability Corporation (NERC) report, “Real-Time Application of Synchrophasors
for Improving Reliability.”10
GPA’s most recent development has been the creation of a framework to facilitate
enterprise-class automated disturbance analysis systems. A research and demonstration project initially funded through the Electric Power Research Institute (EPRI) for
the Tennessee Valley Authority has been extended to facilitate any type of analysis
for disturbance recordings. This analysis can range from near real-time fault location to the assessment of historical records for steady-state system changes. A paper
entitled “Open Source Fault Location in the Real World”11 was developed for the
2013 Georgia Tech Fault and Disturbance Analysis Conference.
GPA software products are developed in the Microsoft .NET framework using the
C# language, and all are managed through Codeplex12 projects and licensed under
the Eclipse Public License.13 As with other OSS projects, there is no licensing fee
for any of our products, but support is offered through either time and materials, or
subscription-based agreements. GPA also hosts an annual users’ forum and tutorial
where attendees can learn directly from the GPA development team and through
presentations from other GPA product users.
The GPA software environment is shown in Figure 8.2.
A brief description of each GPA tool is included below.
PMU Connection Tester: https://pmuconnectiontester.codeplex.com/
The PMU Connection Tester administered by the Grid Protection Alliance
(GPA) veriies that a data stream from synchrophasor measurement device
is being successfully received. The PMU connection tester supports the following phasor data protocols: IEEE C37.118 (Version 1/Draft 7, Draft 6),
IEEE 1344, BPA PDCstream, SEL Fast Message, UTK FNET streaming
data protocol, IEC 61850-90-5 and Macrodyne.
openPDC: http://openpdc.codeplex.com/
The openPDC administered by the Grid Protection Alliance (GPA) is
a complete Phasor Data Concentrator software system designed to process streaming time-series data in real-time. Measured data gathered with
186
Smart Grids
GPA products
OpenPDC
SIEGate
(openPG)
PMU connection
tester
OpenXDA
openFLE
OpenHistorian
...and others
Platform building blocks
Time series processing core
Low-level services
Security services
Grid
solutions
framework
MS framework class library
Common language runtime
Platform
Operating system
Hardware
FIGURE 8.2 GPA software environment.
GPS-time from many hundreds of input sources is time-sorted and provided
to user deined actions as well as to custom outputs for archival. Screen shots
and an overview are shown in Figure 8.3.
openHISTORIAN: https://openhistorian.codeplex.com/
The openHistorian administered by the Grid Protection Alliance (GPA) is a
system for collecting, archiving, retrieving and displaying time-series event
data in real-time from multiple data sources.
A version of the openHistorian (Version 1.0) has been included with the
openPDC since its initial release in 2009. This project is intended to extend and
productize this initial version of the openHistorian to improve its performance
and its lexibility to store and retrieve a broader range of time-series data types.
This work on the openHistorian 2.0 began in early 2012. This site has been
established so the community can follow and participate in this development
activity. A binary alpha release of the openHistorian 2.0 was made available
in 2013.
The openHistorian 2.0 has been optimized for:
•
•
•
•
•
Assurance of archived data integrity/continuity
Broad data source connectivity
High-performance data capture and retrieval
Eficient, high-volume data storage
High availability. An overview of key features is shown in Figure 8.4.
Open Source Software, an Enabling Technology for Smart Grid Evolution
187
Output adapter layer
Measurements
Input adapter layer
Action adapter layer
FIGURE 8.3 openPDC overview.
SIEGate: https://siegate.codeplex.com/
The objectives of the SIEGate (the Secure Information Exchange Gateway
pronounced Psy-gate) project are (1) to improve the security posture and
minimize the external cyber-attack surface of electric utility control centers,
and (2) to reduce the cost of maintaining current control-room-to-controlroom information exchange.
.d
188
Smart Grids
Collect/convert
Configure
Alarm/serve RT data
Save
SDK allows adapters to be developed for data consumption
Recent data served with very low latency
High-efficiency data storage with lossless data compression
Extract
Tools provided for bulk historical data extraction
Publish
Both publish/subscribe API and web services interfaces
Visualize/analyze
Companion openVisN tool allows users to view and export data
FIGURE 8.4 openHistorian features.
Funded by the US Department of Energy, the project team includes the
Grid Protection Alliance (GPA) as the project prime and principal software
developer, with the University of Illinois at Urbana-Champaign (Illinois) as
a research, design, development, and testing partner; the Paciic Northwest
National Laboratory (PNNL) as a design review, testing, and evaluation
partner; PJM as a demonstration partner; and Alstom Grid as a demonstration and commercialization partner. Work on SIEGate began in 2010 and is
expected on conclude in 2013.
As shown below, SIEGate provides a security isolation layer between
critical internal infrastructure and external systems to protect reliability and
market sensitive data. SIEGate reduces the cost of data exchange through
ease-of-coniguration.
SIEGate implements a true publish-subscribe architecture where the sending gateway owner authorizes data as available for subscription by speciic
consuming gateways. Once authorized, the consuming gateway automatically discovers the data that have been made available to it by other SIEGate
nodes and allows the selective subscription to them. SIEGate data available
for publication and subscription includes measurements, such as SCADA or
synchrophasor data; iles, such as SDX iles; and higher-level notiications or
alarms that are of signiicance for overall grid operation. These alarms may
be conigured to promulgate to all interconnected SIEGate nodes so that
global alarms can be raised. See Figure 8.5 for an overview.
openPG: https://openpg.codeplex.com
SIEGate supersedes all future builds of the openPG. All development work
on the openPG has ended and efforts have been transferred to SIEGate as the
future platform for secure synchrophasor data exchange
openXDA: https://openxda.codeplex.com
openXDA is an extensible platform for processing event iles from disturbance
monitoring equipment such as digital fault recorders (DFRs) and power quality
meters. In one implementation of openXDA analytics, based on the waveform
data in these iles and on user-provided transmission line parameters, openXDA
determines the fault type and calculates the line-distance to the fault.
Open Source Software, an Enabling Technology for Smart Grid Evolution
SIEGate
manager
application
SCADA/
EMS
SE/CA
Gateway
input
Data
channel
Reliability
plans
Outage
data
Gateway
output
External
Firewall
Firewall
Command
channel
SIEGate
189
Phasor
data
Control center infrastructure
Internal
FIGURE 8.5 SIEGate overview.
Extensible disturbance analytics
DFR (COMTRADE)
PQmeter (PQDIF)
SCADA data
Phasor data
XDA file
watcher
Data positioning
Analytics
openFLEϟ
Structured data
Time
domain
Analytics
...
Frequency
domain
Analytics
...
Presentation
internal web; tablet
TED.db
Transmission
event
data
Notification
SCADA; email
System integration
reporting, detailed analysis
FIGURE 8.6 openXDA overview.
openXDA is an extension of work conducted in 2012 on openFLE for the
Electric Power Research Institute (EPRI) which is posted under the project http://EPRIopenFLE.codeplex.com/. The openFLE project included
an open source C# parser for power quality iles (IEEE 1159.3-2003). For
more information, see http://www.gridprotectionalliance.org/pdf/openFLE_
Overview_Landscape.pdf. See Figure 8.6 for an overview.
8.5
FUTURE OF THE ELECTRIC UTILITY INDUSTRY
The blackout of 2003 has had an effect on the power grid that has lasted well beyond
those few hours. It was a clear wake-up call and provided strong motivation to
improve grid monitoring and control strategies, and to seriously consider closed-loop
control systems. Since the blackout, a multitude of projects to improve the resiliency
of the grid have been initiated, ranging from the DoE and Smart Grid Investment
Grants, to self-funded projects at forward-thinking electric companies. The recurring theme is the collection, processing, analyzing, archiving, and sharing of data,
and new tools have been developed through these projects.
At the October 16–18, 2012, North American SynchroPhasor Initiative14 conference, Western Electricity Coordinating Council announced a trial run of a closed-loop
190
Smart Grids
control system at a Bonneville Power Administration site. This is the irst experiment
of this magnitude, and GPA products will be used to collect and manage the data.
The demonstrated success of OSS in the utility industry and increasing government
support, combined with increasing pressure for eficient, agile, secure software applications, will result in even wider development and deployment of OSS platforms and tools.
Grid Open Source Software Alliance (GOSSA) is a new nonproit organization
being formed as a resource center for OSS in the electric utility industry. The GOSSA
initiative is supported by the DoE, NERC, EPRI, large utilities, academia, and many
other industry stakeholders. GOSSA will provide a single point of reference for
electric power-focused OSS. An industry-wide GOSSA workshop was held at the
National Rural Electric Cooperative Association headquarters near Washington,
DC, on September 6, 2013.
OSS is a vital and growing sector in today’s computing environment, and is the
perfect platform for evolutionary changes needed to improve tools for monitoring
and managing the electric power grid. Personal research and education in this area
are strongly encouraged, and a list of references is provided at the end of this chapter.
REFERENCES
1. D. A. Wheeler, 2009, Open Source Software (OSS or FLOSS) and the U.S. Department
of Defense (DoD). Essay, November 4, http://www.dwheeler.com/essays/dod-oss.pdf.
2. BusinessDictionary.com,
http://www.businessdictionary.com/deinition/proprietarysoftware.html.
3. GNU operating system, http://www.gnu.org/philosophy/free-sw.html.
4. Open Source Initiative, http://www.opensource.org/history.
5. J. Stogdill, Coding is maneuver. http://www.slideshare.net/jstogdill/coding-is-maneuver.
6. MITRE DoD 2003, http://dodcio.defense.gov/Portals/0/Documents/FOSS/dodfoss_pdf.
pdf.
7. The Linux Foundation, http://www.linuxfoundation.org/.
8. University of Waterloo, UWPFlow, University of Waterloo. http://ece.uwaterloo.
ca/~ccanizar/software/plow.htm.
9. University of Waterloo, Transient Stability and Energy Function Analysis, University of
Waterloo. http://ece.uwaterloo.ca/~ccanizar/software/tefts.htm.
10. NERC (North American Electric Reliability Corporation), 2010, Real-time application of synchrophasors for improving reliability. NERC. http://www.nerc.com/docs/oc/
rapirtf/RAPIR%20inal%20101710.pdf.
11. Open Source Fault Location in the Real World, http://172.21.1.150/pdf/Open%20
Source%20Fault%20Location%20in%20the%20Real%20World%20(R3).pdf.
12. CodePlex, http://www.codeplex.com/.
13. EPL (Eclipse Public License), v. 1.0, http://www.eclipse.org/legal/epl-v10.html.
14. NASPI (North American SynchroPhasor Initiative), http://www.naspi.org/.
9
Contribution of
Microgrids to the
Development of
the Smart Grid
Tine L. Vandoorn and Lieven Vandevelde
CONTENTS
9.1
9.2
Introduction .................................................................................................. 191
Microgrids .................................................................................................... 194
9.2.1 Beneits ............................................................................................. 194
9.2.2 Operating Modes .............................................................................. 195
9.2.2.1 Grid-Connected Mode ....................................................... 195
9.2.2.2 Islanded Mode.................................................................... 196
9.2.2.3 Remote Islanded Mode ......................................................200
9.2.3 Smart Microgrids.............................................................................. 201
9.3 Microgrids in the Smart Grid Paradigm.......................................................202
9.3.1 Similar Beneits and Aims................................................................202
9.3.2 Microgrid Control .............................................................................202
9.3.2.1 Smart Microgrid Control ...................................................202
9.3.2.2 Local Microgrid Control ....................................................202
9.4 Conclusion ....................................................................................................208
Acknowledgments..................................................................................................208
References ..............................................................................................................209
9.1
INTRODUCTION
An approach to dealing with the large increase in decentralized unpredictable
power sources, the aging grid infrastructure, and the increasing (peak) consumption
of electric power, while mitigating the massive investments required for this, is to
add more intelligence to the power system. The use of widespread information and
communication technologies (ICT) to monitor and manage the distributed energy
resources (DER) allows the grid to become more eficient and sustainable [1]. These
ICT and new remote management abilities couple the grid elements via an interactive intelligent electricity network, the so-called smart grid. In brief, a smart grid
involves the use of sensors, communication, computational ability, and control to
191
192
Smart Grids
enhance the overall functionality of the electric power delivery system [2]. The smart
grid offers many advantages, such as increasing the share of clean renewable energy,
reducing costs, improving system reliability in the face of increasing intermittency,
enabling more customer choices and involvement in their energy management, a
more eficient usage of the current assets, and help with using electric vehicles [3].
One barrier to the development of the smart grid is that the individual beneits for
consumers should be clear. One of the main challenges to increasing customer acceptance of smart grids is dealing with customers’ concerns regarding cost, privacy, and
security, together with their fear of possible discomfort when they provide lexibility
to the grid, for example, potential disruptions for peak shaving. It is thus important to
overcome the lack of understanding about the smart grid’s functionality and increase
customers’ knowledge of the beneits for them speciically. As well as lacking awareness about smart grids, consumers are not aware of which appliances are contributing most to their electricity bills. This leads to frustration because, despite the best
of intentions, they are unable to signiicantly lower their bills [4]. Smart devices can
help to address this issue by monitoring their consumption and comparing it with
that of other consumers. End-user acceptance will ultimately determine the success
of many aspects of smart grids [4].
With the massive penetration of distributed generation (DG) units, the current
it-and-forget principle of integrating these units into the electric power system is
no longer a sustainable option, and a coordinated approach is required. One method
to capture the emerging potential of DG, and to cope with the problems caused by
the unconventional behavior and increasing penetration of DG, is to take a system
approach instead of considering each unit separately [5–8]. In the system approach,
the generators and loads are regarded as a subsystem, or microgrid, as depicted in
Figure 9.1. Microgrids are small-scale electricity networks, consisting of an aggregation of (converter-interfaced) DG units, (controllable) loads, and storage elements,
which are connected to the utility network through a single point of connection, the
point of common coupling (PCC), as shown in Figure 9.2. In comparison with a single DG unit, a microgrid has more control lexibility to ensure the system’s reliability
and power quality requirements [9]. By cooperating, the microgrid elements can
jointly achieve certain objectives, such as beneiting from market participation and
Distributed generation units (DG)
Power-electronic
converter
=
Stand-alone mode
PV panel
Wind turbine
CHP
Fuel cell
AC
Grid-connected
Energy storage
mode
(Power-electronically interfaced)
Utility grid
PCC
distributed loads
FIGURE 9.1 Microgrid with (power-electronically interfaced) loads, storage, and DG units
in stand-alone (islanded) or grid-connected mode.
193
Contribution of Microgrids to the Development of the Smart Grid
G
G
Generators
MV distribution network
MV/HV Transmission
HV/MV
DG
transformer network (HV) transformer
MV/LV
transformer
DG
DG
PCC
PCC
LV network
DG
DG
DG
LV microgrid
LV microgrid
FIGURE 9.2 Microgrid connected to utility network through a PCC.
reducing electricity costs through lattening of the aggregated load proile [10,11].
Hence, microgrids are regarded as an important element in the successful integration
of massive amounts of DG units and renewable energy sources [12,13], and help with
power quality issues [11,14,15]. Microgrids also offer the potential to integrate local
electrical and thermal networks to achieve coherent generation, storage, and load
control, and to beneit from combined heat and power (CHP) generation, which is an
important means of eficiency improvement [9].
Smart grids focus on system-wide improvements, with microgrids as the fundamental building blocks of the smart grid [16]. The grid will not become smart at
once, but through a gradual evolution; hence, the smart grid will probably emerge as
a system of integrated smart microgrids [3,17]. Microgrids are better suited than the
overall smart grid to clarify the beneits for consumers, because they are inherently
geographically conined entities in which the microgrid elements are aggregated to
focus on the speciic beneits within the local entity, while smart grids focus on the
more system-wide beneits.
Microgrids can provide power to a small community, which can range from a residential district or an isolated rural community, to academic or public communities
such as universities or schools, to industrial sites. Industrial parks can be managed as
microgrids, for example, to decrease energy dependency, operate as low-carbon business parks,* and increase economic competitiveness (increasing reliability, reducing
the purchase of energy, reducing peak consumption). Microgrids can operate either
connected to the utility network, in the grid-connected mode, or independently of a
main grid, in islanded mode.
The key objectives of microgrids are to provide a coordinated integration of DG in
the electric power system, improve reliability, allow more eficient use of energy, and
become a controllable entity in the network. Hence, the objectives of smart grids and
microgrids are closely related. In this sense, microgrids are fundamental building
blocks of the smart grid that speciically focus on local issues and beneits, making
end users not only consumers but also active participants in the electricity network
by enabling them to control and manage (part of) the energy that they will consume.
* www.ace-low-carbon-economy.eu.
194
9.2
Smart Grids
MICROGRIDS
9.2.1 BeneFIts
Microgrids offer signiicant beneits from both the grid perspective and the consumer
perspective. A key advantage from the grid’s point of view is that the microgrid elements are collectively regarded as a single controllable unit, enabling the microgrid
to deliver the cost beneits of large units [18]. In this way, utilities do not have to
consider each unit separately for their system management. While this is not always
obvious to utilities, microgrids can improve the economics and reliability of their
service, help to meet renewable energy obligations, reduce congestion, improve
power quality under high penetration levels of DG, and support demand response.
For example, microgrids can optimize a “troublesome” business complex or a university campus, making it more economic and reliable to serve [1].
From the customers’ point of view, irstly, the impact of the microgrid on the
reliability of the distribution network is relevant, certainly in the future, with more
unpredictable generation and higher consumption (peaks) [19–21]. Reliability is the
irst requirement of the electric power system. The cost of unreliability can have a
considerable impact on the economy. The outage cost in the United States is estimated to amount to some $79 billion per year, of which momentary outages (<5 min)
account for $52 billion [22,23]. The consequences of grid failure are gigantic,
because industry as a whole, and the transportation, inancial, communication, and
other critical sectors in particular, are largely dependent on a guaranteed and reliable
supply of energy [24]. Most of the reliability improvements will need to be introduced in the distribution system, where most outage and power quality issues arise.
To increase local reliability, backup gensets and other emergency supplies are often
installed near critical loads to provide uninterruptible power supply (UPS) functionality [25]. The costs of these designated devices, which only run for a limited amount
of time, are signiicant. In this context, microgrids can provide UPS functionality by
exploiting locally available DG, storage, and loads.
Secondly, aggregation of mixed assets in a microgrid can also provide considerable economic beneits to the microgrid subsystems by allowing them to participate
in the electric power and ancillary services markets.
Other beneits that can result from the introduction of microgrids are reduced
feeder losses, removal of transmission and distribution bottlenecks, reactive power
and local voltage support, various other ancillary services such as heterogeneous
power quality and reliability (i.e., differentiation based on the speciic needs of the
customers), increased eficiency through CHP, and easier large-scale integration of
DG units [19,26–29].
The main reasons for aggregating units in a microgrid are:
• To deal with power quality challenges. For example, some customers require
a higher power quality than others. In this sense, it is more economically
attractive to increase the power quality locally than to upgrade the power
quality of the entire power system. With respect to voltage quality in the distribution networks, microgrids are especially beneicial in providing voltage
control, as voltage is a local issue in contrast to grid frequency.
Contribution of Microgrids to the Development of the Smart Grid
195
• To increase reliability. The power system is exposed to incidents that may
affect the security of supply, such as lighting strikes, arc lashes on rainy days,
falling trees, man-made faults, and other accidents. Although these incidents
are mostly covered by the protection devices, they can affect the reliability
of the system. The common defense mechanism is to invest in grid assets.
However, grid extensions, for example, in order to avoid the failure of one line
leading to a blackout of a part of the system, are increasingly countered by
social resistance, following the not-in-my-backyard (NIMBY) principle. Even
if these investments are made, natural disasters can completely annihilate
parts of the transmission and distribution infrastructure. Hence, even if a certain area is not directly affected by the disaster, the power supply may be interrupted for a long time. Microgrids can restore the local system more quickly.
• Optimal utilization of DG. Microgrids can coordinate their local generation, consumption, and storage to use the available (renewable) energy
optimally. Also, the connection of DG units in low-voltage (LV) networks
can give rise to overvoltage problems. Rather than using the current full
on-off control of the DG units, the microgrid can coordinate the loads and
generation units to restrict the voltage in the microgrid while capturing as
much renewable energy as possible.
• Peak load limitations. A microgrid can manage its own generation and load,
and hence reduce peak consumption, which can have a signiicant impact
on the electricity bill.
• Reduction of transmission and distribution losses (with the subsequent
energy costs). By locally coordinating its assets, the microgrid can become
more independent of the rest of the power system.
9.2.2 oPeratIng modes
Thanks to the single connection point between the microgrid and the utility network,
microgrids can run in two operating conditions: grid-connected and islanded modes.
Depending on the operating mode, the microgrid can tackle one or more of the aforementioned issues.
9.2.2.1 Grid-Connected Mode
In the grid-connected mode, the microgrid supports the utility grid while exchanging
power with it. In this mode, the microgrid can take economic decisions, such as
controlling the power exchange with the utility network dependent on on-site generation, cost, consumption, and prices on the energy markets. For energy management,
grid-connected microgrids have the following three assets at their disposal [8]:
1. DER controllers (DG and optional storage)
2. Load management
3. Control of the power exchange with the main grid
Generally, the normal condition of the microgrid is grid-connected operation. A irst
advantage is that, for the rest of the network, the microgrid can be seen as a controllable
196
Smart Grids
entity. This provides signiicant beneits both for the microgrid participants, through
scaling and aggregating, and for the distribution network operator, who does not need to
consider all units separately [19,27–29]. A second feature is that, because of the single
connection point between the microgrid and the utility network, the power exchange
can be determined unambiguously and controlled to a predeined value. Hence, in the
electricity markets, there is no need to measure and trade the output power of each unit,
but solely the aggregated exchange. From the customer’s perspective, the main beneit of
a grid-connected microgrid is that it can deliver the required scaling beneits to enable
proitable participation in the electricity markets. First, aggregation of several DG units
can enable units that would be too small individually (a minimum capacity is required
in the markets) to participate in the electricity markets and get better prices for their
energy. Secondly, by aggregating different kinds of units, the risks of deviating from
predictions of production and consumption can be reduced. Thirdly, microgrids can
provide ancillary services to the networks, such as reserve provision or reactive power
support, become more self-reliant, and decrease transmission and distribution losses.
9.2.2.2 Islanded Mode
Although the normal operation mode of a microgrid is grid connected, it offers the
unique characteristic of islanding. By locally generating the consumed energy, and
controlling the consumption to cancel the interchange of energy with the main grid,
the microgrid can operate in a stable islanded mode. In islanded mode, the microgrid
operates independently of the main grid, and, consequently, the DG units, loads,
and storage devices are collectively responsible for maintaining the integrity of the
microgrid without the assistance of a main grid.
There are various reasons to go into islanded mode. Reliability is one of the major
concerns of the electric power system, mainly due to the high outage costs. These
costs are, however, different for different types of customers, for instance:
• Residential consumers mainly face discomfort in the case of a power outage.
• Industrial facilities can be faced with large costs due to loss of production.
• Hospitals are affected signiicantly in regard to personal safety.
Critical systems such as data centers and hospitals have emergency backup power,
mostly from diesel-powered generators, but these are ineficient, environmentally
unfriendly, and costly. Diesel generators are prone to failure, and diesel delivery
can be problematic in the case of large grid outages when the transportation routes
are cut off. Also, unlike microgrids, irstly, there is usually a delay between the grid
outage and the full operation of diesel generation, and, secondly, diesel backup systems are not built to run for a long time. The transition from a purely diesel-powered
backup generator to a hybrid microgrid system, which integrates different on-site
sources with energy storage and load control, signiicantly increases the reliability
of the backup system [30]. Hence, in order to ensure the continuity of supply for the
local load, a (short-term) islanded mode can occur in the case of special situations
such as grid faults, outage of the bulk supply, or power quality problems [12,31–33].
It is important to note that the grid fails more often than most people realize.
In an average year, according to a 2008 Lawrence Berkeley National Laboratory
Contribution of Microgrids to the Development of the Smart Grid
197
(LBNL) study, the power is out for an average of 92 min in the Midwest and 214 min
in the Northeast of the United States. Many of these power outages are weather
related, such as caused by tornadoes, hurricanes or tropical storms, ice storms, lightning, wind, and rain. According to the Center for Research on the Epidemiology
of Disasters, 100–200 million people were affected by weather-related disasters
between 1980 and 2009, with economic losses ranging from $50 billion to $100 billion annually. Another possible reason for grid outages is that much of the infrastructure that serves the US power grid is aging. The average age of power plants is
over 30 years, with most of these facilities having a life expectancy of 40 years [34].
Electric transmission and distribution system components are similarly aging, with
power transformers averaging over 40 years of age and 70% of the transmission lines
being 25 years old or older. LBNL statistics show that 80%–90% of all grid failures
in North America originate at the distribution level of electricity service [3,23]. For
instance, momentary sags and surges on local distribution feeders are a common
cause of breakdowns and work stoppages at high-tech facilities such as semiconductor fabs, research labs, and data centers. The remaining 10% of the grid failures stem
from generation and transmission problems, which can cause wider-scale outages
affecting larger numbers of customers.
Microgrids are able to deal with these power outages by operating in islanded
mode. In this way, microgrids offer a potential improvement of the reliability, quality, and costs of the system. Hence, microgrids are a good option in applications
in which electrical energy must be guaranteed at all times, for example, hospitals,
server centers, and industrial facilities. However, they are a step beyond either emergency backup systems or stand-alone solar panel arrays. They use special software
and power electronics to integrate multiple sources of power and energy storage to
provide electricity around the clock, even when the sun is not shining or regulations
limit the use of diesel generators.
Other than dealing with rare (weather) incidents, islanding is also useful when the
main grid is not robust enough due to factors such as long distances from the main
grid [35]. In this way, in large, highly dispersed countries such as Canada, the United
States, and Japan, major efforts into microgrid research are being made. Microgrids
can also switch into islanded mode due to planned maintenance operations.
In conclusion, true microgrids are much more than backup systems. They include
real-time control to match the microgrid generation, storage, and consumption. In
the normal, grid-connected mode, they interact with the utility network in order to
deliver power when the cost is higher and consume power from the grid when the
cost is lower. In islanded mode, they can achieve independence from the main grid.
Examples
1. Reliability: Hurricane Sandy (United States)
In October 2012, superstorm Sandy reached the eastern shores of the
United States. This led to an electricity outage in many areas. Over two
million connections were de-electriied, of which 750,000 were in New
York City. Two days after the superstorm, six million American people still
had no electricity, of whom two-thirds lived in the states of New York and
198
Smart Grids
New Jersey, according to the U.S. Department of Energy. President Obama
stressed that restoration of the electricity supply was a priority. However,
1 week after Sandy, 1.4 million American people still had no electricity,
according to the Department of Energy in Washington. This shows the
vulnerability of the electricity grid when faced with extreme weather circumstances and the dificulty of restoring this massive system. Microgrids
offer a unique opportunity to provide a more reliable energy supply.
Although many places had no electricity, a number of colleges, including Princeton and New York Universities, relied on their microgrids to keep
power lowing to vital services during the Sandy storm. Princeton University
typically gets its electric power from both the local grid and on-site generation. During Hurricane Sandy, Princeton was able to switch off the grid and
power part of the campus with about 11 MW of local generation.
2. Reliability: 2011 Earthquake and Tsunami (Tohoku, Japan)
After the Great East Japan Earthquake and Tsunami in 2011, around
4.4 million households in northeastern Japan were left without electricity. In
some areas, the outages lasted for weeks. Likewise, the port city of Sendai
was heavily hit, and faced 2 days of outage. But in one small section of the
city the lights stayed on. The Tohoku Fukushi University had an experimental microgrid project fed by three types of energy generators: fuel cells, solar
panels, and natural gas microturbines. The microgrid system was able to
island and operate as an independent energy source, making it less vulnerable to the problems in the larger system. For 2 days, the microgrid continued to provide reliable power to a number of loads, including a hospital,
while the larger macrogrid had halted power supply.
Even if a certain area is not directly affected by a natural disaster, the
power supply may be interrupted for weeks or even months if the connection to the utility network is interrupted by the event. As the microgrid is
not dependent on the utility grid, the immediate construction of microgrids
appears feasible in some areas [36], for example, in Japan after the tsunami.
Microgrids can be planned and assembled in a short time. Furthermore,
a power system disturbance can trigger a cascading outage, in which
microgrids are virtually unaffected.
3. Reliability: Jail (United States)
At the Santa Rita Jail in California, a local microgrid project was oficially
launched in 2012. Since mid-1990, the jail has made a major effort in eficiency upgrades. This has been pushed by the local government to protect
the jail against rolling blackouts and spikes in the electricity prices, due
to the California electricity crises in 2000–2001. Therefore, the jail has a
signiicant number of photovoltaic (PV) panels, a fuel cell power plant, and
some small wind turbines as on-site generators, as well as some diesel generators and batteries for backup. Next to eficiency improvements (lower
costs), power reliability is a primary concern for the jail to ensure that large
amounts of electricity are constantly kept available. Now, the facility forms
a microgrid that can operate in islanded mode. The system keeps the power
on, for example, when storms take down the grid, which is essential for
safety at the maximum security facility.
4. Integration of DG: Jail (United States)
In Alcatraz, near San Francisco, there is a similar story. A feasibility study
conducted by researchers from the University of Washington indicated
that installing an undersea transmission cable from Alcatraz Island to
Contribution of Microgrids to the Development of the Smart Grid
5.
6.
7.
8.
199
San Francisco would be too costly. Hence, at irst, some diesel generators
were responsible for feeding the island. However, as the price of diesel
began to climb, and the cost of solar PV fell, developing a state-of-the-art
microgrid appeared attractive. Thanks to the combination of a PV installation and battery packs, it is expected that the solar-powered microgrid will
reduce the running time of the diesel generators from 100% to 40%.
Power quality: Industrial areas
In highly sensitive industrial areas, for example, chemical or semiconductor manufacturing facilities, reliable power of high quality is required. High
power quality is also very important in modern health care [37]. Electronic
equipment is commonly designed for power supplies with speciic voltage
and current characteristics. When voltage or current levels deviate from the
speciied quality standards, electronic equipment can be damaged or fail,
which may cause high-tech medical equipment to malfunction, resulting in
extended patient discomfort, misdiagnoses, increased equipment downtime
and service costs, and even life-threatening situations [37].
Health-care facilities with extremely sensitive electronic equipment and
high reliability needs are increasingly recognizing the limitations of traditional solutions to costly power quality problems. Microgrids can be tailored
to the speciic needs of these end users and provide so-called premium
power customers with inely calibrated power lows [37]. They can provide
an integrated power supply with different levels of power quality. Hence,
instead of alleviating the power quality levels of the entire grid, microgrids
can ensure the required power quality locally at a more acceptable cost.
Maintenance: Hydro-Québec (Canada)
The islanded microgrid of the Canadian electric utility Hydro-Québec
was the result of an urgent need for the maintenance of the transmission line feeding a substation named Senneterre, to which a privately
owned thermal power plant (Boralex) was connected. The substation
feeds three distribution lines, serving 3000 customers in the municipality
of Senneterre and its surrounding area. Standing on wooden structures,
this line, more than 55 years old, required urgent replacement of its angle
and stop portals [38]. This type of maintenance can only be performed
on a de-energized line [38]. To avoid service interruption to customers
during the restoration of the transmission line, the Boralex power plant
was used for islanding the microgrid. The irst maintenance work of the
transmission line began in fall 2006 [38].
Mobile: Military bases
The U.S. Department of Defense gives high priority to mobile tactical
microgrids. These microgrids need to be highly modular. They need to be
deployed very quickly (in a matter of days), and afterward they need to
be deconstructed and moved to another location. The ability to rapidly
construct and deconstruct is not the only reason for using the microgrid
concept in military systems. Microgrids can signiicantly attenuate the cost
of the transport of oil and gas to military bases, which are often stationed in
remote locations.
Reliability and energy cost: BC Hydro Boston Bar substation (Canada)
In areas with frequent loss of electrical power, for example, in countries with
a weak grid infrastructure, very long lines, or harsh weather circumstances,
islanded systems can reduce the loss of electrical power. An example is
deployed by the Canadian electric utility BC Hydro. In the Boston Bar area,
200
Smart Grids
BC Hydro can switch to the islanded operating mode of the interconnected
feeder during outages in its radial transmission line. This line had proved
to be insuficiently reliable because it is a single radial supply with a long
feeder, which is, moreover, subjected to frequent outages because of falling
rocks and trees, mud, storms, and motor vehicles. By including the ability to
island, the reliability of the system has signiicantly increased.
9. Dealing with the growing demand for electricity: Independence
The increasing use of electricity* demands the construction of new power
plants and power lines. However, due to the common NIMBY attitude,
these investments are being signiicantly delayed. Microgrid structures can
ease this attitude, as they limit the construction of new lines, and, when
they are installed, the beneit to the local consumer is obvious, as opposed
to the case of building long transmission lines. Hence, they can relieve distribution and transmission expansion issues.
9.2.2.3 Remote Islanded Mode
Although not strictly according to the deinition of a microgrid, an islanded microgrid
can also exist in the case of remote electriication, where no main grid is available
due to, for example, geographical issues. According to the World Bank’s 2010 development report, 1.6 billion people in developing countries do not have access to electricity [39]. The most important reason is the extensive investment needed to install
power lines across large distances, across rough terrain, or both, in order to extend
the electricity supply to a few people. It is recognized that electricity is a key driver
of economic growth and combating poverty. Hence, islanded microgrid projects provide great opportunities to realize sustainable human development. According to the
International Energy Agency (IEA 2009), 83% of people who do not have access
to electricity live in rural areas [40], for example, over one-third of India’s rural
population. The Indian government has set the target of providing all Indian households with electricity in the near future [41]. If a conventional grid connection is
not cost effective to accomplish this, consideration is given to decentralized electric
power with local distribution, that is, islanded microgrids, with an example in the
Sundarbans Islands region in India [41].
Two examples of remote islanded microgrids are given below.
1. Serving loads in remote locations that lack infrastructure (Uganda, India)
Uganda has a limited supply of electricity, with less than 10% of the
Ugandan population having access to electric power. Rural areas in particular are not yet electriied, because of the high costs of the grid extensions
this would require. Without electricity, economic growth is constrained,
limiting the chances of combating poverty in these regions. Also, hospitals
are only able to deliver a limited service. A solution can be provided by
small, decentralized (hydro)power stations. Therefore, in Uganda, several
projects are running, such as the Suam project, which enables the use of
small microgrids, often fed by hydropower, for electriication.
* For energy eficiency reasons, both transportation and heating are becoming more and more electriied, for example, through the use of electrical vehicles and heating pumps.
Contribution of Microgrids to the Development of the Smart Grid
201
Remote microgrids are also suitable for electrifying Indian villages that do
not yet have power. Nearly 400 million Indian people, most living in rural
communities, lack access to the grid. To access electricity, these people
need to go to the city. They also generally use kerosene-illed lamps, which
are unhealthy and unsafe, instead of electric lighting. Microgrids offer a
great opportunity to deliver electricity to these communities.
2. Serving islands (Greece)
Many of the Greek Aegean islands are not connected to the mainland grid,
and thus rely on polluting, costly, and ineficient diesel generation to fulill their electricity demands. Fortunately, most islands have optimal wind
potential and solar resources. The exploitation of these resources would
provide a cheaper alternative to the current diesel generation and result in
signiicant savings and lower electricity prices.
9.2.3
smart mIcrogrIds
Realistically, the growth of the smart grid will not be a revolution. Instead, a
gradual evolution is expected [3,17]. Because of their lexibility and scalability,
microgrids are likely to play a key role in the evolution of the smart grid [42],
with smart microgrids as small pilot versions (building blocks) of the future
power system [16]. Hence, the smart grid will probably emerge as a system of
integrated smart microgrids [3]. Smart microgrids are microgrids combined with
an overlying intelligence scheme. The intelligent software implemented in the
microgrid can:
• Contain energy management systems
• Monitor the system (energy supply, storage, and demand) and actively intervene in consumption/generation
• Identify and maximize energy eficiency opportunities
• Use an extra communication and sensor layer to maximize cost savings and
reduce carbon emissions
• Generate more active participation by consumers
A balance between the cost of incorporating intelligence into the grid and the subsequent beneits needs to be achieved. Furthermore, the infrastructure and control
centers for smart grids can be hard to implement on a large scale, and can take many
years. Rather than investing in incorporating intelligence on a large scale, investing
in small smart microgrids (such as business areas) can be done at a lower cost and
more quickly. Also, a high level of intelligence built into all parts of the system is
not necessary: different areas require different levels of smartness. The areas that
allow more and beneit most from high levels of smartness of the system, such as
areas with high penetration of DG, can it directly into the microgrid concept. In
these microgrids, smart features can be installed more quickly than in the rest of the
network, with local levels of intelligence higher than average. In this way, the smart
microgrids can enable a new energy strategy while avoiding the cost of making the
whole system smart.
202
9.3
9.3.1
Smart Grids
MICROGRIDS IN THE SMART GRID PARADIGM
sImIlar BeneFIts and aIms
Essentially, the goals of microgrids and smart grids are the same: to minimize costs,
meet the growing demand, integrate more sustainable generation resources by maximizing generation assets, and increase the eficiency of the power system.
The main difference between microgrids and smart grids is that the microgrid is a
building block of the overlying smart grid. Microgrids aim at local control and optimization of their local assets, with a contribution to the higher-level utility network. Because
of this local scope, microgrids directly focus on the value proposition of their customers.
In contrast, as smart grids focus on more system-wide issues, an important challenge for
smart grids is to make the value of the achieved beneits clear to customers.
9.3.2 mIcrogrId control
As in the conventional power system, microgrid control can be conceived as a hierarchical control structure. Microgrid control can be divided into:
• Local primary control
• Supporting (smart grid) services in an overlying (secondary/tertiary) control;
smart microgrid control
9.3.2.1 Smart Microgrid Control
Smart microgrid control is conceived as a high-level (secondary) control structure,
which deines new set points for the local controllers. Smart microgrid control can
beneit fully from smart grid features, such as communication and metering data.
With respect to this control level, the microgrid is a building block of the smart grid.
Like the primary controller, the secondary controller can tackle technical issues, but
it can deliver more functionalities, such as achieving economic, societal, and environmental objectives. The secondary controller can achieve further optimization of
the microgrid (e.g., fuel savings), coordinate the primary control actions, restore preagreed consumption patterns, restore the voltage to its nominal value (amplitude, frequency, or both), and control the import and export of microgrid energy. For instance,
the secondary or tertiary controller can address the social issues of the producers if
the physical position in the grid impacts the amount of power that can be injected (and
hence also impacts the inancial reimbursement for the energy delivered to the grid).
9.3.2.2 Local Microgrid Control
The local microgrid control is a primary control level operating without communication. In this sense, it goes beyond the smart grid control, as it is faster and depends
on local measurements only. In islanded microgrids, the local microgrid controller
is responsible for the reliability of the system. In grid-connected microgrids, the
local controller can contribute to achieving a proper voltage quality with minimal
communication effort. The contribution of the local microgrid controller to easing
smart grid communication is clear, as the local controller can achieve fast and reliable control actions, not depending on communication and remote measurements.
Contribution of Microgrids to the Development of the Smart Grid
203
An overview of various primary control strategies is given in [43]. For (extended)
islanded microgrids, the droop concept is most promising to ensure reliable system
operation. The active power/grid voltage (P/V) droop concept in general and the voltage-based droop (VBD) controller [44] speciically are highlighted in this chapter.
Droop controllers in microgrids are derived from the well-known active power/grid
frequency (P/f) droop control principle incorporated into the large central synchronous generators in the conventional electric power system. While the P/f droop control
method works well in a microgrid with mainly inductive line impedances, it leads to
concern when implemented on an LV microgrid, without signiicant inertia, and where
the line resistance should not be neglected [9]. In the case of mainly resistive lines, the
active power (P) is linked with the voltage amplitude difference (V), while the reactive
power (Q) is linked with the phase angle difference and hence frequency (f). This leads
to P/V and Q/f droops as opposed to the conventional P/f and Q/V droops [45–47]. The
active and reactive powers are measured and drooped to obtain the root-mean-square
(RMS) voltage and its frequency (ω = 2πf), respectively:
Vg,i = Vg,ref − K P ( Pi − Pi,ref )
(9.1)
ωi = ωref + KQ ( Qi − Qi,ref )
(9.2)
where KP and KQ are the droop coeficients.
The VBD control of [44] is a variant of the P/V droop control that focuses on
the optimal integration of renewables in the network and also incorporates a direct
current (dc)-bus controller. In the VBD control, the P/V droop controller is divided
into two droop controllers, and constant-power bands are included, as depicted in
Figure 9.3. Further details of these control loops are given in [44]. The parameter 2b
determines the width of the constant-power band. When the terminal voltage of the
DG unit is in the constant-power band, that is, (1 − b)Vg,ref < Vg < (1 + b)Vg,ref, the DG
unit delivers its reference power. Otherwise, the power of the DG unit is changed.
A distinction is made between dispatchable and less dispatchable DG units.
For dispatchable DG units, analogously to central large generators, Pref represents
the scheduled power that can be determined in the electricity markets. In this case,
the value of b is small, such that the unit reacts to small variations of the grid voltage.
For less dispatchable DG units, such as many renewables, a wider constant-power
band is used. In this way, these units do not react to all load variations, but change
their output power only in the case of more extreme voltage variations compared with
dispatchable DG units. These extreme voltages only occur when the power limits of
the dispatchable DG units are nearly reached. Consequently, an optimized integration
of renewables is possible, because the VBD control prioritizes the power changes of
the units, by setting different values of b, without the need for communication. In
these units, Pref can, for example, represent the instantaneous maximum power point
of a wind turbine. Also, as voltage is a local parameter, the voltage limits are sometimes already reached in areas where the ratio of dispatchable to less dispatchable DG
units is low. The current it-and-forget strategy of integrating DG solves these voltage
problems by turning the DG units off. In contrast, the VBD controller uses soft power
204
Smart Grids
Vg/Vdc droop controller
Vg
P/Vg droop controller
Pdc
Vg
Vdc
Pdc
Pdc,nom
Vg,nom
bVg,nom bVg,nom
Vdc
Vg
Vg,nom
Vdc,nom
Q/f droop controller
Q
vg,droop
f
f
∫
sin
vg,ref
+
Duty ratio
VSI
Voltage
controller
α
fnom
–
Q
Qnom
ig
zv
FIGURE 9.3 VBD controller and virtual impedance loop.
curtailment to capture more of the renewable energy. In this way, voltage problems
can be overcome, as the renewables also take part in the voltage control.
As shown in Figure 9.3, the Vg/Vdc and Q/f droop controllers, which are digital
controllers, determine the droop voltage vg*,droop,k = Vg,k sin ( α k ) (k is the discrete time
instance). Discrete values are used because pulse width modulation with sampling
period Ts is used in the converter. The amplitude Vg of the droop voltage is obtained
from the Vg/Vdc droop controller and the phase angle is obtained from the frequency
f in the Q/f droop controller. Together, they determine
v*g,droop,k = Vg,k sin ( α k −1 + 2πfkTs )
(9.3)
Figure 9.3 also shows the virtual output impedance loop. A resistive output
impedance zv = Rv is chosen, as this provides more damping in the system [48] and
complies with the power control strategies of the loads and generators, where the
active power is changed based on the grid voltage:
vg* = vg*,droop − Rvig
where:
v*g
is the reference voltage for the voltage controller
*
vg,droop is the voltage obtained by the VBD controller
ig
is the current injected into the microgrid, as illustrated in Figure 9.4
(9.4)
205
Contribution of Microgrids to the Development of the Smart Grid
Rest of microgrid
LC filter
VSI
P
Line
impedance
Primary energy source
Pdc
Pc
T1
ig
T4
Vdc
Pdc
L
Load
C
+
Cdc
vg
T2
T3
FIGURE 9.4 DG unit connected to the microgrid. Parameters (vg, ig, Vdc, P, Pdc).
This chapter illustrates the contribution of the VBD control in a grid-connected
microgrid to operating the grid in a smarter manner.
1. Market operation: The amount of power export/import is preagreed in the
markets, for example, on a day-ahead basis. For example, when the electricity price is high, the microgrid will import as little power as possible from
the grid, or will even export power. One requirement for market participation of a microgrid is thus that the power exchange between the microgrid
and the utility networks can be controlled to an agreed value. This is conventionally achieved by using a microgrid central controller that communicates new set points to all units. However, this may lead to possible delays
and a signiicant communication burden. Therefore, in [49], the usage of a
so-called smart transformer is discussed. The smart transformer is a tapchanging transformer, connected at the PCC of the microgrid, as depicted
in Figure 9.5. The transformer is smart in the sense that the control strategy for changing the microgrid-side voltage is able to control the power
exchange to the set value by changing its taps. In medium-voltage networks,
on-load tap-changing transformers are sometimes already available; hence,
controlling these as smart transformers requires few modiications. In the
lower-voltage networks, most PCC transformers are manual tap-changing
transformers, from which the voltage can only be controlled off-line and
PCC
PCC transformer
Tap changer
Microgrid
PPCC
···
PCC
Utility network
VPCC
DG units
Loads
···
FIGURE 9.5 Smart transformer located at the PCC of a microgrid.
206
Smart Grids
not automatically. This puts a signiicant stress on electrical grids that face
increased penetration of DG units. Historically, the planning of LV grids
is based on a worst-case scenario, ensuring that, in the case of maximum
consumption, the voltage does not drop below the lower voltage limit.
Therefore, many (manual) tap-changing transformers at the beginning of
the LV lines are set somewhat above the nominal voltage. However, with the
increasing presence of DG units, the risk of overvoltages becomes higher.
Also, planning becomes more dificult because of the larger voltage variations (e.g., sunny daytime vs. nighttime). Hence, the ability to change the
tap settings automatically becomes more useful, as it is more effective,
faster, and cheaper to implement than the conventional approach of investing in more lines and larger transformers. Therefore, in future, it may be
beneicial to install an on-load tap changer (OLTC). Moreover, it is well
known that many of the assets in distribution networks are reaching the end
of their life and will have to be replaced in the coming years. Hence, the
manual transformers can gradually be upgraded to OLTCs.
The OLTC transformer with smart control strategy, that is, the smart
transformer, controls the PCC power (PPCC) by changing the microgrid-side
voltage (VPCC). If the microgrid elements are equipped with the VBD control
strategy, these elements automatically react by changing their input/output
power depending on the grid voltage. For example, when the PCC power
exported to the utility network is lower than its preagreed value (PPCC,ref ),
the smart transformer will lower VPCC. The microgrid DG units react to this
voltage drop by increasing their output power P, hence increasing PPCC.
In this way, the PCC power can be controlled by the smart transformer,
without the need to communicate new set points to all grid elements as
they automatically react. A second advantage is that, in this way, a virtual
islanded mode is achieved. The utility grid is not seen as a slack bus, but
is conceived as a constant-power load or generator, that is, generating the
power agreed upon in the markets. The smart transformer can facilitate
distribution network management, as the distribution network operator only
needs to communicate to the microgrid central controller, which, instead
of facing all of the microgrid elements separately, can rely on an automatic
response of the microgrid elements to a changing grid voltage induced by
the smart transformer.
2. Voltage control in LV networks: Although developed for islanded
microgrids, the VBD controller can be used in grid-connected microgrids
as well [50]. Moreover, it offers signiicant advantages, irstly in the voltage
control (avoiding overvoltage) and secondly in avoiding on-off oscillation
of DG units (increasing the capture of renewable energy).
Firstly, a well-known issue when connecting DG units in LV systems is
possible overvoltage due to the injection of active power by the DG unit.
Generally, in LV networks, other than the system planning, there is little or
no means of voltage control. On the other hand, the extensive penetration of
small DG units into these networks offers a unique opportunity to use their
controllability for voltage control in the network.
207
Contribution of Microgrids to the Development of the Smart Grid
The conventional approach to voltage control is by injecting or consuming reactive power, mirroring the voltage control in the transmission
network. In [51], the required amount of reactive power Q that a DG unit
should consume to erase its impact on the voltage is calculated. A minimal
impact is achieved when the DG unit consumes (minus sign) reactive power
according to:
Q=−
R
P,
X
(9.5)
where R and X are the line parameters and P is the generated active power
of the DG unit. Here, a uniform distribution of load along the feeder and
a constant resistance and reactance per unit length are assumed. Still, a
small impact remains due to the power losses associated with the transport
of power over the network, which are not included in Equation 9.5. In this
chapter, LV networks are considered, which are mainly resistive; hence,
they have a high R/X, generally greater than 3. According to Equation 9.5,
in these networks with R ≫ X (LV microgrids), a DG unit needs to consume
a lot of reactive power to minimize the voltage rise due to its active power
injection into the grid. Conversely, reactive power consumption has only
a small effect on the voltage of the system. Hence, it will predominantly
increase the network losses. Furthermore, such large amounts of reactive
power generally cannot be consumed by the generators without signiicant
overrating.
The VBD control achieves an automatic soft curtailment of the DG
units, but because of the usage of constant-power bands, this is done in a
predeined priority order. The renewables are softly curtailed only when
absolutely necessary. In this way, the DG units can control the system voltage by using the VBD controller, without the need for designated devices
for voltage control.
Currently, renewables fully turn off when their terminal voltage exceeds
a certain voltage level. As a single unit can have a large impact on the grid
voltage, this may lead to a signiicant voltage drop, such that the unit may
1,000
245
240
0
200
400
600
800
1,000
1,200
Output power (W)
Voltage (V)
250
0
1,400
FIGURE 9.6 On/off control of DG leading to grid oscillations: measurements of
PV panels in Oostende, Belgium (— = DG unit 1, - - - = DG unit 2, ––––– = VPCC).
208
Smart Grids
TABLE 9.1
Renewable Energy Capture: Comparison
of On/Off Control versus VBD Control
DG 1 (E [J])
DG 2 (E [J])
DG 3 (E [J])
On-Off
Control
VBD
Control
Gain
(%)
9794
9122
4002
11,539
9402
4741
18
3
18
turn on again. This may lead to on-off oscillations of renewables in LV
networks. An example of a measurement of such an oscillation is depicted
in Figure 9.6. The on-off oscillations can also occur between multiple units
that turn on and off in turn. In [50], it is illustrated that the grid-connected
VBD control can, irstly, avoid these on-off oscillations, and, secondly, subsequently achieve a higher renewable energy capture. A result from this
paper is depicted in Table 9.1. The output energy for three DG units, which
are connected to a feeder with multiple RL loads, is shown. The details of
the coniguration are given in [50]. The on-off control of the units leads to
on-off oscillations in the system, which, in turn, lead to a lower renewable
energy capture than with VBD control. Hence, although soft curtailment
instantly results in a loss of the available renewable energy, it may increase
the total energy capture by avoiding on-off oscillations, when the priority is
set, for example, by using the VBD concept.
9.4 CONCLUSION
This chapter illustrates that microgrids can signiicantly contribute to developing the
smart grid. Some examples of microgrids in grid-connected and islanded modes are
given, including the beneits they bring to the smart grid. The microgrid control consists of two layers: a fast and automatic primary controller and an overlying slower
controller. The latter can take full advantage of the communication and metering
of the smart grid, and hence its fully into the scope of the smart grid. The primary
controller contributes signiicantly to the smart grid in the sense that it facilitates the
smart grid control and eases the communication burden. This is illustrated through
a discussion of the contribution of the VBD control, irstly, in controlling the power
exchange between microgrid and utility networks with a minimum of communication, and, secondly, in voltage control.
ACKNOWLEDGMENTS
The work of Dr. Ir. T. Vandoorn is inancially supported by a Fellowship of the
FWO-Vlaanderen (Research Foundation Flanders, Belgium). This research has been
carried out in the frame of the Interuniversity Attraction Poles Programme initiated
by the Belgian Science Policy Ofice (IAP-VII-02).
Contribution of Microgrids to the Development of the Smart Grid
209
REFERENCES
1. K. Iniewski (ed.), Smart Grid Infrastructure & Networking. McGraw-Hill Professional,
New York, 2012.
2. C. W. Gellings, The Smart Grid, Enabling Energy Eficiency and Demand Response.
The Fairmont Press, Lilburn, 2009.
3. H. Farhangi, The path of the smart grid, IEEE Power and Energy Magazine, 8(1), 18–28,
2010.
4. DNV KEMA, Global inventory and analysis of smart grid demonstration projects,
Technical Report, October 22, Netbeheer Nederland, 2012.
5. R. H. Lasseter, J. H. Eto, B. Schenkman, J. Stevens, H. Vollkommer, D. Klapp, E. Linton,
H. Hurtado, and J. Roy, CERTS microgrid laboratory test bed, IEEE Transactions on
Power Delivery, 26(1), 325–332, 2011.
6. M. Popov, H. Karimi, H. Nikkhajoei, and V. Terzija, Modeling, control and islanding
detection of microgrids with passive loads, in Proceedings of 14th International Power
Electronics and Motion Control Conference (EPE/PEMC), September 6–8, Ohrid,
Republic of Macedonia, 2010.
7. A. Engler, O. Osika, M. Barnes, N. Jenkins, and A. Arulampalam, DB1 local micro
source controller strategies and algorithms, European Commission, February, 2004.
www.microgrids.eu/micro2000.
8. J. Driesen and F. Katiraei, Design for distributed energy resources: Microgrid planning
and architectures for improved reliability and integration, IEEE Power and Energy
Magazine, 6(3), 30–40, 2008.
9. Y. W. Li and C. N. Kao, An accurate power control strategy for power electronics
interfaced distributed generation units operating in a low voltage multibus microgrid,
IEEE Transactions on Power Electronics, 24(12), 2977–2988, 2009.
10. R. H. Lasseter, A. Akhil, C. Marnay, J. Stephens, J. Dagle, R. Guttromson, A. Meliopoulous,
R. Yinger, and J. Eto, The CERTS microgrid concept, white paper on integration of distributed energy resources, in California Energy Commission, Ofice of Power Technologies—
U.S. Department of Energy, LBNL-50829, www.certs.lbl.gov, April, 2002.
11. R. H. Lasseter and P. Paigi, Microgrid: A conceptual solution, in Proceedings of IEEE
Power Electronics Specialists Conference (PESC 2004), June, Aachen, Germany, 2004.
12. M. Bollen, J. Zhong, O. Samuelsson, and J. Björnstedt, Performance indicators for
microgrids during grid-connected and island operation, in Proceedings of IEEE
PowerTech Conference, June 28–July 2, Bucharest, Romania, 2009.
13. A. Vaccaro, M. Popov, D. Villacci, and V. Terzija, An integrated framework for smart
microgrids modeling, monitoring, control, communication, and veriication, Proceedings
of IEEE, 99(1), 119–132, 2010.
14. N. Pogaku, M. Prodanović, and T. C. Green, Modeling, analysis and testing of autonomous operation of an inverter-based microgrid, IEEE Transactions of Power Electronics,
22(2), 613–625, 2007.
15. M. A. Pedrasa and T. Spooner, A survey of techniques used to control microgrid generation and storage during island operation, in Proceedings of the 2006 Australian
Universities Power Engineering Conference (AUPEC2006), December 10–13, 2006.
16. M. Savaghebi, A. Jalilian, J. C. Vasquez, and J. M. Guerrero, Secondary control scheme
for voltage unbalance compensation in an islanded droop-controlled microgrid, IEEE
Transactions on Smart Grid, 3(2), 797–807, 2012.
17. M. Kezunovic, V. Vittal, S. Meliopoulos, and T. Mount, The big picture: Smart research
for large-scale integrated smart grid solutions, IEEE Power and Energy Magazine,
10(4), 22–34, 2012.
18. A. G. Tsikalakis and N. D. Hatziargyriou, Centralized control for optimizing microgrids
operation, IEEE Transactions on Energy Conversion, 23(1), 241–248, 2008.
210
Smart Grids
19. P. M. Costa and M. A. Matos, Assessing the contribution of microgrids to the reliability
of distribution networks, Electric Power Systems Research, 79(2), 382–389, 2009.
20. K. De Brabandere, Voltage and frequency droop control in low voltage grids by distributed generators with inverter front-end, PhD Dissertation, KU Leuven, Leuven,
Belgium, October, 2006.
21. R. H. Lasseter and P. Piagi, Control and design of microgrid components, Power Systems
Engineering Research Center (PSERC), Technical Report, January, 2006.
22. X. Tan, Q. Li, and H. Wang, Advances and trends of energy storage technology in
microgrid, International Journal on Electrical Power and Energy Systems, 44(1),
179–191, 2013.
23. K. H. LaCommare and J. H. Eto, Understanding the cost of power interruptions to
U.S. electricity consumers, Ernest Orlando Lawrence Berkeley National Laboratory,
Technical Report, 2004.
24. S. M. Amin and B. F. Wollenberg, Toward a smart grid, IEEE Power and Energy
Magazine, 3(5), 34–41, 2005.
25. C. Marnay and R. Firestone, Microgrids: An emerging paradigm for meeting building
electricity and heat requirements eficiently and with appropriate energy quality,
Environmental Energy Technologies Division, US Department of Energy, presented at
the European Council for an Energy Eficient Economy 2007 Summer Study, Technical
Report, June 4–9, La Colle sur Loup, France, 2007.
26. T. C. Green and M. Prodanovíc, Control of inverter-based micro-grids, Electric Power
Systems Research, 77(9), 1204–1213, 2007.
27. H. Jiayi, J. Chuanwen, and X. Rong, A review on distributed energy resources and
microgrid, Renewable and Sustainable Energy Reviews, 12(9), 2472–2483, 2008.
28. C. Marnay and G. Venkataramanan, Microgrids in the evolving electricity generation
and delivery infrastructure, in Proceedings of IEEE PES General Meeting, June 18–22,
Montréal, QC, 2006.
29. P. Basak, A. K. Saha, S. Chowdhury, and S. P. Chowdhury, Microgrid: Control
techniques and modelling, in Proceedings of the 44th International Universities’ Power
Engineering Conference (UPEC 2009), September 1–4, Glasgow, Scotland, 2009.
30. T. Casey, The spiders microgrid project, 2013 February 12. Available at: www.
cleantechnica.com.
31. European Commission, New ERA for electricity in Europe. Distributed generation: Key
issues, challenges and proposed solutions, 2003. Available at: www.smartgrids.eu.
32. F. Katiraei, M. R. Iravani, and P. W. Lehn, Micro-grid autonomous operation during and
subsequent to islanding process, IEEE Transactions on Power Delivery, 20(1), 248–257,
2005.
33. H. Mohamad and P. Crossley, Islanded operation of UK radial distribution earthing
strategy, in Proceedings of the 44th International Universities’ Power Engineering
Conference (UPEC 2009), September 1–4, Glasgow, UK, 2009.
34. R. J. Campbell, Weather-related power outages and electric system resiliency,
Congressional Research Service Report for Congress, Technical Report, August 28,
2012.
35. E. Planas, A. Gil-de-Muro, J. Andreu, I. Kortabarria, and I. Martínez de Alegría, General
aspects, hierarchical controls and droop methods in microgrids: A review, Renewable
and Sustainable Energy Reviews, 17(5), 147–159, 2013.
36. Siemens, Microgrid white paper, 2011. Available at: www.energy.siemens.com.
37. Center for Energy, Marine Transportation and Public Policy at Columbia University,
Microgrids: An assessment of the value, opportunities and barriers to deployment in New
York state, New York State Energy Research and Development Authority (NYSERDA),
Technical Report, September, 2010.
Contribution of Microgrids to the Development of the Smart Grid
211
38. M. Gauthier, C. Abbey, F. Katiraei, J.-L. Pepin, M. Plamondon, and G. Simard,
Planned islanding as a distribution system operator tool for reliability enhancement, in
Proceedings of 19th International Conference on Electricity Distribution, May 21–24,
Vienna, Austria, 2007.
39. The World Bank, Washington DC, World development report 2010, development and
climate change, 2010. Available at: http://web.worldbank.org.
40. International Energy Agency (IEA), Access to electricity (WEO-2011), 2011. Available
at: www.worldenergyoutlook.org/.
41. I. Mitra, T. Degner, and M. Braun, Distributed generation and microgrids for small
island electriication in developing countries: A review, SESI Journal, Solar Energy
Society of India, 18(1), 6–20, 2008.
42. N. Lidula and A. D. Rajapakse, Microgrids research: A review of experimental microgrids
and test systems, Renewable and Sustainable Energy Reviews, 15(1), 186–202, 2011.
43. T. L. Vandoorn, J. D. M. De Kooning, B. Meersman, and L. Vandevelde, Review of
primary control strategies for islanded microgrids with power-electronic interfaces,
Renewable and Sustainable Energy Reviews, 19, 613–628, 2013.
44. T. L. Vandoorn, B. Meersman, L. Degroote, B. Renders, and L. Vandevelde, A control
strategy for islanded microgrids with dc-link voltage control, IEEE Transactions on
Power Delivery, 26(2), 703–713, 2011.
45. A. Engler and N. Soultanis, Droop control in LV-grids, in Proceedings of International
Conference on Future Power Systems, November 16–18, The Netherlands, 2005.
46. A. Engler, O. Osika, M. Barnes, and N. Hatziargyriou, DB2 evaluation of the local controller strategies, January 2005. www.microgrids.eu/micro2000.
47. C. Sao and P. Lehn, Control and power management of converter fed microgrids, IEEE
Transactions on Power Systems, 23(3), 1088–1098, 2008.
48. W. Yao, M. Chen, J. M. Guerrero, and Z.-M. Qian, Design and analysis of the droop
control method for parallel inverters considering the impact of the complex impedance
on the power sharing, IEEE Transactions on Industrial Electronics, 58(2), 576–588,
2011.
49. T. L. Vandoorn, J. D. M. De Kooning, B. Meersman, J. Guerrero, and L. Vandevelde,
Voltage-based control of a smart transformer in a microgrid, IEEE Transactions on
Industrial Electronics, 60(4), 1291–1305, 2013.
50. T. L. Vandoorn, J. D. M. De Kooning, B. Meersman, and L. Vandevelde, Voltage-based
droop control of renewables to avoid on-off oscillations caused by overvoltages, IEEE
Transactions on Power Delivery, 28(2), 845–854, 2013.
51. M. H. J. Bollen and F. Hassan, Integration of Distributed Generation in the Power
System, IEEE Press Series on Power Engineering, Wiley, Hoboken, NJ, 2011.
10
Microgrids
Mietek Glinkowski, Adam Guglielmo, Alexandre
Oudalov, Gary Rackliffe, Bill Rose, Ernst
Scholtz, Lokesh Verma, and Fang Yang
CONTENTS
10.1 Introduction to Microgrids ........................................................................... 214
10.1.1 What is a Microgrid? ........................................................................ 214
10.1.2 Why Microgrids? .............................................................................. 214
10.1.3 Microgrid Market ............................................................................. 216
10.2 Microgrid Design .......................................................................................... 217
10.2.1 Optimal Microgrid Topology ........................................................... 217
10.2.1.1 Traditional Distribution Network Topologies .................... 218
10.2.1.2 Size and Location of Distributed Energy Resources ......... 220
10.2.2 Microgrid Control ............................................................................. 220
10.2.2.1 Centralized Approach ........................................................ 221
10.2.2.2 Decentralized Approach .................................................... 222
10.2.3 Microgrid Protection ........................................................................ 223
10.2.3.1 Adaptive Protection ........................................................... 223
10.2.3.2 Fault-Current Source.......................................................... 226
10.2.3.3 Fast Static Switch ............................................................... 226
10.2.3.4 Communication Architectures and Protocols for
Adaptive Protection ........................................................... 227
10.2.4 Selecting the Right Communications Technology ........................... 227
10.2.5 Direct Current Microgrids ................................................................ 228
10.2.5.1 Differences Between ac and dc for Microgrids ................. 230
10.2.5.2 Examples of dc Microgrids ................................................ 231
10.2.5.3 Control and Protection of dc Microgrids ........................... 233
10.3 Microgrid Operations ................................................................................... 235
10.3.1 Operation Objectives ........................................................................ 236
10.3.2 Technical Challenges ........................................................................ 236
10.3.2.1 Intermittent Renewable Generation ................................... 236
10.3.2.2 Low Grid Inertia ................................................................ 237
10.3.2.3 Coordination among Distributed Energy Resources ......... 237
10.3.3 Grid-Connected Operation ............................................................... 237
10.3.4 Island-Mode Operation ..................................................................... 238
10.3.4.1 Spinning Reserve ............................................................... 238
10.3.4.2 Minimum Loading of Conventional Generators ............... 238
10.3.4.3 Step-Load Requirements ................................................... 239
213
214
Smart Grids
10.3.4.4
10.3.4.5
10.3.4.6
10.3.4.7
Power System Stability and Oscillations ........................... 239
Reactive Power Requirement and Voltage Control ............ 239
Limitations to Renewable Energy Penetration................... 239
Short-Term Energy Storage Helps Renewable
Energy Penetration .............................................................240
10.4 Microgrid Economic Value Streams ............................................................240
10.5 Reliability ..................................................................................................... 241
10.5.1 Reliability Due to Maintenance ........................................................ 242
10.5.2 Microgrid System Reliability ........................................................... 242
10.5.3 Can Reliability of a Microgrid be Better Than That of a Utility
Grid? ................................................................................................. 243
10.6 Outlook: Microgrids—Energy for the Digital Society ................................. 243
References .............................................................................................................. 247
10.1
10.1.1
INTRODUCTION TO MICROGRIDS
What Is a mIcrogrId?
An integrated energy system consisting of distributed energy resources
(DERs) and multiple electrical loads operating as a single, autonomous
grid either in parallel to or “islanded” from the existing utility power grid.
Pike Research
The microgrid concept is a logical evolution of simple distribution networks and
can accommodate a high density of various distributed generation sources. A typical
microgrid power system consists of: generators; wind turbines; solar photovoltaic
(PV) arrays; other renewable technologies such as geothermal generation; main grid
connection/interconnector switch (optional); energy-storage devices such as lywheels and batteries for short- and long-term storage and stabilization; grid management software; controllable loads; and a robust communication network. The typical
size of a microgrid is 500 kW–15 MW.
10.1.2
Why mIcrogrIds?
Some of the various reasons for increased interest in microgrids include:
• Rising Cost and Burdens of Transmission and Distribution (T&D)
Infrastructure: Building new T&D infrastructure has become dificult in
some areas due to permission issues, public resistance, and the dificulty,
cost, or both of upgrading or building new infrastructure.
• Integration of Renewables and Storage Technologies: The greatest
obstacle to the integration of renewables has been the risk to grid stability. Some renewable energy sources, most notably wind and solar, are
Microgrids
•
•
•
•
•
215
intermittent by nature, which increases the voltage and frequency luctuation on the grids. The risk is greatest when there is a high penetration
of renewables.
However, technology advancements in grid stabilization and energy storage have addressed these concerns in microgrids. Nowadays, the challenge
of renewable energy generation peaking not matching up with demand
peaks can be resolved using energy-storage modules. An energy-storage
module is a packaged solution that stores energy for use at a later time.
The energy is usually stored in batteries for speciic energy demands or to
effectively optimize cost. The modules can store electrical energy and supply it to designated loads as a primary or supplementary source. Moreover,
it provides a stable and continuous power supply regardless of the supply
source status. Voltage and frequency regulation can also be improved by
using storage modules [1].
Power Quality and Reliability: The need for high reliability and good
power quality has increased as more customers install microprocessorbased devices and sensitive end-use machines.
Public Policy: In a full reversal from the past, public policy today is
favoring distributed generation that offers improved eficiency, lower
emissions, enhanced power system security, and other beneits of national
interest. Policies supporting this include tax credits, renewable portfolio
standards, emissions restrictions, grants, and so on.
More Knowledgeable Energy Users: Energy users are becoming more
aware of alternative power approaches and are more willing to consider
on-site generation options than in the past. Many are interested in combined
heat and power (CHP) as well as reliability enhancements.
Storm Recovery and Restoration: Hurricane Sandy and other “superstorms”
of the last decade have highlighted, in painful and expensive ways, the need
to protect consumers from the devastation caused by such natural disasters.
Today, both utilities and industry recognize the need to collaborate to make
the microgrid a reality and ensure our community, place of business, or
entity is insulated from an extended power outage in the case of a natural
disaster. The technology needed to make microgrids feasible for everyday
use, such as that needed for power storage and stabilization, has reached a
point where it is commercially viable and available. The reliability beneits
of the microgrid are clear. When the main grid loses power, the microgrid
can switch to island mode. As long as there is an adequate source of power
at the local level (diesel generator, wind turbine, fuel cell, PV, etc.), the
power continues to low while the rest of the community is enveloped in
darkness [1].
Increased Interest in Security: While cyberattacks often dominate the discussion of grid security, utilities and governments need to ind ways to protect the grid from natural disasters as well as physical sabotage. Experts
agree that any prolonged disruption of power, regardless of the cause, represents a threat to security and economic stability [1].
216
Smart Grids
10.1.3
mIcrogrId market
There are currently four leading end-use applications deployed today:
Institutional/Campus Microgrids: Currently, the largest market for
microgrids is found in institutional campus settings. These installations
often make use of CHP systems to maximize eficiency and reduce dependence on local infrastructure. Reliability can be of extremely high importance, particularly if the campus includes vital loads such as a hospital and
is located in a rural area that is subject to more frequent disruptions in the
power supply [2].
Military Base Microgrids: The most active market now is in military facilities,
both stationary bases and forward-operating bases. The US Department of
Defense (DoD) has mandates to obtain 25% of its electricity from renewable
sources by 2025, and to make US bases net-zero energy users. The Defense
Critical Infrastructure Protocol also directs the DoD to identify critical operations and demonstrate the ability to operate them over a 30-day period without
service from outside electric infrastructure or resupply of fuel [3].
These are formidable objectives, and microgrids will be instrumental
in achieving them. A combination of energy security, reliability, and cost
is driving the military’s activities, which extend to renewable generation,
energy storage, and electric vehicles, among others. According to Pike
Research, there are more than two dozen microgrid implementations going
on now at military facilities in the United States, and there is support in
Congress for more. The military has a long history of pioneering technologies that later become widely accepted—the Internet being the most
compelling example—and the Department of Energy (DOE) is reportedly
looking to leverage the DoD’s work to apply microgrid technology to emergency response and relief efforts [2].
Remote “Off-Grid” Microgrids: Microgrids can provide an alternative to
conventional power grid development to remote facilities that do not have
access to grid power, especially developing countries and isolated islands.
These microgrids never connect to a larger macrogrid, and therefore operate
in an island mode on a 24/7 basis. Many of these microgrids are designed to
reduce diesel fuel consumption by integration of solar PV, distributed wind,
and other potential renewable sources such as hydropower.
Microgrids are easier to build than large power plants serving loads via
large T&D networks, and they are focused on serving the speciic needs of
the area in which they are located. They are also easily scalable. If these
attributes prove to be salient for even a modest number of countries, the
resulting potential market could be quite large [2].
Critical Loads: There is a segment of the microgrid market that is sometimes overlooked, or more often rolled into the larger institutional/campus category, and that is what is referred to as critical load. Historically,
this term has most often been associated with facilities such as hospitals
or telecommunications—things with an essential link to the functioning
217
Microgrids
of modern society. In recent years, however, a new facility has emerged as
“critical” not only from a business perspective but from a societal one as
well: data centers.
Given their explosive growth—they already account for a signiicant
portion of all electricity consumption—and their increasing importance to
every sector of the economy, data centers may well represent their own
segment of the microgrid market. They have tremendous requirements for
power, using up to 100 times as much as a comparably sized ofice building,
and most, if not all, of the load must be served at all times. Top-tier data
centers already go to great lengths to ensure reliability, with redundant grid
connections and extensive on-site backup power capabilities. Going forward, advancing microgrid technologies could offer improvements in both
reliability and eficiency/cost for data centers [2].
10.2
MICROGRID DESIGN
This section addresses several important aspects related to the design of microgrids,
namely:
•
•
•
•
Network topology
Control approaches
Protection approaches
Direct current (dc) microgrid
10.2.1
oPtImal mIcrogrId toPology
Microgrids can operate in parallel with or isolated from the utility grid during
emergency conditions or planned events. This type of distribution grid structure
offers potential for improvements in power supply eficiency and reliability of power
supply in comparison with traditional and passive distribution grids. But what is
the optimal topology for this kind of power distribution network? The well-known
approaches include a radial, a normally open-loop, or a meshed structure for the
distribution systems, which constitute the possible solutions for microgrid optimal
topology (Figure 10.1).
Radial
Open loop
Normally open
FIGURE 10.1 Traditional distribution network topology.
Meshed
218
Smart Grids
Factors determining the selection of optimal microgrid network topology include [4]:
• Size, type, and location of DERs and loads
• Power quality and reliability targets (the minimum power quality that a
microgrid has to provide to its customer affects the selection of an optimal
network structure)
• Economic constraints/available budget
• Investment cost (primary equipment, protection and control, communication, etc.)
• Operating and maintenance costs, including cost of power losses and energy
not supplied due to interruptions
• Technical constraints (e.g., protection system, voltage proile, and physical
equipment dimensions)
• Voltage level (usually medium-voltage networks are open-loop networks
and low-voltage networks are radial, with normally open-loop topologies in
some exceptional cases)
The optimal microgrid topology selection is a complex task with multiple conlicting objectives and uncertainties. Value-based reliability planning provides a rational
and consistent framework for addressing the fundamental economic trade-off question
of how much reliability is adequate from the customer perspective, gives rise to eficient topologies that can meet the consumer requirements with the minimum system
upgrade, makes the system more robust, and complies with consumer requirements [5].
It is well known that in a conventional distribution system the failure frequency
of the medium-voltage network is decisive for the frequency of low-voltage network
interruptions. For that reason, it is also obvious that the adoption of an islanded lowvoltage microgrid leads to a remarkable reduction of “local” interruptions caused by
the medium-voltage network. However, a low-voltage microgrid has a small impact
on the “global” reliability indices of a distribution system.
10.2.1.1 Traditional Distribution Network Topologies
Traditional low-voltage distribution networks have been radial, and the focus of network planning has been primarily on preserving the radiality of the distribution system. For reliability reasons, it is common that medium-voltage distribution networks
use normally open-loop structures (Figure 10.1). In both cases the system operation
is performed radially. The meshed structure is the alternative scheme, and it offers a
large degree of freedom for the optimization procedure, providing a large number of
different topologies according to technical and economic constraints.
Is it still convenient to keep the radial operation of microgrids, considering the large
penetration level of the DER, and verify whether the adoption of advanced topologies
could help overcome the obstacles that limit the penetration level of the DER?
Radial is usually the most cost-effective topology in a traditional, passive distribution grid due to a reduced number of switching devices, shorter cable length, and
simple nondirectional overcurrent (OC) protection systems. However, there are no
options to change the network coniguration if faults occur in the radial topology,
which usually leads to unnecessarily long customer interruptions.
Microgrids
219
The utilities’ increased concern about a reduction of customer interruption time
has led to the addition of connections that provide alternate power supply routes
in the case of outages or scheduled power supply interruptions. These emergency
connections end with an open switch to preserve the radial network structure
during normal operation. In this way, a distribution network has normally open
loops, which can be connected into the meshed structures and increase the reliability of the power supply and the availability of local generating and energystorage devices. Coniguration management is possible with open-loop topology
for service restoration, power loss reduction, and load balancing to relieve feeder
overloads.
Also, it is expected that DERs will play a more important role, and their penetration will reach a high level in the future. In some cases, local generation may exceed
local feeder load and lead to a reverse power low, feeding power into an upstream
grid. Thus, distribution utilities will probably have to dismiss the traditional radial
network operation, adopting more lexible meshed structures used today at higher
voltage levels.
Meshed networks have well-known advantages versus radial/normally open-loop
schemes [4]:
•
•
•
•
•
•
•
Reduction of power losses under the same load demand and DER penetration
Improved voltage regulation
Greater lexibility and ability to cope with the future load growth
Better asset exploitation
Alternative power low roots and increased reliability of power supply
Greater eficiency due to increased availability of local DERs
Improvement of power quality due to fault level increase
However, compared with a radial arrangement, a meshed network also presents
some drawbacks, usually leading to a higher investment cost:
• Increased short-circuit current could necessitate a substitution of the existing circuit breakers (CBs), due to the overcoming of their nominal faultinterrupting capacity.
• Need for more complex protection schemes due to alteration of a nondirectional OC protection logic, which can prejudice its selectivity.
• More complex planning and system engineering.
Therefore, major factors affecting the decision-making process regarding the
selection of optimal microgrid topology are:
•
•
•
•
Penetration and dispersion levels of the DER
Load characteristic (dispersion, reliability requirements, etc.)
Anticipated events and envisaged operation of the microgrid
Availability and suitability of equipment enabling meshed operation (network monitoring, power low and fault level management, protection system, etc.)
220
Smart Grids
10.2.1.2 Size and Location of Distributed Energy Resources
There are two main options for integrating DERs in a microgrid. The irst option can
be called a “single object power supply” [4], consisting of small-scale DERs, each of
which resides with one consumer at a time. The second option is characterized by a
residential area power supply in the form of one larger generator for each residential
area, or a complete microgrid.
In the case of the single object power supply, the DER units can be disconnected
automatically from the main power supply system in the event of any fault and create
a consumer microgrid. The load can be supplied without interruption resulting from
internal microgrid faults; however, it is necessary that the nominal capacity of each
local generator meets the peak of local load. In the second option, a microgrid has
the advantage of using the compensating effect of the consumer’s collective, but a
reduced redundancy in generation is a disadvantage.
The presence of intermittent generation resources requires the use of energy storage to provide a power balance during isolated operation. Energy storage can be
realized as a set of small-scale and distributed devices (located at every household/
building or next to the local DER) or as a community-scale single energy-storage
device for an entire microgrid.
10.2.2
mIcrogrId control
Microgrids are comprised of different components, such as:
• Distributed generators, such as microturbines, fuel cells, PV, diesel
generators
• Energy-storage devices, such as batteries, lywheels, supercapacitors
• Flexible loads, such as heating, ventilation, air-conditioning, and lighting
• Reconigurable feeders, tap-changing transformers, and reactive power
compensation
These components can be controlled in a continuous or discrete way in order
to keep a microgrid running in utility-connected or islanded operating modes, as
well as to guarantee a seamless transition between two modes. Typical control tasks
include:
• Spinning reserve management to cope with emergency conditions
• Generation, energy storage, and demand management for both steady-state
and dynamic microgrid control
• Network coniguration management
• Coordination with distributed generation and feeder overload protection relays
Most of these tasks are executed at different timescales, from the subsecond range
up to hours and days. The short-term control actions are focused on system stability by reaching an instantaneous power balancing and acceptable voltage level. The
mid- and long-term control actions focus on economic objectives and system security.
221
Microgrids
A major control challenge related to microgrids is coordinating a variety of components in the microgrid to facilitate their parallel operations without loss of voltage and frequency stability, and doing so in the most economic way. Techniques
and approaches vary, resulting in a continued wide divergence as to what type of
microgrid control system will work best.
There are several control approaches applicable to microgrids that have been
evaluated by a scientiic community and tested in the ield at several pilots. They
vary between fully decentralized and fully centralized approaches, with various
hierarchical/hybrid schemes in between (Figure 10.2).
10.2.2.1 Centralized Approach
This approach is based on the presence of a centralized controller, which collects
measurements and other necessary information at a single point, where control decisions are taken and communicated back to the actuators. This approach is much more
focused on optimal microgrid operation, including services outside the microgrid [6],
while beneits of reliability may come as a second objective.
Although the centralized approach provides better system observability and optimality of decisions, it has been recognized that it also has several downsides in a
microgrid environment:
• Failure of the central controller can be catastrophic, and redundancy
increases the cost.
• Need for a performing hardware for a central master controller (memory
and central processing unit) and high-speed and low-latency communication network increases the cost.
• System maintenance requires complete shutdowns.
Microgrid control complexity
Comm. requirements and costs
Comm. dependence and latency
Sensors and costs
Responding speed
Microgrid reliability
100%
decentralized
Device control complexity
Performance and effectiveness
Optimization of power flow and
Energy utilization
Efficiency
Standardization
Scalability/modularity
Compatibility
Where is the balance?
100%
centralized
FIGURE 10.2 Microgrid system control: decentralized versus centralized. (From Xu, Y., ORNL
Power Systems and Power Electronics for Utility Applications, ORNL, Workshop on Power
Electronics and Distributed Power systems, UTRC, December 2, 2011. Available at: http://web.
ornl.gov/sci/ees/etsd/pes/pubs/UTRCWorkshop_ORNLPowerElectronics_Microgrid.pdf. [7])
222
Smart Grids
From an economic point of view, the layered/hierarchical system is the more
practical solution. In this case, a local control focused on system stability is delegated to local controllers, while the central controller executes slow control actions
of economic dispatch, calculating schedules for local controllers as well as an
amount of reserve for emergency conditions. A layered centralized microgrid control system is described in [8]. For example, control architecture can consist of the
following levels:
• Local-level controllers
• Microsource controllers use local information to control the voltage
and the frequency of the microgrid in transient conditions.
• Load controllers shed loads in emergency conditions.
• Microgrid central controllers, which are used for
• Forecasting of electrical and heat load and production capabilities
• Economic dispatch of different production and consumption units
• Online security assessment
• Demand-side management
• Black start and grid resynchronization
A practical realization of the centralized microgrid approach is described in [9].
10.2.2.2 Decentralized Approach
The advantages of having the control system distributed and split up into individual
peers are as follows [10]:
• Each electrical device can be itted with a separate, less complex controller.
• The failure of one controller does not have catastrophic impact, since
replacement capacity can be brought online immediately.
• The system is more scalable and extendable, not limited to onboard input/
output of the central master controller.
• It is easier to develop upgrades, thanks to independence and modularity.
• Communication redundancy can be easily achieved.
• Each node is autonomous and yet closely integrated with its peers.
In a fully decentralized approach, the local controllers of different controllable
microgrid components are responsible for satisfying the demand (and possibly
exporting power to the utility grid) autonomously (communication in such a distributed environment works through exchanging messages between the individual controllers) in the most economical way. In addition to economic goals, local controllers
may have local control tasks such as heat generation, voltage control, backup for
critical loads, and so on. These tasks suggest the importance of distributed control
and autonomous operation [8].
The use of multiagent systems in controlling a microgrid has been investigated by
leading academics in [8]. It is especially eficient in the case when different microgrid
components have different owners; in this case, decisions should be taken locally, so
centralized control is dificult. Practical examples of decentralized microgrid control
Microgrids
223
include the Consortium for Electric Reliability Technology Solutions (CERTS) [11]
and Powercorp [10]. Such a system will work for the majority of small microgrids,
whose highest priority is reliability and sustainability during emergency conditions.
10.2.3
mIcrogrId ProtectIon
One of the major technical challenges associated with a wide deployment of
microgrids is the design of their protection system. Protection must respond to both
utility grid and microgrid faults. If the fault is on the utility grid, the desired response
is to isolate the microgrid from the main utility as rapidly as necessary to protect
the microgrid loads. The speed of isolation is dependent on the speciic customer’s
loads on the microgrid, and in some cases may require the use of electronic static
switches. If the fault is within the microgrid, the protection system isolates the smallest possible section of the distribution feeder to eliminate the fault [12]. A further
segmentation of the microgrid during the isolated operation, that is, a creation of
multiple islands or submicrogrids, must be supported by the availability of the DER
and load lexibility.
Time OC protection is the most common method employed in medium- and lowvoltage distribution grids, as it is quite simple to conigure and is a cost-eficient solution. The presence of the DER changes the magnitude and direction of fault currents
and may lead to protection system maloperation [13–16]:
• Reduced fault sensitivity and speed in tapped DER connections.
• Unnecessary tripping of utility breaker for faults in adjacent lines due to
DER contribution.
• Reduced fault contribution of inverter-based DER in isolated mode.
• Increased fault levels may exceed the capacity of the existing switchgear.
• Auto-reclosing and fuse-saving utility policies may fail.
In such circumstances, coordination between relays may fail and generic OC protection with a single setting group in each relay may become inadequate; that is, it
will not guarantee a selective operation for all possible faults. Therefore, it is essential to ensure that protection settings chosen for OC protection relays are dynamically modiied, adapting to changes in grid topology and location, type, and number
of connected DERs.
10.2.3.1 Adaptive Protection
Adaptive protection is deined as “an online activity that modiies the preferred protective response to a change in system conditions or requirements in a timely manner by
means of externally generated signals or control action” [16]. Technical requirements
for the practical implementation of an adaptive microgrid protection system are:
• Use of digital directional OC relays, since fuses or electromechanical and
standard solid-state relays do not provide the lexibility for changing the
settings of tripping characteristics, and they have no current direction sensitivity feature.
224
Smart Grids
• Use of new/existing communication infrastructure and standard communication protocols such that individual relays can communicate and exchange
information with a central computer or between different individual relays
rapidly and reliably to guarantee a required application performance [17].
The parameter adaptation time delay is not critical, because the communication infrastructure is used to collect information about the microgrid coniguration
and to change relay settings accordingly only prior to the fault. The changes of the
microgrid coniguration will be identiied within a few seconds, depending on the
dimension of the network, and the protection system reconiguration has to be completed in times of the same order, provided that basic backup protection functions are
present during the transitory phase.
Next, two approaches to modifying protection settings are addressed. Both
approaches are similar in using a central protection coordination unit that collects
information about microgrid topology and decides whether adaptation of settings is
needed. The difference between the approaches is in how the setting modiication
physically happens after the decision to apply a new setting is taken.
The irst approach is based on online switching between precalculated and “trusted”
setting groups that are kept in the memory of the protection device (today’s intelligent
electronic devices [IEDs] can hold four or more different setting groups). In this case
the central coordination unit sends a command of which setting group number must
be set as active, and the IED implements it locally, reporting back the new value of the
active setting group. There are two main steps in realizing this approach:
• Off-line calculation of settings: A set of meaningful microgrid conigurations as well as feeding-in states of DERs (on/off) is created for off-line
fault analysis. Next, fault currents passing through all monitored CBs are
calculated by simulating different short-circuit faults (three-phase, phaseto-ground, etc.) at different locations of the protected microgrid at a time.
During repetitive short-circuit calculations, a topology or a status of a single
DER or load is modiied between iterations. Based on these results, optimal settings (to achieve targeted sensitivity, selectivity, and speed) for each
directional OC relay and for each particular system state are calculated.
The settings are then uploaded into the memory of protection devices.
• Online operation: During online operation, the central protection coordination unit monitors the microgrid state by receiving information from remote
IEDs or sensors. This process runs periodically or is triggered by an event
(tripping of CB, protection alarm, etc.). The adaptation algorithm then compares the actual microgrid coniguration with previously analyzed cases
and retrieves the setting group number most suitable for the current coniguration. If the actual setting group number (also monitored) is different,
the output of the adaptation algorithm will be a setting switching command,
as shown in Figure 10.3. Alternatively, if the number of settings exceeds the
memory size of the local protection device, a complete new setting group
can be uploaded (replacing one of the settings not used in the memory) into
the memory of the local device and then activated [17].
225
Microgrids
t
Legend
Ik2max = 2.4 kA
seen by CB3.1
100 s
CB3.2, CB6.2
Base trip
curve
CB2.2, CB3.1
CB5.2, CB6.1
10 s
Dynamically
modified trip
curve
Delayed trip
1s
Selective trip
0.1 s
1 kA
10 kA
FIGURE 10.3 Adaptation of the OC trip curves in response to microgrid topological
change. (From Oudalov, A. and Fidigatti, A., International Journal of Distributed Energy,
5, 201–225, 2009.)
The second approach is based on real-time calculation of settings as soon as a
change in microgrid topology is detected by the central protection coordination unit.
The calculated settings are then uploaded into the memory of the local device and
activated. In some speciic implementations, the central unit may also perform primary protection functionality; that is, in the case of a fault, the central algorithm
implemented in the central unit generates selective tripping signals, which are sent
to the respective CBs. This application is limited to a single substation case, since
communication may become a bottleneck for remote CBs, which will require the
presence of high bandwidths and reliable communication channels [16]. Integration
of state estimation routines into a coniguration of the protection system, which monitors the DER operating state and sets up a corresponding data exchange with the
protection system, is also possible [18].
In this case, the prediction data of the availability of the DER are used in order
to review the selectivity of the tripping characteristics for each new operating condition, based on the short-circuit power of the DER and the actual short-circuit power
available from the main grid, and to adapt them accordingly if selectivity is no longer suitably provided. If the adaptation is successful and the boundary conditions
are not violated, the tripping characteristics of respective relays will be matched. If
there is no possible solution without violating the boundary conditions, a signal will
be generated that forbids the acceptance of the operation predicted by the energy
management system (EMS).
226
Smart Grids
10.2.3.2 Fault-Current Source
As mentioned above, one of the major challenges in microgrids is fault detection
in the presence of a DER with power electronics (PE) interfaces, especially in the
isolated mode in the absence of the utility power fault source.
One possible solution is to use an independent system that supplies a fault current
comparable to the level provided by the utility grid. In this case, protection devices
do not need to adjust protection settings. This concept is also suitable in the case of
the active use of fuses as protection devices.
The proposed system, designed as fault-current source (FCS), is based on
PE, but using a simpler and thus less costly technology than is normally used
in inverters in this power range. Its power circuit remains idle during normal
operation of the network, but when a fault is detected (by sensing a local network
voltage level) the FCS is activated. It attempts to restore the original network
voltage by injecting a fault current into the network for a speciied duration,
enough to cause a fuse or CB to clear the fault. The source of the fault current is
an ultracapacitor [19].
10.2.3.3 Fast Static Switch
Depending on the switch technology, momentary interruptions may occur during
transfer from grid-connected to islanded mode. If high-speed isolation is required
by the load, the static switch is a suitable technology. Static switch is a PE-based
solution used for fast and automatic islanding and for grid resynchronization. It is
capable of isolating sensitive loads and microsources from the distribution system in
less than one-quarter of a cycle during faults and other disturbances.
There are two major technologies used in static switches:
• Silicon-controlled rectiiers (SCRs)
• Insulated gate bipolar transistors (IGBTs)
Some lessons learned from testing the static switch prototype are shared in [19].
The differences among the varieties of switching technologies are based on the speed
of response. A wide range of interconnection speeds allows lexibility to match application requirements.
The SCR-based switch can respond in one-half to one cycle. Some SCR switches
in transfer switch coniguration can transfer within one-quarter of a cycle. This is a
relatively expensive solution due to the use of semiconductor materials and the need
for a CB to handle fault conditions.
The IGBT-based switch can respond in the 100 μs range. Also, IGBTs can clamp
the instantaneous currents and turn off in very short time frames.
Key technical questions which must be addressed when using static switch in the
microgrid include:
• Coordination of microsource control with a static switch (especially in the
case of active synchronization)
• Coordination with other protection devices (adjustment of tripping curves
and load-shedding sequences)
Microgrids
227
10.2.3.4 Communication Architectures and
Protocols for Adaptive Protection
Successful implementation of the adaptive protection system requires that protection
IEDs can communicate among themselves or with a central controller. By means
of the communications, IEDs can provide and receive information about the actual
network topology of the microgrid, the status of local generators, energy-storage
devices, and loads, and commands to execute an action (e.g., change of the active
setting group).
Within a range of communication protocols that enable a protection system to
become adaptive, International Electrotechnical Commission (IEC) standard 61850
is of great interest [20]. The most important IEC 61850 advantages are:
•
•
•
•
Guaranteed interoperability between devices from different vendors
Standardized data models
Simpliied engineering process and IED coniguration
Scalable solutions
Its main disadvantage is that it requires an Ethernet network, which is a common
technology for automation at high- and medium-voltage levels, but today still means
extra cost and technical complexity in low-voltage installations. Moreover, it poses
additional complexity for the maintenance personnel, for a system that is used to
working more comfortably with hardwired systems.
For adaptive protection application, this standard has communication services
for usual operations in electrical installations, such as change of settings and setting groups, report of information, commands, and so on. One remarkable service is
known as the generic object-oriented substation event (GOOSE), which makes direct
information exchange between IEDs possible, accepting any type of data (single
signals, double signals, measurements, etc.), guaranteeing that they are transmitted
in less than 3 ms (faster than a cable between digital outputs and inputs), and are
always available, because they are retransmitted indeinitely. Obviously, the application of GOOSE messages for communication between IEDs within the microgrid
provides a valuable tool, allowing their conigurations to be adapted to any operation
scenario, such as the transition from the grid-connected mode to the islanded mode.
Each IED can receive data from all others very rapidly, so the reconiguration process is very fast [16].
10.2.4
selectIng the rIght communIcatIons technology
Perhaps the most important aspect of microgrid communication design involves
deploying suitable and future-proof wired/wireless or hybrid technologies. In general, while wired technologies, particularly optical iber, provide more capacity,
reliability, and security, they can be costly to deploy, and inlexible once they are
deployed. As distribution networks grow more sophisticated, and as more intelligent points are added with a need for communications, this unfavorable cost per
point correlation is only exacerbated. So, while iber might be considered in situations where it is already found close to assets we wish to connect to, or perhaps
228
Smart Grids
in circumstances where a new building or community is going up, it is becoming
apparent that wireless communication technologies may in many cases be a better it
for the needs of microgrids and other distribution systems.
While, in the past, available wireless systems may not have met the requirements
of the most performance-sensitive applications, this is no longer the case. In particular, for ield or distribution-area networks, it is now possible to leverage wireless
mesh technologies to build a network capable of supporting multimegabit throughput, low latency (<20 ms round trip), enterprise-grade security, and resilient routing
architectures supporting up to 99.999% of reliability. These “mesh” wireless systems
typically function in the unlicensed spectrum found in the 2.4 and 5.0 GHz ranges
and can provide coverage over a utility microgrid, along with connectivity into the
various IEDs a utility might wish to connect for the purposes of information and
control (Table 10.1). Complemented by terrestrial communications such as iber found
at local substations or other interconnection points within the microgrid, and possibly by other wireless technologies where low density or lack of iber might dictate,
it is possible to design a network that meets the communication requirements of the
microgrid, both now and in the future.
10.2.5
dIrect current mIcrogrIds
Much has been written recently about dc as applied to power grids. Historically, the
power system started as dc during Thomas Edison’s efforts to electrify New York
City at the end of the nineteenth century. This, however, was short lived, since dc
could not be transformed to higher voltage levels, which prevented the long-range
transmission of dc power. Under Nicolai Tesla and George Westinghouse, alternating current (ac) prevailed, and three-phase ac generation, transmission, and distribution became the choice for power systems. In the 1930s, Swedish company ASEA
TABLE 10.1
Network Requirements Will Increase with Time
Application
Reclosers
Capacitor bank
Remote terminal unit
Motor-operated disconnect
Line regulator
Advanced metering
Interval data recorder
Demand management
Mobile workforce management voice
Mobile workforce management data
Aggregate
Latency (ms)
10 s
100 s
1000
1000
100 s
100 s
100 s
100 s
100 s
100 s
10 s
Bandwidth (kbps)
<56
<56
56
<56
<56
56
<56
<56
<56
1000s
1000s
Source: IBM Presentation for the Utilities Telecom Council on the Utility of the
Future.
Microgrids
229
developed a high-voltage dc transmission, demonstrating that under some conditions
dc could overcome some of the hurdles that ac faced, such as inductive and capacitive effects, which reduce the throughput of the transmission line. In the 1950s, several commercial installations commenced, and the trend of using high-voltage direct
current (HVDC) for applications such as long cable transmission (underground and
marine) and tying otherwise asynchronous systems continues to this day. HVDC
systems are primarily point to point; that is, an HVDC line has an ac/dc converter
station at each end to interconnect with the ac system. The primary reason why
HVDC transmission technology did not develop widely into an HVDC system with
many interconnected dc lines is the lack of dc CBs, devices that could protect the
system against faults and interrupt the resulting short-circuit currents.
Today, dc is experiencing a major comeback. There are several reasons for this.
Let us name just a few:
• More and more electric loads are inherently dc powered. Consumer
electronics (from laptops and smart phones to plasma TVs), information technology (IT) equipment (servers, network switches, storage, etc.),
variable-frequency drives (VFD), and so on, all have dc as the internal supply voltage. It is estimated that the share of the inherently dc electronic
loads will continue to grow.
• The popularity and applications of dc machines (motors and generators)
have grown. With the help of advanced rare-earth magnets, new designs,
and permanent magnet machine technologies, dc machines achieve performance comparable to, if not better than, their ac equivalents. Coupled with
their VFDs, they can provide a new range of functionalities and operational
parameters such as torque characteristics at different speeds.
• Renewable generation resources such as wind energy turbines and solar PV
arrays also use dc before the power is converted to the main frequency ac
and synchronized with the grid.
• Batteries have become the primary means of electric energy storage, and
these technologies are dc based. Other energy-storage technologies, such as
the lywheel and the fuel cell, also use dc within them as an intermediate
stage.
• Advances in PE and power semiconductor technology over the last several
years have made the transformation of current from ac to dc and vice versa
so eficient and inexpensive that it has begun to compete in cost and eficiency with conventional ac transformers.
• New PE equipment opens new functionalities that were never possible
before: controlling the power low, current, and voltage levels, power factors
as seen from the ac side, and sensing and interrupting short-circuit currents
with unprecedented speed, all with no moving parts.
Many believe that, for microgrids, dc will play an even more important role.
Because of its small scale, implementation of dc power in a microgrid could be much
simpler, and therefore one could beneit from the dc technology with much less effort
and cost.
230
Smart Grids
10.2.5.1 Differences Between ac and dc for Microgrids
• The irst difference involves the synchronization and integration of several
energy sources. With dc, one can connect multiple rotating machines without a need for synchronization. Either dc machines or ac machines with rectiiers can be connected to a dc microgrid without expensive and complex
paralleling schemes. Recall that microgrids do not have a so-called ininite
bus, so each connection of a machine to an ac microgrid can cause a voltage
and frequency swing that is dificult to control.
• Operation of dc microgrids, and all dc grids for that matter, involves delivering (generating, distributing, and consuming) real power (P) only; no
reactive power (Q) is present. Since balancing reactive power can often be a
challenge in ac systems, dc is inherently easier to control and operate. There
is no need for power factor correction equipment and control, such as power
factor shunt capacitors or voltage-control shunt reactors.
• System protection for ac is based on mechanical CBs, which cover the
entire range of voltages and currents of today’s systems. CBs for ac are
common and are based on mature technology; coupled with their economy
of scale (tens of millions of devices produced each year for different applications and at various voltage and current levels), they provide an excellent
and low-cost solution for circuit protection. CBs for dc are complex and
limited primarily to low voltage. They are based on the principle of current limiting by developing arc voltage greater than the system voltage or
by injection of transient high-frequency current to create artiicial current
zero crossings. This currently gives ac systems the advantage over dc.
• Operating procedures for disconnecting (islanding) and connecting to the
main grid, as well as operating in parallel with the main grid for bidirectional power low, are sometimes simpler for dc microgrids.
• Single-phase or two-phase dc systems are possible, and, with either two
conductors (+ and −) or three conductors (+, −, and ground), they can offer
signiicant cost reduction compared with ac. In most cases the three-phase
three-wire or four-wire ac conductor systems (overhead lines, underground
cables, or others) could be converted (retroitted) to dc, although the opposite would not be possible. A dc conductor system can deliver sqrt(2) higher
power than the equivalent ac conductor system, primarily because the dc
insulation system can be designed for the operating voltage whereas ac
insulation has to be designed for sqrt(2) times operating ac root-meansquare voltage. Additionally, the conductor resistance for dc is lower than
for ac, since the current density within the conductor is uniformly distributed through its cross section; for ac, the skin effect pushes the current to
the surface. The magnetic skin depth for copper at 60 Hz is only ~8.5 mm.
• Plugs, receptacles, switches, power supplies, sensors, motors, and other dc
equipment are not yet as popular and commercially available as ac equipment; ac equipment has a large historical presence and its installed base is
huge, with multiple vendors and a large commercial selection of brands.
The evolution of the dc industry will take some time.
231
Microgrids
10.2.5.2 Examples of dc Microgrids
In principle, dc microgrid architectures could be similar to their ac counterparts.
They are making inroads into building systems and data centers, among other things
[21,22]. Concepts such as radial, single bus, dual bus (with tie), mesh, or ring main
are equally applicable to dc, depending on the size of the microgrids, that is, the
number of sources and loads to be connected as well as the physical dimensions
(grid length and area to be covered). In most cases so far the microgrids have been
of either single-bus or ring-main (ring-bus) design.
10.2.5.2.1 Single-Bus dc Design
This is the most popular concept, which applies to the smallest installations, such as
ofices, ofice buildings, or small commercial enterprises (Figure 10.4). It can also
be extended to other, noncritical applications such as on board a train or an electric
bus. Some degree of redundancy can be built into such a simple system, but the main
single point of failure, the dc bus itself, will remain.
10.2.5.2.2 Dual-Bus with a dc Tie Breaker or dc Switch
In this particular topology, a dual-bus system on board a ship [23] is used for a degree of
redundancy. The concept is that if one bus is not available, for example, due to faults or
maintenance, the other bus can still power enough loads to keep mission-critical operation running. In traditional marine electrical propulsion systems, ac VFDs typically
account for more than 80% of the installed power. In the new approach, the onboard
dc grid concept is based on taking a multidrive system and distributing it so that the
distributed rectiiers can be eliminated. In this way, generation sources and major loads,
in this example diesel generators and electric propulsion motors, can be connected to
two common buses. Such a system can be optimized from the energy production and
consumption point of view, resulting in substantial energy savings, up to 20%.
The new microgrid can be seen as one large multidrive system merging the various dc links that would otherwise reside in individual VFDs around the vessel.
Power is distributed through a dual-bus single 1000 V dc circuit, thereby eliminating
the need for main ac switchboards, distributed rectiiers, and converter transformers.
ac grid (macrogrid)
ac
source(s)
ac
dc
dc
source(s)
ac/dc
ac/dc
dc bus
dc loads
dc/ac
dc/ac
ac loads
FIGURE 10.4 Simple diagram of a single-bus dc microgrid.
Bidirectional
ac/dc conversion
232
Smart Grids
All generated electric power is fed either directly or via a rectiier, depending on the
nature of the source, into a common dc bus. Each main load is then fed by a separate
inverter unit (Figure 10.5). If an ac distribution network is still needed, for example,
for auxiliary loads, it is fed using dc/ac converters. Additional converters for energy
storage in the form of batteries, PV arrays, or supercapacitors for leveling out power
variations can be added to the dc grid. In some applications, one can envision connecting dc energy sources or storage devices directly to the dc bus(es); however,
adequate protection would have to be included against short-circuit currents.
The onboard dc grid is lexible and can be conigured in several different ways
depending on the physical constraints (space), length of feeders for sources and
loads, dc currents, losses, and reliability of different components, among other factors. For example, with a centralized approach all converter modules are located in
one or multiple lineups within the same space in which the main ac switchboard
would normally be installed. In another implementation, the individual converter
modules could be located close to the devices they serve, and the dc bus(es) extended
throughout the ship.
10.2.5.2.3 Ring Main
For even better reliability and to provide some redundancy of power delivery, a dc
ring bus (or ring main or loop) provides an alternative path for supplying power while
any section of the bus is out, either faulted or for routine maintenance (Figure 10.6).
Such a system can be implemented in several different variants. For example, for
more critical loads a more dedicated source can be placed on the same section of the
ring; one can also use fewer separation points (fewer breaks in the ring) for lowercost implementation. In yet another variant, the ac grid can be connected in more
than one point around the loop.
The operational principle is rather simple and almost identical to the ac ring mains
popular in many distribution systems in large European cities. Whenever a bus fault
occurs, this section of the bus is isolated by the respective CBs (or other protective
devices) and the remaining portions of the ring main operate as usual. There are no
G
G
∼
∼
=
G
G
∼
=
∼
=
=
dc bus
dc bus
=
=
∼
=
∼
=
∼
=
∼
=
∼
ac bus
ac bus
M
FIGURE 10.5
∼
M
Simpliied diagram of a dual-bus dc system with a tie-switch.
233
Microgrids
ac grid (macrogrid)
ac
source(s)
ac
dc
dc
source(s)
ac/dc
Bidirectional
ac/dc conversion
ac/dc
dc bus
dc loads
dc/ac
dc/ac
dc ring
(loop)
ac loads
FIGURE 10.6
Simpliied diagram of a ring-main dc microgrid.
downstream or upstream devices. Each load and source can be operated from two
directions of the ring main.
10.2.5.3 Control and Protection of dc Microgrids
As with ac microgrids, the most critical aspect of designing, operating, and maintaining the dc microgrid is control and protection [24,25]. Even when the individual
devices (energy sources, loads, conductors) are available, the design of control and
protection logic has to be carefully thought out.
Controlling dc microgrids involves primarily three aspects:
1. Switching microgrids from islanded to utility network operations, due to
the availability of power, economics, utility outages, or other factors. The
switching can be either hard, that is, instantaneous, as in the case of sudden, unplanned utility outage, or soft, as in the case of planned outages,
economic considerations, or approaching a natural disaster such as a storm,
hurricane, or looding.
2. Managing (balancing) the generation sources based on their availability
(e.g., solar PV and wind), generating cost, and—in some instances—their
location; this managing could involve different strategies depending on
whether the microgrid is islanded or on grid.
3. Managing (controlling) load demand. Loads should be irst categorized as
to their priority (mission critical, critical, essential, nonessential); their level
of criticality can be dynamic, for example, a function of time, weather, or
major external events.
What makes the dc microgrid different is the fact that switching dc equipment
(generation or load) on or off is potentially simpler, since no point-on-wave synchronization is necessary—no out-of-phase conditions, no reactive power low or
phase angle differentials—and no frequency luctuations occur. At the same time,
the dc challenge today involves the lack of established, documented, and wellproven dc control algorithms and processes, so that they have to be developed
on a case-by-case basis. This will likely change as microgrids and the use of dc
continue to grow.
234
Smart Grids
Protecting dc microgrids against system disturbances such as faults, arcs, and
open circuits can be quite different from the case of ac. All the aspects of fault detection, isolation, and restoration require different treatment. First of all, the waveforms
of dc short circuit are fundamentally different. Depending on which devices are
connected to the microgrid, it may have one of several types, as shown in Table 10.2.
In most cases the short-circuit current of a dc microgrid will be a combination of the
three generic waveforms shown in Figure 10.7, with different inductance to resistance (L/R) ratios, oscillating frequencies, and damping factors.
Detecting a dc fault is, on the one hand, a little simpler, and can be a faster process, since any excursion of the current above a certain threshold for even a short
period of time can indicate a fault. Therefore, the detection algorithm does not
need to wait for a quarter of a cycle (5 or 4.166 ms for 50 and 60 Hz, respectively)
as in ac systems. On the other hand, with a potential overshoot of short-circuit
current when various dc link capacitors discharge (waveforms (b) and (c)), the dc
system and its components have to withstand these instantaneous stresses comparably to ac.
TABLE 10.2
Types of Short Circuits in dc Microgrids
Source
1. Battery bank
Load
Nonbackfeeding loads
2. Rotating generator (dc)
Nonbackfeeding loads
3. Rectiied ac from the
utility grid
Nonbackfeeding loads
4. Combination of sources
Nonbackfeeding or
backfeeding loads, or both
5. Ungrounded (loating)
or high impedance
grounded
Ungrounded or high
impedance grounded
Short-Circuit Current Waveform
Expected
Exponential waveform from load current
level to short-circuit current (Isc)
determined by resistance of the faulty
path, including internal battery
resistance, conductors, and electrical
arc, if any
Depending on the type of machine
Exponential waveform as in case 1, with
superimposed contribution from the
discharging capacitors
Waveform will be complex, with several
superimposed contributions from the
discharging capacitors of dc links of
various PE devices. Depending on the
resistances involved, these contributions
can be either underdamped or
overdamped and with different L/R
ratios or frequency and damping
factors, or both.
Single line to ground fault is limited to
either small transient current (due to
parasitic capacitance/inductance) or
small dc current via grounding
resistances
235
Current
Microgrids
L/R
Steady-state
fault current
Prefault current
Time
(a)
Time
(b)
Time
(c)
FIGURE 10.7 Three generic waveforms of short-circuit current in dc microgrids: (a) simple
L-R connection with constant dc voltage; L/R ratio determines the rate of rise of the fault current; (b) an overdamped L-R-C oscillatory circuit; overshoot occurs due to the presence of a
discharging capacitor(s); (c) underdamped L-R-C oscillatory circuit; frequency of oscillations
is determined by a combination of L, C, and R, and decay of the oscillations is determined
by R/sqrt(L/C).
Isolating the fault means fault-current interruption. As mentioned before, interrupting the dc current by conventional means is not possible due to the lack of (natural) current zeros. The two possible methods of dc fault isolation are by means of dc
breakers, PE converters, or both. Depending on the design and ratings, dc breakers
can take up to several milliseconds to force the short-circuit current to zero. IGBT or
similar PE-based devices can bring the current low to zero much faster, in a fraction
of a millisecond, and without any internal arcing. The other advantage of handling
the faults with PE is that the current does not have to be interrupted completely;
for example, a small dc current could be maintained for further fault location or
for the purpose of protection coordination with dc fuses, downstream and upstream
dc breakers, and ac devices outside the microgrid. When the system is ready to be
restored, PE devices can also “reclose” or re-energize the system almost instantaneously without recharging, and, if the reclose is at fault again, the current does not
have to rise to its previous, original fault level. In fact, the system can be re-energized
in a soft manner. This can result in virtually arcless, arcproof, and fault-tolerant systems, increasing safety for personnel and equipment alike. The availability of such
fault-tolerant systems will also increase.
In summary, dc microgrids offer unique advantages over their ac counterparts.
With the growth of inherently dc loads, the advent of battery technologies, PE, dc
machines, renewable energy, and a growing interest in plug-in electric vehicles, dc
not only simpliies the design, integration, and operation of the microgrids, but also
provides new functionalities and increased safety and availability.
10.3
MICROGRID OPERATIONS
A microgrid can operate in either the grid-connected or islanded mode, and may
experience mode transition. In the grid-connected operation mode, the microgrid
is connected to the main grid through the point of common coupling and operates
in parallel with the main grid to deliver power to the load. From the distribution
236
Smart Grids
system perspective, a grid-connected microgrid can be treated as an individual controllable entity, acting like a generator to supply power or a load to consume power.
On the other hand, the microgrid operating in the islanded mode is independent of
or isolated from the main grid, due either to disturbance in the main grid or to its
geographical isolation, such as a remote island. In the islanded mode, the microgrid
operation, such as the voltage and frequency control and regulation, is implemented
in a stand-alone system. The transition between the two modes can occur during
the operation. For example, when a fault happens to the main grid, the microgrid is
disconnected from the main grid to continue operation in the islanded mode, and,
after the fault has been cleared from the main grid, the microgrid is resynchronized
and connected back to the main grid via the common coupling point. Some types of
microgrids, such as institutional/campus, industrial/commercial, community/utility,
and military base microgrids, operate in the grid-connected mode most of the time,
and only switch to the islanded mode when disturbances happen to the main grid.
Some other types of microgrids, that is, remote off-grid and weakly grid-connected
microgrids (such as military forward installations and remote mining industrial
sites), operate in the islanded mode all the time, or very often.
10.3.1
oPeratIon oBJectIves
The deployment of microgrids generally aims to increase the penetration level of
DERs (especially the green energy resources), improve energy eficiency, enhance
service reliability and power quality, and reduce greenhouse gas emission. In order
to achieve these objectives, it is critical that the microgrid can maintain stable,
economic, and secure operation in different modes. A stable operation relies on
control of the voltage and frequency within certain limits when disturbances
occur. Such disturbances may be load demand variation, renewable generation
output luctuation, islanding from the main grid, and so on. Economic operation
refers to the supply of the local load with minimum generation cost, which can be
achieved by coordinated control among different devices, such as renewable generation, conventional distributed generators, energy storage, and the power import
from the utility grid. Secure operation aims to guarantee the power supply for
critical loads all the time and is very important for hospital, military base, and
data center microgrids, in which the disruption of power supply for critical loads
can cause a huge impact.
10.3.2
technIcal challenges
Some of the technical challenges that must be overcome to achieve stable, economic,
and secure microgrid operational status must deal with several aspects.
10.3.2.1 Intermittent Renewable Generation
One of the major incentives to deploy microgrids is to facilitate the integration of
renewable generation in the distribution system. The power output of renewable
generation (such as solar and wind) is signiicantly inluenced by the season and
weather, and these are characterized as intermittent power resources. That is, the
Microgrids
237
power output of these resources can vary abruptly and frequently and imposes challenges on maintaining microgrid stability, especially in the islanded mode.
10.3.2.2 Low Grid Inertia
A microgrid may include both conventional and modern DERs. Conventional distributed generators, such as diesel generators, usually utilize synchronous generators
that directly connect to the grid. Modern distribution grids use the most renewable resources and energy-storage devices that are connected to the grid indirectly
through the PE interface. These DERs either have a low inertia or are inertialess, and
bring dynamic problems to microgrid operation.
10.3.2.3 Coordination among Distributed Energy Resources
A microgrid may have various types of DERs, such as diesel generator, microturbine, fuel cell, CHP, energy-storage device, and so on. These DERs usually have
different operation characteristics in their generation capacity, startup/shutdown
time, ramping rate, operation cost/eficiency, energy storage charging/discharging
rate, and intertemporary control limitations. The microgrid operation should consider the characteristics of different components and provide appropriate control
strategies accordingly to improve economic and secure operation.
10.3.3
grId-connected oPeratIon
In the grid-connected mode, since the microgrid is connected to the main grid, stable
operation is not a major concern. Therefore, the operation objective is mainly to
reduce operation cost and improve energy eficiency. The microgrid management
can be implemented in either a centralized or a decentralized manner. This section
illustrates the centralized microgrid management applications for grid-connected
mode operation.
Centralized control enables optimal operation via coordinated control among different types of controllable DERs. The centralized microgrid EMS performs the
coordinated management of the microgrid in a central controller, which monitors
overall system operating conditions, makes optimal control decisions in terms of
minimizing operation cost, reducing fossil fuel consumption, providing services for
the utility grid, and so on, and then communicates the setting points to DERs and
control commands to controllable loads.
Most of the current centralized EMS solution technologies fall into two categories: (1) day-ahead unit commitment combined with online economic dispatch (ED)
and (2) online ED across multiple time intervals. These solutions have attempted to
provide optimized operation strategy over a predeined period of time while taking
into account the renewable generation and load forecast.
The method of day-ahead unit commitment with online ED approach [26–31]
generates an optimal operation plan for next day based on the day-ahead renewable
generation and load forecast, and the ED is executed in real time using the day-ahead
unit commitment results. The method of online ED over multiple intervals [32–43]
can take the latest forecast on the renewable generation and load into account for
the operation decision. However, it provides the control decision not only for the
238
Smart Grids
current interval but also for the future intervals at each execution interval (e.g., every
5–15 min) in real time. Due to the heavy computational burden, simpliied optimization is usually deployed, considering only the power balance instead of more detailed
operating constraints.
10.3.4
Island-mode oPeratIon
In island-mode operation, the microgrid is operating independently of or isolated
from the main grid. This mode of operation may be the result of a disturbance that
has occurred in the main grid, or the microgrid may be geographically isolated from
the main grid. Examples of this would include a remote island without a mainland
link or a remote location without access to the grid, such as a forward military installation or remote mining industrial site.
One of the drivers for isolated microgrids is the ability of these systems to integrate renewable energy sources. These renewable energy sources include generation
sources such as wind turbines, PV solar systems, solar-thermal power, biomass power
plants, fuel cells, gas microturbines, hydropower turbines, CHP microturbines, and
hybrid power systems. The renewable energy sources are normally paired with fossil
generation as part of the power generation systems for isolated microgrids.
The exploitation of renewable energy sources, even when there is a good potential
resource, can be problematic due to their variable and intermittent nature, especially
when these resources are connected to isolated grids.
When using renewable generation in isolated power systems, some of the major
dificulties involve power quality and reliability, due to the stochastic nature of
the sources. However, the inclusion of renewable energy resources in an islanded
microgrid is attractive, since they can reduce the use of fossil fuel. Unit power costs
for fossil generation are usually very high for islanded microgrids because power is
typically supplied by diesel-generating sets with inherently high costs due to transportation of the fuel to remote islands and locations.
There are a number of limitations to the attainable level of renewable energy
penetration in isolated grids. These limitations are described below and include spinning reserve requirements, conventional generator minimum loading, conventional
generator step-load response, system stability, and reactive power and voltage control
requirements [44].
10.3.4.1 Spinning Reserve
Spinning reserves are needed to account for unscheduled or scheduled loss of DER
[45]. The intermittent nature of wind and solar requires slightly higher spinning
reserves. However, 1:1 reserves are typically not needed. Depending on the degree
of wind/solar penetration, these reserves could reach 18%.
10.3.4.2 Minimum Loading of Conventional Generators
Some rotating machines require minimum loading depending on the type of prime
mover used. This particularly applies to diesel engine–driven generators, which
might require up to 40% minimum load to function properly.
239
Microgrids
10.3.4.3 Step-Load Requirements
Different generating sources can tolerate and respond to step-load changes (sudden
increases or drops in active power demand). The mechanical inertia of prime movers
and the speed at which prime movers can be adjusted to the changing electrical load
play an important role. Hydrogenerators respond rather slowly (orders of seconds or
tens of seconds), whereas PE-connected sources (wind or solar) are extremely fast
(orders of milliseconds). In case of a sudden increase in active power demand, one has
to take into account the maximum available power from the prime movers, which in
the case of wind and solar is changing and could be below the maximum-rated power.
10.3.4.4 Power System Stability and Oscillations
The stability of power systems depends on the inertia of the prime movers (as
described above) as well as the level of attenuation (damping) that the system exhibits during a disturbance. Large power-generating stations utilize power system stabilizers (PSS), which help in damping out possible system swings. Smaller machines
(diesel gensets, for example) typically do not employ these devices. On the other
hand, the PE-connected sources (wind and solar) can support the system stability if
their controls can perform the PSS equivalent functions.
10.3.4.5 Reactive Power Requirement and Voltage Control
In large-scale power grids, the typical power factor is lagging (inductive) and of the
order of 0.8–0.9 or better. In microgrids, the variety of generating sources, shorter
or no transmission/distribution lines, and more complex PE loads make the reactive power requirements more complex. It does not necessarily mean that the power
factor is lower than in the macrogrid. It only implies that the controls of microgrids
(either distributed or centralized) have to be able to monitor, calculate, and respond
to the more complex reactive power controls and voltage regulation.
10.3.4.6 Limitations to Renewable Energy Penetration
It is believed that the larger the degree of renewable penetration the higher the
requirements for spinning reserves (see also Section 10.3.4.1). In [44] the formula
given is
δmax = 1 –  Pmin (1 − PSR ) 
where:
δ represents the percentage of renewable energy penetration
δmax is the maximum possible percentage of renewable energy penetration,
given unlimited renewable power available
Pmin is the minimum conventional generator loading
PSR is the spinning reserve requirement of the power system
Figure 10.8 [44] illustrates an example of the achievable penetration δ for a system
where PSR = 20%, Pmin = 40%, and the conventional generator size is 50 kW. In this
case, δmax is only 50%.
240
Smart Grids
RE penetration (%)
60
50
40
30
20
10
0
0
50
100
150
Pload (kW)
200
250
300
FIGURE 10.8 δ versus Pload for a system where PSR = 20% and Pmin = 40%.
10.3.4.7 Short-Term Energy Storage Helps Renewable Energy Penetration
Understandably, storing the energy generated by the intermittent sources (wind,
solar) contributes to the smoothing effect and helps ride-through the power peaks
and valleys [46]. An electric battery, or a compressed gas facility, a lywheel or a
fuel-cell technology could be used for this purpose. Typically, the battery storage is
capable of several minutes of active power generation, and the lywheel can typically
cover 30 s or so [47]. For longer periods of energy storage, diesel gensets or fuel cells
can be utilized. One has to remember here that these different sources have also different ramp-up times: batteries have the shortest time (milliseconds) where fuel cells
require several hours to come online.
10.4
MICROGRID ECONOMIC VALUE STREAMS
The cost element involved in the implementation of microgrids is a widely discussed
topic among consumers and utilities. As with any capital project, implementing a
microgrid involves an investment in infrastructure, including the power sources and
the technologies needed to manage and connect the microgrid to the main grid.
However, the capital outlay required for a microgrid is often much less and payback is signiicantly faster than for other initiatives intended to improve the availability and reliability of power. The payback for a microgrid is often measured in years,
if not months, whereas the payback for a new utility-scale power generation might
be measured in decades and the project might face a barrage of regulatory hurdles
before it can even begin.
A number of factors can help offset the cost of a microgrid:
• Reduced Overall Energy Costs: One of the most important economic value
streams for microgrids is certainly the reduced dependence on grid-sourced
electricity and utility T&D services, although this ability to reduce dependence will result only if “average cost of internally generated power is less
than grid sourced power” [26].
• Sales of Excess Power to the Macrogrid: In some regions, energy markets
allow microgrid operators to sell the excess power generated, making the
241
Microgrids
•
•
•
•
10.5
microgrid a potential source of revenue. In addition, the heat generated from
the source powering the microgrid can be used to create an additional stream
of revenue for a business. For example, steam might be used to power up
additional generators, or hot water could be used for absorption chilling [26].
Greater Fuel Choice: While wind turbines and diesel generators are probably the two most common sources of power for established microgrids,
there are many choices available. New technologies and new discoveries are
driving down the cost of natural gas in some areas. Solar is becoming more
eficient and less costly. While fuel cells are still relatively expensive, they
are low maintenance, and research continues to ind new ways to make them
commercially viable. Leveraging a biomass fuel source may allow the company to claim carbon credits. These and other sources of energy can be combined as the consumer’s needs, technology, and market dynamics evolve [1].
Reduced T&D Losses: As much as 10% of energy is lost in transmission and
distribution. By their nature, microgrids are local, and the power consumed
has less distance to travel from generation to consumer, thus reducing a
signiicant percentage of T&D losses [1].
Participation in Organized Demand Response Markets: Demand response can
be understood as a change in net energy consumption (loads, distributed generation, storage) by a customer in response to a signal (price, load, emissions).
By virtue of the microgrid’s ability to precisely control sources of supply
and demand in response to market prices or other signals, interconnected
microgrids may be able to participate in organized demand response programs. This helps in reducing the cost of energy by charging the system with
low-priced energy and discharging later when the energy prices are high [26].
Enhanced Electricity Price Elasticity: By reducing consumption of energy
from the macrogrid, microgrids may be able to reduce the output from high
marginal cost or “peaking” plants, thereby reducing the clearing price for
electricity in wholesale energy markets. All power consumers—including
utility ratepayers—will beneit from this value stream.
RELIABILITY
Reliability is deined as the probability that a device will perform its intended function during a speciied period of time under stated conditions. Mathematically, this
may be expressed as
∞
R ( t ) = Pr {T > t} =
∫ f ( x ) dx
t
where f(x) is the failure probability density function and t is the length of time (which
is assumed here to start from time zero).
Reliability is therefore a function of time—the longer the time, the more likely
it is that it will fail—and, of course, a function of the initial (factory) quality and
design robustness of the device. Availability is the opposite of reliability; it is the
probability that a device will be available (will not fail) for a given period of time.
242
Smart Grids
Since it is dificult to know the reliability of a device before it is actually put into
service, another, more practical way of describing reliability is by two parameters:
failure rate (FR), measured in (number per period of time), and mean time to repair
(MTTR), measured in (minutes, hours, days, etc.). A commonly used parameter,
mean time between failures (MTBF), is the reciprocal of FR. Availability related to
failures is called inherent availability and is expressed as
Ai =
MTBF
MTBF + MTTR
Although a product of FR and MTTR (if MTBF≫MTTR) could yield a measure
of unavailability (i.e., [1−R(t)]*t), and it is often used in this way to describe devices
and systems, this calculation could be easily misinterpreted. For example, if a device
fails once over a period of 1 year and its MTTR is 1 day, its impact on the operational
availability of the system could be dramatically different from that of a device that
fails 24 times in 1 year with an MTTR of 1 h, or a device that fails 1000 times per
year with an MTTR of 86.4 s.
Systems such as microgrids are made up of components (devices), which all contribute to reliability. However, the degree to which these components play a role
in the overall reliability depends on a few factors, such as (a) topology, that is, the
location of the device within the microgrid, and the degree of redundancy built into
the system; and (b) the degree of criticality of the device for the mission (failures of
some devices, such as CBs, could result in immediate loss of service, whereas failures of some other devices, such as power quality meters, could be tolerated until the
next scheduled maintenance, for example).
10.5.1
relIaBIlIty due to maIntenance
So far, reliability has been deined in terms of occurrence of failures, that is, unexpected events. To complete the analysis, one must also include maintenance of devices
(expected, planned events), since this also contributes to the total system (un)availability. This can be easily done by quantifying maintenance frequency (MF) and
maintenance duration (MD), parameters equivalent to FR and MTTR. MF is often
expressed as the reciprocal of the mean time between maintenance (MTBM) and MD
is also called mean down time (MDT). Operational availability is deined similarly as
Ao =
10.5.2
MTBM
MTBM + MDT
mIcrogrId system relIaBIlIty
There is no one single reliability number for a whole microgrid as a system. Instead,
the reliability should be deined at speciic points, typically where loads are connected [27]. One has to realize that the same microgrid can have different reliability
performance at different load locations. Topology, load requirements, and state of
the microgrid all play a role.
Microgrids
243
In order to deine the reliability at any given point of the microgrid, one has to
combine the reliability data of all the devices and components as seen from the point
of interest. There are basically two approaches. The irst approach is to use analytical formulae to combine all the component data based on their interconnectivity, in
series or in parallel. The entire network can be then simpliied to a single set of two
numbers, outage frequency (OF) and outage duration (OD). If alternative, nonconcurrent paths exist to the given point of interest (where the load is connected), one
has to perform this process for both paths and weigh the OF and OD according to the
relative share of each path in which the microgrid will operate, for example, 30% in
path A and 70% in path B.
10.5.3
can relIaBIlIty oF a mIcrogrId Be Better
than that oF a utIlIty grId?
There is no simple answer to this question [28–32]. First of all, renewable resources such
as wind and solar have mixed availability due to natural factors such as weather [30,31].
Although they might not experience many equipment failures, the natural factors can
be interpreted as unavailability of a resource. On the other hand, the fact that the
microgrid is rather small and consists of a relatively low number of components can
have a positive inluence on reliability; fewer components mean fewer possibilities of
failure. Yet, in the islanded mode, a microgrid does not have the support (or backup) of
a utility grid behind it, and this can decrease the reliability of microgrid service. Yet
another aspect has to do with the energy storage [29]. If such equipment is used, it can
really improve the reliability of the service if it is properly designed and located within
the microgrid, and if the appropriate control logic is implemented.
10.6
OUTLOOK: MICROGRIDS—ENERGY
FOR THE DIGITAL SOCIETY
The notion of a “microgrid” has been around for more than a decade [33,34] in
the power engineering community. Yet these grids are far from being commercially prevalent (see DOE data, circa 2012, on installed microgrids), even though
the improved supply reliability—by disconnecting from a stronger utility grid if this
grid experiences operational problems—for end customers that own or are part of
the microgrid is technically sound. T&D utilities and medium- to large-sized consumers have been recognizing the technical beneit for some time, but the economics, in combination with existing utility grid supply reliability, have not yet been right
for microgrids to lourish.
The idea of splitting up a grid into islands for increased reliability is not new, and
it is employed, coupled with continuing research, to secure—or limit the negative
impact on—an interconnected transmission grid in emergency situations such as
a spreading blackout. By splitting up the system, parts of the network can survive,
as was illustrated in the 2003 northeast blackout in North America, when New
England disconnected from its neighbors and kept the lights on for its customers.
The T&D space has been slow to embrace microgrids, and so far all that has materialized are generally either small examples where disconnection was not explicitly
244
Smart Grids
2010
Washington
Maine
Montana
North Dakota Minnesota
VT NH
Oregon
ldaho
Wisconsin
South Dakota
Michigan
Wyoming
Nebraska
Nevada
Utah
CT
RI
Pennsylvania New Jersey
Iowa
Ohio
Delaware
West
DC
Virginia Virginia
Kentucky
North Carolina
Tennessee
Illinois Indiana
Colorado
Kansas
California
Arizona
MA
New York
Oklahoma
New Mexico
Missouri
Arkansas
Mississippi
Alabama
South Carolina
Georgia
Texas
Louisiana
Alaska
Florida
DOE R&D
DOE/DoD
Peak reduction
ARRA SGDP
Industry led
Hawaii
2012
Maine
Washington
Montana
North Dakota Minnesota
VT NH
MA
New York
CT
RI
Oregon
ldaho
South Dakota
Wisconsin
Michigan
Wyoming
Nebraska
Nevada
Utah
New Mexico
Ohio
Delaware
West
Virginia
DC
Virginia
Kentucky
North Carolina
Tennessee
Illinois Indiana
Colorado
Kansas
California
Pennsylvania New Jersey
Iowa
Missouri
Oklahoma
Arkansas
Arizona
South Carolina
Alabama
Georgia
Mississippi
Texas
Louisiana
Alaska
Hawaii
Florida
R&D test beds (DOE)
R&D test beds
(Non-DOE)
DOE/DoD
(Assessments and
SPIDERS)
DoD ESTCP
Peak load reduction
ARRA SGDP
Industry/utility/
university/other fed led
FIGURE 10.9 US microgrid landscape in 2010 and 2012. (From Smith, M., Ofice of
Electricity Delivery and Energy Reliability smart grid R&D program, in 2012 DOE Microgrid
Workshop, Summary Report, 30–31 July, Chicago, IL, 2012. For 2010: http://der.lbl.gov/sites/
der.lbl.gov/iles/Smith_2010.pdf; for 2012: http://e2rg.com/microgrid-2012/lnt%27l_USA_
Smith.pdf.)
245
Microgrids
allowed, but a net-zero power/energy exchange was demonstrated, or examples of permanently islanded systems (i.e., mini power systems such as the Azores Islands, where
ABB/PowerCorp showcased their solution, as reported in [35]). Figure 10.9 shows that
the US DOE has been tracking microgrid installations, and as of 2012 it is clear that
the numbers have been steady.
The United States is the leading microgrid market today, and is expected to remain
that way until at least 2020. Another microgrid hotspot is the Asia Paciic market,
where efforts have intensiied with the aim of ending energy poverty through remote/
off-grid microgrids in China and India. The market for microgrid technologies in
Europe is expected to expand in the near future as the region updates an aging electricity infrastructure and integrates a higher share of renewable energy generation.
Over the past 5 years, an increased interest in microgrids in scientiic and
industrial arenas has been observed, which can be seen in the upward trending of
“Google” hits (Figure 10.10).
The same accelerated trends have been identiied in [36] for a total microgrid
revenue by forecast scenario in 2013–2020 (Figure 10.11).
The main drivers for the increasing number of microgrid installations are as follows (the ranking may differ for different regions):
• Increased penetration of local DERs (motivated by government incentives
and falling costs of technology)
• Increased need for reliability and energy independence, even in the presence of natural disasters—extreme weather events causing massive power
outages will continue in the future—or homeland security events, such as
military microgrids and data centers
• Push from federal and state regulators to develop, implement, and uphold
various renewable portfolio standards
• Awareness of opportunities to participate in competitive energy markets by
selling lexible demand, ancillary services, etc.
• Technology push by a large number of companies offering hardware and
software control and protection technologies applicable to microgrids
One can think of a microgrid as a ixed-boundary virtual power plant, which by
itself can participate in the grid management of T&D systems. Due to this active participation, microgrid operators or owners would like to be adequately compensated
for their contribution. The business case input and evaluation methodology for these
100
80
60
40
20
2008
2009
2010
2011
FIGURE 10.10 Microgrid-related hits in Google trends.
2012
2013
246
Smart Grids
70,000
Base scenario
Average scenario
60,000
Aggressive scenario
($ millions)
50,000
40,000
30,000
20,000
10,000
2013
FIGURE 10.11
2014
2015
2016
2017
2018
2019
2020
Total microgrid revenue by forecast scenario, world markets: 2013–2020.
types of microgrids is quite similar to evaluating other interfacing grid technologies,
such as utility-scale energy-storage systems.
Some of the interesting applications of microgrids mentioned recently are:
• Data Centers: where the data center owner would be interested in predominantly net-zero energy operation, but would also have the ability to isolate
from the utility supply grid to maintain a connection to the information cloud.
• Buildings: owners are interested in net-zero operation and maintaining reliable energy supply to critical loads such as elevators. Disconnection from a
grid could be a possibility.
• Military Bases: these are interested in reliable power supply for critical loads and reduced dependency on imported fossil fuel (especially for
remote and isolated bases). For the bases connected to utility-scale grids,
due to policy on enforced renewable portfolio standards, the military aim to
self-supply and limit their dependence on the utility grid.
• Suburbs or Planned Community Developments: these might want to form their
own microgrid in order to survive or get up and running more quickly after
storms. This could make sense, seeing that the infeed to the community might
be overhead lines, prone to circuit damage from falling trees, whereas inside the
community the distribution might be underground and less prone to damage.
Hence, after an event it is conceivable that a community with suficient local
generation could self-supply while the utility distribution grid is being repaired.
• Cities: in China, due to overproduction of solar panels, the government is
starting to mandate more local installation of PV and is requesting cities to
be designed as microgrids.
From the above list, which is by no means exhaustive, it is clear that a microgrid
owner or operator would in general aim for net-zero, or, if suficient local capacity is
Microgrids
247
available, net-supplier energy operation with the larger grid. Once this is achievable,
operators would also aim to disconnect from the grid when necessary, if the larger
grids are experiencing issues, in order to increase their overall reliability.
It is fair to say that there are indications that microgrids have graduated from a mainly
academic topic to one that is of interest to various industries. Technology building blocks
exist, and reinements of these technologies are set to continue, but one of the biggest
breakthroughs comes when the economics are starting to make more sense and novel
business models are emerging. Understanding the economics of developing and operating a microgrid in detail is essential for success in this new and rapidly evolving market.
REFERENCES
1. M. MacDonald, Microgrids: Weathering the storm, June 2013. Available at: http://
www05.abb.com/global/scot/scot299.nsf/0/f49786ab08e927ff85257c4000554a40/$f
ile/abb-800-wpo_microgrids_weathering_the_storm_11-8-13.pdf.
2. Microgrids: From data centers to developing countries, an enticing solution for
self-suficient power, January 2012. Available at: http://www.lead-central.com/
AssetManager/e0bc9e75-84f1-4ca1-9c65-21c74a8b75fe/Documents/New%20WP%20
ile/ABB-73-WPO_Microgrids.pdf.
3. US Congress, National Defense Authorization Act for Fiscal Year 2007, 16 January,
2012. Available at: https://www.govtrack.us/congress/bills/109/s2507.
4. EU More Microgrids Project, Alternative microgrids network structures, Project
Deliverable DC1, November 2007. Available at: http://www.microgrids.eu/documents/643.pdf.
5. A. Chowdhury, D. Koval, Current practices and customer value-based distribution system reliability planning, IEEE Transactions on Industry Applications, 40(5), 1174–1182,
2004.
6. A. Dimeas, N. Hatziargyriou, Operation of multiagent system for microgrid control,
IEEE Transactions on Power Systems, 20(3), 1447–1455, 2005.
7. Y. Xu, “ORNL Power Systems and Power Electronics for Utility Applications”, ORNL,
Workshop on Power Electronics and Distributed Power systems, UTRC, December
2, 2011. Available at: http://web.ornl.gov/sci/ees/etsd/pes/pubs/UTRCWorkshop_
ORNLPowerElectronics_Microgrid.pdf.
8. A. Tsikalakis, N. Hatziargyriou, Centralized control for optimizing microgrid operation,
in Proceedings of the IEEE Power and Energy Society General Meeting, pp. 1–8, 24–29
July, San Diego, CA, 2011.
9. GE Digital Energy, Microgrid control system—Optimization solution for permanently
Islanded and grid-connected microgrids, Technical Brochure, 2012. Available at: http://
www.gedigitalenergy.com/products/brochures/MicrogridControlSystem.pdf.
10. ABB, Renewable microgrid controller, Technical Brochure, 2012. Available at:
http://www05.abb.com/global/scot/scot221.nsf/veritydisplay/d66c5a59934a10b7c1
257c37002cc707/$ile/9AKK100580A2552_Microgrid%20Controller%20600_EN_
LR(Dic2013).pdf.
11. P. Piagi, R. Lasseter, Autonomous control of microgrid, in Proceedings of the IEEE
Power Engineering Society General Meeting, 8–22 June, Montreal, QC, 2006.
12. W. Feero, D. Dawson, J. Stevens, Protection issues of the microgrid concept. Available
at: http://certs.lbl.gov/pdf/protection-mg.pdf.
13. H. Al-Nasseri, M.A. Redfern, R. O’Gorman, Protecting microgrid systems containing solid-state converter generation, in Proceedings of the International Conference on
Future Power Systems, p. 5, 18 November, Amsterdam, The Netherlands, 2005.
248
Smart Grids
14. J. Driesen, P. Vermeyen, R. Belmans, Protection issues in microgrids with multiple
distributed generation units, in Proceedings of the 4th Power Conversion Conference,
pp. 646–653, 2–5 April, Nagoya, Japan, 2007.
15. H. Nikkhajoei, R.H. Lasseter, Microgrid protection, in Proceedings of the IEEE Power
Engineering Society General Meeting, pp. 1–6, 24–28 June, Tampa, FL, 2007.
16. EU More Microgrids Project, Novel protection systems for microgrids, Project
Deliverable DC2. Available at: http://www.microgrids.eu/documents/688.pdf.
17. A. Oudalov, A. Fidigatti, Adaptive network protection in microgrids, International
Journal of Distributed Energy, 5(3), 201–225, 2009.
18. N. Schaefer, T. Degner, A. Shustov, T. Keil, J. Jaeger, Adaptive protection system for
distribution networks with distributed energy resources, in Proceedings of the 10th IET
International Conference on Developments in Power System Protection (DPSP), pp.
1–5, 29 March–1 April, Manchester, UK, 2010.
19. F. van Overbeeke, Fault current source to ensure the fault level in inverter-dominated
networks, in Proceedings of the 20th International Conference and Exhibition on
Electricity Distribution—Part 1 (CIRED), pp. 1–4, 8–11 June, Prague, Czech Republic,
2009.
20. D. Ishchenko, A. Oudalov, J. Stoupis, Protection coordination in active distribution
grids with IEC 61850, in Proceedings of the IEEE PES Transmission and Distribution
Conference and Exposition (T&D), pp. 1–6, 7–10 May, Orlando, FL, 2012. Available at:
http://www.informatik.uni-trier.de/~ley/db/journals/tie/tie58.html.
21. D.J. Becker, B.J. Sonnenberg, DC microgrids in buildings and data centers, in
Proceedings of the IEEE 33rd International Telecommunications Energy Conference
(INTELEC), pp. 1–7, 9–13 October, Amsterdam, The Netherland, 2011.
22. C. Marnay, S. Lanzisera, M. Stadler, J. Lai, Building scale DC microgrids, in Proceedings
of the IEEE Energytech, pp. 1–5, 29–31 May, Cleveland, OH, 2012.
23. J.-F. Hansen, J.O. Lindtjorn, T.A. Myklebust, K. Vanska, On-board DC grid, newest
design for marine power and propulsion systems, ABB Review, 2, 28–33, 2012.
24. H. Kakigano, Y. Miura, T. Ise, Distribution voltage control for DC microgrids using
fuzzy control and gain-scheduling technique, IEEE Transactions on Power Electronics,
28(5), 2246–2258, 2013.
25. A. Kwasinski, C.N. Onwuchekwa, Dynamic behavior and stabilization of DC microgrids
with instantaneous constant-power loads, IEEE Transactions on Power Electronics,
26(3), 822–834, 2011.
26. M.A. Hyams, Microgrids: An assessment of the value, opportunities and barriers to
deployment in NY State, in New York State Energy Research and Development Authority
(NYSERDA), pp. 77–83, September, 2010.
27. E.N. Dialynas, Evaluating the approximate probability distributions of load point reliability indices in power distribution networks, IEE Proceedings C, 135(5), 450–460,
1988.
28. Certs Microgrid Laboratory Test Bed, prepared for: California Energy Commission
Public Interest Energy Research Program, prepared by: Lawrence Berkeley National
Laboratory, February 2009, CEC-500-2009-004.
29. M. Stadler, C. Marnay, A. Siddiqui, J. Lai, B. Coffey, H. Aki, Effect of Heat and
Electricity Storage and Reliability on Microgrid Viability: A Study of Commercial
Buildings in California and New York States, Environmental Energy Technologies
Division, Lawrence Berkeley National Laboratory, LBNL-1334E-2009, Berkeley, 2009.
30. E.N. Dialynas, L.G. Daoutis, C. Toufexis, I. Charalambous, Reliability and reserve
capacity assessment of isolated power systems with increased penetration of renewable
energy sources, in Proceedings of the 7th Mediterranean Conference and Exhibition on
Power Generation, Transmission, Distribution and Energy Conversion, pp. 1–7, 7–10
November 2010, Agia Napa, Cyprus (Paper No. MED10/242).
Microgrids
249
31. E.N. Dialynas, D.J. Papakammenos, N.C. Koskolos, Integration of nonutility generating
facilities into the reliability assessment of composite generation and transmission power
systems, IEEE Transactions on Power Systems, 12(1), 464–470, 1997.
32. H. Liang, J. Su, S. Liu, Reliability evaluation of distribution system containing microgrid,
in Proceedings of the China International Conference on Electricity Distribution, pp.
1–7, 13–16 September, Nanjing, China, 2010.
33. T.E. Hoff, H.J. Wenger, C. Herig, R.W. Shaw Jr, Distributed generation and microgrids,
in Proceedings of the 18th Annual North American Conference of the USAEE, San
Francisco, CA, September 1997.
34. R. Lasseter, P. Paigi, Microgrid a conceptual solution, in Proceedings of the IEEE 35th
Power Electronics Specialists Conference, vol. 6, pp. 4285–4290, 20–25 June, Aachen,
Germany, 2004.
35. M. Smith, Ofice of Electricity Delivery and Energy Reliability smart grid R&D program, in 2012 DOE Microgrid Workshop, Summary Report, 30–31 July, Chicago, IL,
2012. Available at: (a) http://energy.gov/oe/downloads/2012-doe-microgrid-workshop-summary-report-september-2012, (b) http://der.lbl.gov/sites/der.lbl.gov/iles/
Smith_2010.pdf, (c) http://e2rg.com/microgrid-2012/lnt%27l_USA_Smith.pdf.
36. P. Asmus, A. Lauderbaugh, M. Lawrence, Navigant Research, Market data: Microgrids,
Executive Summary, 2013. Available at: http://www.navigantresearch.com/wp-assets/
uploads/2013/03/MD-MICRO-13-Executive-Summary.pdf.
37. G.D. Rockefeller, C.L. Wagner, J.R. Linders, K.L. Hicks, D.R. Rizi, Adaptive transmission relaying concepts for improved performance, IEEE Transactions on Power
Delivery, 3(4), 1446–1458, 1988.
38. B. Kroposki, C. Pink, J. Lynch, V. John, S.M. Daniel, E. Benedict, I. Vihinen, Flexible
DER utility interface system, Final Technical Report NREL/TP-560-39876, August
2006. Available at: http://www.nrel.gov/docs/fy06osti/39876.pdf.
39. EU More Microgrids Project, Decentralized control concepts for microgrids, Project
Deliverable DB3, February, 2008. Available: http://www.microgrids.eu/documents/641.pdf.
40. C. Yuen, A. Oudalov, A. Timbus, The provision of frequency control reserves from multiple microgrids, IEEE Transactions on Industrial Electronics, 58(1), 173–183, 2011.
Available at: http://www.informatik.uni-trier.de/~ley/pers/hd/t/Timbus:Adrian.html.
41. A. Bagherian, S.M.M. Tafreshi, A developed energy management system for a microgrid
in the competitive electricity market, in Proceedings of the IEEE Bucharest Power Tech
Conference, pp. 1–6, 28 June–2 July, Bucharest, Romania, 2009.
42. H. Kanchev, D. Lu, B. Francois, V. Lazarov, Smart monitoring of a microgrid including gas turbine and a dispatched PV-based active generator for energy management and
emission reduction, in Proceedings of the IEEE PES Innovative Smart Grid Technologies
Conference Europe (ISGT Europe), pp. 1–8, 11–13 October, Gothenburg, Sweden, 2010.
43. D. Lu, B. Francois, Strategic framework of an energy management of a microgrid with a
photovoltaic-based active generator, in Proceedings of the 8th International Symposium
on Advanced Electromechanical Motion Systems and Electric Drives Joint Symposium,
pp. 1–6, 1–3 July, Lille, France, 2009.
44. N. Hamsic, A. Schmelter, A. Mohd, E. Ortjohann, E. Schultze, A. Tuckey, J. Zimmermann,
Increasing renewable energy penetration in Isolated grids using a lywheel energy storage
system, in Proceedings of the International Conference on Power Engineering, Energy
and Electrical Drives (POWERENG), pp. 195–200, 12–14 April, Portugal, Setúbal, 2007.
45. Y. Rebours, D. Kirschen, What is Spinning Reserve?, The University of Manchester,
Manchester, UK, 2005.
46. H. Bindner, Power Control for Wind Turbines in Weak Grids, Riso National Lab,
Roskilde, Denmark, 1999.
47. A. Ruddell, G. Schönnenbeck, R. Jones, Flywheel Energy Storage Systems, Rutherford
Appleton Lab, UK, 1998.
11
Integrating Consumer
Advance Demand
Data in Smart Grid
Energy Supply Chain
Tongdan Jin, Chongqing Kang, and Heping Chen
CONTENTS
11.1 Introduction .................................................................................................. 252
11.2 Smart Grid Technology: The State of the Art .............................................. 252
11.2.1 Distribution Automation ................................................................... 253
11.2.2 Distributed Energy Generation and Integration ............................... 253
11.2.3 Optimal Asset Management ............................................................. 254
11.2.4 Advanced Metering Infrastructure ................................................... 254
11.2.5 Demand-Side Management .............................................................. 254
11.3 Principle of Virtual Energy Provisioning ..................................................... 255
11.3.1 Revisiting Manufacturing Supply Chain Model ............................... 255
11.3.2 Virtual Energy Supply Chain Model ................................................ 256
11.4 System Architecture and Operational Conditions ........................................ 257
11.4.1 System Architecture ......................................................................... 257
11.4.2 Operational Conditions ..................................................................... 258
11.5 Beneiting Operation of Smart Grid System................................................. 258
11.5.1 Facilitating Operation of Distributed Generation Systems .............. 258
11.5.2 Optimizing Asset Management ........................................................ 259
11.5.3 Enhancing Load Forecasting ............................................................ 259
11.5.3.1 Discrete Bayesian Prediction .............................................260
11.5.3.2 Continuous Bayesian Prediction ........................................ 262
11.5.3.3 Bayesian Prediction with Normal Distributions ................ 263
11.6 Consumer Performance Assessment ............................................................ 263
11.6.1 Minimizing Mean Square Error .......................................................264
11.6.2 Composite Performance Assessment Metrics ..................................264
11.6.3 Illustrative Example .......................................................................... 265
11.7 Communications and Automation Technology ............................................266
11.8 System Preliminary Design .......................................................................... 269
11.9 Conclusion .................................................................................................... 271
References .............................................................................................................. 272
251
252
11.1
Smart Grids
INTRODUCTION
Smart grids are enabling technologies featuring high reliability, self-healing, full
controllability, and optimal asset management. By combining intelligent control
algorithms with advanced communication and information technology, smart grids
allow utility suppliers to manage and control energy generation and distribution in
real time. As such, the network security, reliability, and eficiency are signiicantly
improved, beneiting all stakeholders, including generators, distributors, marketers,
and end consumers. A key technology for smart grid deployment is demand-side
management (DSM) technology. Leveraging communication and information technology, the purpose of DSM is to allow consumers to directly participate in the grid
operation through various means such as load shifting, direct load control, dynamic
pricing, and other demand response programs.
This chapter will introduce a novel DSM concept, virtual energy provisioning
(VEP), to manage the energy supply chain from generation to delivery and inal consumption. In a manufacturing supply chain system, the production typically is driven
by advance demand information from the retailers or end customers. Similarly, the
power system can be treated as an energy supply chain system in which the decision on electricity generation and distribution is made based on the consumption of
end consumers. However, existing energy systems usually do not take into account
the consumers’ advance demand information. In other words, electricity is generated according to forecasts based on historical consumption. Since consumers are
the ultimate decision makers regarding future consumption, it is imperative for
the utility supplier to let consumers participate in the forecasting process. To that
end, we propose an online VEP system in which consumers are able to provide
advance demand information so that better and more realistic forecasts can be made
by the utility supplier. The utility company, after aggregating consumers’ advance
demand data, can decide on an optimal generation and distribution scheme. This
chapter will discuss the VEP concept, the operational condition, and the preliminary
implementation.
The remainder of the chapter is organized as follows. Section 11.2 provides an
overview on recent advances in smart grid initiatives, including DSM technology.
Section 11.3 describes the principle of the VEP concept. Section 11.4 explains the
operational condition of the VEP system. In Section 11.5, the system beneits will
be discussed, focusing on asset management and distributed generation (DG) integration. Section 11.6 elaborates on the consumer performance assessment metrics.
Section 11.7 investigates modern communication technology and its impact on
VEP implementation. In Section 11.8, the preliminary implementation is presented.
Section 11.9 concludes the chapter with some remarks on future research.
11.2
SMART GRID TECHNOLOGY: THE STATE OF THE ART
Smart grids are enabling technologies aiming to resolve major challenges arising
from increased energy demands, the aging utility infrastructure, and growing concerns over the use of fossil fuels. Smart grids have been characterized or deined by
many organizations as Intelligent Grid (or Intelligrid) or Future Grid [1]. In order
Integrating Consumer Demand Data in Smart Grid Energy Supply Chain
253
to build intelligent features into the existing electric network, a comprehensive
technological suite comprised of real-time monitoring, forecasting, decision making, controls, and optimization algorithms is necessary [2]. From the implementation
perspective, smart grid technologies include (1) distribution automation; (2) distributed energy generation and integration; (3) optimal asset management; (4) advanced
metering infrastructure (AMI); and (5) DSM. In conjunction with the deployment
of information technology and telecommunication networks, smart grids allow utility companies to monitor and optimize the production and distribution of electric
energy in real time. As a result, the system, reliability, power quality, and energy
eficiency are consistently guaranteed, beneiting all the stakeholders, including generators, utilities, market traders, and end consumers.
11.2.1
dIstrIButIon automatIon
Distribution automation (DA) includes monitoring, control, and communication
functions located on the electrical feeder [1]. Protection and switching functions are
among the most important aspects when we design an effective DA mechanism.
Nowadays, various intelligent DA devices have been deployed in distribution lines
to track current and voltage states, exchange device information, and reconigure the
network to meet the changing environment. The topology in a smart grid is anticipated to be more lexible and adaptive, allowing the reconiguration of the topology
upon the occurrence of faults, excessive load demands, and malicious attacks.
Another important DA feature is automated fault diagnosis by monitoring the grid
fault, identifying the root cause, and restoring the system. Automated fault and rootcause identiication has also been investigated using cause–effect network, fuzzy
rules, artiicial intelligence, or Bayesian inference [3–5]. The deployment of wireless
sensor networks, distributed actuators, and ubiquitous communication technologies
on smart grids would generate more accurate real-time data about the system states,
making automated fault diagnoses feasible and applicable in future grids.
11.2.2
dIstrIButed energy generatIon and IntegratIon
DG produces and supplies electricity from many small energy sources, which are
often referred to as distributed energy resources (DERs). Typical DER units include
wind turbines, solar photovoltaics (PV), fuel cells, microturbines, and internal combustion engines. The capacity of a DER unit ranges from 3 kW to 10 MW, while
the capacity of a centralized generation unit usually exceeds 500 MW. Since energy
generated from DER units is often consumed locally, there is no need to build long
transmission lines to transport the electricity to end consumers, reducing costs
signiicantly.
Wind power and solar PV emerge as sustainable energy resources to meet the
increased electricity demand in the next two decades. It is anticipated that by 2030
about 20% of US electricity will be supplied from wind turbines, compared with
the current 2% market share [6]. However, wind and solar energies are quite intermittent, and the power output depends on the weather conditions, the geographical location, and the season of the year. The techniques for integrating renewable
254
Smart Grids
energy resources into the distribution networks have been extensively studied in the
literature, with the objective of designing robust and adaptive DG systems. For a
comprehensive survey, readers are referred to [7].
11.2.3
oPtImal asset management
The strategy of asset management can be perceived as a process to maximize the
return on investment of the equipment over its life cycle [8]. Decision makers often
leverage operations research and probability theory to identify the best asset management strategy that maximizes the long-term proits or minimizes the equipment
life-cycle cost. Utilities have aggressively devised various asset maintenance strategies to monitor, predict, and extend the remaining useful lifetime of the equipment
[9,10]. Advanced sensory technology combined with data mining provides ample
opportunities to characterize and forecast the remaining useful life of wind turbines,
solar PV, and transformers [11,12]. In smart grids, more intelligent sensors will be
built into the equipment to monitor the performance parameters so that conditionbased maintenance will be carried out immediately before the occurrence of equipment failures.
11.2.4
advanced meterIng InFrastructure
Communication systems play a pivotal role in transmitting metering data, publishing price information, and sending grid protections and control commands. Power
line communication, broadband Internet, and wireless meshes are three mainstream
communication technologies currently used in the power industry. Ostertag and
Imboden [13] propose a hybrid power line communication system to support both the
DA and the DSM functions using a priority-based medium access protocol. In [14],
it is reported that Texas New Mexico Power is using AT&T’s digital wireless cellular
network to collect metering data from 10,000 residential smart meters.
Debates still continue with respect to the selection of communication infrastructures for AMI deployment. Power line communication has been demonstrated to be
a reliable and secure communication technology, yet the transmission rate is quite
limited. Internet-based AMI systems seem very promising because of the maturity
of the technology and the ubiquity of network infrastructure. The concern with the
Internet is reliability, security, and robustness when faced with high-volume metering transmission and malicious cyberattacks.
11.2.5
demand-sIde management
DSM is the key to the effective implementation of the smart grid concept. The purpose of DSM is to develop and deploy new technologies for achieving power eficiency, energy conservation, and demand responsiveness. Many studies have been
published to explore various types of DSM concept along with the implementation
technology. These technologies include optimal usage scheduling [15], load shifting
[16], direct load control [17], dynamic pricing [18], and smart metering and appliances [19], among others. In [20], Zhong et al. reviewed the successes and challenges
Integrating Consumer Demand Data in Smart Grid Energy Supply Chain
255
during the deployment of DSM technology in China, one of the fastest-growing
countries in the world in terms of energy consumption.
Dynamic pricing such as real-time pricing (RTP) and time-of-use (TOU) encourages consumers to alter their energy usage patterns in order to reap inancial beneits.
The implementation of AMI makes dynamic pricing possible. Recently, many studies have been carried out to evaluate the impacts of dynamic pricing on electricity
consumption patterns. For a comprehensive review, interested readers are referred
to [21]. These methods usually apply game theory or develop inancial incentives to
guide consumers to reach a delicate balance between generation and consumption
in peak hours.
11.3
PRINCIPLE OF VIRTUAL ENERGY PROVISIONING
The power system is considered healthy and reliable only when all demands are
appropriately satisied or the interruption is kept at a minimum level. It is almost
impossible to ind a global optimal algorithm to control such a complex and dynamic
system based on today’s knowledge [2]. One viable approach is to leverage information technology and proactively collect future demand information from consumers.
The utility company, after analyzing and aggregating consumers’ advance demand
information, decides an optimal generation and distribution scheme to meet their
actual needs. In fact, the idea of consolidating consumers’ advance demand information is not new in the literature. In a manufacturing setting, the production and
distribution of goods are often driven by advance orders placed by end customers
or retailers. However, it is quite novel when this order-then-production or make-toorder concept is extended and applied to the energy industry. In the following, we
describe an Internet-based demand management system, called On-line Purchase
Electricity Now (OPEN) system, which is genially proposed in [22]. Through the
OPEN system, consumers are able to send advance demand information to the utility
supplier before they actually consume the electricity.
11.3.1
revIsItIng manuFacturIng suPPly chaIn model
Figure 11.1 depicts a typical manufacturing supply chain system comprising customer orders, production, inventory, and delivery process. Obviously, the supply
chain is driven or pulled by customers’ advance demand information. Upon receipt
of customer orders, the manufacturer starts to produce the necessary goods to meet
the demand. The inventory is widely used in manufacturing industries for temporarily storing inished goods before they are shipped to the retailers or end customers.
Customer
order
Production
Inventory or
warehouse
Delivery
FIGURE 11.1
Manufacturing supply chain driven by customer orders.
256
Smart Grids
An inventory usually involves three decision variables: the order quantity, the
stocking level (i.e., quantity of goods), and the lead time. The lead time is the time
interval between the arrival of the order and the instance of receiving the inished
goods by the customer. Obviously, this closed-loop product supply chain mechanism
remains stable as long as the inventory is not backordered. This stability can be easily maintained and sustained as long as the lead time is suficiently long. As such, the
manufacturer has enough time to produce the goods to meet the customer demand.
11.3.2
vIrtual energy suPPly chaIn model
Many similarities can be found between a manufacturing enterprise system and an
energy production system. In the former, goods are produced and sold to the end
customers, and the inventory is used as a buffer to mitigate any potential backorders
in the presence of demand uncertainty. Similarly, electricity is generated by power
plants or wind farms, and it is delivered to the end consumers via transmission/
distribution lines. Power interruptions or outages will occur if the load exceeds the
generation. The problem can be mitigated or partially resolved if large storage systems can be created between the generation and the distribution lines. These storage
systems, such as inventory systems, provide backup electricity to end consumers
when the load exceeds the generation. At present, research is under way to develop
eficient conversion or large storage technologies using lying wheels, batteries, and
supercapacitors [23]. However, commercially large and inexpensive storage systems
are still not mature enough for wide-scale applications.
We propose a VEP concept to complement and enhance the physical storage
technology. Conceptually, the energy provisioning system is operated as follows (see Figure 11.2). The utility company will design and implement an online
database, the OPEN system, in which an individual account will be assigned to
each consumer. Consumers can conveniently log on and access their web-based
account via desktops, laptops, and mobile devices. The system allows consumers
to place demand orders based on expected needs. These orders, representing future
Supply chain under production-then-consumption
Demand
forecasting
Generation
Transmission
and distribution
Consumption
Supply chain under order-then-production
Consumer
order
Utility
supplier
Virtual energy
provisioning
Actual
generation
Transmission
and distribution
FIGURE 11.2 Generation-then-consumption versus order-then-generation.
257
Integrating Consumer Demand Data in Smart Grid Energy Supply Chain
electricity demand, can be speciied on an hourly, daily, weekly, or monthly basis.
The utility company, after aggregating all consumers’ orders, can decide an optimal generation and distribution scheme to meet the consumers’ needs. Just like a
manufacturing supply chain system, this managed order-then-generation system is
able to deliver the required electricity to the end consumers in a timely and reliable
manner. In Section 11.6, various incentive programs are devised to motivate the
consumers to place orders and then consume exactly the amount of energy that they
ordered.
The VEP model aims to achieve the same objective as the physical storage system, by delivering reliable and secure energy to end consumers. However, VEP does
not store the electricity in physical media such as batteries or capacitors; rather, it
leverages information technology to solicit advance demand information from consumers. The lead time between the demand order and the actual generation (i.e.,
consumption) allows the utility supplier to plan the optimal production scheme to
meet the future need.
11.4
11.4.1
SYSTEM ARCHITECTURE AND OPERATIONAL CONDITIONS
system archItecture
Laptop
Internet
Power line
Loading 1
SM
Data mining
Database
Decision
making
Power
generation
Wireless meshes
The OPEN system is a realization of the VEP concept described in the previous
section. The OPEN system will be built on a hybrid architecture integrating existing utility infrastructure with the Internet platform. Figure 11.3 depicts a conceptual
coniguration of the hybrid architecture, duplicated from [22]. Given the ubiquitous
Internet technology, the deployment of the OPEN system only requires several costeffective software programs and hardware components. Software programs include
an online database, a data mining engine, and a decision-making tool. The purpose
of the database is to collect the demand information from end consumers as well as
to store the actual consumption transmitted by the smart meter. The data mining
engine is capable of forecasting the future load condition by combining consumers’
Smart
phone
PC
Loading 2
Loading 3
(e.g., household)
SM
SM
Smart
meter
Internet
Power line
FIGURE 11.3 A hybrid architecture for OPEN system implementation. (From Jin, T. and
Mechehoul, M., IEEE Transactions on Smart Grid, 1, 302–310, 2010.)
258
Smart Grids
orders with their historical consumption. The decision-making tool synthesizes the
information and determines the optimal generation and distribution scheme.
The major hardware component is the smart meter, which transmits the data
from the consumer site to the utility supplier. The consumption data are primarily in a short message or packet format, representing hourly or daily energy
consumption by the consumer. Since the transmission of the discrete data is not
time-critical, transmission control protocol/Internet protocol (TCP/IP) should be
suficient and reliable enough to undertake the communication task. Research
[13,14] shows that such a communication function will be relatively easy to implement by directly integrating the smart meter with the Internet system or the wireless meshes.
11.4.2
oPeratIonal condItIons
Consumer participation is the key to the effective operation of the OPEN system.
Various incentive packages can be developed so that consumers will participate in
the program and provide demand information. Fahrioglu and Alvardo [24] show that
inancial incentives, when appropriately implemented, will motivate consumers to
participate in the energy management system. For example, monetary discounts or
energy credits could be used to reward the consumer whose demand order is equal
or close to the actual consumption. In addition, the supplier can design customized
usage portfolios to accommodate different users’ consumption behaviors. These
incentive programs can be easily posted and distributed to individual consumers via
the online database system.
Another assumption is that dynamic pricing, such as TOU or RTP, is an effective tool to inluence and guide the consumer’s usage behavior. Studies [19,25] have
shown that dynamic pricing is an effective tool to control and minimize demand
uncertainty. Like an adaptive controller, dynamic pricing establishes a closed-loop
feedback between the suppliers and their consumers, and brings the actual demand
close to the generation via game theory [26]. Eventually, a long-term balance between
generation and consumption is reached.
11.5
BENEFITING OPERATION OF SMART GRID SYSTEM
The utility supplier is able to gain substantial beneits from the deployment of the
OPEN system. These beneits include (1) facilitating the integration and operation
of DG systems; (2) optimizing asset management; and (3) improving demand forecasting. These beneits will be explained in the following paragraphs, and, for more
detailed discussions, readers are referred to [27].
11.5.1
FacIlItatIng oPeratIon oF dIstrIButed generatIon systems
Wind turbines and solar PV have emerged as clean and sustainable resources to
meet growing electricity needs in the next 20–30 years. These renewable technologies, however, are quite intermittent due to the stochastic nature of wind speed and
solar irradiation. Integrating renewable energy into existing power grids potentially
Integrating Consumer Demand Data in Smart Grid Energy Supply Chain
259
creates additional uncertainty on top of the load variability. When planning and
operating DG systems, the uncertainties in power generation and load proile must
be taken into account at the same time. Otherwise, the grid reliability and stability
could be jeopardized, as more variable energy resources are penetrating into the
electricity market.
The OPEN system can play a substantial role in managing DG systems. Built
upon the order-then-production principle, the OPEN system is capable of collecting
and consolidating consumers’ future demands ahead of time. The lead time allows
the utility supplier to perform demand predictions, plan optimal generation, and
determine distribution schemes. Proactive measures can be adopted in advance to
avoid the occurrence of power outage or interruption due to lower wind speed or
cloudy weather. In this case, the OPEN system serves as an advance information system that allows the supplier to acquire future load conditions to mitigate the power
uncertainty in renewable energy resources.
11.5.2
oPtImIzIng asset management
The OPEN system creates a truly interactive environment in which the supplier and
the consumers can engage in real-time communications. The system is quite beneicial in situations when a sudden change in demand occurs, such as a family leaving
for a long vacation or a surge in energy demand due to manufacturing ramp-up.
Consumers can easily send their reduced or increased energy demand information
through the OPEN system.
The OPEN system facilitates the deployment of plug-in hybrid electric vehicles
(PHEVs) or other electrical apparatus across the grid system. Charging a PHEV
means pumping 20–40 kWh electricity from the grid within 2–4 h, causing a signiicant impact on the stability of the distribution grid [28]. Functions can be built into
the OPEN system that allow PHEV users to notify utility suppliers when and where
the charging will be performed, and how much energy will be needed. This allows
the supplier to take the necessary measures to mitigate the impact of charging on the
local distribution network.
The OPEN system can assist utility companies in optimizing the asset maintenance time based on consumers’ demand information. Maintenance activities on
transformers, power lines, and other utility facilities can be carried out at times when
the consumers’ demands are relatively low, based on their placed orders. If an urgent
repair must be performed in a timely manner, the supplier can notify the affected
consumers about the pending repair actions by broadcasting the message through the
OPEN system. Consumers can then temporarily stop or reduce their energy usage to
streamline the maintenance activity.
11.5.3
enhancIng load ForecastIng
Utility companies are able to perform more realistic load forecasting, as they can
combine historical data with advance demand orders. Bayes’ theorem is capable
of synthesizing historical usage with advance orders to estimate the future load.
Although the theorem has been widely used in various domains, its application in
260
Smart Grids
the power industry for predicting future consumption is relatively new. Some studies
[29,30] have reported construction of the prediction model using Bayes’ theorem.
These works primarily focus on the utilization of Bayes’ theorem to obtain a better
model structure.
In the OPEN system, the Bayesian forecasting model is adopted to predict future
electricity demand without making any assumptions about the model structure. The
probability function representing historical usage is referred to as the prior distribution, and the probability function representing future consumption is called the
posterior distribution. As shown in Figure 11.4, Bayes’ theorem allows us to directly
estimate the posterior distribution by synthesizing the prior information with the
demand orders. Two Bayesian prediction methods are available: discrete Bayesian
and continuous Bayesian models. In the former, the prior and the posterior distributions are represented by discrete probabilities. In the latter, continuous probabilities
are used to model the prior or the posterior distribution or both.
11.5.3.1 Discrete Bayesian Prediction
The discrete Bayesian model can be applied in situations in which the demand data
are characterized by attributes, or categorical data. The mathematical expression for
the model is given in Equation 11.1:
P { Xi | Y } =
P { X jY }
P {Y }
=
P {Y | Xi } P { Xi }
∑
k
j =1
(11.1)
P {Y | X j } P { X j }
where:
k
is the total number of events or categories
P{Xi} is the prior probability for Xi with i = 1, 2, …, k
P{Y|Xj} is the conditional probability for consuming Y given Xj
P{Xi|Y} is the posterior probability for Xi given Y
The following example, taken from [27], is used to explain the application of this
model. Table 11.1 lists the daily energy consumption of a consumer over 15 days.
The actual consumption of energy, denoted as X, is listed in the second column,
while the forecast or ordered quantity, Y, provided by the consumer is shown in the
last column. For example, on Day 10, the consumer’s original order is 13 kWh, while
the actual consumption is 14 kWh on that day.
Prior
distribution
Customer
demand
orders
Historical consumption
FIGURE 11.4
The Bayesian forecast process.
Posterior
distribution
Future consumption
Integrating Consumer Demand Data in Smart Grid Energy Supply Chain
261
TABLE 11.1
Daily Energy Usage versus
Prediction
Actual Usage
(X) (kWh)
Consumer Order
(Y) (kWh)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
10
11
13
13
12
11
10
14
14
14
12
14
12
12
11
12
12
14
12
10
12
14
12
13
13
13
11
12
15
11
11
Day
TABLE 11.2
Conditional Probability of P{Y|X}
Y
kWh
10
11
X 12
13
14
10
0
0.333
0
0
0
11
0.5
0.333
0.25
0
0
12
0.5
0.333
0.5
0.5
13
0
0
0.25
0
0.25
0.5
14
0
0
0
0.5
0.25
From Table 11.1, we can easily compute the prior probability for X, and the result
is given as P{X = 10, 11, 12, 13, or 14} = {0.13, 0.20, 0.27, 0.13, or 0.27}, respectively.
Using Equation 11.1, we are able to estimate the conditional probability for Y given
X by using the data in Table 11.1, and the results are summarized in Table 11.2. For
example, P{Y = 13|X = 14} = 0.5, as shown in the last row. Now the posterior probability for X can be extrapolated based on Table 11.2 along with the prior probability
of X. The result is summarized in Table 11.3. For instance, the last row of the table
shows P{X = 14|Y = 13} = 0.67, meaning that the probability of consuming 14 kWh
electricity is 0.67, given that the consumer’s order is 13 kWh.
Given the consumer’s order, we can, further, compute the conditional mean
demand and the variance based on the conditional probability in Table 11.3. For
262
Smart Grids
TABLE 11.3
Posterior Probability of P{X|Y}
Y
kWh
10
11
X
12
13
14
10
0
1.0
0
0
0
11
0.33
0.33
0.33
0
0
12
0.17
0.17
0.33
0.17
0.17
13
0
0
0.33
0
0.67
14
0
0
0
0.5
0.5
instance, if the consumer’s order quantity is Y = 13 kWh, then the conditional mean
demand and its variance can be estimated as
5
E  X Y = 13 =
∑ xP {X = x Y = 13} = 13.33 kWh
(11.2)
i =1
(
(
)
var X Y = 13 = E 2  X Y = 13 − E  X Y = 13
)
2
= 0.89 ( kWh) 2
(11.3)
This numerical example, though relatively simple, clearly demonstrates how the
discrete Bayesian inference model can be utilized to predict consumers’ actual consumption by synthesizing the historical data with their demand perspectives.
11.5.3.2 Continuous Bayesian Prediction
If the actual consumption data are continuous and the demand order is placed in
a discrete format, forecasting can be performed by using the Bayesian model for
density functions:
(
)
fX x Y = y =
{
}
P Y X = x fX ( x )
P {Y }
=
{
}
P {Y X = x} f ( x ) dx
P Y X = x fX ( x )
∫
+∞
0
(11.4)
X
When both X and Y are represented by continuous density functions, the Bayesian
prediction model can be expressed by
( )
f X |Y x y =
( )
f X ,Y ( x, y ) fY | X y x f X ( x )
=
fY ( y )
fY ( y )
where:
X is the actual consumption, a continuous random variable
Y is the consumer demand order, a discrete random variable
P{Y} is the probability for Y
f X(x) is the prior distribution for x
f X(x|Y = y) is the probability density function for x given y
P{Y|X = x} is the conditional probability for Y given x
(11.5)
263
Integrating Consumer Demand Data in Smart Grid Energy Supply Chain
f Y (y) is the marginal distribution for y
f X,Y (x,y) is the joint distribution for x and y
f Y|X(y|x) is the conditional density function for y given x
f X|Y (x|y) is the posterior density function for x given y
The likelihood function f Y|X(y|x) can be derived from consumers’ previous
orders along with their actual consumption. All other notations are the same as in
Equation 11.4.
11.5.3.3 Bayesian Prediction with Normal Distributions
Normal distributions are often used to represent random electricity loads [31]. If X
and Y follow the normal distributions with X ∼ N µ x , σ2x , and Y ∼ N µ y , σ2y , their
joint density function is given as
(
f X ,Y ( x, y ) =
1
2πσ x σ y

1

exp −
2
2
1− ρ
 2 1 − ρ
(
)
)
(
)
 ( x − µ x )2 ( y − µ y )2 2ρxy  


+
−

σ2x
σ2y
σ xσ y 


(11.6)
where ρ is the correlation coeficient with −1 ≤ ρ ≤ 1. By substituting Equation 11.6
into Equation 11.5, the posterior distribution for X is given as
f X |Y ( x | y ) =
1
(
σ x 2 π 1 − ρ2
)

1

exp −
2
 2 1 − ρ
(
)
 ( x − µ x )2 ρ2 ( y − µ y )2 2ρxy  


+
−

σ2x
σ2y
σ xσ y 


(11.7)
In summary, Bayes’ theorem is able to combine historical usage with the demand
order to predict the future load. The discrete Bayesian model is preferred if the consumer usage and order are speciied in discrete format. The continuous distributions
are more appropriate in applications in which the consumption and demand data are
expressed as continuous variables.
11.6
CONSUMER PERFORMANCE ASSESSMENT
Incentive programs are implemented to encourage consumers to participate in the
OPEN system and place demand orders via their online accounts. Prior to constructing incentive programs, it is important to deine the performance metrics in order to
assess the quality of orders. One reasonable approach is to compare the difference
between the order and the actual consumption. Then various inancial incentive programs can be implemented, based on whether the actual usage matches the original
order. The following set of performance measures to assess the quality of orders is
based on [32].
264
Smart Grids
11.6.1
mInImIzIng mean square error
Minimum mean square error (MMSE) is often used to measure the difference
between the prediction and the actual result. Let xi be the actual consumption and yi
be the demand order for time ti; MMSE can be computed as
MMSE =
1
m
m
∑( x − y )
i
2
(11.8)
i
i =1
where m is the sample size. A small MMSE indicates that actual consumption closely
matches the demand order. Obviously, energy credits or monetary discounts should
be used to reward the consumer who maintains a small MMSE over the consumption horizon.
MMSE is a simple statistical tool, which can be easily implemented in the OPEN
system. However, MMSE cannot effectively capture the discrepancy between xi and
yi for a particular time. In other words, consumers with the same MMSE may have
a very different usage proile. Table 11.4 shows a list of 7-day energy orders versus
the actual usages collected from three consumers. All consumers have the same
MMSE = 4. The variation for consumer C on Day 7 is more pronounced than those
of A and B. In addition, the actual computation for A is consistently smaller than
the order, while the actual consumption for B is consistently larger than the order.
Therefore, additional metrics are required to capture the consumer performance
variations.
11.6.2
comPosIte PerFormance assessment metrIcs
Three major factors need to be considered when we perform the assessment on the
quality of the demand order: (1) the mean error between the order and the actual
TABLE 11.4
Demand Order versus Actual Consumption
Day
Consumer A (kWh)
i
xi
yi
1
2
3
4
5
6
7
18
20
17
20
21
23
16
20
22
19
22
23
25
18
MMSE
(xi − yi)2
4
4
4
4
4
4
4
4
Consumer B (kWh)
xi
yi
12
14
11
9
11
12
13
10
12
9
7
9
10
11
(xi − yi)2
4
4
4
4
4
4
4
4
Consumer C (kWh)
xi
yi
(xi − yi)2
32
31
29
34
32
29
38
32
31
29
33
31
28
33
0
0
0
1
1
1
25
4
Integrating Consumer Demand Data in Smart Grid Energy Supply Chain
265
consumption; (2) the variation between the order and the consumption; and (3) the
largest deviation across the entire period. Based on these criteria, the following composite metrics are proposed:
q1 =
q2 =
{
1
m
1
m
m
∑( x − y )
i
(11.9)
i
i
m
∑( x − y )
i
2
(11.10)
i
i =1
q3 = max ( xi − yi )
2
}
for i = 1, 2, … , m
(11.11)
These metrics form a set of composite assessment instruments to address the concerns raised previously. q1 calculates the mean error between the order data and the
actual usage. The result is able to indicate whether the order is always below or above
the actual usage. q2 measures the deviation between the order and the actual usage.
q3 identiies the largest discrepancy between xi and yi for all i. All these performance
metrics have a simple analytical form, and together they are able to assess consumer
performance by simply comparing the advance demand order and actual usage data.
11.6.3
IllustratIve examPle
The following example, taken from [32], is used to demonstrate the application of
the performance assessment instrument. Three sets of 15-day orders from three consumers have been collected and listed in Table 11.5. The actual daily consumption is
also recorded by the smart meters, and the results are also listed in the table.
Table 11.6 summarizes the performance metrics computed using Equations 11.9
through 11.11. According to the mean error q1, Consumer A overestimates the
demand while B and C underestimate the needed electricity. In terms of variation,
Consumer B has the smallest MMSE, meaning that the actual usage for B is close to
what she ordered. Consumer A has the largest MMSE, indicating a high daily variation between the order and the actual usage. Although the MMSE for C is not the
largest, the largest daily deviation occurred for Consumer C. By examining the dataset for C, we found that the difference between the order and actual usage is 7 kWh
on Day 15, resulting in the largest daily discrepancy among all the consumers. These
examples show that, based on the composite assessment instruments, the utility supplier is able to determine consumer performance in a fairly easy way.
Since the target value for q1, q2, and q3 is zero, the consumer performance can be
easily evaluated in terms of the proximity to the target value. Based on this, various
incentive packages can be devised to reward consumers who achieve good performance. In [22], a quadratic price model combining the reward function is proposed
to incentivize the consumer to participate in the OPEN system. The model allows a
266
Smart Grids
TABLE 11.5
Demand Order versus Actual Consumption
Consumer A (kWh)
Order (yi)
Consumer C (kWh)
Actual (xi)
Order (yi)
Actual (xi)
Order (yi)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
24.5
22.1
26.9
24.8
27.3
28.9
32.6
30.2
38.5
29.3
31.0
27.1
32.6
29.5
28.6
25.7
24.3
27.9
32.0
32.0
33.8
25.9
34.9
33.0
29.1
32.7
29.8
31.3
25.2
30.3
22.5
26.7
18.7
20.3
24.3
18.8
24.0
31.3
24.6
24.6
19.7
26.5
25.7
27.3
21.4
25.9
18.3
19.1
25.3
17.4
24.1
29.4
26.4
25.0
18.0
27.0
38.4
41.7
36.4
39.6
33.9
39.8
41.4
43.1
43.8
38.2
37.8
40.2
40.2
39.4
15
26.5
26.1
26.0
26.5
50.0
39.2
40.6
37.5
40.3
35.1
40.6
40.8
41.7
41.5
37.5
40.1
40.4
40.2
37.8
42.6
Day (i)
Actual (xi)
Consumer B (kWh)
TABLE 11.6
Performance Assessment for Three
Consumers
Metrics
q1
q2
q3
A
B
C
Target Value
−1.0
11.4
32.0
0.5
1.7
9.3
0.5
5.0
54.8
0
0
0
consumer to save up to 30% of the monthly bill if all performance metrics meet the
targets. Notice that other options are available, and they can be built into the reward
functions. Interested readers are referred to [33,34].
11.7
COMMUNICATIONS AND AUTOMATION TECHNOLOGY
To help utility companies and consumers to generate a power usage proile, the actual
power consumption has to be measured and available to both the utility company and
the consumers. There are many advantages if the real-time power consumption is
available:
• The transparency of power usage between utility companies and consumers
is increased. Hence, the utility companies have a better understanding of
consumers’ demand.
Integrating Consumer Demand Data in Smart Grid Energy Supply Chain
267
• The utility companies can schedule their power supply according to consumers’ demand. Thus, the power generation cost can be reduced and pollution can be decreased.
• Consumers will be aware of their real-time power usage rather than obtaining a monthly utility bill, or at even longer intervals. Thus, their power
consumption can be altered and controlled.
• If dynamic pricing is available, consumers can schedule their power consumption according to the price proile. For example, they can schedule
their washing machine when the price is lowest if they are not in a hurry to
wash their clothes.
To measure the actual power consumption and make it available to both utility
companies and consumers, smart meters and data communication networks have to
be implemented. Since the Ethernet is widely used at both consumer sites and utility
companies, using the existing communication infrastructure reduces the infrastructure cost and eliminates many implementation-related issues. Figure 11.5 shows a
power consumption data collection and distribution scheme.
In Figure 11.5, the local smart sensors form a local network. The local servers
are installed at the consumer sites. The utility servers are installed at the utility
companies. As computers are becoming more powerful and have many different
built-in interfaces, such as wireless, Ethernet, USB, and serial ports, it is convenient
to connect the smart sensors with different communication interfaces to the computers at the consumer sites. Therefore, there are many possible ways for the local smart
sensor network to communicate with the local server. The local servers exchange the
data with the utility servers via Ethernet. The main reason for isolating the smart
meters with their network from the local servers is security considerations. As the
Smart
sensor
Smart
sensor
Smart
sensor
Smart
sensor
Smart
sensor
Smart
sensor
Local
server
Local
server
Local
server
Utility
server
FIGURE 11.5
Utility
server
Utility
server
A sensor network interconnecting utility companies with consumers.
268
Smart Grids
smart sensor network becomes more complex, network security will become a major
issue in the near future. Since the local servers and the utility servers in the utility
companies are connected via Ethernet, they can communicate with each other without sacriicing the security of smart meters.
As the amount of data collected increases with time, servers require vast storage
capacity, especially in the utility companies, because there are many smart meters
connected via the local servers. There are several problems with storage of vast
amounts of data: (a) extra hardware is needed to save the data; (b) data mining is
becoming more dificult; and (c) the processing time is becoming longer. To deal
with these problems, distributed data storage and processing techniques are being
developed to reduce the amount of data transmitted to the utility servers.
Under the distributed processing framework, data from each smart meter are collected and saved in the smart meter, if it has a storage device, or in the local server.
Since the storage capacity of a smart meter is typically limited, the data should be
transferred to the local server for storage. As the local server is only connected to one
or several local smart meters, the storage capacity of a local server should be large
enough to store the data from the local smart meters. After the data are transferred
to the local server, they are processed to reduce the data size. Here we propose a
B-spline curve-itting method. Suppose we have m power consumption data points
Pi (i = 1, 2, …, m) during a cycle (e.g., 1 h or a day), a B-spline curve can be itted to
the m points:
S (t ) =
m−n
∑C b
i i,n
( t ),
t ∈ t n , t m − n 
(11.12)
i =1
where:
t
S(t)
Ci
bi,n(t)
is the sampling time during a cycle
is the B-spline curve
are the control points
is the B-spline basis function, which is deined as
if t j ≤ t ≤ t j +1
otherwise
1
b j ,0 ( t ) = 
0
b j ,n ( t ) =
t − tj
t
−t
b j +1,n −1 ( t )
b j ,n −1 ( t ) + j + n +1
t j+n − t j
t j + n +1 − t j +1
(11.13)
For example, if a uniform cubic B-spline is used, the function can be represented
as
Sk ( t ) = 1 6 t 3
t2
t
 −1

3
1 
 −3

1
3
−6
0
4
−3
3
3
1
1   Ck −1 


0   Ck 
0   Ck +1 


0  Ck + 2 
(11.14)
Integrating Consumer Demand Data in Smart Grid Energy Supply Chain
269
Therefore, instead of transmitting m data points, only four points are needed to
represent the power consumption during a cycle if a common basis function is used.
This will greatly reduce the data storage requirement and decrease the data processing time. For the local server, the control points have to be computed. Here we developed a least square method to obtain the optimal control points:
f = min
∑ || S(t ) − P ||
i
i
(11.15)
i
After the objective function f is minimized, the control points are obtained. Since
this optimization is performed in the local servers, this will greatly mitigate the data
congestion issue.
11.8
SYSTEM PRELIMINARY DESIGN
Based on the concept proposed in Section 11.3, we have started system implementation in the laboratory setting. This section will summarize our preliminary designs
and development of the user-friendly interface that is an integral part of the OPEN
system. More detailed discussions are given in the recent work in [27].
In our VEP system, the user interface behaves like a bridge between users and
service provider. Similarly to an online shopping web page, users can easily browse
their historical usage data and input new usage proiles and other related consumption information, such as number of persons living in a house, appliance information,
and daily activities. Hence, designing an interactive, functional, and user-friendly
interface is very important. Figure 11.6 shows a web-based interface allowing consumers to specify the power capacity of major household appliances.
Our designed interface consists of four components: (1) User information. When
the new user logs into the system, he or she needs to set up an account (user name
and password) and input personal information, including the number of persons in
the house and major appliances, for example, air conditioner model. This information will be stored in the database, and will be very useful for the OPEN system in
predicting the future electricity load. The above example is taken from residential
homes, but the same principle can be applied to industrial and business entities.
(2) Visualization. In order for consumers to plan well, it is necessary for them to
visualize past electricity usage (e.g., hourly, daily, weekly, or monthly), usage patterns of the same period in previous years, and usage patterns of similar households.
Figures 11.7 and 11.8 graphically show the daily consumption proile by apparatus
type and the cumulative usage, respectively. (3) Search. The major function of the
“Search” module is to help users ind the electricity consumption information for
their appliances, such as air conditioners or refrigerators. For example, consumers
can input their refrigerator model and the search module will ind the average electricity consumption of this type of refrigerator. This information will be very helpful
for them to make an accurate prediction. (4) Feedback. Through the feedback component, the system will provide the user with a predicted usage amount. Consumers
can also input their estimated amount and give feedback to the system. For example,
based on previous usage information, the system will automatically generate an
270
FIGURE 11.6
Smart Grids
Interface for specifying power capacity of apparatus.
Watts
800
700
600
500
400
300
[Total]
Sum of consum_AC
Sum of consum_TV
Sum of consum_music
Sum of consum_oven
200
100
Days 1
FIGURE 11.7
Consumer online account.
2
3
4
5
6
271
Integrating Consumer Demand Data in Smart Grid Energy Supply Chain
Watts
600
500
400
300
200
100
Sum of max_consum
Sum of min_consum
Sum of actual_consum
1
2
3
4
5
6
Day
FIGURE 11.8
Consumer’s actual usage proile.
estimated demand. It is up to the consumer to modify the recommended usage and
place the order through the OPEN system. If the consumer will be out of town next
month, for instance, he or she can input a very small amount of electricity usage.
This advance demand information, once appropriately analyzed by the data mining
engine at the utility side, can be used as the baseline to plan future generation.
11.9
CONCLUSION
This chapter presented a novel DSM technology, VEP, which allows smart grid consumers to predict and order electricity via the broadband Internet as if performing
an online purchasing transaction. This new energy management concept intends to
shift the electricity delivery service from production-then-consumption mode to an
order-then-production paradigm. The system architecture, operational conditions,
and potential beneits to energy stakeholders are elaborated from various aspects,
including generation, marketing, delivering, and end consumption. We also investigated several statistical instruments to assess the quality of demand orders. The
purpose is to evaluate consumers’ performance and further develop incentive programs to increase their participation. The preliminary research shows that the OPEN
system is technically feasible, theoretically justiiable, and inancially achievable.
The ultimate goal of the VEP concept is to make the consumers “order the electricity
they need, and consume exactly what they ordered.” As a result, the stochastic load
proile is turned into a deterministic demand curve, reducing operational uncertainty
and improving system robustness. Many research opportunities will be opened up
272
Smart Grids
by implementing this new DSM concept. These areas include, but are not limited
to, dynamic pricing, optimal generation and distribution planning, and consumer
incentive management. Future efforts will be focused on software development and
laboratory testing, and will eventually move to real-world applications.
REFERENCES
1. Brown, R.E., Impact of smart grid on distribution system design, in Proceedings of
the IEEE Power and Energy Society General Meeting, pp. 1–4, July 20–24, 2008,
Pittsburgh, PA.
2. Venayagamoorthy, G.K., Potentials and promises of computational intelligence for
smart grids, in Proceedings of the IEEE Power and Energy Society General Meeting,
pp. 1–6, July 26–30, 2009, Calgary, AB.
3. Chen, W.-H., C.-W. Liu, and M.-S. Tsai, On-line fault diagnosis of distribution substations using hybrid cause-effect network and fuzzy rule-based method, IEEE Transactions
on Power Delivery, 15, 710–717, 2000.
4. Xu, L. and M.-Y. Chow, A classiication approach for power distribution systems fault
cause identiication, IEEE Transactions on Power Systems, 21, 53–60, 2006.
5. Cai, Y. and M.-Y. Chow, Exploratory analysis of massive data for distribution fault diagnosis in smart grids, in Proceedings of the IEEE Power and Energy Society General
Meeting, pp. 1–6, July 26–30, 2009, Calgary, AB.
6. US DOE, The 20% wind energy by 2030: Increasing wind energy’s contribution to the
U.S. electricity supply, US Department of Energy Report, p. 7, 2008.
7. Jin, T. and J. Jimenez, A review on planning and automation for distributed generation systems, in Proceedings of Conference on Automation Science and Engineering,
pp. 269–274, August 21–24, 2010, Toronto, ON.
8. Tor, O. and M. Shahidehpour, Power distribution asset management, in Proceedings
of the IEEE Power Engineering Society General Meeting, pp. 1–7, June 18–22, 2006,
Montreal, QC.
9. Garrity, T.F., Innovation and trends for future electric power systems, in Proceedings of
the Power Systems Conference (PSC ‘09), pp. 1–8, March 10–13, 2009, Clemson, SC.
10. Gao, Z., J. McCalley, and W. Meeker, A transformer health assessment ranking method:
Use of model based scoring expert system, in Proceedings of North American Power
Symposium (NAPS), pp. 1–6, October 4–6, 2009, Starkville, MS.
11. Nilsson, J. and L. Bertling, Maintenance management of wind power systems using
condition monitoring systems—Life cycle cost analysis for two case studies, IEEE
Transactions on Energy Conversion, 22(1), 223–229, 2007.
12. Tian, Z., T. Jin, B. Wu, and F. Li, Condition-based maintenance optimization for wind
power generation systems under continuous monitoring, Renewable Energy, 36, 1502–
1509, 2011.
13. Ostertag, M. and Ch. Imboden, High data rate, medium voltage powerline communications for hybrid DA/DSM, in Proceedings of the IEEE Transmission and Distribution
Conference, vol. 1, pp. 240–245, April 11–16, 1999, New Orleans, LA.
14. Heidell, J. and H. Ware, Is there a case for broadband utility communications networks?
Valuing and pricing incremental communications capacity on electric utility smart grid
networks, The Electricity Journal, 23(1), 21–33, 2010.
15. Molderink, A., V. Bakker, M.G.C. Bosman, J.L. Hurink, and G.J.M. Smit, A three-step
methodology to improve domestic energy eficiency, in Proceedings of the Innovative
Smart Grid Technologies (ISGT), pp. 1–8, January 19–21, 2010, Gaithersburg, MD.
16. Babu, C.A. and S. Ashok, Peak load management in electrolytic process industries,
IEEE Transactions on Power Systems, 23(2), 399–405, 2008.
Integrating Consumer Demand Data in Smart Grid Energy Supply Chain
273
17. Ramanathan, B. and V. Vittal, A framework for evaluation of advanced direct load control with minimum disruption, IEEE Transactions on Power Systems, 23(4), 1681–1688,
2008.
18. Moholkar, A., P. Klinkhachorn, and A. Feliachi, Effects of dynamic pricing on residential electricity bill, Power Systems Conference and Exposition, 2, 1030–1035, 2004.
19. Sui, H., H. Wang, M.-S. Lu, and W.-J. Lee, An AMI system for the deregulated electricity markets, IEEE Transactions on Industry Applications, 45(6), 2104–2108, 2009.
20. Zhong, J., C. Kang, and K. Liu, Demand side management in China, in Proceedings
of the IEEE Power and Energy Society General Meeting, pp. 1–4, July 25–29, 2010,
Minneapolis, MN.
21. Zeng, S., J. Li, and Y. Ren, Research of time-of-use electricity pricing models in
China: A survey, in Proceedings of the IEEE International Conference on Industrial
Engineering and Engineering Management, pp. 2191–2195, December 8–11, 2008,
Singapore.
22. Jin, T. and M. Mechehoul, Ordering electricity via Internet and its potentials for smart
grid systems, IEEE Transactions on Smart Grid, 1(3), 302–310, 2010.
23. Kusko, A. and J. DeDad, Stored energy—Short-term and long-term energy storage
methods, IEEE Industry Applications Magazine, 13(4), 66–72, 2007.
24. Fahrioglu, M. and F.L. Alvardo, Designing incentive compatible contracts for effective
demand managements, IEEE Transactions on Power Systems, 15(4), 1255–1260, 2000.
25. Tsoukalas, L.H. and R. Gao, From smart grids to an energy Internet: Assumptions,
architectures and requirements, in Proceedings of the Third International Conference
on Electric Utility Deregulation and Restructuring and Power Technologies, pp. 94–98,
April 6–9, 2008, Nanjing.
26. Guan, X., Y.-C. Ho, and D.L. Pepyne, Gaming and price spikes in electric power markets, IEEE Transactions on Power Systems, 16(3), 402–408, 2001.
27. Jin, T., L. Lu, and C. Kang, Energy management system for smart grid consumers using
advance demand information, in Proceedings of the IASTED European Conference on
Power and Energy Systems, pp. 300–306, June 22–24, 2011, Greece.
28. Clement-Nyns, K., E. Haesen, and J. Driesen, The impact of charging plug-in hybrid
electric vehicles on a residential distribution grid, IEEE Transactions on Power Systems,
25(1), 371–380, 2010.
29. Barghinia, S., S. Kamankesh, N. Mahdavi, A.H. Vahabie, and A.A. Gorji, A combination
method for short term load forecasting used in Iran electricity market by NeuroFuzzy,
Bayesian and inding similar days methods, in Proceedings of the 5th International
Conference on European Electricity Market, pp. 1–6, May 28–30, 2008, Lisbon.
30. Niu, L., J. Zhao, and M. Liu, Application of relevance vector regression model based on
sparse Bayesian learning to long-term electricity demand forecasting, in Proceedings of
the International Conference on Mechatronics and Automation, pp. 2363–2367, August
9–12, 2009, Changchun.
31. Couchman, P., B. Kouvaritakis, M. Cannon, and F. Prashad, Gaming strategy for electric
power with random demand, IEEE Transactions on Power Systems, 20(3), 1283–1292,
2005.
32. Jin, T. and Y. Yu, Online virtual electricity provisioning system and its preliminary
design, Automation of Electric Power Systems, 35(4), 35–40, 2011.
33. Ferrero, R.W. and S.M. Shahidehpour, Short-term power purchases considering uncertain prices, IEE Proceedings—Generation, Transmission and Distribution, 144(5),
423–428, 1997.
34. Mohsenian-Rad, A.H., V. Wong, J. Jatskevich, and R. Schober, Optimal and autonomous incentive-based energy consumption scheduling algorithm for smart grid, in
Proceedings of the Innovative Smart Grid Technologies (ISGT), pp. 1–6, January 19–21,
2010, Gaithersburg, MD.
12
Photovoltaic Energy
Generation and Control
for an Autonomous
Shunt Active Power Filter
Ayman Blorfan, Damien Flieller, Patrice
Wira, Guy Sturtzer, and Jean Mercklé
CONTENTS
12.1 Introduction .................................................................................................. 275
12.2 Consequences of Harmonic Pollution .......................................................... 277
12.3 Principle of Sliding Mode Control ...............................................................280
12.3.1 Choice of Sliding Surface ................................................................. 282
12.3.2 Determination of the Reference Current .......................................... 285
12.4 Characteristics of the System ....................................................................... 286
12.4.1 The Electrical Equations .................................................................. 286
12.4.2 Projection in the Rotation Reference ................................................ 288
12.4.3 Control Vectors ................................................................................. 288
12.4.4 Description of the Control Laws....................................................... 289
12.4.5 Sliding Mode Existence Conditions ................................................. 290
12.4.6 Choice of Vector Field ...................................................................... 294
12.5 The Photovoltaic Source ............................................................................... 294
12.5.1 Solar Cell Model Using Orcad.......................................................... 296
12.5.2 Maximum Power Point Tracking ...................................................... 297
12.5.3 Direct Current Voltage Optimization with the Photovoltaic Source .... 301
12.5.4 Control of the Switching Frequency .................................................302
12.6 Simulation Results ........................................................................................304
12.7 Conclusion .................................................................................................... 305
12.8 Nomenclature ................................................................................................308
References ..............................................................................................................309
12.1 INTRODUCTION
The past several decades have seen a rapid increase of power electronics-based loads
connected to the utility system [1]. However, the proliferation of these nonlinear
loads has raised concern with regard to the effect of the resulting harmonic distortion
275
276
Smart Grids
levels of the supply current on the power system [2]. Passive iltering is the traditional method of harmonic iltering. However, it is well known that the application
of passive ilters creates new system resonances that are dependent on the speciic
system conditions. Although this solution is simple, it gives rise to several shortcomings. Furthermore, since the harmonics to be eliminated are of low order, large ilter
components are required [3]. The active power ilter (APF) is an eficient solution to
solve the problems created by nonlinear loads; the concept is shown in Figure 12.1.
We will use it in this work.
Hybrid APFs (HAPFs) have been developed in which a passive ilter is connected
parallel to a conventional APF [4]. The HAPF coniguration is effective in improving the damping performance of high-order harmonics. Our system consists of the
following ive parts:
1. Source: alternating current (ac) line sinusoidal, balanced or not
2. Photovoltaic (PV) array: a nine-module (85 W each) PV array with
1000 W/m2 insolation and a boost direct current-direct current (dc-dc)
converter
3. Nonlinear load: a bridge diode rectiier feeding a resistive inductive load
4. Three-phase inverter
5. Resistor/inductor/capacitor (RLC) passive power ilter
Recently, there has been increasing concern about environmental pollution. The
need to generate pollution-free energy has triggered intensive efforts to ind alternative sources of energy. Solar energy, in particular, is a promising option. Some
researchers have worked on developing a combined APF and PV system [5,6].
However, existing HAPF conigurations, with all their beneits, are not yet implemented for PV applications.
iS
iL
Source
Active
power
filter
Photovoltaic array
FIGURE 12.1 Circuit diagram of a shunt APF connected to a nonlinear load.
Nonlinear load
iinf
Photovoltaic Energy Generation and Control for Active Power Filter
12.2
277
CONSEQUENCES OF HARMONIC POLLUTION
Many of the effects of harmonics on electrical equipment and installations can be
cited [5]. The greatest effects are overheating and interference with networks.
• Heating: Total losses due to the joule effect are the sum of the losses due to
the fundamentals and harmonics; it is deined by Equation 12.1:
∞
PLosses =
∑I R
2
h
(12.1)
h =1
•
•
•
•
where Ih is the harmonic current, which represents the fundamental range
for h = 1, and R is the resistance traversed by the current. Harmonics also
increase iron and copper losses. They become more important in devices
based on magnetic circuits (motors, transformers, etc.). In power electronic
equipment, these harmonics can shift the voltage zero crossing and lead
to malfunctioning of medical instruments and other electronic devices.
Furthermore, losses and malfunctions caused by the presence of harmonic
currents markedly reduce the performance of equipment such as motors,
transformers, and so on.
Aging of insulation: Insulation aging often results from tensile stress due
to the presence of harmonics and thus a local increase in leakage current,
or to excessive heating in the conductors. The most spectacular example of
this type of effect is the destruction of equipment (condenser, breaker, etc.).
Interference with telecommunication systems: The electromagnetic coupling
between electric and telecommunication networks can cause interferences.
The importance of the interference depends on the amplitude and the frequency of electric currents and the importance of electromagnetic coupling
between the networks. For example, parts of the telecommunication networks may be rendered unusable in the case of resonance. The unintentional
generation, propagation, and reception of the electromagnetic energy and the
effect of these harmonics on the equipment items or systems must be analyzed and characterized. This is referred to as electromagnetic compatibility.
Malfunctioning of electrical equipment: In the presence of harmonics,
the voltage, the current, or both can change sign several times in a half
period; therefore, sensitive equipment at the zero crossing of these electrical
parameters is perturbed.
Risk of resonance excitation: The instantaneous effects of resonance are
due to capacities or inductances that can have resonant frequencies close to
harmonic frequencies. This is the case when the batteries are connected to
the network capacity to meet the power factor; resonance frequencies can
be quite low, and thus coincide with those of the harmonics generated by
static converters. In this case, there will be an ampliication of harmonic
phenomena; surges occur and damage the cables, transformers, protection
systems, battery capacity, and so on.
278
Smart Grids
Various quantities are deined to quantify this disturbance:
• The total harmonic distortion (THD) rate deined by:
THDh =
Ih
I1
(12.2)
where Ih is the h-order harmonic component and I1 represents the
fundamental component
• The overall harmonic distortion: we characterize the pollution of a whole
network by the THD in the voltage or current as
THD ( % ) = 100
∞
∑
h=2
I h2
I12
(12.3)
Generally, the harmonics included in electrical networks are lower than
2500 Hz. They correspond to harmonic terms from rank 2 to 50 and are
considered low-frequency disturbances. The higher-frequency harmonics
are greatly attenuated by the skin effect and the presence of the inductances
of lines. In addition, the devices that generate harmonics mostly have an
emission spectrum below 2500 Hz; this is why the harmonics study ield
generally extends from 100 to 2500 Hz, that is, from rank 2 to 50 [1].
• The power factor: For a signal wave, the power factor is given by the ratio
between the active power and the apparent power S. Generators, transformers, transmission lines, and control devices inluence the measurement of
voltage and current ratings. A low value of the power factor often results
from poor use of these facilities.
In the case of harmonics, an additional power called the deformans power is
deined mostly when the power is produced by a nonsinusoidal source. The active
power in periodic nonsinusoidal signals is deined by
P=
∑ P = ∑U I cos φ
n
n
n n
n
(12.4)
n
where Un and In are the root mean square values of the voltage and current harmonic
of order n, and ϕn is the phase shift between them. The reactive power can then be
deined by
Q=
∑ Q = ∑U I sin φ
n
n
n n
n
(12.5)
n
However, these deinitions do not conform to the equation of the power triangle:
S = P 2 + Q2
(12.6)
279
Photovoltaic Energy Generation and Control for Active Power Filter
The apparent power is deined by
S 2 = U 2I 2 =
∑U ∑ I
2
n
n
2
n
(12.7)
n
Then
S =
2

+


∑U ∑ I ≥ ( ∑U I cos φ ) ∑
2
n
n
2
2
n
n n
n
n
n

U n I n sin φn 


2
(12.8)
The quantity D is added using the following equation:
D2 = S 2 − P 2 − Q2
(12.9)
We can write the apparent power as follows:
(12.10)
S 2 = P 2 + Q2 + D2
The Fresnel diagram of the powers is shown in Figure 12.2.
Distorting power consists mainly of the cross-product of the voltage and current
harmonics of different orders and will be reduced to zero if the harmonics are
reduced to zero, that is, under sinusoidal conditions:
∞
D=
∑U I
2 2
i j
(12.11)
i, j =2
i≠ j
If only the current is nonsinusoidal, then:
∞
D = U1
∑I
2
k
(12.12)
2
S
D
Q
γ
φ
φ1
FIGURE 12.2 Fresnel diagram of the power deformans.
P
280
Smart Grids
The major drawback of this deinition is that it is not certain that the power factor
is equal to unity if the reactive power (by deinition) is reduced to zero and the reactive power can be fully compensated by inserting inductive or capacitive components.
• The power factor (F.P.) becomes:
F.P. =
P
2
P + Q2 + D2
= cos ϕ ⋅ cos γ
(12.13)
It is clear that the harmonics affect the power factor.
12.3
PRINCIPLE OF SLIDING MODE CONTROL
The sliding mode design approach is one of the effective nonlinear robust control
laws. Sliding mode control has been widely applied to power converters due to its
characteristics, such as speed, robustness, and stability to large load variations. It
provides system dynamics with an invariance property to uncertainties once the system dynamics are controlled in the sliding mode. It consists of two components. The
irst involves the design of a switching function so that the sliding motion satisies
the design speciications. The second is concerned with the selection of a control
law that will make the switching function attractive to the system state. It can be
seen that this control law is not necessarily discontinuous. Let the nonlinear system
be in the following form with a scalar control:
Xɺ = f ( X , U )
(12.14)
where X is the system error and U is the control issued by the corrector with variable
structure. The second member of the equation is discontinuous due to the discontinuity imposed on the control side of the surface σ(x) = 0. Depending on the order
applied, the system will take one of the following two forms, as shown in Figure 12.3:
(
X = f − X ,U −
Sliding
surface
σ=
0
)
if
σ( x) < 0
σ>0
U = U+
σ<0
U = U−
FIGURE 12.3 System behaviors on the sliding surface.
(12.15)
281
Photovoltaic Energy Generation and Control for Active Power Filter
(
X = f + X ,U +
)
if
σ( x) > 0
(12.16)
A nonlinear system is given by the differential equation:
Xɺ = f ( X , t ) + g ( X , t ) ⋅ U
n
m
n
(12.17)
n× m
where X ∈ ℜ, U ∈ ℜ f ( X , t ) ∈ ℜ g ( X , t ) ∈ ℜ .
How should the variable structure for the command be applied? Two questions
can be asked:
1. What is this control?
2. How can it be synthesized?
The synthesis of a variable structure, represented by Figure 12.4, involves two
phases [5]:
• The irst is to deine the switching surface σ(x) = 0 of the desired dynamics.
• The second aims to synthesize the control function u(x, t), in order to
make sure that [6] any state outside the switching surface will be joined
to the states inside the sliding tranche (with bandwidth 2δ surrounding
the sliding surface; see Figure 12.5) during a inite time (limited by the
switching frequency of the insulated gate bipolar transistor). Once it
Umax
Actuator
Controlled
system
Umin
Switching controller
S (X)
FIGURE 12.4 Schematic diagram of variable structure control.
2.δ
σ>0
σ>0
0
0
σ<0
σ<0
Surface S
σ=0
FIGURE 12.5 Trajectories of sliding mode, ideal and real.
Surface S
σ=0
282
Smart Grids
reaches them, a sliding mode occurs. The system adopts the dynamic of
the surface and the point representing the evolution of the system reaches
the equilibrium point.
The dynamic slip is independent of the control, which only serves the evolution of
the intersection of surfaces and ensures their attractiveness. From any initial condition,
the representative point of the movement meets the hypersurface slip in a inite time.
12.3.1
choIce oF slIdIng surFace
Generally, the sliding surfaces are chosen as hyperplanes through the origin of
space, and their control is shown in Figure 12.4.
The general shape of these surfaces is given by the following equations:
σ ( X ) = CX = 0
σ(X ) =
(12.18)
n
∑C X
i
(12.19)
i
i =1
Although the surface may be nonlinear, the linear surfaces are more practical for
the design of a corrective variable structure sliding mode in the case of a multivariable system. We consider the following nonlinear system:
Xɺ = f ( X , t ) + g ( X , t ) ⋅ U
n
m
n
(12.20)
n ×m
where X ∈ ℜ, U ∈ ℜ f ( X , t ) ∈ ℜ ( X , t ) ∈ ℜ .
The problem is more complex due to the existence of several switching surfaces
(σ1, σ2, σ3, …, σm). The sliding mode occurs on the intersection of these surfaces.
The two switching surfaces are chosen as
σ ( X ) = CX
(12.21)
σ ( X ) = σ1 ( X ) , σ2 ( X ) , … , σm ( X )  = 0
T
C = c1, c2 , … , cn 
T
(12.22)
(12.23)
The switching surface σ(x) = 0 is a surface of dimension m, and σi(x) is the ith
switching surface. The intersection of the surface σ(X) is a sliding surface dimension
(n−m). Each entry has the form Ui:
U + with σi ( X ) > 0
U ( X, t ) =  −
i = 1, … , m
U with σi ( X ) < 0
(12.24)
Photovoltaic Energy Generation and Control for Active Power Filter
283
The switching surface is chosen such that, when the system is in sliding mode,
the behavior is meeting the desired objectives. After selection of the sliding surface,
an important phase remains to be considered. Can we ensure the existence of a sliding mode? If the tangents to the trajectories or velocity vectors near the surface
are directed toward the switching surface σ(x) = 0, then a sliding mode occurs on
this plane.
Let us consider the canonical form:
 xɺi = xi +1 i = 1, … , n − 1

n

ɺ
=
−
x
ai xi + f ( t ) + U
 n
i =1

∑
(12.25)
where U is the command with discontinuities in the vicinity of σ(x) = 0. It should be
noted that the disturbance f(t) and the parameters may be unknown.
The switching surface is of the form:
n
σ( x) =
∑C X
T
i
i
(12.26)
i =1
where the coeficients are constants. The following pair of inequalities constitutes
the conditions to the existence of the sliding mode:
lim σɺ > 0 and lim+ σɺ < 0
s →0−
s →0
(12.27)
To prove the invariability with respect to the coeficient ai and f(t), solve
Equation 12.26 to equal 0, and substitute it into Equation 12.25.
Once the system is in sliding mode, the closed-loop system becomes
 xɺi = xi +1 i = 1, … , n − 2

n −1

ɺ
=
−
x
ci xi
1
−
n

i =1

∑
(12.28)
The system no longer depends on f(t) or on the ai coeficients; on the contrary, it
depends on the parameters ci, which gives the system an invariance to disturbances
and parametric variations.
Let the system be described by Equation 12.15, and a surface σ separate the
hypersurface into two parts, G+ and G−:
dX
= F ( t , x1, …, xn )
dt
(12.29)
with X = [x1, …, xn]T; F = [f1, …, f n]T. The functions f1, …, f are piecewise continuous
and have discontinuities on the hypersurface σ of equation σ(x) = 0.
284
Smart Grids
f N+ and f N− are the f-projections on the normal orientation of σ. These projections
are made of both sides of the surface; we can show that if, at any point on the surface
σ, ∂fi/∂xj is bounded, f N+ − f N−( f N+ < 0 and f N− > 0) is negative and σσɺ < 0 .
The sliding condition can take two equivalent forms, given by Equations 12.30
through 12.32:
σσɺ < 0
(12.30)
f N+ < 0 being σ > 0 and σɺ < 0
(12.31)
f N− > 0 > 0 being σ < 0 and σɺ > 0
(12.32)
( f N+ < 0 and f N− > 0) ⇒ σσɺ < 0
(12.33)
where
In the ideal sliding mode case, the switching frequency is assumed to be ininite.
In practice, the frequency is limited by the power electronic components. As a consequence, some hysteresis effects are introduced. They increase the phenomenon of
“chattering” embodied by the oscillations around the sliding surface, but do not call
into question the theory of sliding. Figure 12.5 shows the effect of hysteresis bandwidth 2δ. The balance point can be arbitrarily chosen, but an origin allows faster
convergence to the sliding surface. Commutations move freely within the bandwidth
around the surface. An oscillating phenomenon appears around the equilibrium
point. These trajectories are shown in Figure 12.5.
Consider the monovariable state system described by Equation 12.20; the equivalent control of the ideal sliding mode (neither threshold, nor delay, nor hysteresis) is
obtained on σ(x, t) = 0 with σɺ = ( x, t ) = 0.
T
T
∂σ
 ∂σ  dx ∂σ  ∂σ 
σɺ = ( x, t ) = 
 dt + dt =  ∂x   f ( x, t ) + g ( x, t ) U eq  + dt
x
∂




 ∂σ T

,
U eq ( x t ) = − 
g ( x, t ) 

 ∂x 

−1
 ∂σ T
∂σ 
f ( x, t ) +



∂t 
 ∂x 
(12.34)
(12.35)
The condition of the sliding mode is
−1
 ∂σ T

g ( x, t )  ≠ 0


 ∂x 

(12.36)
Photovoltaic Energy Generation and Control for Active Power Filter
285
This condition is also called the transversality condition. In reality, the fact that
the product is zero means that the vector ield g is orthogonal to the surface normal
vector. So g has no vertical component and system status cannot be brought to the
surface. This condition is necessary but not suficient to ensure the sliding mode.
By replacing the expression of the equivalent control in the system of Equation
12.20, we obtain the following expression:
−1
−1

 ∂σ T
  ∂σ T 
 ∂σ T
  ∂σ T

xɺ =  I − g ( x, t ) 
 g ( x, t )   ∂x   f ( x, t ) − g ( x, t )  ∂x  g ( x, t )   ∂x 

 

 ∂x 
 

 


(12.37)
The equivalent control shown in Figure 12.6 maintains the system state on the
sliding surface σ = 0. The average value that takes control is deined as follows
{
}
{
}
min U − ( x, t ) , U + ( x, t ) ⟨ueq ⟨ max U − ( x, t ) , U + ( x, t )
(12.38)
The necessary and suficient condition for the sliding mode’s existence is given by
{
}
{
}
U min = min U − ( x, t ) , U + ( x, t )
(12.39)
U max = max U − ( x, t ) , U + ( x, t )
12.3.2
determInatIon oF the reFerence current
In an APF application, the dynamic performance of the sliding mode control will
also depend on the algorithm used for identifying the current harmonics [7]. The
method adopted to extract the reference currents and the synthesis of different regulators (control loops of voltage and current) will ensure a sinusoidal current on the
network. Different methods of identifying the harmonic (or the fundamental) current components of a nonlinear load can be used. In the following study, we will
present and analyze, respectively, the p−q method of instantaneous powers and the
direct method. Their use will depend on the targets: harmonic compensation, its
Umax
Umin
FIGURE 12.6 Representation of the equivalent control.
Ueq
Voltage
of the source
286
Smart Grids
Vsα
L.P.F
Vs1
Vs2
T32t
Vs3
p
iref1
p
Vsβ
~
p
p=?
irefα = ?
irefα
iref2
Current
of the load
T32
icα
iL1
iL2
iL3
T32t
L.P.F
q=?
icβ
q
irefβ = ?
q
irefβ
iref3
~
q
FIGURE 12.7 Overall system coniguration and control block diagram.
application to three-phase installations (under balanced or unbalanced conditions),
or single-phase installations. The p−q technique is illustrated in Figure 12.7.
In order to generate the compensation current that follows the reference current,
a ixed-band hysteresis current control method is adopted. The proposed scheme is
designed to inject the reactive and harmonic current of the nonlinear load and the
reactive current of the passive ilter. Furthermore, a current must be drawn from the
utility source to maintain the voltage across the dc-bus capacitor to a preset value that
is higher than the amplitude of the source voltage. Under normal operation, the PV
system will provide active power to the load and the utility. Under poor PV power
generation conditions, the utility source supplies the active power to the load directly.
12.4
CHARACTERISTICS OF THE SYSTEM
The power circuit of the proposed HAPF [7] consists of a PV system that is in parallel with a nonlinear load and supplied by a source voltage connected at the point
of common coupling. The proposed HAPF is a passive high-pass ilter (HPF) associated with a shunt APF constructed by a three-phase full-bridge voltage source
inverter connected to a dc-bus capacitor, and PV arrays in parallel with the dc-bus
capacitor. In the proposed scheme, the low-order harmonics are compensated using
the shunt APF, while the high-order harmonics are iltered with the passive HPF.
This coniguration is effective to improve the iltering performance of the high-order
harmonics. The inverter structure that is adopted is shown in Figure 12.8.
12.4.1
the electrIcal equatIons
The voltages supplied by the transformer on the secondary side are assumed to be
a sinusoidal three-phase balanced system with a constant frequency of 50 Hz. They
are expressed by
2⋅π 

Vsi = Vm ⋅ cos  ωt − ( i − 1)
3 

with i = 1, 2, 3 and ω = 2·π·F.
(12.40)
287
Photovoltaic Energy Generation and Control for Active Power Filter
Rf
Lf
Rf
Lf
Rf
Lf
Vc
FIGURE 12.8 Structure of the inverter.
If we denote the output voltages of the inverter, the following model provides an
initial pattern of the inverter’s behavior:
 dI ond1
 L f dt + R fI ond1 = Vs1 − Vond1

 dI ond2
+ R fI ond2 = Vs2 − Vond2
L f
dt

 dI ond3
 L f dt + R f I ond3 = Vs3 − Vond3

C
dVc
=
dt
3
∑ (u I
i ondi
)
(12.41)
(12.42)
i =1
ui is deined as the switching function on the state of the switch on or off of the
inverter. A switch is characterized by the value 1 if it is closed and 0 if it is open.
The output voltages of the inverter Vondi based on the dc voltage can be expressed
by the following relationship:
Vond1 

 Vc
Vond2  = 3
Vond3 
2

 −1
 −1
−1
2
−1
−1  u1 
 
−1 u2 
2  u3 
(12.43)
The differential system of Equations 12.20 through 12.22 is strongly coupled. So,
we apply a change of variables on all orders:
 uɶ1 
2
ɶ  1 
u2  = 3 ⋅  −1
uɶ3 
 −1
−1
2
−1
−1  u1 
 
−1 u2 
2  u3 
(12.44)
288
Smart Grids
This allows us to obtain a decoupled model of Equation 12.20; each switching
function uɶi affects only the current I ondi:
 dI ond1
ɶ1 − RfI ond1
 L f dt = V1 − uV

 dI ond2
= V 2 − uɶ2V − RfI ond2
L f
dt

 dI ond 3
 L f dt = Vs 3 − uɶ3Vc − Rf I ond 3

(12.45)
The relationship of the inverter dc side remains unchanged:
C
12.4.2
dVc
=
dt
3
∑ (uɶ I
i ondi
)
(12.46)
1
ProJectIon In the rotatIon reFerence
The projection of Equation 12.45 in the Park reference frame (dq) leads to the following equation:
dI ondd

Vsd = R f I ondd + L f dt − L f ω I ondq + Vcud

dI ondq
V = R I
+ L f ω I ondd + Vcuq
f ondq + L f
 sq
dt
(12.47)
Considering the rotating reference linked to the voltage of phase 1, we can write
Vsd = Vm

 Vsq = 0
(12.48)
The inverter model expressed in the Park reference frame is given by
Vc
 dI ondd Vm R f
 dt = L − L I ondd + ω I ondq + L ud
f
f
f

 dI ondq
Rf
Vc
I ondq − ω I ondd −
=−
uq

Lf
Lf
 dt
 dVc
2
I ondd * ud + I ondq * uq
=

d
3
t
C

(
12.4.3
(12.49)
)
control vectors
The logic states of the inverter are given in Table 12.1. There are six vectors of control. These vectors form the vertices of a hexagon, shown in Figure 12.9.
Photovoltaic Energy Generation and Control for Active Power Filter
289
TABLE 12.1
Inverter States
(uɶ1, uɶ2, uɶ3)
(u1, u2, u3)
(uα, uβ)
Vector
A
(0,0,0)
(0,0,0)
(0,0)
(0,0,1)
(−1/3,−1/3,2/3)
(−1/3,−1/ 3)
B
(0,1,0)
(−1/3,2/3,−1/3)
(−1/3,1/ 3)
C
(0,1,1)
(1,0,0)
(−2/3,1/3,1/3)
(−2/3,−1/3,−1/3)
(−2/3,0)
(2/3,0)
D
E
(1,0,1)
(1/3,−2/3,1/3)
(1/3,−1/ 3)
F
(1,1,0)
(1/3,1/3,−2/3)
(1/3,1/ 3)
G
(1,1,1)
(0,0,0)
(0,0)
H
β
1
C: (0,1,0)
D: (0,1,1)
A: (0,0,0)
G: (1,1,0)
3
H: (1,1,1)
1
E: (1,0,0)
α
3
B: (0,0,1)
F: (1,0,1)
FIGURE 12.9 Representation of different control vectors.
12.4.4
descrIPtIon oF the control laWs
In order to insert the control in the (dq) reference frame, we set the following functions:
*
σd = Iondd − Iond
d

*
σ
=
I
−
I
q
ond
ond
q
q

(12.50)
*
*
The components I ond
and I ond
represent the reference currents that will serve as
q
d
inputs for the inverter. The two corresponding switching surfaces are thus deined by
*
Sd : σd = I ondd − I ond
d

*
S
:
σ
=
I
−
I
ondq
ondq
 d q
(12.51)
290
Smart Grids
They can be calculated from Equation 12.34. In our case, the derivatives of the
reference currents are not zero and the model in the Park reference frame gives
 dI fd VSd R f
Vc
 dt = L − L I fd + ωI fq − L ud
f
f
f

 dI fq VSq R f
V
I fq − ωI fd − c ud
=
−

d
t
L
L
L
f
f
f

 Vc
3
I fd ud + I fq ud
=

 Lf 2C
(
(12.52)
)
The equivalent command is then given by
udeq =

L f  dI fd VSd Rf
I fd + ωI fq 
+
−
−
Vc  dt L f L f


L  dI f VS R
uqeq = f  − q + q − f I fq − ωI fd 
Vc  dt L f L f

12.4.5
(12.53)
slIdIng mode exIstence condItIons
The convergence of the sliding surface refers to the following conditions:

σd

 σq

σd
⟨0
dt
σq
⟨0
dt
(12.54)
Now, according to relations 12.51 through 12.53, we have
Vc

σd = L ( udeq − ud )
s

 σq = Vc ( uqeq − uq )

Ls
(12.55)
The convergence condition is equivalent to the following equations:
sgn ( σd ) = sgn ( ud − udeq )

 sgn ( σq ) = sgn ( uq − uqeq )
(12.56)
The expression of controls ud and uq is given by the Park transformation:
ud = cos ( θ ) ⋅ uα + sin ( θ ) ⋅ uβ

uq = − sin ( θ ) ⋅ uα + cos ( θ ) ⋅ uβ
(12.57)
Photovoltaic Energy Generation and Control for Active Power Filter
291
Given that the equivalent commands must be bounded, leaving the minimum and
the maximum, one can deduce them from Equation 12.58 in order to regulate the dc
voltage of the inverter:
dVc2 ( t )
1
Pf = − ⋅ C f ⋅
2
dt
(12.58)
The orders of uα and uβ are conined to ± 23 ; udq = ud2 + uq2 must be less than or
equal to the standard uαβ = uα2 + uβ2 .
The vector ueq must therefore be inside the circle inscribed in the hexagon in
Figure 12.10. The suficient condition for the existence of sliding regimes therefore requires that the two commands are equivalent bounded by the two following
equations:
 ud ⟨ 23

2
 uq ⟨ 3
(12.59)
This condition depends on the values chosen for the dc-bus voltage Vc and the
coupling inductance of the inverter network (Lf).
The worst case occurs between two phases of the rectiier conduction, because
it is during this phase that the gradient of the load current is highest. Consider, for
example, the switching phase between phases 2 and 3, which occurs during the time
interval [0, time].
q
β
q′
C: (0,1,0)
G: (1,1,0)
d′
d
θ
D: (0,1,1)
E: (1,0,0)
α
B: (0,0,1)
F: (1,0,1)
FIGURE 12.10 Example of choice of control vectors.
292
Smart Grids
Assuming that the currents change linearly during the encroachment, the load
currents are deined by the following system:

 Ich = I
 1 d

Id
t
 Ich2 = − I d +
time


Id
t
 Ich3 = −
time

(12.60)
According to the Concordia transform, this leads to
 Ichα = I d

2 * Id
1 

 Ichβ = 3  − I d + t
ime



t

(12.61)
If the current Id is assumed to be perfectly smooth, then the derivatives of the
currents Ichα and Ichβ are
 dIchα
 dt = 0


 dIchβ =  2 * I d 
 dt
 3time 
(12.62)
They can be written in the Park reference frame as
2 * Id
 dIchd
 dt = ω Ichq + 3t sin ( ωt )

ime

d
Ich
2
Id
*
q

= −ω Ichd +
cos ( ωt )
 dt
3time
(12.63)
From Equation 12.63, we deduce
 dIfd*
= ωIfq* −

d
t

 *
 dIfq = −ωIf * −
d
 dt

2Id
sin(ωt )
3time
2Id
sin(ωt )
3time
(12.64)
Photovoltaic Energy Generation and Control for Active Power Filter
293
Assuming that Ifd* = Ifd and Ifq* = Ifq, and Rf can be neglected, the equivalent
commands are expressed as Equation 12.65:
Vm
2 * Lf * I d

sin(ωt )
udeq 0 = V + Vm
c
+ Vc 3time

Vc



Lf  2 * I d
cos(ωt ) 
uqeq 0 =

V
3
t
c 

ime

(12.65)
The phase duration is [0, time] if t = time and τ = time/ω:
Vm 2 * Lf * ω * I d sin(ωtime )

udeq 0 = V +
ω * time
Vc 3
c


Lf  2 * ω * I d cos(ωt ime ) 
u
=

 qeq 0 Vc 
ωtime 
3
(12.66)
We consider the following simpliication with
sin ( ω t ime )
≈1
ω * t ime
Vm 2 * Lf * ω * Id

 udeq0 = V +
V0 3
c


L
 2 * ω * I d cos(τ) 
u
= f

 qeq0 Vc 
τ 
3
(12.67)
(12.68)
The conditions of Equation 12.59 give

2 * L f * ω* Id 
3
 Vc ⟩  Vm +

2
3





V ⟩ 3 L *  2 * ω * I d cos(τ)  

 c 2  f 
τ 
3


(12.69)
To ensure sliding mode, the value of Vc must take the maximum of these two
values.
294
12.4.6
Smart Grids
choIce oF vector FIeld
The appropriate vector to ensure the convergence to the desired sliding surfaces must
be chosen. We recall that the convergence condition is subject to the direct:
sgn ( σd ) = sgn ( ud − udeq )

 sgn ( σq ) = sgn ( uq − uqeq )
(12.70)
To represent the magnitude (ud − udeq) and (uq − uqeq), we work in another rotating
reference frame (dʹ, qʹ) that is deduced from (d, q) by the translation vector ueq. The
axes of the rotating reference frame are oriented according to the sign of (ud − udeq)
and (uq − uqeq).
Thus, in the rotating reference frame (dʹ, qʹ), the sign of (u d − u deq) and (uq − uqeq)
is directly determined by the quadrant or located (see Figure 12.10). For example,
if the surfaces σd and σq are both positive and if the position θ is given arbitrarily,
then the convergence condition (Equation 12.59) requires that the terms (u d − u deq)
and (uq − uqeq) must be positive. This is only possible for a vector located in the
irst quadrant of the reference frame (dʹ, qʹ), and only the vector G can guarantee
this.
We set the two following equations:
λ = arc sin ( 23 uqeq ) and µ = arc cos ( 23 udeq )
(12.71)
The different values of θ are determined by the passage of the axis dʹ or qʹ in one
of the six vertices of the hexagon, which provides a boundary of θ. The choice of
switching controls is directly determined by the choice of vector B, C, D, E, F, or G,
which depends on three parameters: the sign of σd and σq, the vector ueq (and there*
*
fore the instructions Iondd and I ondq), and position θ. Finally, Table 12.2 provides the
values of the switching vector.
Table 12.2 allows the switching sector to be selected in the ideal case, that is
to say, when π 3 − µ + λ = 0. However, this condition is valid only for a particular
operating point. In the general case, two situations may arise, depending on the
values of λ and μ. The operating point to be determined in the working space
depends at the same time on the values of λ and μ. Actually, two lookup tables are
used, depending on the sign of π 3 − µ + λ = 0. The regions of θ, mod 2π, are given
in Table 12.3.
12.5
THE PHOTOVOLTAIC SOURCE
Recently, there has been increasing concern about environmental pollution. The
need to generate pollution-free energy has triggered intensive efforts in search
of alternative sources of energy. In particular, solar energy is a promising option.
295
Photovoltaic Energy Generation and Control for Active Power Filter
TABLE 12.2
Table of Switching Vectors
Region θ
Module 2π
σd > 0 and σq > 0
σd > 0 and σq < 0
σd < 0 and σq > 0
σd < 0 and σq < 0
θ
2π


2π − λ, 3 − µ 


G
E
D
B
θ1
π

 3 − λ, π − µ 


C
G
B
F
θ2
4π
 2π

 3 − λ, 3 − µ 


D
C
F
E
θ3
5π


 π − λ, 3 − µ 


B
D
E
G
θ3
 4π

 3 − λ, 2π − µ 


F
B
G
C
θ4
7π
 5π

 3 − λ, 3 − µ 


E
F
C
D
θ5
Some researchers have worked on developing a system combining an APF and a PV
panel [8]. However, the existing HAPF coniguration has not yet been implemented
in an experimental PV application. The power output of a solar PV module changes
with the solar elevation angle, changes in the solar insolation level, and varying temperature [9]. Hence, the maximization of power improves the utilization of the solar
PV module.
The simplest solar cell model consists of a diode and a current source connected
in parallel. Source current is directly proportional to the solar radiation. The diode
represents the PN junction of a solar cell.
iPV = I sc − id
(12.72)
i = I 0 (eeV / KT − 1) − I L ⇒
i = I 0 (eVD /VT − 1) − I d
(12.73)
The proposed model of a PV cell using MATLAB® is shown in Figure 12.11.
The PV system consists of:
1. A nine-module (85 W each) PV array with full sun (1000 W/m2 insolation)
2. A boost dc-dc converter that steps up a dc input voltage
3. A capacitor C under dc voltage that provides the necessary energy storage
to balance instantaneous power delivered to the grid
296
Smart Grids
TABLE 12.3
Improved Vectors of Choice for Switching
π −µ + λ < 0(disunion)
3
12.5.1
π −µ + λ > 0(overlap)
3


7π

 2π
− µ ,
− µ
min  −λ,
3
3






7π

 2π
− µ ,
− µ
max  −λ,
3
3






2π
π

− µ  , π − µ
min  − λ,
3
3





2π
π

− µ  , π − µ
max  − λ,
3
3





 2π
 4π
− λ, π − µ  ,
− µ
min 
3
3






 2π
 4π
− λ, π − µ  ,
− µ
max 
3
3






4π

 5π
− µ ,
− µ
min  π − λ,
3

 3




4π

 5π
− µ ,
− µ
max  π − λ,
3

 3




5π
 4π

− λ,
− µ  , 2π − µ 
min 
3
3






5π
 4π

− λ,
− µ  , 2π − µ 
max 
3
3






 5π
 7π
− λ, 2π − µ  ,
− µ
min 
 3
 3




 5π
 7π
− λ, 2π − µ  ,
− µ
max 
 3
 3


solar cell model usIng orcad
The characteristics of the photodiode are provided by the producer of the PV array.
Figure 12.12 shows the characteristics of the photodiode that will be used.
Id = f(Vd)
Looking at Figure 12.12, we can conclude that a geometric approximation represented by four segments will be suficient for our needs, so the new model will be as
shown in Figure 12.11.
The values of the resistance are calculated using Equation 12.72 [10].
Ri =
1
Ei
mi − mi −1
m0 = 0
(12.74)
Vi = Ei
The new approach is modulated using Orcad 15.7, which is considered a new
version of PSpice with a high ability to produce an electronic board from the simulated circuit. Figure 12.11 shows this new model with four parallel branches.
The characteristic of the PV is represented by the four segments, the irst graph
is the current (Ipv) as a function of the voltage of the PV (Vpv), and the second is the
power (P) as a function of the same variable (Vpv).
297
Photovoltaic Energy Generation and Control for Active Power Filter
m4
rox
im
atio
nb
y se
Id (A)
gm
en t
s
Diode I-V curve
m3
Ap
p
m2
m1
E1
E2
E3
E4
Vd (V)
FIGURE 12.11 Real model of the PV.
PV power
Vpv
X
ppv
Product
1
Insolation
1–9 ∗(e(u/26∗1 )–1)
PN-junction characteristic
Isc
–
1
+
Ipv
To workspace
I-V characteristic
–3
Vpv
PV
Id
Ipv
Insolation to
Isc current gain
FIGURE 12.12 Model of PV cell using Simulink.
12.5.2
maxImum PoWer PoInt trackIng
The power output of a solar PV module changes with the changing direction of the sun,
with the solar insolation level, and with varying temperature. Hence, in order to get the
maximum power from the solar PV, an interface should be added. A maximum power
point tracker (MPPT) is used for extracting the maximum power from the solar PV module and transferring that power to the load. A dc/dc converter (step up/step down) serves
the purpose of transferring the maximum power from the solar PV module to the load.
298
Smart Grids
We choose the hill climbing method, which consists of observing the current and
voltage at the output of the PV generator. Multiplying these two signals allows the power
to be obtained. The value of the power is stored in two integrators in order to be used
as two successive values in memory. Two successive values of the power can thus be
compared to determine the power’s derivative. Then, using a JK toggle with a third integrator and a triangular signal allows the pulse width modulation control to be created
and sent to the driver of the transistor (we use a P-channel metal–oxide semiconductor
ield-effect transistor [MOSFET]). Figure 12.13 shows a diagram of this technique.
The MPPT controller represents a good compromise between high performance
and robustness. However, to achieve convergence in good conditions, whatever the
power range and the algorithm, it is necessary that the power curves issued by the
generator are constant or slowly varying. If this assumption is not respected (i.e., if
there are some abrupt changes in operating conditions), then the system can diverge.
The principle of the MPPT controller is often based on the “elbow” of the PV
characteristics. This is more or less trial and error, as shown by Figure 12.14. Indeed,
the algorithm starts from a point located on the curve, let us say X1. After a while, the
actual value on the curve (X2) is compared with the previous one (X1). If X2 > X1,
then we move to X2, and this is repeated until the next term (Xn) is lower than the
previous (Xn−1). At this time, we take an interval value from every lower point, and
repeat from (Xn−1) to obtain the maximum power point (MPP) (X).
The MPPT controller consists of two distinct parts:
1. The control part, whose purpose is to determine the operating point, where
the solar cell can transfer the power toward the batteries
2. The power part, which transfers energy between the solar panels and batteries
These two parts are also shown in Figure 12.15.
Now it is easy to transfer to the electronic schematic from the Orcad library, as
shown in Figure 12.16.
Idc
{Idc}
R1
5.88
V1
9
R2
12.5
V2
13.5
R3
0.86
V3
15
R5 0.02
D1N4001
D4
D1N4001
D3
D1N4001
D2
Parameters:
Rs = 10
D1N4001
D1
Parameters:
Idc = 3.1
R4
0.38
Rp
50
V4
18
0
FIGURE 12.13 Block diagram of the proposed MPPT controller.
Rs
{Rs}
299
Photovoltaic Energy Generation and Control for Active Power Filter
Model
PV
Converter
dc-dc
R1
Rs
Rsh
Driver
MOSFET
R2
Quick
integrator
T1
K
+
JK
Compare
Multiplier
Quick
integrator
T2
Inverter
amplifier
–
Integrator
(R0,C0)
Compare
Clear
Triangular generator
FIGURE 12.14 Principle of operation of MPPT.
X (MPP)
Xn
P (PV)
Xn–1
X2
X1
V (PV)
FIGURE 12.15 Block diagram of this type of control.
Some techniques for implementing an MPPT depend on the changing bias point
and measuring changes in the output, while others use predetermined PV models.
Table 12.4 shows all the MPPT methods that can be employed.
The perturbation and observation (P&O) method is one of the most popular
MPPT methods because of its simplicity. The P&O method operates by making
small incremental changes in voltage and measuring the resulting power change. By
comparing the current power measurement with the previous power measurement,
the P&O method selects the direction for the next perturbation. The P&O MPPT
method can be implemented using a minimal number of components; however, its
300
Smart Grids
P
Solar array
P0
Power part
Battery
PWM
P
Control part
V0
FIGURE 12.16 Proposed controller circuit diagram.
TABLE 12.4
Table of Various MPPT Methods
MPPT Method
1
2
3
4
5
6
7
8
9
10
P&O
Hill climbing
Incremental conductance
Power feedback control
Fractional short-circuit current
Fractional open-circuit voltage
Computational and lookup table
Current sweep
Fuzzy logic
Array reconiguration
speed is limited by the size and the period of the perturbation [11]. Figure 12.17
shows the lowchart of this algorithm.
In a situation where the irradiance changes rapidly, the MPPT also moves along
the right-hand side of the curve (see Figure 12.14). The algorithm takes this as a
change due to perturbation, and in the next iteration it changes the direction of perturbation, hence moving away from the MPP, as shown in Figure 12.18.
The proposed MPPT scheme employing the P&O algorithm has been modeled
and simulated using the MATLAB/Simulink® software. Figure 12.19 shows a block
diagram of the developed PV system, consisting of a PV array and a boost converter
circuit connected to a load and an MPPT controller. The PV array, with a capacity
of 1 kW, consists of six PV modules associated with an MPPT controller. All this is
simulated with conventional P&O under operating conditions of temperature 25°C
and with varying insolation (0–1000 W/m2). The MPPT controller is based on the
P&O method.
301
Photovoltaic Energy Generation and Control for Active Power Filter
Parameters:
Parameters:
Idc = 25
D17
MUR150
Rs = 50
R1
5.88
V1
9Vdc
D4
R2
12.5
R3
0.86
R4
0.38
V2
13.5Vdc
V3
15Vdc
V4
18Vdc
D13
D1N4001
1
IRFP9140 M1
L2
2
1mH
20m
D1N4001
D3
D1N4001
D2
D1N4001
D1
Idc
{Idc}
D1N4001
R5
D15
Rp
50
Rs
{Rs}
C6
3.5u
MUR150
12 V11
100
C5
R19
0.1
R27
0
100
V
4
3
6
R11
-VCC
AD633/AD
VCC
220k 220n
0
C3
R33
VCC
6
V+
7
220n C2
1k
+
U12A
1
LM139 OUT
1
2
– 12
12
8
0
0
R10
22k
8.2k
V-VCC
3
U15
J
Q
CLK
K
Q
JKFF
5
R12
V
1k
7
C4
R22
1k
5
R21
1k
15
1
R34
1k
-VCC
0
–
7
LM324
OUT
Jcc
5
15
OUT
+ 3
U13A V+
V-
6
LM239
VCC
3.3u
0
R20
1k
15
–
4
11
0
+ 4
U6C V+
VCC
5
10
–
R16
LM324 OUT 8 5k
-VCC
R39
VCC
VCC
X1
X2
7
Y1 W
Y2
Z
7
+ 8
U19B V+VCC
0
10k
V- U16 V+
11
9
1k
1
2
3
4
6
V-
R29
OUT
5
V-
VCC
R17
22k
4
2.2k
+ 8
U18A V+ VCC
R38
0
-VCC
V-
1
3
R28
6n
TL082
6
–
R37
OUT
0
100k
-VCC
C8
-VCC
V-
TL082
2
–
+ 4
U7B
0
V+ VCC
0
0
-VCC
FIGURE 12.17 Flowchart of the P&O algorithm.
The whole PV system associated with an APF, as shown by Figure 12.19, has
been simulated. The performance of the PV system using the P&O algorithm under
a constant temperature of 25°C and a changing irradiance is shown by Figure 12.20.
The irradiance values are 0, 400, 850, 950, 1000, 950, 850, 400, and 0 W/m2 at times
0, 0.01, 0.02, 0.03, 0.04, 0.05, 0.06, 0.07, and 0.08, respectively. It can be seen that the
power issued by the converter is close to the PV power (Figure 12.21).
12.5.3
dIrect current voltage oPtImIzatIon
WIth the PhotovoltaIc source
The shunt APF and the PV array are connected back-to-back with a dc-bus capacitor.
The dc-bus voltage controller maintains the average voltage across the dc-bus capacitor constant against variations in distribution source. As a result, the reference current
injected into each phase allows the grid to gain more useful active power. The voltage across the capacitor must be constant and ixed at a predetermined value. Active
power losses in the active ilter (switches and output ilter) are the main factors that
may change the voltage. The average voltage across the capacitor energy storage must
be regulated by adding the current fundamental asset in the reference currents.
302
Smart Grids
Begin P&O
algorithm
Measure
V(k), I(K)
P(k) = V(k)×I(k)
∆P = P(k)–P(k–1)
No
Yes
∆P>0
V(k)–V(k–1)>0
V(k)–V(k–1)>0
No
Decrease
module
voltage
Yes
Yes
No
Increase
module
voltage
Decrease
module
voltage
Increase
module
voltage
Update history
V(k–1) = V(k)
P(k–1) = P(k)
FIGURE 12.18 Curve showing incorrect tracking of MPP by P&O algorithm under rapidly
varying irradiance.
This stabilizes the dc voltage of the capacitor C of the three-phase voltage power
inverter close to a continuous value. This has been veriied by simulations, and the
result can be seen in Figure 12.22.
12.5.4
control oF the sWItchIng Frequency
The switching frequency strongly depends on the coupling inductance Lf and on the
width of the hysteresis band. The objective of this part is to propose a method to
limit the frequency switching and make it independent of the parametric variations.
For this, we propose to add to the expression of the surface a triangular signal with
constant frequency and ixed amplitude. The principle of the method is illustrated
in Figure 12.23.
Photovoltaic Energy Generation and Control for Active Power Filter
160
1000 W/m2
D
140
Module output power (W)
303
120
750 W/m2
C
∗
G
100
∗
F
80
500 W/m2
B
∗
E
60
250 W/m2
40
A
20
0
0
5
10
15
20 25 30 35
Module voltage (V)
40
45
50
FIGURE 12.19 Curve showing wrong tracking of MPP by P&O algorithm under rapidly
varying irradiance.
Photovoltaic array
Vpv
+
Boost converter
Input
Output
Vx
Vin
Vo
Active power filter
Vout
Cpv
Ipv
Idc
Iref
Perturb and
observe
FIGURE 12.20 Functional block of the whole system using the P&O technique.
500
400
300
200
100
0
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
FIGURE 12.21 Power of the converter Pout-Boost, photovoltaic Ppv, and the ideal output Pideal.
304
Smart Grids
0.1
0.05
0
FIGURE 12.22 Evolution of the dc-bus voltage around its reference value at 300 V.
σ+
u
+
m
–m
1
Fdec
2δ
FIGURE 12.23 Principle of limiting the switching frequency.
12.6
SIMULATION RESULTS
The proposed HAPF was simulated using the MATLAB/Simulink software. The
system parameters that were used are summarized in Table 12.5. In the simulation,
a diode rectiier with a dc-link capacitor and a smoothing inductor was used as a
nonlinear load producing harmonics.
The complete scheme of the control structure that has been developed in this study
to govern the APF supplied by a PV array and a capacitor is shown in Figure 12.23.
Furthermore, the high-order harmonics are iltered from the source voltage and the
source current.
The simulation results show that the source currents after compensation are
almost sinusoidal and in phase with the voltage at the connection point. The harmonic spectra of the source current before and after compensation are presented in
Figures 12.24 and 12.25, respectively. The THD of the current source (calculated
on the irst 30 harmonic orders) has decreased from 30.09% to 1.95%, and the load
current harmonic components are greatly reduced. The dc-bus voltage is regulated
to 300 V.
All results are shown in Figure 12.26 (the source, the current of load before
and after the compensation, the current of the ilter, and the dc voltage of the
capacitor).
Figure 12.27 shows the spectrum of the source current Is in decibels. This spectrum allows us to see that the switching frequency varies between 5 and 10 kHz (see
Figure 12.25). The average switching frequency is equal to 9.07 kHz. Low-frequency
peaks are located between −40 and −60 dB. Above 4 kHz, the amplitudes of the
Photovoltaic Energy Generation and Control for Active Power Filter
305
TABLE 12.5
Matlab/Simulink Simulation Parameters
Utility voltage
220 2 (50 Hz)
Source inductance
0.65e-3 H
Source resistance
5−3 Ω
Rectiier smoothing inductor
1−3 H
Rectiier smoothing resistance
1.5 Ω
Load inductance
100−3 H
Load resistance
29*1.5 Ω
Load capacitance
2000−6 F
Current control band
1.8 A
Sampling time
10−6 s
APF inductor
2−3 H
APF resistor
0.005 Ω
APF dc-bus capacitor
450 V
HPF capacitor
5.01e-6 F
HPF resistor
20.46 Ω
PV
9 arrays
Insolation
1000 W/m2
Short-circuit current Isc
5.45 A
Open-circuit voltage Voc
22.2 V
Rated current IR at MPP
4.95 A
Rated voltage VR at MPP
17.2 V
Capacitance of the PV output
2−3 F
Resistance of PV output
72.78 Ω
lines are increasing, and once the frequency 7.5 kHz is reached, they are decreasing. Beyond 16 kHz, the spectrum is more regular, and rays have relative amplitudes
below −50 dB.
12.7 CONCLUSION
This chapter has detailed a new technique for improving the eficiency of the sliding
mode controller that is used to manage the PV energy in an autonomous shunt APF.
In order to improve the reference current for compensating harmonics, a method
306
Vs123 (V)
Smart Grids
200
0
–200
0.2
0.205
0.21
0.215
0.22
0.225
0.23
0.235
0.24
0.205
0.21
0.215
0.22
0.225
0.23
0.235
0.24
0.205
0.21
0.215
0.22
0.225
0.23
0.235
0.24
–20
0.2
0.205
0.21
0.215
0.22
0.225
0.23
0.235
0.24
702
700
698
0.2
0.205
0.21
0.215
0.22
Time (s)
0.225
0.23
0.235
0.24
iL (A)
20
0
vc (V)
if, irefi (A)
is (A)
–20
0.2
20
0
–20
0.2
20
0
FIGURE 12.24 Compensation of harmonic currents with the sliding mode control.
30
Harmonic spectrum
25
20
15
10
5
0
0
500
1,000
Frequency f (Hz)
1,500
FIGURE 12.25 Harmonic spectrum of source current before compensation.
to limit the switching frequency has been added and veriied. The THD has been
improved and the average switching frequency has only slightly increased. To reduce
the switching frequency without changing the system settings, a technique based on
the addition of a triangular signal to the switching surface has been proposed and
tested (see Figure 12.23).
307
Photovoltaic Energy Generation and Control for Active Power Filter
Reference
value
Vc
Regulator
of voltage
Pl
Low pass filter
u1,2,3
Inverter
λ
µ = arccos
3
2
udeq
Vαβ
Equivalent
control
λ = arcsin
3
2
Source of voltage
Vs123
µ
Park
uqeq
θ
PLL
Hysteresis
bloc
Ifa,1,2,3
Ifad,q
Park
Gissant surface
Ĩchd,q
Harmonic
references
calculation
Nonlinear
load
Ich,1,2,3
Vs123
Source
FIGURE 12.26 Complete diagram of the control structure.
A P&O MPPT has been implemented to maximize the PV output power.
Through a boost dc-dc converter combined with the fastness and robustness
of sliding mode control, the HAPF has good robustness and stability. A loop
for capacitor voltage regulation is introduced into the sliding mode control; a
good regulation effect is obtained for load variation in the dc bus. The simulation results confirm the effectiveness of the proposed scheme for wideband
harmonics compensation and PV power handling capability (see Figures 12.27
and 12.28).
308
Relative amplitude in decibels
Smart Grids
10
0
–10
–20
–30
–40
–50
–60
–70
–80
–90
–100
0
0.5
1
1.5
2
Frequency f (Hz)
2.5
3
× 104
FIGURE 12.27 Spectrum of the source current Is in decibels.
30
Harmonic spectrum
25
20
15
10
5
0
0
500
1,000
Frequency f (Hz)
1,500
FIGURE 12.28 Harmonic spectrum of source current after the compensation.
12.8
HAPF
APF
PV
σ(x)
X
u
U
2δ
f, g
C
NOMENCLATURE
Hybrid active power ilter
Active power ilter
Photovoltaic
Surface of sliding
System error
Control variable
Control vector
Bandwidth around the sliding surface
Functions of the nonlinear system
Coeficients of σ(x)
Photovoltaic Energy Generation and Control for Active Power Filter
309
REFERENCES
1. H. Akagi, A. Nabae and A. Satoshi, Control strategy of active power ilters using multiple voltage-source PWM converters, IEEE Transactions on Industry Applications,
IA-22, 460–465, 1986.
2. D. Ould Abdeslam, P. Wira, J. Merckle, D. Flieller and Y.-A. Chapuis, A uniied artiicial
neural network architecture for active power ilters, IEEE Transactions on Industrial
Electronics, 54, 61–76, 2007.
3. S. Rahmani, A. Hamadi, N. Mendalek and K. Al-Haddad, A new control technique for
three-phase shunt hybrid power ilter, IEEE Transactions on Industrial Electronics, 56,
2904–2915, 2009.
4. A. Hamadi, S. Rahmani and K. Al-Haddad, Sliding mode control of three-phase shunt
hybrid power ilter for current harmonics compensation, in Proceedings of the IEEE
International Symposium on Industrial Electronics (ISIE), pp. 1076–1082, 4–7 July,
Bari, 2010.
5. A. Blorfan, J. Mercklé, D. Flieller, P. Wira and G. Sturtzer, Sliding mode controller for
three-phase hybrid active power ilter with photovoltaic application, Smart Grid and
Renewable Energy, 3, 17–26, 2012.
6. B. Cheng, P. Wang and Z. Zhang, Sliding mode control for a shunt active power ilter,
in Proceedings of the Third International Conference on Measuring Technology and
Mechatronics Automation (ICMTMA), pp. 282–285, 6–7 January, Shanghai, 2011.
7. D. Wenjin and Yu. Wang, Active power ilter of three-phase based on neural network,
in Proceedings of the Fourth International Conference on Natural Computation, ICNC
‘08, pp. 124–128, 18–20 October, Jinan, 2008.
8. A. Blorfan, D. Flieller, P. Wira, G. Sturtzer and J. Merckle, A new approach for
modeling the photovoltaic cell using Orcad comparing with the model done in Matlab,
in International Review on Modelling and Simulations (IREMOS), vol. 3, pp. 948–954,
October 2010.
9. G. Sturtzer, Projet Simulation Panneaux Photovoltaïques, p. 8, 2009.
10. S. Chowdhury, S.-P. Chowdhury, G.A. Taylor and Y.H. Song, Mathematical modelling and performance evaluation of a stand-alone polycrystalline PV plant with
MPPT facility, in Proceedings of the 2008 IEEE Power and Energy Society General
Meeting—Conversion and Delivery of Electrical Energy in the 21st Century, pp. 1–7,
20–24 July, Pittsburgh, PA, 2008.
11. M. Veerachary, T. Senjyu and K. Uezato, Voltage-based maximum power point tracking
control of PV system, IEEE Transactions on Aerospace and Electronic Systems, 38,
262–270, 2002.
13
Self-Tuning and SelfDiagnosing Simulation
Jin Ma
CONTENTS
13.1 Introduction .................................................................................................. 311
13.2 Framework of Self-Tuning and Self-Diagnosing Simulation ........................ 312
13.3 Integrate the Synchronous Measurements with Digital Simulation ............. 315
13.4 Model Data Calibration ................................................................................ 320
13.5 Locate the Disturbance Source for Low-Frequency Oscillation .................. 325
13.6 Conclusion .................................................................................................... 331
Acknowledgments.................................................................................................. 331
References .............................................................................................................. 331
13.1
INTRODUCTION
The bulk power system is usually divided into the transmission network and the
distribution network. The transmission network is responsible for transferring the
bulk electrical energy from the generator side to the demand center, while the distribution network connects the end users. There are two ways to design the generation
layouts. One is to build a large power plant in an area with abundant raw energy, for
example, near the coal base or the hydrobase. Consequently, the electrical energy
produced by this kind of plant is transferred long distance on the high and the ultrahigh voltage levels, usually above 110 kV. The other way to plan the generation is
to disperse small-scale generation facilities in the demand center, which not only
minimizes the risk of blackout for the users, but also serves as a backup for the
transmission network.
One of the main targets of the smart grid is to provide clean and secure electrical energy with high quality. But how? The transmission network will be strengthened; advanced electronic technologies will be applied to better integrate the
distributed generation into the demand side; more lexible and interactive methods
of energy utilization will be attempted. But, when we do this, we have to understand how this system is operated and what will happen under our smart controls.
The power grid, as built so far by human beings, is too complex to be easily understood. So, one may ask why this giant machine has been operated for such a long
time although we do not understand it very well. Actually, it is not operated very
well. We manage it very cautiously and operate it very conservatively because we
do not know the capability limits of this machine. Yet we have witnessed several
311
312
Smart Grids
major blackouts in the last decade. We have to say NO to this traditional operation
mode with the advent of the smart grid, since it requires high eficiency in energy
utilization without sacriicing security. So, a compromise between security and
economy has to be made. At least, we have to know how much potential transfer
capability this network has before we take full advantage of it. But, again, we have
to ask ourselves how.
Performing tests on this system is certainly one solution. For things that we do
not know, we have to employ tests. However, there are three major challenges to
performing tests in the real power system. First, system security is always top priority. Therefore, a major category of tests, especially those that may cause serious
damage, such as short-circuit tests, cannot easily be applied to the main power system. Second, the operation scenarios are always changing. The load is different at
any time. The generation units and the controllers in operation vary with time. One
cannot design a test case covering all the scenarios. The operation case is always
different from the test case. Third, the possible contingencies are numerous, and
it is impractical to perform all these contingency tests in the grid. Consequently,
digital simulation becomes a key tool for the operators in understanding the power
system operation mechanisms and designing control strategies. Reliable simulation
is crucial to system security, which is especially important to the modern, largescale power system. It is the prior condition for inding the balance point between
economy and security in the coming smart grid era.
The traditional simulation is one-way. The smart grid is equipped with an
advanced global measurement system in the whole grid consisting of accurate and
smart meters. Based on this, a two-way simulation is possible, that is, self-tuning
and self-diagnosing simulation. It would be possible to combine advanced measurement technology with digital simulation, greatly enhancing the simulation reliability; thus, the security of the whole power system could be ensured. This chapter will
present the framework of the self-tuning and self-diagnosing simulation and propose a speciic approach to combining wide-area measurements (WAMs) with the
digital models of the power system component. This is called the variable impedance method. This chapter also introduces two cases of applying this self-tuning
and self-diagnosing simulation in the real power grid. One concerns simulation
validation and model data correction, while the other concerns locating the disturbance source to minimize the possibility of resonant oscillation in a large power
grid. The chapter is organized as follows: the general framework of the self-tuning
simulation is given in Section 13.2; the speciic methods to incorporate the real
measurements into the simulation are discussed in Section 13.3; two application
ields, as stated before, are discussed in Sections 13.4 and 13.5; and Section 13.6
concludes the chapter.
13.2
FRAMEWORK OF SELF-TUNING AND
SELF-DIAGNOSING SIMULATION
Power system simulation can be deined as using models, either digital or physical, to
simulate the real operation scenarios of the power system. Due to the rapid development of computer technologies and the greatly increased complexity of the power
313
Self-Tuning and Self-Diagnosing Simulation
Model data preparation
WAMs
Selected contingency list
Recorded contingency list
Digital simulation
CM/M&M simulation
Results output
Match?
Y
N
End
Model parameters calibration
FIGURE 13.1 Flowchart of closed-loop power system simulation.
grid, digital simulation has been playing an increasingly important role. The usual
way of carrying out power system digital simulation consists of four steps, as shown
in the left part of Figure 13.1. Before starting the simulation, it is necessary to collect
all the model data from either the ield tests or the manufacturers. Usually, this is
done repeatedly when the new devices are put into operation in the utility. The data
are then stored in the database and will be retrieved whenever the simulation is to be
implemented. The second step of the digital simulation is to build the contingency
lists based on the operators’ knowledge of the power grid. Since numerous possible
faults exist, the operator will select the most serious ones based on various system
operation scenarios. The digital simulations are carried out case by case for all the
contingencies in the list to ensure that the system is securely operated under all possible serious faults. For those faults that may cause serious damage to the system, such
as blackouts, additional digital simulations must be performed to ind proper control
measures to minimize the risks. All the information obtained from the simulation
will be sent to the system operator to aid decision making on the system operation.
This is a typical open-loop simulation, starting from the data preparation and
ending with the output of results. However, one key element of the simulation is
missing, that is, the simulation validation. Sometimes it is far too easy to trust the
output of the simulation, but think too little about its idelity. The computer may
give a completely incorrect image of the operated system, partly due to inappropriate algorithms and incorrect model parameters. False simulation results could also
come from the uncertain components of the power system, such as the load. Since
314
Smart Grids
the load is always changing, both in its composition and in its amplitude, the difference between the real load and its simpliied model will affect the simulation results.
Typical lessons in unreliable simulation can be learned from the 1996 Western
Systems Coordinating Council (WSCC) blackouts [1]. After cascading contingencies, the WSCC system witnessed undamped oscillations and consequently broke
into several islands. The postfault simulation analysis based on WSCC’s operation
database indicated that the blackout could not be reproduced. Even worse, the simulation showed a very stable system, in contrast to the real situation. It is dangerous to
make an optimistic simulation, which gives the system operator false conidence and
puts the power system under unknown potential risks. On the other hand, if the simulation is too pessimistic and exaggerates the fault hazards, the power system will be
operated very conservatively. The eficiency of the energy transfer will drop and the
utilization rate of the power lines and power transformers will be low, which causes
great waste. To enhance the simulation validity, the model and its parameters have
to be calibrated frequently. However, due to various reasons, including obstacles on
the operation side, the technique side, and the management side, routine ield tests in
order to correct the model parameters are often delayed.
For those models with great uncertainty, for example, the load model, their parameters should be updated even more frequently. As the recorded data accumulate, the
law behind the supericial randomness may be revealed gradually. By continually
analyzing these recorded data, we can ind the laws behind the uncertainty and apply
them to adjust the model parameters. Currently, however, this mostly relies on offline data collection and analysis. The updating rate for the model parameters cannot
keep up with the change of the component to be modeled.
Thanks to the advent of advanced measurement technologies and communication
technologies, which will be further developed in the coming smart grid era, the once
open-loop digital simulation can be closed, as shown in Figure 13.1. The deployment
of WAMs [2–4] provides new opportunities for the traditional simulation and widens its application ields. As shown in the other chapters of this book, WAMs could
provide synchronized measurements, which are just like taking snapshots of the
dynamics for the whole power system, whatever its scale. It is even more important
that the WAMs make this closed-loop simulation more adaptive. This helps to ind
the real problem of the power system in operation, by locating either the erroneous
model parameters or the disturbance source, as will be illustrated in detail in the
following sections. By fully utilizing the newly emerged measurement technologies,
this closed-loop simulation could increase the reliability of the power system in operation and provide more authentic guides to the system operators. So, the simulation
becomes smart; and, in turn, it helps to make a smart power grid.
This self-tuning and self-diagnosing simulation has two parallel branches. The
conventional simulation is still carried out based on the prepared contingency list.
However, when the WAMs record the real faults or the disturbances in the power
grid, the other branch of the simulation initiates itself. This simulation consists of
two aspects: the complete model (CM) simulation and the model and measurement
(M&M) simulation. CM simulation is based on the computer models in the operation database. Since it is carried out on the speciic fault recorded by the WAMs, the
sequence of events (SOE) has to be set to be the same as the recorded one, which
Self-Tuning and Self-Diagnosing Simulation
315
is the basic difference between this CM simulation and the conventional digital
simulation. To fulill this, the recorded SOE must be programmed into the computer
codes for piecewise simulations. This can be realized by interactive simulation packages, which are often equipped with user programming tools and automatic simulation functionality.
The basic target of the CM simulation is to make a comparison with the WAMs.
Thus, the validity of the current model database for system operation can be evaluated. The error criteria have to be carefully designed to relect not only numerical
errors but also the dynamic characteristic differences between the simulation and
the WAMs. So, this evaluation could facilitate searching for the error sources to the
greatest extent.
The M&M simulation can be classiied as a type of hybrid simulation. It mixes the
computer model data and the measurements to carry out the simulation. It is a newly
proposed method to utilize the measurements in the digital simulation. The principle
and the advantage of the M&M simulation will be elaborated further in the following section. If the CM simulation or the M&M simulation matches the measurements, the models and the respective data will be reliable for use in the simulation
to aid the system operation. However, if it does not, the models or the parameters, or
both, have to be calibrated based on the WAMs. For power components with uncertainty, the WAMs can record their inputs and outputs, which therefore can be used to
decouple these components from the power system. Then the model parameters can
be identiied and updated to adapt to the changing situation. The sensitivity analysis
method can be used to detect erroneous parameters. If the discrepancy between the
simulation and the WAMs is caused by an unknown disturbance in the system, the
WAMs can be utilized to ind the disturbance source at this stage, which may reduce
disturbance hazards in future system operation. On all occasions, the M&M simulation may have to be performed repeatedly to minimize the searching space and
alleviate the computational burden caused by the complexity of the system. It should
be noted here that, although the self-tuning simulation and self-diagnosing simulation improve the simulation reliability and reduce the operation risks, this cannot be
done once and for all. The power system is a highly nonlinear dynamic system with
a great number of models and parameters. Therefore, many sets of parameters exist
to make the simulation meet the WAMs, but not all of them are correct ones that can
be used in the system operation. The best way to discriminate between all these candidate parameters is to take advantage of their sensitivities to the changing operation
scenarios. The true set of parameters is robust to the various operation situations,
both in the steady state and in the presence of faults, while the false parameters are
sensitive to them. So, with increased WAMs, the self-tuning simulation can converge
gradually to the correct parameter values and the real system operation scenarios.
13.3
INTEGRATE THE SYNCHRONOUS MEASUREMENTS
WITH DIGITAL SIMULATION
One key step in self-tuning simulation is the M&M simulation, which combines the
model and the measurements to simulate the power system dynamics. Before the
wide application of WAMs, this was impossible, because the measurements are not
316
Smart Grids
all synchronized and consequently cannot be put into the digital simulation at the
same time. The basic idea of the M&M simulation is to replace part of the variables
in the digital simulation with the real measurements; therefore, the simulation data
will consist of the model data and the real measurements, which can be called the
hybrid data dynamic simulation.
The power system experiences transients when it changes its operation state
due to either faults or operation actions. The transients are usually classiied as
electromagnetic transients and electromechanic transients. The electromagnetic
transients involve the current and the voltage dynamics, while the electromechanic
transients involve the dynamics of the generator rotor position and the rotation
speed. Obviously, because of the considerable inertia of the generator rotors, the
electromechanic transients last much longer than the electromagnetic transients. So
these two kinds of transients are usually decoupled and analyzed individually. Since
WAMs only record the electromechanic dynamic states, we will apply them here to
the electromechanic dynamic simulation. The mathematical representation of the
power system electromechanic dynamics can be described as
xɺ = f ( x, y, p, u )
0 = g ( x, y, p, u )
(13.1)
where:
x denotes the electromechanic dynamic states, such as the rotor angle and the
rotor speed
y denotes the algebraic states, such as the bus voltage and the phase angle
p denotes the parameters
u denotes the controls
In Equation 13.1, the transients of the transmission network are neglected, since
they fall into the domain of the electromagnetic transients and last for a much shorter
time; consequently, they are represented by a set of algebraic equations, as shown in
the second equation of Equation 13.1. The irst equation of Equation 13.1 represents
the power system electromechanic transients. In the M&M simulation, since parts
of the state variables are available from the WAMs, they will be replaced by the
measurements, so that the power system dynamic equation takes the form of
xɺ1 = f ( x1, y1, p, u, x2 , y2 )
0 = g ( x1, y1, p, u, x2 , y2 )
(13.2)
Compared with Equation 13.1, the dynamic states x and the algebraic states y
are decomposed into two parts. The unknown states that have to be solved by the
digital simulation are represented by x1 and y1, respectively, while the known measured states are represented by x2 and y2, correspondingly. Equation 13.2 represents dimension-reduced power system electromechanic dynamics compared with
Equation 13.1, since only a subset of the power system states has to be solved by
Self-Tuning and Self-Diagnosing Simulation
317
the step-by-step integration. Naturally, there is less computational burden in solving
Equation 13.2 than in solving Equation 13.1. The other advantage of this hybrid data
dynamic simulation is that the measured states x2 and y2 are independent of the system models and parameters. Thus, if the WAMs data are reliable and well calibrated,
x2 and y2 represent true images of the power system dynamics. This fact is very
important, since it provides a possible way to decouple a very big system model into
various small pieces with the real measurements at the boundaries.
The one machine ininite bus (OMIB) system can be taken as an example to illustrate the idea of the hybrid data dynamic simulation. Assuming the phasor measurement unit (PMU) is installed at the high-voltage bus side, the whole OMIB system
can be decoupled into two subsystems. One of them models the generator as well
as its controllers in detail, while the system outside the machine is equalized by
PMU measurements; on the other hand, the whole system outside the machine will
be reserved in the simulation database while the machine is equalized by the PMU
measurements. Each subsystem has fewer dimensions than the original whole system. The advantages of this measurement-based decoupling work lie not only in this
dimension reduction, but also in the independence of the measurements from the
models, as well as their parameters. For example, when simulating the subsystem in
which only the generator is modeled in detail, we do not need to consider the validity of the models and their parameters outside the generator; nor do we worry about
unknown disturbance in the outside system that may affect the simulation results.
This is because the synchronized measurements from the WAMs represent the real
dynamics of the outside system.
The conigurations to decouple a big system into small subsystems vary, but rely
on the deployment of WAMs. With the wide application of modern measurement
technologies, the alternatives to decouple the system become more lexible. Three
typical characteristics of the power system can be utilized in decoupling the original
large system. The irst characteristic of the power system is its hierarchy. The power
system consists of the transmission system and the distribution system. The transmission system is operated on higher voltage levels, while the distribution system is
operated on lower voltage levels. Thus, the big system can be decoupled based on
the voltage levels by injecting the measurements from one side of the transformers. If the measurements are injected from the lower-voltage bus of the transformer,
the grid on the lower voltage level is equalized by the WAMs. If they are injected
from the higher-voltage bus of the transformer, the grid on the higher voltage level
is equalized by the WAMs. The second characteristic of the power system that can
be used in the hybrid data dynamic simulation is its geographic composition. The
large power system consists of different utilities connected by interties. In general,
the interties are equipped with advanced meters. If measurements from these meters
are synchronized, for example, the WAMs, the whole system model can be decomposed into individual utility models by replacing the electrical states on the interties
with these measurements. So, the simulation on a very large system becomes several
simulations on small subsystems. In real applications, it is not necessary to simulate
all these subsystems, but only the system that is causing trouble. This will be further illustrated later. The third characteristic of the power system is that it is composed of various components, and these components have clearly deined interfaces.
318
Smart Grids
So, if the inputs and outputs of each component are well recorded and synchronized,
the measurements can be utilized to decouple them from the large system, and the
hybrid data dynamic simulation, therefore, can be implemented on each individual
component.
To realize the hybrid data dynamic simulation, the interfaces of the model data
and the real measurements have to be well designed. The core of the algorithm is
how to inject the measurements into the digital simulation model. There have been
three state-of-the-art methods: the fast-response generator method [5], the ideal tapchanger transformer method [6,7], and the variable impedance method [8,9].
The fast-response generator method realizes the injection of the measurements by
an assumed generator. The WAMs are coded as a preset sequence of the reference
values to the assumed generator controller. The dynamic response of the generator
is fast enough that the outputs of the generator follow the control reference values
immediately. For example, the bus voltage measurements from the WAMs are input
to the generator automatic voltage regulators as the target voltage references. Due
to the fast response characteristics of the generator and its excitation system, the
terminal voltage of this assumed generator takes the exact preset reference values at
each simulation point. In this way, the measurements and the model parameters are
merged for the hybrid data dynamic simulation.
The ideal phase-shifting transformer method uses the changeable ratio between
the primary winding and the secondary winding to realize the hybrid data dynamic
simulation. The primary winding of the phase-shifting transformer connects to the
ininite bus with constant voltage and constant frequency. The measurements from
the WAMs are taken to modulate the tap ratio. At each simulation step, the tap ratio
will be corrected by the quotient of the constant ininite bus voltage with respect to
the measured voltage phasor. In this way, the voltage of the secondary-winding bus
takes the exact values of the measurements.
Among the three methods to implement the hybrid data dynamic simulation,
the variable impedance method takes the simplest form. It applies a value-changed
impedance to replace the system outside the boundary, as shown in Figure 13.2.
The equivalent complex impedance Z or Y = 1/Z can be calculated from the measurements as follows:
P (t )
1
= Y ( t ) = G ( t ) + jB ( t ) = 2
+
U (t )
Z (t )
 Q (t ) 
j− 2
 U ( t ) 


Boundary bus
Subsystem
Z
FIGURE 13.2 The variable impedance method.
(13.3)
Self-Tuning and Self-Diagnosing Simulation
319
where P(t), Q(t), and U(t) represent the boundary bus voltage, the transferred active
power, and the reactive power. Since these measurements change with time, the
value of Z is not constant, but is also changed at every simulation time step. That is
why it is called the variable impedance method. Since the value of the impedance Z
is changed according to the WAMs, the digital simulation integrates the model data
and the real measurements together.
It should be noted here that the simulation step and the sampling period of the
measurements may not be consistent. Thus, the WAMs must irst be synchronized
with the simulation before being injected into the digital simulation system using
any of the above three methods. No standards have been speciied for the digital
simulation step length, or the time interval when the WAMs data are expected to be
sent to the data server in the control center. In general, the typical electromechanic
transient simulation package takes half a cycle as the simulation step, that is, 0.01 s
for a 50 Hz system, while the time interval for two consecutive measured phasors
is one cycle, that is, 0.02 s for a 50 Hz system. Thus, the measurements cannot be
directly incorporated into the reserved digital model to implement the hybrid data
simulation. One solution to this problem is to interpolate the measured points so
that a one-to-one correspondence can be set up between the measured point and the
simulation point [8].
We have applied the variable impedance method in many benchmark test systems
and real power grids. Its eficiency has been well proved [9]. A typical three-area
system in a Power System Simulator for Engineering (PSS/E) package [10] is used
here to exemplify the hybrid data dynamic simulation. Assume that the synchronized PMUs are installed at the interties among Area 1 and the other two areas.
Since this is a test system, we take the simulation results as the pseudomeasurements
in order to derive the variable impedances connected to Area 1’s boundaries. In this
way, we have a case of hybrid data dynamic simulation, with Area 1 represented
in detailed models and parameters while its boundaries are injected by the timediscrete measurements.
An assumed three-phase short circuit on one load bus is simulated based on the
CMs and parameters of this three-area system. The active power dynamics on one
generator bus is shown in Figure 13.3. The simulation results from the reduced network are also shown in Figure 13.3. It can be seen clearly that the M&M simulation
using the variable impedance method produces the same dynamic trajectories as the
CM simulation does. However, if we change some parameter values in the reduced
network, Area 1 in this case, such as the parameters of Gen. NUC_A and Gen.
NUC_B, and make them different from their values in the original complete network, the hybrid data dynamic simulation results will become erroneous compared
with the simulation results for the full network, which has also been clearly shown
in Figure 13.3.
The hybrid data dynamic simulation curves in both cases of Figure 13.3 verify
that the hybrid data dynamic simulation can relect the real dynamics of the whole
network from a reduced network model. If the hybrid data dynamic simulation does
not match the real measurements, it can be concluded that there is a discrepancy
between the reserved models of the scale-reduced network, in this case the models in Area 1, and the real situation of the complete network. Since the injected
320
Smart Grids
780
Active power on bus 101 (MW)
760
740
720
700
680
660
Simulation from the whole network
Hybrid simulation from the reduced
network
Hybrid simulation with wrong
parameters in the reduced network
640
620
600
0
0.5
1
1.5
2
Time (s)
2.5
3
3.5
4
FIGURE 13.3 Active power on one generator bus.
measurements are independent of the models and parameters in the simulation
database, the discrepancy between the measurements and the hybrid data dynamic
simulation results could only be caused by the differences between the reserved network and the corresponding part of the real power system. This information is very
important, since it reduces the error location ranges, which greatly enhances the
simulation eficiency, especially when simulating a very large system with many
components.
The hybrid data dynamic simulation presented in this section is the core technique in self-tuning and self-diagnosing simulation, since it provides a realizable
approach to constituting a feedback closed-loop simulation using measurements,
which makes self-correcting parameters possible in the simulation.
13.4
MODEL DATA CALIBRATION
As stated in the prior section, when the hybrid data dynamic simulation produces
different results from the measurements, there must be errors in the simulated scalereduced system. There are two major error sources. One is incorrect models or
parameters, or both; the other is the unknown disturbance that happens in the real
power system but without being simulated in the computer. This section addresses
the irst problem, that is, the false model and parameters. The second problem is
addressed in the next section.
There are few incorrect model structures in the power system simulation database,
since all the models are built based on the components’ physical mechanisms and
conirmed by academic researchers as well as electrical engineers. Although there
are electric components with uncertainty, such as the load, widely accepted models
of these are used in the operation database. As time goes on and the measured data
accumulate, the models can be veriied. If the hybrid data dynamic simulation is
applied to an individual component, with the entire external network equalized by
Self-Tuning and Self-Diagnosing Simulation
321
the variable impedance, and indicates signiicant errors between the simulations and
the measurements, the validity of the model structure should be suspected. With the
synchronously recorded inputs and outputs of this component, its operation mechanism can be reanalyzed and consequently its model structure can be reselected.
After the parameters of the new model have been identiied from the measurements,
the simulation database will be updated with the newly built model and will wait for
the next recorded contingency for the model validation.
In most cases when the hybrid data dynamic simulation does not match the measurements from the WAMs, this is due to incorrect parameters or changes of parameter values under different operation situations. The best solution to ind these key
parameters is to use the proposed hybrid data dynamic simulation tool to decouple
the system into as many subsystems as possible. This greatly reduces the searching
space of each subsystem for the incorrect parameter. To cut down the searching
scale further, the sensitivity method can be utilized to ind the parameter that has
the greatest effect on the system simulation results. Corrections to these most sensitive parameters have the highest eficiency in itting the simulation results to the
measurements. As pointed out in Section 13.2, due to the high nonlinearity of the
power system and large number of components, the parameter corrections made may
not be right even if they do match the simulations with the measurements. Thus, the
self-tuning simulation has to repeat itself continually whenever WAMs records new
dynamics, and validate the parameter constantly.
A real application case in a very large power system will be presented here to
illustrate the hybrid data dynamic simulation as well as the model parameter correction work [9]. The Northeast (NE) power grid is a very large interconnected power
system with more than 45,000 MW installed generation capacity in 2006. The north
part of the NE grid has an abundance of generators, while the south part of the NE
grid is load-concentrated. So, a considerable amount of energy has to be transferred
from the north to the south through long transmission lines. Meanwhile, the NE grid
also outputs the energy to another large power grid through 500 kV tie-lines. So, the
available transfer capability calculation is important for the system’s security and
economic operation, which relies on an authentic simulation.
In order to enhance the transfer capability of the transmission corridors and
validate the models as well as their parameters, artiicial three-phase short circuits at
the 500 kV buses were intentionally created four times in the NE power grid during
2004 and 2005. The WAMs in the NE power grid recorded the disturbed dynamics
reliably during these four short-circuit tests, and therefore provided the measured
database for the model parameter tuning work.
Following the procedure in Figure 13.1, CM simulations are performed on the
whole NE power system for the tested fault scenarios. The results show that the system dynamics cannot be reproduced by the computer simulation. However, it is not
easy to locate the erroneous parameters that cause the mismatch between the simulation and the WAMs for such a large power grid. To solve this problem, an M&M
simulation is carried out. The WAMs are deployed on all the 500 kV buses in the NE
power grid; thus the whole system is decomposed into several subsystems from the
500 kV buses. Each subsystem has fewer components than the original large system,
which minimizes the parameter correction space to a great extent.
322
Smart Grids
Take the YM plant as an example. It is a key fossil power plant in the NE power
grid, equipped with two 500 MW generators and producing more than 4 trillion kWh
electric energy each year. The terminal voltage of these two generators is 20 kV,
which is transformed to 500 kV at 2M51 buses where both generators are connected
to the 500 kV power grids. PMUs are installed at the 500 kV 2M51 buses to monitor the energy transfer from the YM plant to the NE grid and the local network.
Applying the variable impedance method, the whole NE grid and the local network
can be replaced by the equivalent admittances with
G ( t ) + jB ( t ) =
P1 ( t ) + P2 ( t ) + P3 ( t )
+
V 2 (t )
 Q1 ( t ) + Q2 ( t ) + Q3 ( t ) 
j−


V 2 (t )


(13.4)
where V denotes the voltage of bus 2M51; and P1(t), Q1(t), P2(t), Q2(t), P3(t), and Q3(t)
represent the active power as well as the reactive power recorded by PMU1, PMU2,
and PMU3. The active power outputs of the YM plant from both the WAMs and the
hybrid data dynamic simulations are shown in Figure 13.4.
It can be clearly seen that the simulated active power cannot match the measurements. This means that some model parameters in the YM power plant, especially
those that have great effects on the active power dynamics, need to be calibrated. In
this way, the searching space for the incorrect parameters is limited to the YM plant,
and the effects of the incorrect parameters in the YM plant on the CM-based simulation can also be evaluated very clearly. To ind the incorrect parameters and correct
them, the sensitivity method is applied. Considering
P0 = Vd I d + Vq I q
(13.5)
1,200
Active power (MW)
1,000
800
600
400
Real measurements
200
Hybrid dynamic simulation
before model validation work
0
0
0.5
1
1.5
2
Time (s)
2.5
3
3.5
FIGURE 13.4 Real measurements and hybrid data dynamic simulation of active power.
4
323
Self-Tuning and Self-Diagnosing Simulation
Q0 = Vq I d − Vd I q
(13.6)
V0 = Vd2 + Vq2
(13.7)
where
V0
denotes the generator terminal voltage
Vd, Vq and Id, Iq are the projections of the generator terminal voltage and the stator
current at the d-axis and the q-axis, respectively
P0 and Q0
are the generator active power output and the reactive power output
The trajectory sensitivities P0 and Q 0 with respect to the model parameter λ can
be derived as
∂I
∂V
∂P0
∂I
∂V
= Vd d + Vq q + I d d + I q q
∂λ
∂λ
∂λ
∂λ
∂λ
(13.8)
∂I
∂V
∂Q0
∂I
∂V
= Vq d − Vd q − I q d + I d q
∂λ
∂λ
∂λ
∂λ
∂λ
(13.9)
∂V0 Vd ( ∂ Vd ∂λ ) + Vq ( ∂ Vq ∂λ )
=
∂λ
Vd2 + Vd2
(13.10)
since V, P, Q on bus 2M51 can be calculated as
P ≈ 2 P0
(13.11)


P 2 + Q2
Q ≈ 2  Q0 − 0 2 0 XT 
V
0


(13.12)

QX
V =  V0 − 0 T
V0


K

(13.13)
where K is the transformer voltage ratio and XT is the transformer leakage reactance.
The trajectory sensitivities of PMU measurements with respect to the parameter λ
can be calculated as
∂P
∂P
≈2 0
∂λ
∂λ
(13.14)

∂Q
Q X  ∂Q
P X ∂P
P 2 + Q2
∂V 
= 2  1 − 2 0 2 T  0 − 2 0 2 T 0 + 2 0 3 0 X T 0 
∂
∂λ
V
V
V
∂
∂λ 
λ
λ
0
0
0


(13.15)
324
Smart Grids
 Q X  ∂V
∂V
X ∂Q0
= K 1 + 0 2 T  0 − K T
∂λ
V0  ∂λ
V0 ∂λ

(13.16)
Substituting Equations 13.8 through 13.10 into Equations 13.14 through 13.16, the
sensitivities of the measured active power, the reactive power, and the voltage from
PMUs with respect to the generator parameters can be calculated. Equations 13.8
through 13.10 indicate that it is very important to calculate the partial derivatives of
the d-axis and q-axis components of the generator terminal voltage and the current
with respect to the selected parameters; nevertheless, it should be noted here that the
result is dependent on the model structure. So this is again a repeating model tuning
process. The model in the operation database will be taken as the initial model to
start the sensitivity calculation. If the corrected parameters based on this sensitivity
information cause the simulation to match the measurements, the conidence level
of the current model structure will increase; if not, a more detailed model should
be considered to replace the initial model. In this way the simulation reliability is
improved gradually, allowing more credible operation decisions on the real power
system.
The generator model in the YM plant for the power system operation and control
includes the rotor dynamics, the dynamics of the ield winding, and the dynamics
of three damping windings. The discrepancy between the PMU measurements and
the simulation is caused mainly by parameters. After the most sensitive parameters
were corrected based on the sensitivity analysis, the simulations closely itted the
measurements.
To verify the parameter corrections, the calibrated model parameters were applied
to other recorded faults. Figure 13.5 shows the PMU measurements in another threephase short-circuit test in the NE power grid, and the active power simulated using
the hybrid data dynamic simulation.
1,100
Active power (MW)
1,000
900
800
700
PMU
600
500
Model before validation work
Model after validation work
0
0.5
1
1.5
2
Time (s)
2.5
3
FIGURE 13.5 Real measurement and model outputs on active power.
3.5
4
Self-Tuning and Self-Diagnosing Simulation
325
From the igures, it can be seen clearly that the simulation based on the calibrated
models its the PMU measurements more closely. The model’s adaptive capability to
the different faults indicates good reliability of the model database.
13.5
LOCATE THE DISTURBANCE SOURCE FOR
LOW-FREQUENCY OSCILLATION
As well as incorrect parameters, unknown disturbances happening in the power grid
can also cause inaccurate simulation results. Finding the disturbance is very important to the system operators, so that they can eliminate the disturbance source in
time to avoid it happening again. Generally, a large disturbance can be detected and
separated from the grid by the relays. However, it is sometimes very hard to ind and
eliminate a small disturbance in a very large network.
One of the most widely encountered problems caused by a small disturbance in
a large power system is low-frequency oscillation. This is a long-standing problem.
Theoretically, a small disturbance can excite low-frequency oscillation in two general ways. If the system is marginally stable or unstable, the small disturbance can
push the operation states away from their equilibriums. In this case, the voltage,
the current, and the power transferred in the grid begin to oscillate in increasing
magnitudes or equal magnitudes, depending on whether the system equilibrium lies
on the imaginary axis or the right half plane. This type of low-frequency oscillation
is known as negative damping low-frequency oscillation, since the real part of the
unstable equilibrium is greater than zero. This phenomenon was irst noticed during
the 1960s after the wide application of fast high-gain exciters. These increase the
power system transient stability by offsetting the power imbalance caused by the
fault; however, they shift the eigenvalues of the linearized power system to the right
half plane. The best way to quench the negative damping oscillation is to compensate
the damping so that the total damping of the dynamic system becomes positive. The
power system stabilizer (PSS) is designed as an auxiliary exciter control strategy
to make up for system damping, and has been widely applied in the power grid
worldwide.
The second way in which a small disturbance excites the low-frequency oscillation of the whole power grid is resonance. Resonance-raised oscillation has become
a serious problem with the increased interconnections of the power grids. As the
power grid grows larger, the chances of being affected by the disturbance increase.
On the other hand, the intrinsic oscillation frequency of the interconnected power
system becomes smaller, which makes it very close to the oscillation frequency
existing in the disturbance. Unlike negative damping low-frequency oscillation,
resonant low-frequency oscillation appears and disappears in a very short time. It
can reach its oscillation peak in several oscillation cycles, and constitutes a great
danger to system security. The key to quenching resonant low-frequency oscillation
lies in inding the oscillation source, that is, the disturbance that excites the resonant
oscillation, and preventing the disturbance from happening again. However, it is not
easy to ind an unknown disturbance in the large power grid. In practice, the digital
simulation is performed repeatedly in a trial-and-error way. The pseudodisturbance
is initiated at various locations of the power grid to match the simulation with the
326
Smart Grids
PMU measurements. Due to the large scale of the power system and the number of
electrical components involved in the simulation, great effort is required to search
for the small disturbance.
A self-diagnosing simulation could take the measurements from WAMs and
apply hybrid data dynamic simulation to minimize the searching space for the disturbance. In the following, a real resonant oscillation case will be presented to show
how modern measurement technologies and analysis tools are applied to minimize
the computational burden of locating the disturbance source [11].
Figure 13.6 is a schematic diagram of the South Power Grid of China, interconnecting the provincial power grids GX, GZ, GD, and YN. It is a very large power grid,
covering 1,000,000 km2, and nearly 2000 km long from east to west. Hydropower
and fossil power are concentrated in the west of the South Power Grid; however,
the load is mainly located in the east. Consequently, the electric energy has to be
transferred from the west to the east through long transmission lines. Low-frequency
oscillations tend to happen on these heavily loaded long transmission lines, which
constitutes a major threat to the secure operation of the South Power Grid.
One low-frequency oscillation happened in the South Power Grid of China in
2005. The oscillation spread quickly to the whole grid and affected the energy transfer on the interties. Based on the recorded dynamic trajectories, it can be seen that the
low-frequency oscillation happened twice between 00:55:00 and 00:59:30. The irst
oscillation started from 00:54:54 and ended at 00:57:37, lasting about 4 min. It took
about 30 s for the oscillation to reach its peak. Although the oscillation peak-to-peak
value reached 140 MW in the irst 30 s, the average power transfer remained nearly
constant over this period. The oscillation amplitude of the active power remained
unchanged. Meanwhile, the average power transfer increased during the period of
the oscillation, which exacerbated the loading of this intertie. Nevertheless, it can
be observed that the increased loading of the transmission line had few effects on
GZ
ASH
Guiyang
GX
Qingyan
Nayong
GD
GZH
Anshun
Beijiao
Hechi
Anshun
Yantan
Hezhou
TSQ
Pingguo
TSQ 2
Luoping
Shatang
Luodong
Wuzhou
Lubuge
YN
TSQ 1
Baise
ZHQ
Nanning
Xijiang
Yulin
Maoming
Hydro power plant
Power plant
Jiangmen
Converter station
FIGURE 13.6 Schematic diagram of 500 kV network of South Power Grid, China.
327
Self-Tuning and Self-Diagnosing Simulation
the oscillation. The oscillation continued until 00:56:48. Then it disappeared very
quickly from 00:56:48 to 00:57:15, taking about 30 s, as it had begun. Meanwhile,
the grid control center reduced the power generation in the west of the South Power
Grid to alleviate the loading of the grid transfer interfaces. Even though the transferred energy was continually decreasing, the second low-frequency oscillation still
happened.
The second low-frequency oscillation took place from 00:57:40 to 00:59:30.
During the oscillation, the average energy transferred continuously decreased. It
also took about 30 s to reach its oscillation peak. The peak-to-peak value of the
second low-frequency oscillation was about 100 MW, which was a little lower than
that of the irst oscillation. Following this, the active power oscillated with equal
amplitude, as had happened in the irst oscillation. The equal amplitude oscillation
continued until 00:59:00. Then it quenched very quickly, also in 30 s, from 00:59:00
to 00:59:30.
By linearizing the power system model and calculating its eigenvalues, we can
ind the major electromechanic oscillation modes during this contingency, which are
shown in Table 13.1. The oscillation frequency ranges from around 0.55 to around
1.21 Hz. It should be noted here that all these oscillation modes are very well damped.
The analysis directly on the PMU measurements via the Prony method conirms the
results derived from the linearized system model. The Prony analysis decomposes
the signal into the summation of a series of exponential functions; thus the principal oscillation mode can be found. When the steady-state intertie power oscillation
undergoes Prony analysis, the real measurements from PMU and the Prony analysis
results in both the irst and the second oscillation it very well.
Based on the Prony analysis, the principal frequency mode of the irst oscillation
is 0.854 Hz and that of the second oscillation is 0.9117 Hz. They are slightly different, but both can be found in Table 13.1, which indicates that these two oscillation
TABLE 13.1
Electromechanic Oscillation Modes of
South Power Grid, China
No.
1
2
3
4
5
6
7
8
9
10
11
12
Real
Image
Frequency
(Hz)
Damp
−5.1679
−5.4668
−4.3498
−8.8456
−8.8883
−9.1937
−9.2359
−0.3420
−0.2911
−4.3851
−0.7818
−1.7864
3.4812
3.7609
4.3003
4.8572
5.0454
5.2298
5.3949
5.7577
5.9420
6.1588
7.0740
7.6212
0.5541
0.5986
0.6844
0.7730
0.8030
0.8323
0.8586
0.9164
0.9457
0.9802
1.1259
1.2130
0.8294
0.8239
0.7111
0.8766
0.8697
0.8692
0.8635
0.0593
0.0489
0.5800
0.1098
0.2282
328
Smart Grids
modes are intrinsic oscillation modes of the South Power Grid, China. To sum up,
this low-frequency oscillation has the following characteristics. First, it reaches its
peak oscillation magnitude in a very short time and disappears in the same manner.
The time duration is 30 s for both. This is typically different from negative damping
oscillation, which usually takes a much longer time to reach its peak value or quench,
as observed from several prior low-frequency oscillations in the same power grid.
Second, there is no obvious oscillation start level for this low-frequency oscillation.
Neither increased nor decreased average power transfer level on the transmission
interties could quench the oscillation. In contrast, negative damping oscillation
has, in general, an interchange power threshold. When the transferred power of the
interties is above the threshold, low-frequency oscillation with increasing amplitude starts; on the other hand, if the transferred power of the interties is below this
threshold, the oscillation disappears.
To prevent low-frequency oscillation in the South Power Grid, China, PSS has
been widely applied in this system; however, this time it was unable to stop the lowfrequency oscillation. Posterior simulation after the fault could not reproduce the
oscillation. Since the oscillation modes are intrinsic to the South Power Grid, it is
suspected that some unknown disturbance in the grid coincidentally has a frequency
component close to the intrinsic oscillation frequency of the grid, which therefore
excites forced low-frequency oscillation in the whole grid. Because the location is
unknown, the disturbance cannot be properly simulated. Consequently, the simulation could not ind the root of the oscillation problem and effectively erase it.
To ind the oscillation cause, the pseudodisturbance is assumed at different points
in the South Power Grid. Due to the large scale of the grid, the computational burden
is quite heavy. In addition, another problem closely related to locating the disturbance source by the trial-and-error method is that the simulation based on the CM
data of the South Power Grid could possibly produce similar simulation curves even
when the disturbance is assumed at different locations. For example, simulations
based on the imaginary disturbance either in a power plant of GZ or in another power
plant of GX produce nearly the same power oscillations in the Luoma intertie. So we
cannot judge the location of the disturbance from the simulations alone. This aggravates the dificulty of inding the disturbance source.
Let us take a two-area system to explain the idea of applying hybrid data dynamic
simulation to locate the disturbance source for the forced low-frequency oscillation.
Assuming there is a disturbance in Area 1, we could not judge where the disturbance lies from PMU measurements alone. But the PMU measurements could be
used to decompose the original two-area system into two one-area subsystems by
hybrid data dynamic simulation, as shown in Figures 13.7 and 13.8. The dashed line
in Area 1 indicates that the disturbance is actually in Area 1 but is not simulated
because we do not know where it is.
Analyzing Figure 13.8, we ind that its simulation scenario is completely the same
as for the two-area system, except that Area 1 in the original two-area system is
replaced by the real measurements. Since the measurements relect the real dynamics of the power system and are irrelevant to the system conigurations in Area 1, the
unknown disturbance in Area 1 does not affect the hybrid data dynamic simulation
results of Figure 13.8. So, the hybrid data dynamic simulation on Figure 13.8 would
329
Self-Tuning and Self-Diagnosing Simulation
Area 1
PMU
FIGURE 13.7 Subsystem with Area 1 modeled in detail.
PMU
Area 2
FIGURE 13.8 Subsystem with Area 2 modeled in detail.
be expected to reproduce the PMU measurements. This match between the simulation and the measurements indicates that the disturbance does not lie in Area 2.
However, the scenario of the hybrid dynamic simulation in Figure 13.7 is different
from the true scenario. In Figure 13.7, PMU measurements dynamically equalize
the models and parameters of Area 2, but they cannot represent the unknown disturbance in Area 1. Due to the lack of simulation of the unknown disturbance in Area 1,
the hybrid data dynamic simulation could not reproduce the PMU measurements,
which indicates that the disturbance lies in this area. In this way, the searching
range of the disturbance shrinks to Area 1 only, greatly reducing the computational
burden. Since the PMU measurements are applied to decouple the very large system into small ones, the probability of very similar simulation results from different
pseudodisturbance locations is also minimized.
Self-diagnosing simulation to ind the disturbance source for the forced lowfrequency oscillation is applied to the South Power Grid, China. Due to data conidentiality agreements with the utility, simulated phasors will replace the PMU
measurements in the following to illustrate the idea. However, the procedure is completely the same as when the real measurements are applied.
Assume that a small disturbance in one power plant in the South Power Grid,
China, has excited the resonance in the whole grid and we are trying to ind it
from the pseudomeasurements. The South Power Grid is divided into four subsystems, as shown in Figure 13.6 by the dashed lines. Each subsystem is actually an
administrative province in south China and is connected to the others by heavily
loaded interties. The hybrid dynamic simulation can be applied to each subsystem
as well as any combination of the subsystems. If the subsystem GZ is replaced by
the measurements and we simulate the remaining three subsystems (GX, YN, and
GD) together using the hybrid data dynamic simulation method, the measurements
and the simulation results match each other very well. However, if the GX, GD, and
YN subsystems are substituted by the measurements on the boundaries to GZ and
we simulate the GZ power grid individually, the voltage measurement on 500 kV bus
Anshun and the respective hybrid data dynamic simulation result will be quite different. Therefore, the disturbance must lie in the GZ power system. Then the searching space can be reduced to GZ power system, with the power grids outside GZ
330
Smart Grids
equalized by the phasor injections. The hybrid data dynamic simulation is further
applied to the individual power plant in GZ. The active power outputs of the two
power plants in the inner power grid of GZ are shown in Figures 13.9 and 13.10,
respectively, with the hybrid data dynamic simulation results.
Since, in Figure 13.9, the measurements and the hybrid data dynamic simulation
results match each other, the disturbance is not in the NY power plant. On the other
Measurement
Simulation
301.0
Active power (MW)
300.5
300.0
299.5
299.0
298.5
115
116
117
118
119
120
121
122
Time (s)
FIGURE 13.9 Active power outputs of NY power plant in GZ power grid.
235
Measurement
Simulation
230
Pe (MW)
225
220
215
210
205
200
115
116
117
118
119
120
Time (s)
FIGURE 13.10 Active power outputs of QB power plant in GZ power grid.
121
122
Self-Tuning and Self-Diagnosing Simulation
331
hand, Figure 13.10 shows obvious differences between the measurements and the
hybrid data dynamic simulations on the active power outputs in the QB power plant.
It is, therefore, likely that the disturbance lies in the QB power plant. Then the hybrid
data dynamic simulation is carried out in the QB power plant with a small disturbance initiated in its prime mover. In this way, the forced oscillation is reproduced by
the hybrid data dynamic simulation, which veriies the disturbance location.
13.6
CONCLUSION
Simulation has always played a very important role in power system operation and
control. On some occasions it is the only tool for understanding the fault mechanisms
of the power system, due to the large scale and complexity of the system as well as
the security concerns. The traditional simulation is an open loop. With the wide
application of modern and smart measurement technologies in the power grid, the
simulation loop can be closed; in this way, simulation will assume even more responsibility for aiding system operation. The smart grid has revolutionized the concepts
and the visions of the future power system. As one important block in building the
smart grid, digital simulation will also make great advances in adaptability and lexibility. Together with fast-growing computation technologies, self-tuning and selfdiagnosing simulation is coming.
ACKNOWLEDGMENTS
The author would like to dedicate this chapter in memory of Professor Alain
Germond of Ecole Polytechnique Federale de Lausanne, Switzerland, in thanks for
all his encouragement and selless help to the author.
This work is supported in part by the National Science Foundation of China (Grant
No. 51077049 and No. 50707009), in part by the Beijing Nova Program, and in part by
the ABB Research Center in China. Part of the research was carried out when the author
was with North China Electric Power University, Beijing. The author would like to
thank his postgraduate students in this research ield: Mr Sheng Wen-Jin, Mr Bo Bo, and
Ms Zhang Pu. The author wants to express his hearty thanks to Professor Dong ZhaoYang at the University of Sydney, Australia, for his insightful suggestions and discussions. The author appreciates the discussions on smart grid development with Professor
Rehtanz of the University of Dortmund and Dr Yue Cheng-Yan of the ABB Research
Center in China, and would like to thank them for their support and suggestions.
REFERENCES
1. D.N. Kosterev, C.W. Taylor, W.A. Mittelstadt, Model validation for the August 10, 1996
WSCC system outage, IEEE Transactions on Power Systems, 14(3), 967–979, 1999.
2. R.O. Burnett Jr, M.M. Butts, T.W. Cease, V. Centeno, G. Michel, R.J. Murphy, A.G.
Phadke, Synchronized phasor measurements of a power system event, IEEE Transactions
on Power Systems, 9(3), 1643–1650, 1994.
3. J. Hauer, D. Trudnowski, G. Rogers, B. Mittelstadt, W. Litzenberger, J. Johnson,
Keeping an eye on power system dynamics, IEEE Computer Applications in Power,
10(4), 50–54, 1997.
332
Smart Grids
4. M. Zima, M. Larsson, P. Korba, C. Rehtanz, G. Andersson, Design aspects for wide-area
monitoring and control systems, Proceedings of IEEE, 93(5), 980–996, 2005.
5. D. Kosterev, Hydro turbine-governor model validation in Paciic Northwest,
IEEE Transactions on Power Systems, 19(2), 1144–1149, 2004.
6. Z. Huang, R.T. Guttromson, J.F. Hauer, Large-scale hybrid dynamic simulation employing ield measurements, in Proceedings of the IEEE Power Engineering Society (PES)
General Meeting, June 5–10, 2004, Denver, CO.
7. Z. Huang, D. Kosterev, R.T. Guttromson, T. Nguyen, Model validation with hybrid
dynamic simulation, in Proceedings of the IEEE Power Engineering Society (PES)
General Meeting, June 18–22, 2006, Montreal, QC.
8. J. Ma, W.-J. Sheng, R.M. He, Research on WAMs-based power system dynamic simulation strategy, Automation of Electric Power Systems, 31(18), 11–15, 2007.
9. J. Ma, D. Han, W.-J. Sheng, R.M. He, C.Y. Yue, J. Zhang, Wide area measurements-based
model validation and its application, IET Generation, Transmission and Distribution,
2(6), 906–916, 2008.
10. Shaw Power Technologies, Program Operation Manual: PSS/E-30, 2004.
11. J. Ma, P. Zhang, H.-J. Fu, B. Bo, Z.-Y. Dong, Application of phasor measurement unit on
locating disturbance source for low-frequency oscillation, IEEE Transactions on Smart
Grid, 1(3), 340–346, 2010.
14
A Consensus-Based
Fully Distributed Load
Management Algorithm
for Smart Grid
Yinliang Xu, Wei Zhang, and Wenxin Liu
CONTENTS
14.1 Introduction .................................................................................................. 333
14.2 Consensus-Based Global Information Discovery Algorithm ....................... 334
14.2.1 Introduction to Average Consensus Theorem................................... 335
14.2.2 Consensus-Based Information Discovery Algorithm Design .......... 337
14.2.3 Convergence Speed of Information Discovery Algorithm ............... 339
14.2.4 Considerations during Implementation.............................................340
14.2.4.1 Distributed Synchronization ..............................................340
14.2.4.2 Plug and Play ..................................................................... 341
14.3 Applications .................................................................................................. 343
14.3.1 MAS-Based Load Management ....................................................... 343
14.3.2 Load Shedding ..................................................................................344
14.3.2.1 Normal Operating Conditions ........................................... 345
14.3.2.2 Reliability........................................................................... 347
14.3.3 Load Restoration ............................................................................... 350
14.4 Conclusion .................................................................................................... 356
References .............................................................................................................. 356
14.1 INTRODUCTION
Global concern over greenhouse gas production, increasing demand for better
reliability and security, widespread use of intermittent energy resources, charging of plug-in hybrid electric vehicles, and so on, has led to new requirements
being proposed for the power grid. However, the aging power grid, designed
according to decades-old techniques, can hardly address these new requirements. Driven by these new requirements, the concept of the smart grid was
proposed to upgrade current power grids for the next century. The smart grid is
the electric delivery network integrated with the latest advances in digital and
333
334
Smart Grids
information technology to improve the electric system’s reliability, eficiency,
security, and resiliency.
The power system is probably the most complex engineering system ever to have
existed. In [1–3], various kinds of centralized methods were proposed to deal with
problems such as load shedding, load restoration, optimal power low, and so on. It
is known that a centralized control scheme requires a complicated communication
network to collect global operating conditions and a powerful central controller to
process the huge amount of information. Thus, centralized schemes are costly to
implement and susceptible to single-point failures. Due to the uncertainty of intermittent renewable energy resources, unintentional changes of system structure will
further increase the burden on a centralized control scheme. Since distributed control solutions are lexible, reliable, and low cost to implement [4,5], they are a promising choice for the next generation of power systems.
As one of the most popular distributed control solutions, multiagent systems (MAS)
have been widely applied for the operation and control of power systems. Advantages
of MAS include the ability to survive single-point failures [6] and decentralized data
processing, which leads to eficient task distribution and eventually a faster operation and decision-making process [7]. In artiicial intelligence, agent-based technology has been hailed as a promising paradigm for conceptualizing, designing, and
implementing software systems. Recently, MAS-based solutions have been proposed
for power system reconiguration and restoration [5,8,9]. However, it is still dificult
to apply existing algorithms to systems with complex structure. Most importantly,
most methods were validated through simulation instead of rigorous analysis. The
convergence and stability of these algorithms have not been suficiently investigated.
To address the needs of the smart grid and the problems with existing research,
this chapter introduces a stable, fully distributed, multiagent-based solution. “Fully
distributed” means that each bus in a power system has an associated bus agent,
and two agents communicate with each other only if their corresponding buses are
electrically coupled. The design of the distributed algorithm is based on a consensus-based stable global information discovery algorithm. The information discovery
algorithm can guarantee convergence for power grids of any size and topology, and
is robust against various faults.
The distributed optimization algorithm has been successfully applied to load
management (load shedding [10] and load restoration [11]). The information discovery
algorithm is used to discover net active power and load connection information.
Based on the information, centralized optimization algorithms, such as greedy algorithms and dynamic programming (DP), can be used to maximize energy utilization.
The rest of the chapter is organized as follows. In Section 14.2, a consensus-based
global information discovery algorithm is introduced. In Section 14.3, the proposed
algorithm is applied to load management. Section 14.4 concludes this chapter.
14.2 CONSENSUS-BASED GLOBAL INFORMATION
DISCOVERY ALGORITHM
According to the fully distributed solution, calculation and distribution of global
information are achieved through an average-consensus-based information discovery
A Consensus-Based Fully Distributed Load Management Algorithm
335
algorithm. For convenience, the average consensus theorem is briely introduced as
follows.
14.2.1
INTRODUCTION TO AVERAGE CONSENSUS THEOREM
The average consensus theorem relies on the local information of agents to guarantee that important information is shared in a distributed way [12]. Recently, the
consensus algorithm has attracted great interest in various areas; a recent survey can
be found in [13].
According to the consensus theory, the overall information discovery process is
modeled as a discrete time-linear system, as in Equation 14.1.
X k +1 = DX k
(14.1)
where Xk and Xk + 1 are the vectors of discovered information at the kth and (k + 1)th
iterations, respectively, and D is a sparse iteration matrix.
∑
1 −
a1 j

j∈N1

 ⋯

D =  ai1


 ⋯

 an1

⋯
a1i
⋯
⋯
⋯
⋯
⋯
1−
∑a
⋯
ij
j∈Ni
⋯
⋯
⋯
⋯
ani
⋯




⋯

ain



⋯


1−
anj 
j∈N n

a1n
∑
If all elements of D are nonnegative and the sums of each row and column are
all ones, that is, D*e = e and DT*e = e, with e = [1, 1, …, 1]T, then D is called a doubly stochastic matrix [14]. According to the theorem of Gerschgorin’s Disks [15], it
can be concluded that all eigenvalues of D are ≤1. Based on the Perron–Frobenius
lemma [15], we have
J = lim D k =
k →∞
eeT
n
(14.2)
where n is the dimension of D.
According to Equations 14.1 and 14.2, the system will reach consensus as k
approaches ininity, as represented in Equation 14.3.
lim X k = lim D k X 0 =
k →∞
k →∞
eeT 0
X
n
(14.3)
From Equation 14.3, one can see that the speed of the consensus-based information discovery algorithm is determined by D, and the coeficients aij need to be
336
Smart Grids
properly designed to guarantee stability. For bilateral networks such as power systems, D can be designed to be symmetrical, like Ybus and Zbus, that is, aij = aji.
For convergence analysis, a positive deinite Lyapunov function is deined as
follows:
T
( )
Vk = Xk
∆V = V k +1 − V k = ⎡ X k
⎣⎢
T
( ) (D D − I ) X
T
n
≤−
∑∑(
aij + aij2
)(
x kj − xik
)
i =1 j∈Ni
2
+
(14.4)
⎤
⎦⎥
k
1
2
Xk
n
∑ ∑{( n + n − 2 ) a ( x
i
i =1 j∈Ni
n
≤−
∑ ∑ a ⎡⎣1 − ( n + n ) a
ij
i
ij
j
i =1 j∈Ni
2
ij
j
2 ⎤⎦ x kj − xik
(
)
2
k
i
− x kj
)}
2
(14.5)
where ni and nj are the numbers of agents in the neighborhood of agents i and j,
respectively. From Equation 14.5, it can be seen that ΔV ≤ 0, provided that aij satisies
the following condition:
0 < aij <
2
ni + n j
(14.6)
Since it may take too long to reach the exact equilibrium, it is necessary to deine
a terminating criterion. For this purpose, error at iteration k can be deined as in
Equation 14.7 [16]:
E k = X k − X ∞ = Dk − J X 0 ≤ D − J
(
)
k
X 0 − JX 0
(14.7)
where Ek is the error at the kth step. If Ek satisies a predeined precision requirement,
it can be concluded that a consensus has been reached and the number of steps needed
for convergence is approximately determined according to Equation 14.8 [16].
K=
log1/ E
1
1 D−J
(
)
=
−1
log E 1 λ 2
(
)
(14.8)
where λ2 is the second largest eigenvalue of D and E is the error tolerance.
Equation 14.8 indicates that λ2 decides the converging speed and gives the asymptotic number of steps for the error to decrease by a factor of E. Thus, λ2 can be used
for evaluation of algorithm speed and comparison of algorithms. As will be introduced later, the optimal converging speed can be obtained by designing a D with
minimum λ2.
A Consensus-Based Fully Distributed Load Management Algorithm
337
Since only the second largest eigenvalue is of interest to us, power method with
delation can be used to calculate the second largest eigenvalue directly. During the
process, calculation of the other eigenvalues is avoided. For convenience, the delation method is introduced as follows.
The n eigenvalues of D can be arranged as 1, λ2, …, λn, in decreasing order of
magnitude. One is the largest eigenvalue of D, and e is the corresponding eigenvector.
Now consider the following matrix:
D1 = D − λ1
e1e1T
e1e1T
D
= D−J
=
−
e1T e1
n
(14.9)
The eigenvectors of a symmetrical matrix are orthogonal, that is, eiT e j = 0, for
i > 1,
D1ei = Dei − λ1
e1e1T
ei = Dei
e1T e1
(14.10)
Therefore, the eigenvalues of (D−J) are λ2, …, λn, 0. Now λ2 becomes the largest
eigenvalue of (D−J) and can be calculated directly according to the power method [17].
14.2.2
CONSENSUS-BASED INFORMATION DISCOVERY ALGORITHM DESIGN
In this chapter, it is assumed that each bus in a microgrid is assigned a node agent
(NA). Each NA knows local generation and load information, but does not have
direct access to global information. An agent can only communicate with its neighboring agents if they are connected through distribution lines during normal operating conditions.
The consensus algorithm is used to design the communication law for global
information discovery. The information discovery process for agent i is represented
as Equation 14.11.
xik +1 = xik +
∑a (x
ij
k
j
− xik
)
(14.11)
j∈Ni
where:
i
n
xi0
xik and xik +1
aij
Ni
= 1, 2, …, n
is the number of agents (buses) in the microgrid
is the predicted local net power of agent i
are the information discovered (average net power or average agent
index) by agent i at iteration k and k + 1, respectively
is the coeficient for information exchanged between neighboring
agents i and j (0 < aij < 1 if agents i and j are connected through a
distribution line, otherwise, aij = 0)
is the number of indexes of agents that are connected to agent i
338
Smart Grids
The global information discovery process of the whole microgrid can be represented using matrix format, as in Equation 14.12.
X k +1 = X k + A ∗ X k = ( I + A ) X k = DX k
(14.12)
where:
X k = [ x1k , …, xik , … , xnk ]T,
I is an identity matrix.
By making aij = aji, the number of independent coeficients in D is equal to the
number of distribution lines in the microgrid. To achieve good converging speed,
the independent variables aij need to be properly designed. For practical power systems, the iteration matrix D is usually a sparse matrix. According to Equation 14.8,
by minimizing |λ2|, the optimal coeficient setting can be obtained [17]. If the system seldom suffers from loss of distribution lines, the coeficients can be optimized
off-line. However, since off-line optimization requires global information and is
time-consuming, the method is not suitable for online applications during changes
of system coniguration. Thus, there is a need for algorithms that can adjust the coeficients automatically and achieve near-optimal performance.
The Uniform and Metropolis method is used to set the coeficients [16]. An
improved algorithm called Mean Metropolis is proposed, which is inspired by the
Lyapunov stability analysis. Similarly to the Metropolis method, the new algorithm
has the properties of stability and adaptivity. The improvement is in the near-optimal
converging speed, as will be demonstrated later through simulation studies. The
updating rule of the method is represented in Equation 14.13.
2 ( ni + n j + ε )


aij = 1 −
2 ( ni + n j + ε )
j∈Ni

0
∑
j ∈ Ni
i= j
(14.13)
otherwise
where ɛ is a very small number (for large complex systems, ɛ can be set to zero).
The above information discovery algorithm can be applied to calculate Pnet indirectly. If xik in Equation 14.11 is initialized with local net power (PGi−PLi), where PGi
and PLi are the local active power generation and load, respectively, of agent i, xik
will converge to average total net power through information exchange. However,
the numerator, that is, the total net power, is the value that is going to be used. Thus,
the denominator n, which is the number of agents participating in the information
exchange, needs to be determined.
According to the proposed algorithm, n is discovered together with the discovery
of indexes of loads that are connected in the power system. An agent in the power
system is assigned a unique index beforehand. Figure 14.1 illustrates the index discovery process during normal operating conditions. Initially, agent i’s N-dimensional
index vector vi only has one nonzero element at (i, 1) with the value of i. By applying
the above information discovery algorithm to the same elements of the index vectors,
the ith elements of the index vectors will converge to i/N.
339
A Consensus-Based Fully Distributed Load Management Algorithm
Index: 1 2
1 0
i
0
N
0
1
1/N
2
1/N
i
1/N
N
1/N
0
2
0
0
2/N
2/N
2/N
2/N
0
0
i
0
i/N
i/N
i/N
i/N
0
0
0
N
N/N
N/N
N/N
N/N
Initial index vectors
Converged index vectors
FIGURE 14.1 Initial and inal index vectors during normal operating conditions.
14.2.3
CONVERGENCE SPEED OF INFORMATION DISCOVERY ALGORITHM
The proposed information discovery algorithm is independent of system coniguration and size. Theoretically, the algorithm can be applied to large-scale power
systems. The authors limit the application to microgrids because of the following considerations. First, microgrids are smaller and distribution lines are shorter.
Thus, the algorithm has a better chance to provide real-time solutions, depending
on how “real time” is deined. Second, because of the intermittent renewable energy
resources, the need for load shedding and restoration is more frequent, especially
when microgrids are operating in the islanded mode.
The proposed algorithm is tested with the 162-bus power systems to investigate
the possibility of applications to large-scale power systems. The coeficients are
designed based on the Mean Metropolis method. The second largest eigenvalue of
the corresponding iteration matrices D for the 162-bus systems is 0.9792. According
to Equation 14.8, the estimated number of steps needed for achieving consensus with
tolerance 0.03 is 165, which is supported by Figure 14.2.
According to Equation 14.8, one can see that the speed of information discovery
process is decided not by the size of the system, but by how the buses are connected
and how the coeficients are decided. However, Equation 14.8 is not accurate enough,
because some assumptions had to be made during the analysis. For example, the
estimated number of steps (71 steps) is less accurate for the 16-bus system. This is
Average net power
50
0
–50
0
20
40
60
80
100
120
Number of iterations
140
160
180
FIGURE 14.2 Average total net power discovery process of the IEEE 162-bus system.
200
340
Smart Grids
because λ2 is not large enough, such that the remaining eigenvalues have more impact
on the convergence.
From the algorithm design, one can see that the converging speed of the information discovery algorithm is independent of the initial values. This property can also
be observed from Equation 14.8. For veriication, the local net power of each agent
is set to a random value as opposed to the rated value. Simulation results show that
the converging speed of the system with random loads is the same as that with rated
loads. This property makes converging speed analysis and implementation easy.
The speed of the information discovery algorithm is evaluated using number of
iterations instead of time. Even though the actual time is decided by speciic implementation of hardware and software, it is preferable to make a rough estimation. By
neglecting the time used for information update, the time needed for information
discovery can be estimated according to the following equation:
T=
na ∗ ( 3 ∗ nb ) ∗ nc
nd
(14.14)
where:
na is the number of iterations needed for convergence
(3*nb) is the size of the information matrix
nc is the number of bits used to represent each element of the information matrix
nd is the communication speed in terms of bits per second
For example, assume that the information discovery process of the Institute of
Electrical and Electronics Engineers (IEEE) 162-bus system requires 160 iterations
to converge and each element of the information matrix is represented using 16 bits.
For a network speed of 10 Mbit/s, the time used for information discovery can be
calculated as 165(3 × 162)16/10,000,000 = 0.1283 s. Thus, even if the algorithm is not
fast enough for real-time control, it is suficient for most planning and optimization
problems in large-scale power systems.
It should be noted that the above estimation is not accurate, but it can give a
rough idea of the speed of the information discovery process. With the development
of network techniques, the algorithm can be implemented faster and faster, so that
application to large-scale power systems will no longer be a problem.
14.2.4
CONSIDERATIONS DURING IMPLEMENTATION
14.2.4.1 Distributed Synchronization
It should be noted that synchronization of autonomous agents is a challenging problem. For a fully distributed MAS solution, no centralized coordinative mechanism
can be assumed, not even a common clock signal. The synchronization has to be
accomplished through communication between distributed agents. Thus, it is impossible to rigorously synchronize the activities of autonomous agents in terms of time.
Due to the randomness of autonomous activities, the speed of the presented algorithm
is evaluated using number of iterations. Below are the preliminary solutions for three
A Consensus-Based Fully Distributed Load Management Algorithm
341
major synchronization problems: coordinating the information update of neighboring agents, terminating information discovery for decision making, and terminating
decision making for information discovery.
To synchronize the information updates of agents, each agent maintains an index
for its current step of information discovery. Two neighboring agents will exchange
and update information only if their indexes are the same. If their indexes are not
the same, the agent with the larger index will wait till the agent with the smaller
index catches up. After a successful information update, both agents increase their
indexes by one. In this way, the information exchange and update activities of
autonomous agents can be synchronized. Due to unexpected changes of operating
conditions, the required number of iterations for a successful information discovery is unpredictable. Thus, the upper limit of the index can be set to a very large
value. Alternatively, to save memory requirement, a smaller upper limit can be
used and the index will be reset to one once it hits the upper limit.
Since the information discovery instances of agents may converge at different
times, there is a need to terminate the overall MAS’s information discovery process
and start the load-shedding process as soon as possible. Similarly, the information
discovery process needs to be restarted once a load management decision has been
deployed. The above operations can be accomplished by distributing signs of convergence and decision making, respectively.
Once agent i concludes that it has found the global information based on Equation
14.11, the agent will set its sign of convergence to one and propagate the nonzero
value to other agents. If agent j notices that the ith element of the convergence status
vector is nonzero, agent j knows that the information discovery process of agent i
has converged. Finally, if an agent notices that all elements of the convergence status
vector are nonzero, it concludes that the overall system’s information discovery process has converged. The agent will start the load-shedding decision-making process
immediately.
The last converged agent will make the load-shedding decision based on the discovered global information, deploy its corresponding local decision, and set its sign
of decision making to one. Once an agent notices a nonzero value in the decisionmaking vector, it will set its own sign of decision making to one while making and
deploying the load-shedding decision. The process proceeds until an agent inds that
all elements in the decision-making vector are nonzero. After deploying its local
load-shedding decision, the agent will restart the information discovery process by
initializing the convergence status vector to all zeros and the index for information
exchange to one.
14.2.4.2 Plug and Play
The proposed algorithm can be applied to power systems with capability for
plug-and-play operation. If a new generator or load is attached to an existing
agent, only local net power needs to be updated. Thus, some redundant buses
and agents can be designed for the purpose of plug-and-play operations. Since no
new agent is involved during plug-in operation, no ID needs to be dynamically
assigned. Similarly, IDs of existing agents do not need to be updated during
plug-outs.
342
Smart Grids
However, special attention is required during changes of network topology. If
a new bus is added to a power system, the corresponding new agent will require a
unique ID to interact with existing agents. If an existing agent is removed, indexes
of the remaining agents may need to be updated to simplify the plugging-in of new
agents. The agent ID acquisition and updating process is introduced in the following
paragraphs.
First, let us consider the scenario of a new agent/bus being plugged in. The new
agent will start with a zero vector to participate in the information discovery with
other agents. The size of the zero vector is the same as the sizes of existing index
vectors, which are larger than N to accommodate future network expansion. For
simpliication, only useful information is illustrated in the following igures. Since
N + 1 agents participate in information discovery, the ith element of the index vector
will converge to i/(N + 1) instead of i/N as before. Based on the converged vector, the
new agent will know that its ID should be N + 1. The ID discovery process of the
new agent is illustrated in Figure 14.3, and the information discovery process after
ID acquisition is shown in Figure 14.4.
Second, let us consider the scenario in which agent i was disconnected from
the system. To simplify the future ID acquisition process, the IDs of the remaining agents may need to be updated. The detection of agent disconnection and the
updated information discovery processes are illustrated in Figures 14.5 and 14.6,
respectively.
1 2
1 0
i
0
New
N agent
0 0
0 2
0
0
0 0
i
0 0
0
New
agent
1
1/(N + 1)
2
1/(N + 1)
i
1/(N + 1)
N
1/(N + 1)
1/(N + 1)
0
2/(N + 1)
2/(N + 1)
2/(N + 1)
2/(N + 1)
2/(N + 1)
0
0
i/(N + 1)
i/(N + 1)
i/(N + 1)
i/(N + 1)
i/(N + 1)
N
0
N/(N + 1)
N/(N + 1)
N/(N + 1)
N/(N + 1)
N/(N + 1)
FIGURE 14.3 ID discovery process of the new plugged-in agent.
1 2
i
N N+1
1
2
i
N
N+1
1 0
0
0
0
1/(N + 1)
1/(N + 1)
1/(N + 1)
1/(N + 1)
1/(N + 1)
0 2
0
0
0
2/(N + 1)
2/(N + 1)
2/(N + 1)
2/(N + 1)
2/(N + 1)
0 0
i
0
0
i/(N + 1)
i/(N + 1)
i/(N + 1)
i/(N + 1)
i/(N + 1)
0 0
0
N
0
N/(N + 1)
N/(N + 1)
N/(N + 1)
N/(N + 1)
0 0
0
0 N+1
(N + 1)/(N + 1) (N + 1)/(N + 1)
N/(N + 1)
(N + 1)/(N + 1)
(N + 1)/(N + 1) (N + 1)/(N + 1)
FIGURE 14.4 Information discovery process after ID acquisition.
343
A Consensus-Based Fully Distributed Load Management Algorithm
1 2
i–1 i+1
1 0
0
0
N
1
2
i–1
i+1
N
0
1/(N – 1)
1/(N – 1)
1/(N – 1)
1/(N – 1)
1/(N – 1)
0 0
i–1
0
0
0 0
0
0
0
0 0
0
i+1
0
0 0
0
0
N
(i – 1)/(N – 1) (i – 1)/(N – 1)
0
(i + 1)/(N – 1) (i + 1)/(N – 1)
N/(N – 1)
(i – 1)/(N – 1) (i – 1)/(N – 1)
0
N/(N – 1)
0
0
(i + 1)/(N – 1) (i + 1)/(N – 1)
N/(N – 1)
(i – 1)/(N – 1)
0
(i + 1)/(N – 1)
N/(N – 1)
N/(N – 1)
i+1
N
FIGURE 14.5 Detection of the disconnection of agent i.
i+1
N
N–1
1
2
i–1
i
N–1
1 0
0
0
0
1/(N – 1)
1/(N – 1)
1/(N – 1)
1/(N – 1)
1/(N – 1)
0 2
0
0
0
2/(N – 1)
2/(N – 1)
2/(N – 1)
2/(N – 1)
2/(N – 1)
i–1 0
0
1 2
0 0
i–1 i
0 0
0
i
0
0 0
0
0
N–1
(i – 1)/(N – 1) (i – 1)/(N – 1)
i/(N – 1)
i/(N – 1)
(N – 1)/(N – 1) (N – 1)/(N – 1)
(i – 1)/(N – 1) (i – 1)/(N – 1)
i/(N – 1)
i/(N – 1)
(N – 1)/(N – 1) (N – 1)/(N – 1)
(i – 1)/(N – 1)
i/(N – 1)
(N – 1)/(N – 1)
FIGURE 14.6 Updated information discovery process.
14.3
14.3.1
APPLICATIONS
MAS-BASED LOAD MANAGEMENT
The introduced algorithm is implemented using NAs, the number of which is equal
to the number of buses. The function modules in an NA and the MAS-based load
management operation are illustrated in Figure 14.7.
In Figure 14.7, the initialization module will initialize local information vectors based on the operating condition of the corresponding load. The information
exchange module will exchange the current information vectors with its neighboring agents and provide required information to the information update module. The
information update module will update aij during change of conigurations and keep
updating the information vector till it converges. The converged information vector is then sent to the information analysis module to obtain global operating conditions (global net power and operating conditions of loads) for load management
decision making. A negative total net power means that the microgrid cannot support the load demand, and load shedding is required to avoid further damage. On
the other hand, a positive average total net power means that the microgrid may be
able to satisfy the demands of some unfaulted but out-of-service loads. In that case,
the load restoration process will be activated.
344
Smart Grids
Load controller
Load
management
decision
Node agent i
Loads
selection
Information
analysis
System operating
condition
Initialization
Consensus
xi∞ =
xi0
Neighboring
agents j
j ∈Ni
xik+1
xjk, j ∈Ni
Information
exchange
1
n
n
∑ xi0
i=1
xik+1
Information update
xjk, j ∈Ni
xik+1 = xik +∑ aij xik–xik
n
j=1
(
)
FIGURE 14.7 Function of NA in MAS load management.
The optimal load management decision is then deployed by load controllers. An optimal load management decision can be obtained by satisfying the following two constraints:
• Maximizing the amount of loads while ensuring loads with higher priorities
are served irst, that is,
N
max
∑P ⋅P
Li
i
(14.15)
i =1
where Pi is the priority index associated with the ith load.
• Loads to be served must be less than or equal to excessive generation, that is,
N
∑
i =1
N
PGi −
∑P
Li
≥0
(14.16)
i =1
Identifying the loads to be served can be modeled as a 0–1 knapsack problem, which
is an NP-hard combinatorial optimization problem. DP is one of the commonly used
techniques for this kind of problem [18]. For large-scale systems, it is very dificult to ind
the optimal solution within a limited time. Greedy algorithms, such as greedy by proit,
greedy by weight, and greedy by proit density, can be applied to reduce the computational burden [19]. According to these greedy algorithms, items with higher proits, larger
weights, and larger ratio of proit over weight have higher priorities. Greedy algorithms
are usually faster than DP but cannot guarantee optimal solutions. Performances of these
algorithms will be compared through an example in this section.
14.3.2
LOAD SHEDDING
Faults, sudden dramatic load change, and insuficient generation can create power
mismatch between generation and loads. If generation in power systems is insuficient
A Consensus-Based Fully Distributed Load Management Algorithm
4
2
3
5
d23
d24
G
d35
d12
345
G
d13
1
FIGURE 14.8 IEEE 5-bus power system.
TABLE 14.1
Data of the IEEE 5-Bus System
Neighbors
PGi
PLi
X i0
Priority
4
2
100
0
100
—
5
3
150
0
150
—
3
1, 2, 5
0
150
−150
P1
1
2, 3
0
50
−50
P2
2
1, 3, 4
0
100
−100
P2
No.
to power all loads, eficient load-shedding operations may need to be deployed to
maintain the supply–demand balance. Load shedding is the process of tripping a
certain amount of load with lower priority to maintain the stability of the rest of the
system.
The proposed MAS-based load-shedding algorithm [10] is tested with the IEEE
5-bus shown in Figure 14.8. The original data on system conigurations are modiied
and summarized in Table 14.1.
14.3.2.1 Normal Operating Conditions
The ive independent variables a12, a13, a23, a24, and a35 are the coeficients for information exchange between neighboring agents. Eigenvalues of D after optimization
are summarized in Table 14.2.
The net power discovery process is shown in Figure 14.9. From the igure, it can
be seen that the agents will reach consensus within 10 iterations.
From Figure 14.9, it can be seen that all agents’ discovered information converged
to the same value of −10, which is the average net power of the system. To ind the
total net power in the system, the number of agents that are currently connected has
to be found. The process is illustrated in Table 14.3.
During the process, every agent is initialized with a 5 × 1 index column vector.
For agent i, the initial index vector only has one nonzero element at (i, 1) with a
value of i. Since there is no agent/bus loss, the ith elements of all index vectors will
converge to i/5. According to the converged index vector, each agent can ind out that
346
Smart Grids
TABLE 14.2
Independent Coeficients and Eigenvalues
after Optimization
Independent Coeficients
Eigenvalues of D
1/3
1/3
1/3
1/2
1/2
a12
a13
a23
a24
a35
1.0000
0.7071
0.4082
λ1
λ2
λ3
λ4
λ5
−0.4082
−0.7071
150
Agent 1
Agent 2
Agent 3
Agent 4
Agent 5
Net power
100
50
0
–50
–100
–150
0
2
4
6
8
10
12
14
16
18
20
Number of iterations
FIGURE 14.9 Average total net power discovery process of the IEEE 5-bus system.
the number of agents in the system is ive and their indexes are 1–5. Thus, the total
net power can be calculated as −10 × 5 = −50. Now all necessary information for
load shedding has been obtained.
In Table 14.1, the three loads are classiied into two priority levels, with P1 > P2
in terms of priority. In the same priority level, the loads are sorted according to the
amount of local net power. The results of different load-shedding algorithms are
shown in Table 14.4.
Compared with DP, the greedy algorithms may not be able to guarantee the best
energy utilization but are more computationally eficient. Thus, DP is a better choice
for smaller systems and greedy algorithms are a better choice for large-scale systems.
Since all agents maintain the same load table and use the same load-shedding algorithm, the load-shedding decisions of distributed agents will be the same and the
load-shedding activities of the overall system will be coordinated. The information
discovery process after load 1 has been disconnected is shown in Figure 14.10.
From Figure 14.10, one can see that the initial local net power of Agent 1 is set
to zero because its corresponding load has been disconnected. Since the newly
discovered net power is 0, no additional load shedding is necessary.
347
A Consensus-Based Fully Distributed Load Management Algorithm
TABLE 14.3
Discovery of Number of Agents in Power System
Agent
Initial Index
Vector
Converged Index
Vector
Number
of Agents
1
2
3
4
5
[1, 0, 0, 0, 0]T
[0, 2, 0, 0, 0]T
[0, 0, 3, 0, 0]T
[0, 0, 0, 4, 0]T
[0, 0, 0, 0, 5]T
[0.2, 0.4, 0.6, 0.8, 1.0]T
[0.2, 0.4, 0.6, 0.8, 1.0]T
[0.2, 0.4, 0.6, 0.8, 1.0]T
[0.2, 0.4, 0.6, 0.8, 1.0]T
[0.2, 0.4, 0.6, 0.8, 1.0]T
5
5
5
5
5
TABLE 14.4
Comparison of Different Algorithms for
Load Shedding
Algorithm
Loads To Be Shed
Dynamic programming
Greedy by proit
Greedy by weight
Greedy by proit density
1(−50)
1(−50)
1(−50)
2(−100)
150
Agent 1
Agent 2
100
Net power
Agent 3
50
Agent 4
Agent 5
0
–50
–100
–150
0
2
4
6
8
10
12
Number of iterations
14
16
18
20
FIGURE 14.10 Average total net power discovery process after load 1 is disconnected.
14.3.2.2 Reliability
To test the robustness of the proposed algorithm, the three operating conditions analyzed during robustness analysis are investigated, which are loss of distribution line,
disconnection of load, and loss of agent during an uninished information discovery
process.
First, consider the loss of distribution line between Buses 1 and 2, which can be
represented as a12 = 0. Within 16 steps, the agents will reach the same consensus
348
Smart Grids
as before. Since the coeficients are no longer optimal, the information discovery
process takes more iterations to converge. This example proves that the algorithm is
robust to loss of distribution lines. The information discovery process for this case
is shown in Figure 14.11.
Second, assume the load that was originally connected to Agent 2 is disconnected.
In response to the change of operating condition, Agent 2 reinitializes local net
power to 0 and index vector to 0 s in all cases. The information discovery processes
of average net power and indexes of connected loads are illustrated in Figures 14.12
and 14.13, respectively.
From Figure 14.12, it can be seen that Agent 2 still works in the same way as
before, even though its associated load is disconnected. From Figure 14.13, one can
see that all agents except Agent 2 will converge to the same value (i/N) as before.
Since i is the predeined index of agent i, the number of agents that are currently
150
Agent 1
Agent 2
100
Net power
Agent 3
50
Agent 4
Agent 5
0
–50
–100
–150
0
FIGURE 14.11
2
4
6
8
10
12
Number of iterations
14
16
20
Average total net power discovery process after a distribution line loss.
150
Agent 1
Agent 2
Agent 3
Agent 4
Agent 5
100
Net power
18
50
0
–50
–100
–150
0
2
4
6
8
10
12
Number of iterations
14
16
18
FIGURE 14.12 Average total net power discovery process after load 2 is disconnected.
20
349
A Consensus-Based Fully Distributed Load Management Algorithm
5
1st row
2nd row
3rd row
4th row
5th row
Index
4
3
2
1
0
0
2
4
6
8
10
12
Number of iterations
14
16
18
20
FIGURE 14.13 Network structure information discovery process after load 2 is
disconnected.
connected in the power system can be easily determined as ive. By checking the
positions of the zeros, all agents know that the load on Bus 2 is now disconnected.
The obtained information on total net power and load connection is suficient for
distributed load shedding.
Last, consider the loss of Agent 4 during an ongoing information discovery process. Since the working agents do not know that a reinitialization is necessary, they
will continue the uninished information discovery process until convergence. By
inspecting the converged index vectors, the agents can easily identify the fault and
restart the information discovery process with updated information. The above process can be observed in Figures 14.14 and 14.15.
It should be noted that, the earlier a fault happens during the information discovery process, the larger is its impact on accuracy. If the fault happens approaching the
end of the information discovery process, its impact may be negligible.
150
Agent 1
Agent 2
Agent 3
Agent 4
Agent 5
Net power
100
50
0
–50
–100
–150
0
5
10
15
Number of iterations
20
FIGURE 14.14 Average total net power discovery process of agent loss.
25
30
350
Smart Grids
5
1st row
2nd row
3rd row
4th row
5th row
Index
4
3
2
1
0
0
5
10
15
Number of iterations
20
25
30
FIGURE 14.15 Network structure information discovery process of agent loss.
14.3.3
LOAD RESTORATION
When a fault occurs in a microgrid system, corresponding circuit breakers will be
opened to isolate the fault so as to protect the power equipment. During the process,
some unfaulted loads are forced to be out of service. Restoring power supply to
these de-energized but unfaulted loads as soon as possible is of great importance.
Eficient restoration methods are very important for energy use and service quality. Restoration may require the investigation of massive combinations of switching
operations that satisfy multiple objectives under certain constraints.
The proposed MAS-based load restoration algorithm [11] is now applied to the
16-bus system [18] shown in Figure 14.16. Data are modiied as shown in Table 14.5.
In Table 14.4, PGi is the local generation, PLi is the local load, Xi0 is the local net
power, and P1–2 is the priority of the loads, with P1 > P2.
2
3
4
10
1
5
11
6
7
8
9
12
14
15
Connected
Disconnected
FIGURE 14.16 The 16-bus test power system.
13
16
351
A Consensus-Based Fully Distributed Load Management Algorithm
TABLE 14.5
Data of the 16-Bus System
Neighbors
PGi
PLi
Xi0
Priority
1
2
3
4
5
6
7
8
9
10
11
12
2, 5
1, 3, 5, 10
2, 10
10, 11
1, 2, 6
5, 7, 9
6, 8, 9
7, 9
6, 7, 8, 11, 15
2, 3, 4
4, 9, 12
11, 13
40
40
50
0
0
0
0
0
0
0
40
0
0
0
0
20
35
20
15
25
5
30
0
45
40
40
50
−25
−35
−20
−15
−25
−5
−30
40
−45
—
—
—
P2
P2
P1
P2
P1
P1
P2
—
P2
13
14
15
16
12, 14
13, 16
9
14
0
0
30
50
10
40
0
0
−10
−40
30
50
P2
P2
—
—
No.
2
3
4
Fault
10
1
5
11
6
7
8
9
12
14
15
Connected
13
16
Disconnected
FIGURE 14.17 Postfault coniguration of the 16-bus system.
Assume that a sudden fault occurs to the generator on Bus 11. The protection
system reacts immediately to isolate the fault by opening ive switches: S2,10, S3,10,
S5,6, S9,15, and S12,13 [20]. The postfault system after protective action is illustrated
in Figure 14.17. For the postfault system, the generator at Bus 11 cannot be restored
immediately and the eight loads attached to Agents 4 and 6–12 should be restored as
soon as possible. The initial information matrices of the 16 agents and the converged
matrix are shown in Table 14.6.
352
TABLE 14.6
Initial and Converged Information Matrices
Initial Matrices M
T
0

0
0

⎡0
⎢
⎢0
⎢⎣0
0
5
−35
0⎤
⎥
0⎥
0 ⎥⎦
⎡0
⎢
⎢0
⎢⎣0
0
9
−5
⎡0
⎢
⎢0
⎢⎣0
13
0
−10
0⎤
⎥
0⎥
0 ⎥⎦
T
5
T
0

0
0  2
⎡0
⎢
⎢0
⎢0
⎣
0
6
−20
⎡0
⎢
⎢0
⎢⎣0
9
T
0⎤
⎥
0⎥
0 ⎥⎦13
T
2
0
40
⎡0
⎢
⎢0
⎢⎣0
0
10
−30
14
0
−40
0

0
0
0
0
0
T
0⎤
⎥
0⎥
0 ⎥⎦ 6
0⎤
⎥
0⎥
0 ⎥⎦
T
⎡0
⎢
⎢0
⎢0
⎣
⎡0
⎢
⎢0
⎢
⎣0
10
T
0⎤
⎥
0⎥
0 ⎥⎦14
T
⎡0
⎢
⎢0
⎢0
⎣
0

0
0

0

0
0  3
3
0
50
T
0⎤
⎥
0⎥
0 ⎥⎦ 7
0
7
−15
T
15
0
30
⎡0
⎢
⎢0
⎢0
⎣
⎡0
⎢
⎢0
⎢0
⎣
0⎤
⎥
0⎥
0 ⎥⎦11
0
0
0
T
0⎤
⎥
0⎥
0 ⎥⎦15
T
0

0
0  4
0
4
−25
0
12
−45
⎡0
⎢
⎢0
⎢0
⎣
T
0⎤
⎥
0⎥
0 ⎥⎦ 8
0
8
−25
T
0⎤
⎥
0⎥
0 ⎥⎦12
T
16 ⎤
⎥
0⎥
50 ⎥⎦16
⎡ 1 / 16
⎢
⎢ 2 / 16
⎢ 3 / 16
⎢
⎢ 0
⎢ 5 / 16
⎢
⎢ 0
⎢
⎢ 0
⎢ 0
⎢
⎢ 0
⎢ 0
⎢
⎢ 0
⎢
⎢ 0
⎢13 / 16
⎢
⎢14 / 16
⎢15 / 16
⎢
⎣⎢16 / 16
0
0
0
4 / 16
0
6 / 16
7 / 16
8 / 16
9 / 16
10 / 16
0
12 / 16
0
0
0
0
40 / 16 ⎤
⎥
40 / 16 ⎥
50 / 16 ⎥
⎥
−25 / 16 ⎥
−35 / 16 ⎥
⎥
−220 / 16 ⎥
⎥
−15 / 16 ⎥
−25 / 16 ⎥
⎥
−5 / 16 ⎥
−30 / 16 ⎥
⎥
0 / 16 ⎥
⎥
−45 / 16 ⎥
−10 / 16 ⎥
⎥
−40 / 16 ⎥
30 / 16 ⎥
⎥
50 / 16 ⎥⎦
Smart Grids
0

0
0 1
1

0
 40

Converged Matrix M
353
A Consensus-Based Fully Distributed Load Management Algorithm
For the eight agents in the working area, that is, Agents 1–3, 5, and 13–16, Mi,1–3
are initialized with the corresponding index, 0, and local net power. For Agent 11,
since it is not ready for restoration yet, Mi,1–3 are initialized with 0, 0, and 0. Since
Agents 4 and 6–12 are ready for restoration, the Mi,1–3 in the initial matrices are
initialized with 0, index, and load demands. By applying the proposed information
discovery algorithm to every element of the 16 initial matrices, all 16 information
matrices will converge to the same matrix, as shown in Table 14.5. For example,
the (1, 1) elements of the 16 matrices are 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.
According to Equation 14.11, the (1, 1) element of the converged matrix will be 1/16.
Since load restoration is possible only when excessive generation becomes available, total net power should be checked irst. The average total net power can be
calculated by summing up the elements in the third column that have a nonzero element in the irst column. For convenience, the information discovery process of the
average total net power in the microgrid is shown in Figure 14.18.
From Figure 14.18, it can be seen that all agents’ discovered information converged to the same value of 7.81 = (40 + 40 + 5 – 35 – 10 – 40 + 30 + 50)/16. Since the
average total net power is positive, it is possible to restore part or all of the unfaulted
but out-of-service loads. To identify loads to restore, it is necessary to ind the indexes
and demands of the ready-to-restore loads.
The indexes and demands of loads that are ready for restoration are discovered
simultaneously, as shown in Figures 14.19 and 14.20. For simpliication, only the
(i, 2) and (i, 3) elements of M are displayed for the ith ready-to-restore agent. During
the information discovery process, the information matrices are initialized with
local information and all converge to the corresponding average values.
From Figures 14.18 through 14.20, it can be seen that the information discovery process converges well within 90 iterations. Considering the complexity of the
16-bus microgrid, the speed of the algorithm is more than acceptable. It should be
noted that the information discovery process is evaluated based on number of iterations, not time. This is because actual communication time is decided by speciic
implementation (hardware and software) and is thus hard to quantify. Since the
exchanged information is very simple (index and local net power), it can be predicted
that the time used to reach a consensus is very short, especially with the development
of computer and communication techniques.
Average net power
60
40
20
0
–20
–40
0
10
20
30
40
50
60
Number of iterations
70
80
90
FIGURE 14.18 Average total net power discovery process of the 16-bus system.
100
354
Smart Grids
12
Agent 4
Agent 6
Agent 7
Agent 8
Agent 9
Agent 10
Agent 11
Agent 12
10
Index
8
6
4
2
0
0
10
20
30
40
50
60
Number of iterations
70
80
90
100
FIGURE 14.19 Indexes discovery process of out-of-service loads.
Average net power
10
0
Agent 4
Agent 6
Agent 7
Agent 8
Agent 9
Agent 10
Agent 11
Agent 12
–10
–20
–30
–40
0
10
20
30
40
50
60
Number of iterations
70
80
90
100
FIGURE 14.20 Demands discovery process of out-of-service loads.
From the converged vector, all required information can be easily calculated.
Since the number of agents in the system is ixed, each agent can easily ind the true
values from the average values. By checking the nonzero elements in the irst column
and the corresponding elements in the third column, the indexes and local net power
of working generators and loads can be obtained. Similarly, by checking the nonzero
elements in the second column and the corresponding elements in the third column,
every agent knows the indexes and demands of loads that are ready for restoration.
By checking Tables 14.5 and 14.6, it can be seen that loads connected to Buses 6,
8, and 9 have higher priority compared with loads connected to Buses 4, 7, 10, and
12. For this example, the priority indexes in Equation 14.15 are selected to be 10 and
1 for P1 and P2, respectively. In addition, all agents know the demands of the outof-service loads and the available power, which is 7.81 × 16 = 125. Now, all required
information for load restoration has been obtained and the loads can be selected by
solving the 0/1 knapsack problem. Here, DP and three greedy algorithms are investigated. Load selection results according to different algorithms are summarized in
Table 14.7 [19].
Compared with DP, greedy algorithms may not provide the best energy utilization. However, greedy algorithms usually require less time for calculation, which is
a very important property, especially for large or complex systems. For smaller-scale
355
A Consensus-Based Fully Distributed Load Management Algorithm
TABLE 14.7
Comparison of Different Algorithms for Load Restoration
Loads Selected for Restoration
Index (Demand)
Algorithm
Dynamic programming
Greedy by proit
Greedy by weight
Greedy by proit density
Utilization
Ratio (%)
6(−20), 8(−25), 9(−5), 10(−30), 12(−45)
8(−25), 6(−20), 9(−5), 12(−45), 10(−30)
12(−45), 10(−30), 4(−25), 8(−25)
9(−5), 6(−20), 8(−25), 7(−15), 4(−25), 10(−30)
100
100
100
96
systems like the example used here, DP is a good choice. For large and complex
systems, greedy algorithms will be a better choice.
The microgrid after restoration is shown in Figure 14.21, in which the generator and load in circles are disconnected by opening the corresponding circuit
breakers. The information discovery process of average total net power is shown in
Figure 14.22. Since the converged value is 0, no more load restoration is possible.
2
3
4
10
1
5
11
6
7
8
9
12
14
13
15
16
Connected
Disconnected
FIGURE 14.21 Coniguration of the 16-bus system after restoration.
Average net power
60
40
20
0
–20
–40
0
10
20
30
40
50
60
70
80
Number of iterations
FIGURE 14.22 Average total net power discovery process after restoration.
90
100
356
Smart Grids
14.4 CONCLUSION
In this chapter, a stable MAS-based load management algorithm is proposed for
microgrids based on the average consensus theorem. Even though each agent only
communicates with its direct neighbors, global information that is needed for distributed load management can be discovered. Compared with existing research on
MAS-based power system applications, the proposed algorithm has the following
advantages. First, stability of the algorithm can be rigorously guaranteed. Second,
the algorithm can be applied to power systems of any coniguration and size. Third,
an improved parameter-setting algorithm, Mean Metropolis, is proposed to adjust
the coeficients for exchanged information. Simulation results demonstrate the effectiveness of the proposed algorithm.
REFERENCES
1. S. Borg, G. Andersson, D.J. Hill, and I.A. Hiskens, On inluence of load modeling
for undervoltage load shedding studies, IEEE Transactions on Power Systems, 13(2),
395–400, 1998.
2. J. Mendoza, R. López, D. Morales, E. López, Ph. Dessante, and R. Moraga, Minimal
loss reconiguration using genetic algorithms with restricted population and addressed
operators: Real application, IEEE Transactions on Power Systems, 21(2), 59–67, 2006.
3. J.G. Vlachogiannis and N.D. Hatziargyriou, Reinforcement learning for reactive power
control, IEEE Transactions on Power Systems, 9(3), 1317–1325, 2004.
4. J.M. Solanki, S. Khushalani, and N.N. Schulz, A multi-agent solution to distribution
systems restoration, IEEE Transactions on Power Systems, 22(3), 1026–1034, 2007.
5. K. Huang, D.A. Cartes, and S.K. Srivastava, A multiagent-based algorithm for ringstructured shipboard power system reconiguration, IEEE Transactions on Power
Systems, 37(5), 1016–1021, 2007.
6. M. Wooldridge, An Introduction to Multiagent Systems, John Wiley & Sons, New York,
2002.
7. T. Logenthiran, D. Srinivasan, and D. Wong, Multi-agent coordination for DER in
MicroGrid, in Proceedings of the IEEE International Conference on Sustainable Energy
Technologies, pp. 77–82, 24–27 November, 2008, Singapore.
8. J.M. Solanki and N.N. Schulz, Using intelligent multi-agent systems for shipboard
power systems reconiguration, in Proceedings of the 13th International Conference
on Intelligent Systems Application to Power Systems, 6–10 November, 2005, Arlington,
VA.
9. A. Dimeas and N.D. Hatziargyriou, Operation of a multiagent system for microgrid
control, IEEE Transactions on Power Systems, 20(3), 1447–1455, 2005.
10. Y. Xu, W. Liu, and J. Gong, Stable multi-agent based load shedding algorithm for power
systems, IEEE Transactions on Power Systems, 26(4), 2006–2014, 2011.
11. Y. Xu and W. Liu, Novel multi-agent based load restoration algorithm for microgrids,
IEEE Transactions on Smart Grid, 2(1), 152–161, 2011.
12. F. Xiao, L. Wang, and Y. Jia, Fast information sharing in networks of autonomous agent,
in Proceedings of the American Control Conference, pp. 4388–4393, 11–13 June, 2008,
Seattle, WA.
13. R. Olfati-Saber, J. Fax, and R. Murray, Consensus and cooperation in networked multiagent systems, Proceedings of the IEEE, 95(1), 215–233, 2007.
14. R.A. Horn, Doubly stochastic matrices and the diagonal of a rotation matrix, American
Journal of Mathematics, 76, 620–630, 1954.
A Consensus-Based Fully Distributed Load Management Algorithm
357
15. R.A. Horn and C.R. Johnson, Matrix Analysis, Cambridge University Press, Cambridge,
UK, 1987.
16. L. Xiao and S. Boyd, Fast linear iterations for distributed averaging, Systems and Control
Letters, 53, 65–78, 2004.
17. R. Olfati-Saber, Algebraic connectivity ratio of Ramanujan graphs, in Proceedings of
2007 American Control Conference, pp. 4619–4624, 11–13 July, 2007, New York.
18. R. Perez-Guerrero, G.T. Heydt, N.J. Jack, B.K. Keel, and A.R. Castelhano, Optimal
restoration of distribution systems using dynamic programming, IEEE Transactions on
Power Delivery, 23(3), 1589–1596, 2008.
19. T.H. Cormen, C.E. Leiserson, and R.L. Rivest, Greedy Algorithms, pp. 370–404. MIT
Press, Cambridge, MA, 2001.
20. T. Nagata, Y. Tao, and H. Fujita, An autonomous agent for power system restoration,
IEEE Power Engineering Society General Meeting, 1, 1069–1074, 2004.
15
Expert Systems
Application for the
Reconiguration of
Electric Distribution
Systems
Horacio Tovar-Hernández and
Guillermo Gutierrez-Alcaraz
CONTENTS
15.1 Introduction .................................................................................................. 359
15.2 Expert System ............................................................................................... 362
15.2.1 Optimal Power Flow ......................................................................... 363
15.2.2 Volt/VAr Control ..............................................................................364
15.3 Case Study .................................................................................................... 365
15.3.1 Scenario 1 ......................................................................................... 366
15.3.2 Scenario 2 ......................................................................................... 371
15.3.3 Scenario 3 ......................................................................................... 372
15.4 Summary ...................................................................................................... 373
References .............................................................................................................. 373
15.1
INTRODUCTION
In the past three decades, intensive effort has been devoted to formulating and solving
distribution network optimal planning and operating problems. Nowadays, advances
in communication and computer technologies have made feasible the development
of more sophisticated methodologies to solve some of the particular problems deriving from those two general challenges. For example, in an operational framework,
advances in distribution automation (DA) technology have substantially improved
control and management capabilities, so that it is possible to reconigure distribution
electric systems directly from an operation control center [1]. Also, time-varying
consumer demand, always an issue, can be addressed by very speciic feeder reconiguration to enable load transfers from heavy or overloaded feeders to feeders with
available load capacity [2].
359
360
Smart Grids
The concept of the smart grid is based on a distribution management system (DMS)
comprising a set of hardware and software applications installed in a control center for
monitoring and controlling distribution systems with a high degree of automation. An
important objective is to enhance system operation eficiency and reliability. For this
purpose, many utilities rely on two-way communications to monitor and control equipment at distribution substations and feeders. A DMS incorporates the measurement
data from DA and advanced metering infrastructure to provide more precise load estimations of feeder sections for integrated Volt/VAr and service restoration controls [3].
The DMS, which is able to maintain information about the current and past topologies
of the electric distribution system, facilitates and automates the creation of switching
orders for planned and unplanned work that operators or ield personnel can execute
according to traditional procedures, or that can be executed automatically [1].
The software infrastructure associated with a DMS can be classiied into the sets of
basic or conventional applications, and advanced or intelligent applications. The former set
includes conventional algorithms such as load lows, short-circuit analysis, state estimation, and load forecasting, while the latter combines these applications with database and
operational criteria management to help operators to obtain the optimal operational set
points of each network component [4]. The DMS is superior to older technology in terms
of lexibility and handling operation planning horizons by acquiring, iltering, and manipulating data for eficient decision making. Within this environment, operators are able to
determine the optimal state of the distribution system as much as an hour in advance [2].
Distribution systems are subject to operational changes that can be either smooth or
critical, depending on the magnitude of the perturbation experienced. If load variations
or less severe failures occur, resulting in some overloads or voltages out of limits, then
a transition from the normal to the emergency state occurs, and some corrective actions
can be performed to restore the system to a normal operational state. A different situation
occurs when a perturbation is a permanent failure; in this case, the distribution system
moves from normal to an emergency state, in which fault-clearing actions are carried out
and some subsequent restorative actions can be performed to take the distribution system
to a new normal operational state. These processes are illustrated in Figure 15.1.
In the normal state, relatively small or smooth changes occur, for example, hourly
variations of load, and normal-state operational conditions may vary. Therefore,
operators can make some decisions to take the distribution system from a nonoptimal
Optimal system
reconfiguration
Normal
state
Operational
changes
Emergency
state
Fault
clearing
Restorative
state
Corrective actions
Restorative actions
FIGURE 15.1 Operational states, transitions between them, and remedial actions in electrical distribution systems.
Expert Systems Application for the Reconiguration of Distribution Systems
361
operational condition to a better operational state by using a short-term operation
planning process comprising the following strategies [4]:
1.
2.
3.
4.
5.
Detection of restriction violations
Corrective action planning
Maintenance outage programming
Loss reduction by reconigurations
Volt/VAr control
A combination of these strategies allows the development of sophisticated
methodologies with the capacity to obtain an optimal operative state. For example,
loss reduction can be optimized by reconiguration and Volt/VAr control.
Under the smart grid framework, a great number of algorithms and methodologies
have been developed to optimize distribution system operations. In particular, solving the
problem of loss minimization is a key factor in laying out circuits as well as predicting
a desired system coniguration for different normal-operation or contingency cases. For
example, many proposals solve the reconiguration problem from a normal-operation
condition perspective involving either one or several load scenarios in order to follow
load variations more accurately and to obtain minimum distribution losses for each load
scenario [5]. Both of these approaches normally change the open/closed status of tie
switches in a progressive manner, until they ind a radial coniguration of the distribution
system with minimum losses, while keeping nodal voltages within limits and avoiding
overloads on each feeder section, including substation transformers. In some cases,
such proposals are based on conventional power low solvers combined with topological search algorithms [6]; in others, they are based on evolutionary programming techniques, including power low solvers [7,8] and sometimes topological search methods [9].
Nevertheless, solving the more complex problem of reconiguration with Volt/VAr
control has not been addressed with the same interest, even though its solution should
result in a greater reduction of distribution losses while still maintaining better nodal
voltage proiles. It has been proposed to solve this problem by multiobjective/multistep
evolutionary programming techniques [9–11] or optimal power low (OPF) Benders
decomposition algorithms [12,13], both including the radial-structure restriction.
Furthermore, expert systems (ESs) have been shown to be useful for dealing with
distribution system operation. ESs have been developed as decision-making support methodologies for distribution system operators to solve problems of restoration, feeder load balancing, voltage control, and loss minimization in distribution
systems [14,15]. These tools make use of computational algorithms such as load
low, topological search, sensitivities, and optimization techniques, in order to obtain
solutions that are feasible and optimal in some sense to the system operator [16].
This combination of artiicial intelligence tools and numerical algorithms evolves
into methodologies that could be more eficient than those based only on very complex numerical algorithms or evolutionary multiobjective methodologies for solving
mixed-integer nonlinear programming [17]. In many cases, however, ESs lack solid
mathematical foundations, so there is no guarantee that they reach an optimal solution; rather, it may be only a good approximation [16]. Nevertheless, due to the discrete nature of tie switching, transformer and regulator tap adjustment, and capacitor
362
Smart Grids
bank commutation, the solutions obtained with ESs and those based on optimization
algorithms or evolutionary programming will be very similar in practical terms.
In this chapter, we propose an ES for optimal loss reduction that uses a coordinated application for reconiguration and Volt/VAr control in the distribution system.
For this purpose, the ES improves a set of solutions proposed by an OPF-based
reconiguration algorithm, executing reactive power control actions by capacitor
bank commutations and regulator tap adjustments.
15.2
EXPERT SYSTEM
As mentioned, intelligent methodologies are needed to support operational decision
making in an information-rich environment. In a dynamic electrical distribution system, the system operator must make decisions in real time to maintain the distribution system in an optimal operational state. ESs, or knowledge-based systems, were
developed approximately 30 years ago to assist system operators in deining appropriate control actions to keep transmission or distribution systems operating under
security restrictions [14,16]. Some ESs have been designed as aids for transmission
and distribution system operators for voltage and reactive power control. In general,
ES applications in electrical engineering are designed as knowledge-based systems
based on heuristics and numerical calculations that give more intelligent solutions to
complex problems than those given by conventional algorithms [16].
Our proposed ES is a rule-based system with a forward-chaining reasoning
that involves two general stages: (1) inding one set of distribution system reconigurations with voltages within limits and no overloads by means of an OPF, and
(2) performing a reactive Volt/VAr control to reduce losses given by the set of reconigurations found in stage (1).
We organized the If-Then rules into steps for making orderly and eficient decisions. In our hypothetical real-time environment where the ES must be able to act for
the next load scenario, each step represents a level of the following solution process:
Step 1. Identify the actual state of switches between feeders, capacitor banks,
and nodal loads for the current hour, deined as h−1.
Step 2. For the next load demand scenario, D(h), ind all n options for reconiguration and classify the options as feasible reconigurations (FRs).
Calculate each FR by the OPF algorithm, setting all capacitor banks in
a disconnected state, so that each FR found does not present overloads or
voltage deviations out of limits. If n = 0, the ES modiies the voltage limits
until the OPF inds at least one FR.
Step 3. For each FR, perform a reactive power control by means of capacitor
bank commutations to investigate their impact over distribution losses until
there is no capacitor bank commutation that reduces losses.
Step 4. Find the reconiguration with the minimum distribution losses, and
name it as the new operating state, E(h).
Step 5. Verify that E(h) ≠ E(h−1). If true, deine the changes needed in switches
and capacitor banks to pass from the E(h−1) state to the E(h) state, and convey
this information to the system operator.
Expert Systems Application for the Reconiguration of Distribution Systems
15.2.1
363
OPTIMAL POWER FLOW
The OPF algorithm obtains reconigurations with voltages remaining within certain
limits [12,13,18]. However, at this stage the solutions of the OPF are allowed to present slight voltage deviations out of limits, which can be corrected by reactive power/
voltage control by capacitor bank switching and adjusting regulator taps. Interfaces
between transmission and distribution are treated as a voltage-controlled node with
one generator. These generators will have the same cost production function, such that
the power sharing among them is not inluenced by the most economical generation
contribution, but by the least generation needed to satisfy all the loads connected to
the distribution system. Hence, OPF solves an economical dispatch for each coniguration r ∈ R, where R is the set of all possible radial network reconigurations. Each
reconiguration r without overloads and voltages out of limits is included in a subset of
FRs. Mathematically, the OPF can be formulated as a nonlinear optimization problem:
ng
Min C T ( r ) =
{Pgi }
∑ ( d + e Pg + f Pg )
i
i
i
i
2
i
(15.1)
i =1
N
S. to Pgk ( r ) − Pdk ( r ) =
∑ V V (G
k
j
kj
cos ( θk − θ j ) + Bkj sin ( θk − θ j )
)
∀k ∈ N
j =1
(15.2)
N
Qgk ( r ) − Qdk ( r ) =
∑ V V (G
k
j
kj
sin ( θk − θ j ) − Bkj cos ( θk − θ j )
)
∀k ∈ N
(15.3)
j =1
∑S
pk
( r ) ≤ S pMax
∀ p ∈ Nt
(15.4)
pk
( r ) ≤ S pMin
∀ p ∈ Nts
(15.5)
k∈K out
∑S
k∈K out
VkMin
Vk ,r
S j ,r ≤ S Max
j
VkMax
(15.6)
j ∈ Nb
(15.7)
Nrs = N − ng
where:
N
Nb
Nt
Kout
Spk
is the total number of nodes
is the total number of branches
is the set of all substation transformers
is the set of feeders receiving power from transformer p
is the power lowing from transformer p to feeder k
(15.8)
364
Smart Grids
S pMax
is the rating of transformer p
θk
is the voltage angle at node k
Pdk
is the active load at node k
Qdk
is the reactive load at node k
Pgk
is the active power injected at node k
Qgk
is the reactive power injected at node k
|Vk|
is the voltage magnitude at node k
VkMin , VkMax are the allowed minimum and maximum voltage magnitude at node k
Sj
is the loading of feeder j
S Max
is the rating of feeder j
j
di, ei, and fi are the production cost factors
Gkj and Bkj are the real and imaginary parts of the admittance matrix element
Ykj, respectively
ng
is the total number of substations or generators
The objective function (Equation 15.1) is the minimization of the production
cost of power provided by the utility. Constraints 15.2 and 15.3 represent the
nodal power balance equations. Constraints 15.4 and 15.5 represent the maximum
and minimum power output limits of the substation or generator, respectively.
Inequality constraints 15.6 and 15.7 are the system operative and security constraints (upper and lower bounds), respectively [13]. Restriction 15.8 is related to
the radial structure of the distribution system, where Nrs is the number of branches
included in the spanning tree representing each one of the possible distribution
system reconigurations [2].
In our formulation, the OPF does not perform reactive power control to minimize distribution losses, chiely because this process is carried out by capacitor bank
switching and regulator tap adjustments, which are discrete control actions, and the
OPF algorithm would have to solve a more complex mixed nonlinear and integer
programming problem. Traditionally, reactive power dispatch for transmission systems has been formulated assuming that these devices are continuous variables in
the OPF formulation. However, commercial capacitor bank capacities are relatively
large (e.g., 300 kVAr modules), so that modeling them as continuous variables in
OPF for distribution systems would lead to impractical results.
15.2.2
VOLT/VAR CONTROL
Traditionally, reactive power control is performed to minimize losses and control
voltages through capacitor bank switching and regulator tap adjustments. The resulting operational state of these devices is maintained during a long period of time
deined by considerable changes in seasonal load patterns. Capacitor bank switching
can be ixed or commutable. In the latter case, commutation is made by local criteria
such as temperature changes, low voltage, and so on. Furthermore, regulator taps are
adjusted by measuring the voltage on a speciic distribution node. However, under
the smart grid paradigm, it is assumed that the control center must monitor and control these devices by considering their beneit to the entire distribution system, that
is, minimization of loss and voltage deviation.
Expert Systems Application for the Reconiguration of Distribution Systems
365
Capacitor bank commutations and regulator tap adjustments should strictly be
modeled as discrete events. To ind the best combination of capacitor bank commutations and regulator adjustments to reduce the distribution losses to a minimum,
the ES performs a dynamic programming process. Experimental results show that
a good practice is irst to analyze the effect of capacitor banks, and then realize the
tap adjustments in regulators, because capacitor banks have a great impact on loss
reduction, while regulator taps have a minor effect, but can be used to make inal
changes in voltage magnitudes, which also help to diminish losses.
In particular, for our proposed computational methodology, we consider that
capacitor banks have a modular capacity based on 300 kVAr blocks. Once the OPF
has obtained the subset of FRs, the ES performs the Volt/VAr control following the
next steps:
1. Select an FR from the subset FR.
2. For the capacitor bank set, select one capacitor block and switch it in or out,
depending on its actual state. Observe its impact on distribution losses; if
they augment, leave this capacitor bank in its actual state; if they drop, add
the capacitor block commutation to a feasible control action (FCA) list and
turn back the capacitor block to its original status. To avoid capacitor bank
commutations with an almost null impact on loss reduction, it is convenient
to establish a threshold value, for example, 5 kW.
3. If there are more capacitor bank blocks to investigate, go back to Step 2. If
not, go to next step.
4. If the FCA is empty, go to Step 6. If not, go to the next step.
5. Order the control actions in the FCA from greater to lesser impact. Once
the FCA has been ordered, the control action that causes the greatest loss
reduction is put into a selected control action (SCA) list.
6. For each regulator, investigate adjustments in its tap, which can reduce the
distribution loss value reached in Step 2. Select the adjustment with the
greatest loss reduction, and add it to the SCA list. To avoid regulator tap
adjustments with an almost null impact on loss reduction, it is convenient to
establish a threshold value, for example, 5 kW.
7. Return to Step 1 if there are more FRs to investigate. If not, create the
deinitive control action (DCA) list and end the process. The inal DCA list
will be the SCA list corresponding to the least-loss reconiguration.
The process with OPF and Volt/VAr control, as described above, is designed for a
load demand scenario h. If there are more scenarios to be analyzed, then h and h−1
are updated, and the whole process is repeated until all the load scenarios have been
analyzed.
15.3
CASE STUDY
A practical distribution network with two contiguous substations is used to illustrate
the ES functioning. It is considered that the distribution system operates under threephase balanced conditions; therefore, the positive sequence network represents the
366
Smart Grids
system network. The complete network data are given in [17], and the corresponding
single-line diagram is depicted in Figure 15.2. Both substations operate independently as radial feeders, but they can be interconnected by one of the normally open
tie-line switches.
Optimal network coniguration problem is solved by brute force combined with
graph theory implemented in MATLAB®. For illustrative purposes, only switches
connecting tie-lines are considered for load transfer (S84, S85, S87, S88, and S91).
However, to preserve a radial structure, when switch S91 operates, switch S21 or
switch S83 should operate simultaneously, depending on where the load transfer takes
place; switches S5, S7, S11, S12, S55, S64, S72, and S76 are in the same situation.
Intra-area switches, for example, switches 89 and 97, are mainly used to achieve load
balancing between feeders belonging to a substation, depending on the diversity and
amount of each feeder demand. Under this consideration, it is possible to operate 15
switches and generate 32,768 possible reconigurations. A subset of this large number
of possible reconigurations will be feasible and the rest will not be feasible due to
overloads, out-of-limit voltages, or both. Table 15.1 shows individual switch operation states for 11 possible reconigurations, which were selected randomly, the only
purpose being to show how our proposal works. However, in a real-life environment,
there should be an algorithm that helps to ind the more promising subset of FRs.
Table 15.2 shows the location and capacity of capacitor banks; they can be connected or disconnected by 300 kVAr blocks. In this case study, regulators do not exist
in the distribution system, and thus they are not included in the reconiguration process.
Also, it is assumed that the ES acts to propose a new reconiguration depending
on the load scenario for the next hour, while taking the current scenario into account.
For illustration purposes, we analyze three of the hourly load scenarios that could
occur during a period of time, namely, a day ahead.
To initiate the ES process, we consider that the distribution system is operating
under the reconiguration 00 shown in Table 15.1, and the nine capacitor banks of
Table 15.2 are disconnected. Additionally, voltage limits are speciied as 0.95 and
1.05 p.u. in all nodes, and we will assume that all feeder sections have the very high
overload limit of 100 MVA.
15.3.1
SCENARIO 1
Scenario 1 presents the loads shown in Table 15.3, for all the nodes of the distribution system. The substation voltages are set to 1.01 p.u. The ive general steps are
developed below:
Step 1. The ES identiies the actual operation state E(0) based on the information presented in Tables 15.1 through 15.3, also considering that the distribution system operates under the 00 reconiguration, and all capacitor
banks are disconnected.
Step 2. The OPF obtains FRs observing overloads and voltages for the 11
reconigurations. For scenario 1, reconigurations 01, 09, and 10 are classiied as unfeasible, while reconigurations 00, 02, 03, 04, 05, 06, 07, and 08
are classiied as feasible, that is, they are included in the subset FR.
Expert Systems Application for the Reconiguration of Distribution Systems
367
Sub 1
B6
B5
B4
B3
B2
B1
Bus 1
43
11
90
12
26
32
92
46
94
28
89
18
29
6
7
21
93
22
38
10
9
8
S84
24
41
39
42
40
S85
S91
95
S88
S87
64
72
55
96
53
83
76
73
77
B11
56
65
B10
B9
47
B8
B7
Sub 2
FIGURE 15.2 Distribution system with two substations and 11 feeders operating in radial
structure.
368
TABLE 15.1
Individual Tie-Line Switch States for the 11 Possible Conigurations
Reconiguration
S5
S7
S11
S12
S21
S55
S64
S72
S76
S83
S84
S85
S87
S88
S91
00
01
02
03
04
05
06
07
08
09
10
1
0
1
1
1
1
1
1
1
1
1
1
1
1
0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0
1
1
1
1
1
1
1
1
0
1
1
1
1
1
1
1
1
1
1
1
1
1
0
1
1
0
1
1
1
1
1
1
1
1
1
1
1
1
0
1
1
1
1
1
1
1
1
1
1
1
1
0
1
1
1
1
1
1
1
1
1
1
1
1
0
1
1
1
1
1
1
1
0
1
1
1
1
1
0
1
1
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0
1
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
1
0
0
0
0
1
Smart Grids
Expert Systems Application for the Reconiguration of Distribution Systems
TABLE 15.2
Capacitor Bank Location
and Capacity (kVAr)
No.
1
2
3
4
5
6
7
8
9
Node
Capacity
6
21
28
37
51
63
71
75
81
2400
2100
2400
1800
2100
1500
2400
2400
2400
TABLE 15.3
Active and Reactive Nodal Demands for Scenario 1
Node
MW
MVAr
Node
MW
MVAr
Node
MW
MVAr
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
0.08
0.24
0.28
0.176
0.88
0.32
0.24
0.24
0.24
0
0.96
0.64
0.56
0
0.24
0.4
0.56
0.96
0.24
0.32
0.04
0.04
0.04
0.04
0.08
0.4
1.44
0.16
0.04
0.16
0.2
0.08
0.4
−0.224
0.16
0.104
0.208
0
0.32
0.48
0.4
0
0.12
0.28
0.32
0.8
0.24
0.28
0.016
0.016
0.008
0.024
0.048
0.28
1.28
0.096
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
0
1.44
0.16
0.16
0.64
0.08
0.08
0.016
0.016
0.016
0.016
0.16
0.04
0
0.024
0.64
0.16
0
0
0
0.16
0.64
0.4
0.4
0.4
0.16
0
0.024
0
1.28
0.12
0.08
0.48
0.048
0.048
0.008
0.008
0.008
0.008
0.128
0.024
0
0.016
0.56
0.12
0
0
0
0.128
0.48
0.24
0.04
0.24
0.064
0
0.016
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
0.48
0
0.016
0.016
0.16
0.24
0.24
0
0.04
0
0.32
0
0
1.6
0.16
0.64
0.4
1.6
0.72
0
0.32
1.6
0.16
0.4
0.08
0.32
0
0
0.096
0
0.008
0.008
0.024
0.192
0.16
0
0.024
0
0.288
0
0
0.96
0.12
0.16
0.096
1.08
0.304
0
0.288
0.8
0.112
0.288
0.024
0.288
0
0
369
370
Smart Grids
Step 3. With the results of Step 2, the ES performs the reactive power control
to minimize distribution losses with the connection of 300 kVAr capacitor
bank blocks for each of the FRs of scenario 1. Once the ES has found all
the capacitor bank blocks that should be connected in order to reduce the
distribution losses to a minimal value, it goes to the next step.
Step 4. Distribution losses for all FRs and the corresponding order with and
without reactive control are shown in Table 15.4. Based on this information,
the ES deines 07 as the best reconiguration, because it has the minimum
distribution losses.
The results from this table show the different order before and after
applying the reactive power control; this strategy, performed by the ES
to minimize losses, holds promise for helping distribution companies to
achieve substantial savings, as is evident from the columns without and
with reactive power control.
Step 5. The ES inds the control actions to be performed to reach the new operating state E(1). The initial and inal network conigurations are coniguration 00 and coniguration 07, respectively.
It can be seen that the only changes needed to pass from the actual reconiguration to the new one are the following:
Control action 1: Close switch S88
Control action 2: Open switch S12
Note that the control action to close tie-line switch S88 is executed irst, in order
to prevent the isolation of some portion of the electrical network. Since we assumed
that capacitor banks were initially in disconnected status, they are connected as
speciied by the next control actions:
Control action 3: Connect 1200 kVAr at node 6
Control action 4: Connect 1200 kVAr at node 21
TABLE 15.4
Distribution Losses for All FRs of Scenario 1
Without Reactive Power Control
Reconiguration
00
02
03
04
05
06
07
08
With Reactive Power Control
Losses (MW)
Ordering
Losses (MW)
Ordering
0.3952
0.4010
0.3930
0.4065
0.3956
0.3921
0.3949
0.3984
4
7
2
8
5
1
3
6
0.3005
0.3040
0.2995
0.3081
0.2993
0.2997
0.2943
0.3042
5
6
3
8
2
4
1
7
Expert Systems Application for the Reconiguration of Distribution Systems
371
Control action 5: Connect 1500 kVAr at node 28
Control action 6: Connect 900 kVAr at node 37
Control action 7: Connect 1200 kVAr at node 51
Control action 8: Connect 1200 kVAr at node 63
Control action 9: Connect 1500 kVAr at node 71
Control action 10: Connect 1500 kVAr at node 75
Control action 11: Connect 1500 kVAr at node 81
In this case, the control action order to connect capacitor bank blocks has been
established according to Table 15.2, because no capacitor-connection sequence will
cause overloads or voltages out of limits, and the inal result of realizing control
actions 4–11 will be the same independently of the execution order.
Note that there are many control actions for reaching the optimal reconiguration.
However, when there are small changes in reconiguration between scenarios, the
control actions will be few, mainly because we already have an optimal solution, as
will be observed when scenario 2 is analyzed.
15.3.2
SCENARIO 2
The loads of scenario 2 are obtained by multiplying the active and reactive loads of
scenario 1 by 1.125. Once again, the substation voltages are set to 1.01 p.u. The ES
carries out Step 1 considering scenario 2, actual reconiguration 07, and the corresponding capacitor bank status shown through the control actions 3–11 listed above.
The results of Step 2 show that reconigurations 01, 02, 04, 09, and 10 are now
unfeasible, while 00, 03, 05, 06, 07, and 08 are feasible. Application of Steps 3 and
4 produces the results shown in Table 15.5, where minimum losses are reached once
again for reconiguration 07, and the three capacitor bank control actions.
For Step 5, the ES displays the next control actions, which are the only changes
with respect to the operational state E(1) for reaching E(2):
Control action 1: Connect 300 kVAr at node 28
Control action 2: Connect 300 kVAr at node 37
Control action 3: Connect 300 kVAr at node 81
TABLE 15.5
Distribution Losses for All FRs of Scenario 2
Without Reactive Power Control
Reconiguration
00
03
05
06
07
08
With Reactive Power Control
Losses (MW)
Ordering
Losses (MW)
Ordering
0.5054
0.5025
0.5060
0.5013
0.5052
0.5096
3
2
5
1
4
6
0.3819
0.3808
0.3811
0.3816
0.3748
0.3874
5
2
3
4
1
6
372
Smart Grids
According to these results, capacitor banks should be connected with the capacity
listed below:
Node 6: 1200 kVAr
Node 21: 1200 kVAr
Node 28: 1800 kVAr
Node 37: 1200 kVAr
Node 51: 1200 kVAr
Node 63: 1200 kVAr
Node 71: 1500 kVAr
Node 75: 1500 kVAr
Node 81: 1800 kVAr
15.3.3
SCENARIO 3
The loads for scenario 3 are obtained by multiplying the loads of scenario 1 by 1.25
and substituting the loads at nodes 73, 74, 75, and 76 with the values of (0.8,0.2),
(0.5,0.12), (2.0,1.35), and (0.9,0.38), MW and MVAr, respectively. In this case, substation voltages are set to 1.02 p.u.
For the realization of Step 1, E(2) is taken as the actual operational state, but considering the corresponding loads of scenario 3.
The results of Step 2 show that feasible and unfeasible reconigurations are the same as
for scenario 2. Table 15.6 shows the solutions obtained by the ES after carrying out Steps
3 and 4. It is clear that now the reconiguration 03 is the new coniguration for scenario 3.
For scenario 3, the control actions displayed by the ES to the distribution system
operator would be the following:
Control action 1: Close switch S12
Control action 2: Close switch S85
Control action 3: Open switch S88
Control action 4: Open switch S7
Control action 5: Disconnect 300 kVAr at node 6
Control action 6: Connect 300 kVAr at node 21
Control action 7: Connect 300 kVAr at node 28
TABLE 15.6
Distribution Losses for All Reconigurations of Scenario 3
Without Reactive Power Control
Reconiguration
00
03
05
06
07
08
With Reactive Power Control
Losses (MW)
Ordering
Losses (MW)
Ordering
0.6881
0.6847
0.6888
0.6829
0.7137
0.6849
4
2
5
1
9
3
0.5192
0.5181
0.5184
0.5187
0.5291
0.5244
5
1
2
3
4
6
Expert Systems Application for the Reconiguration of Distribution Systems
373
Control action 8: Connect 300 kVAr at node 51
Control action 9: Connect 600 kVAr at node 63
Control action 10: Connect 300 kVAr at node 71
Control action 11: Connect 600 kVAr at node 75
Control actions 1–4 are executed to change from reconiguration 07 to 03. Control
action 5 shows that the disconnection of a 300 kVAr capacitor bank block should
take place, while control actions 6–11 indicate the connection of 300 or 600 kVAr
capacitor bank blocks relative to their status in scenario 2. This result appears to be
because the load of scenario 3 is greater than the load of scenarios 1 and 2. The inal
connection statuses of the capacitor banks are:
Node 6: 900 kVAr
Node 21: 1500 kVAr
Node 28: 2100 kVAr
Node 37: 1200 kVAr
Node 51: 1500 kVAr
Node 63: 1800 kVAr
Node 71: 1800 kVAr
Node 75: 1800 kVAr
Node 81: 1800 kVAr
15.4
SUMMARY
According to the results shown in this chapter, it is clear that considering only reconiguration techniques for loss reduction may result in suboptimal reconigurations.
Also, the reactive power/voltage control signiicantly reduces distribution losses
obtained with a reconiguration methodology alone. Thus, we believe that this key
feature for optimizing smart grid performance will give operators the ability to make
“smarter” operational decisions, once the planning for optimal location of capacitors
has been done. We suggest that system operators will ind that the proposed approach
of the ES integrated by the OPF and Volt/VAr control is highly effective for minimizing losses and achieving superior voltage proiles.
Furthermore, it is important to note that the ES-based methodology can be applied
in planning operational scenarios for large periods of time, that is, seasonal load
conditions, because we are considering three-phase balanced distribution system
modeling, speciic load proiles, and normal-state operational conditions for reconiguration, capacitor commutation, and regulator tap adjustments. Also, in the realtime frame, coordinated network reconiguration and Volt/VAr control can handle
the time-varying effects of feeder bus loads and distributed generation outputs.
REFERENCES
1. F. Borjas, A. Espinosa, A. Quintero, B. Sierra, and R. Torres-Abrego, An architecture
for integrating an expert system with NEPLAN in a DMS/EMS operational environment, in Proceedings of the 2009 IEEE International Conference on Systems, Man, and
Cybernetics, 11–14 October, San Antonio, TX, 2009.
374
Smart Grids
2. E. López, H. Opazo, L. García, and P. Bastard, Online reconiguration considering
variability demand: Applications to real networks, IEEE Trans. Power Syst., 19(1),
549–553, 2004.
3. S.-Y. Su, C. Lu, R.-F. Chang, and G. Gutiérrez-Alcaraz, Distributed generation interconnection planning: A wind power case study, IEEE Trans. Smart Grid, 2(1), 181–189,
2011.
4. E. Lakervi and E. J. Holmes, Electric Distribution Network Design, 2nd edn., IEE Power
Engineering Series 21, Peter Peregrinus Ltd, Exeter, 1995.
5. S.-A. Yin and C.-N. Lu, Distribution feeder scheduling considering variable load proile
and outage costs, IEEE Trans. Power Syst., 24(2), 652–660, 2009.
6. S. Civanlar, J. J. Grainger, and S. S. H. Le, Distribution feeder reconiguration for loss
reduction, IEEE Trans. Power Deliver., 3(3), 1217–1223, 1988.
7. C.-C. Liu, S. J. Lee, and K. Vu, Loss minimization of distribution feeders: Optimality
and algorithms, IEEE Trans. Power Deliver., 4(2), 1281–1289, 1989.
8. D. Jiang and R. Baldick, Optimal electric distribution system switch reconiguration and
capacitor control, IEEE Trans. Power Syst., 11(2), 890–897, 1996.
9. D. Zhang, Z. Fu, and L. Zhang, Joint optimization for power loss reduction in distribution systems, IEEE Trans. Power Syst., 23(1), 161–169, 2008.
10. C.-F. Chang, Reconiguration and capacitor placement for loss reduction of distribution
systems by ant colony search algorithm, IEEE Trans. Power Syst., 23(4), 1747–1755,
2008.
11. C. T. Su and C. S. Lee, Feeder reconiguration and capacitor setting for loss reduction of
distribution systems, Electr. Power Syst. Res., 58(2), 97–102, 2001.
12. H. M. Khodr, J. Martínez-Crespo, M. A. Matos, and J. Pereira, Distribution systems
reconiguration based on OPF using Benders decomposition, IEEE Trans. Power
Deliver., 24(4), 2166–2176, 2009.
13. H. M. Khodr, J. Martínez-Crespo, Z. A. Vale, and C. Ramos, Optimal methodology for
distribution systems reconiguration based on OPF and solved by decomposition technique, Euro. Trans. Electr. Power, 20(6), 730–746, 2009.
14. M. M. Salama and A. Y. Chikhani, An expert system for reactive power control of a
distribution system part 1: System coniguration, IEEE Trans. Power Deliver., 7(3),
940–945, 1992.
15. J. R. P.-R. Laframboise, G. Ferland, A. Y. Chikhani, and M. M. A. Salama, An expert
system for reactive power control of a distribution system part 2: System implementation, IEEE Trans. Power Syst., 10(3), 1433–1441, 1995.
16. T. S. Dillon and M. A. Laughton, Expert System Applications in Power Systems, Prentice
Hall, New York, 1990.
17. C. Wang and H. Z. Cheng, Optimization of network coniguration in large distribution systems using plant growth simulation algorithm, IEEE Trans. Power Syst., 23(1),
119–126, 2008.
18. C. J. Dent, L. F. Ochoa, and G. P. Harrison, Network distributed generation capacity
analysis using OPF with voltage step constraints, IEEE Trans. Power Syst., 25(1),
296–304, 2010.
16
Load Data Cleansing
and Bus Load
Coincidence Factors*
Wenyuan Li, Ke Wang, and Wijarn Wangdee
CONTENTS
16.1 Introduction .................................................................................................. 375
16.2 Load Curve Data Cleansing ......................................................................... 377
16.2.1 Basic Concepts .................................................................................. 377
16.2.2 Method for Cleansing Outliers ......................................................... 378
16.2.2.1 Kernel Function Representation of Load Curve ................ 378
16.2.2.2 Conidence Interval for Identifying y-Outliers .................. 379
16.2.2.3 Candidates for x-Outliers ................................................... 380
16.2.2.4 Identifying x-Outliers ........................................................ 382
16.2.3 Repairing Outliers ............................................................................ 384
16.2.4 Selection of Parameters .................................................................... 384
16.3 Bus Load Coincidence Factors ..................................................................... 385
16.3.1 Basic Concept of BLCF .................................................................... 386
16.3.2 Procedure for Calculating BLCF ...................................................... 387
16.4 Application of BLCF to System Analysis..................................................... 389
16.4.1 Studied System and Load Curve Data .............................................. 389
16.4.2 Bus Load Coincidence Factors at the Three Substations ................. 390
16.4.3 Power Flow Analysis Using the Bus Load Coincidence Factors ...... 393
16.5 Conclusions ................................................................................................... 397
References .............................................................................................................. 397
16.1
INTRODUCTION
Load curve data refer to power consumptions recorded by meters at certain time
intervals at buses of individual substations. Load curve data are one of the most
important datasets collected and retained by utilities. The analysis of load curve data
would greatly improve day-to-day operations, system analysis in smart grids, system
visualization, system performance reliability, energy saving, and accuracy in system
planning [1–4].
* This work is partly supported by a collaborative research and development grant from the Natural
Sciences and Engineering Research Council of Canada.
375
376
Smart Grids
The load forecast generally provides annual peak values for the whole system, each
region, and individual substations. Loads at substations (buses) never reach their peaks
at the same point in time. This phenomenon is called noncoincidence. A system analysis, such as a power low study, is often conducted at the system peak, or a regional
peak, or a percentage level of the annual system peak, which may be a seasonal peak.
It is a dificult task to accurately determine the load at each bus that corresponds to the
annual system or regional peak or the percentage level of the peak load. Because hourly
load curve data at all individual buses are generally not available, the majority of utilities have used a so-called worst-case approach by ignoring the noncoincidence among
bus loads, such that the annual peaks of all bus loads are assumed to appear at the same
time as the annual system peak in power low studies. Such an assumption leads to
overestimation of loading levels on the transmission network, which in turn most likely
results in overinvestment in system planning or overacting in system operation. For a
system with its annual peak in winter, the system summer peak is usually assumed to
be a percentage (such as 60%–80%) of the system winter peak, and the same percentage
for the system is applied to all buses. In other words, the bus loads corresponding to the
system summer peak are obtained by proportionally scaling down the bus loads at the
system winter peak. If a system has its annual peak in summer, a similar approach is
used to calculate the bus loads at the system winter peak. This proportional approach to
power lows at a seasonal system peak may lead to underestimation of loading levels on
some network branches and overestimation of loading levels on others. An underestimation of loading levels on branches results in a system risk in planning and operation.
Some utilities may consider the coincidence among several regional loads if they have
collected chronological regional load curve data. However, this is not suficient, as the
coincidence among substations has not been captured.
In recent years, many utilities have implemented smart metering projects. Smart
meters accurately record time-varying loads at end users or on feeders at intervals of
seconds or minutes. This enables chronological load curve data to be automatically
acquired at all buses (substations). With load curve data for system or region and at
all buses, the bus load coincidence factors (BLCFs) at all bus loads can be easily
estimated to accurately calculate bus loads corresponding to any reference point in
the system or regional load curve. The concept of BLCFs will be explained in detail
in Section 16.3.
Even for smart meters, however, invalid load data in the process of information
collection and transfer are still unavoidable. Missing or corrupted data can be caused
by various factors, including meter problems, communication failures, equipment
outages, lost data, and unknown factors. Also, unexpected interruption or shutdown
of power due to strikes, unscheduled maintenance, and temporary closure of production lines can produce signiicant, irregular deviations from the usual load pattern,
resulting in the related load data records being unrepresentative of actual usage patterns. Poor quality of load curve data due to invalid information can cause errors in
data analysis for capturing the coincidence relationship among bus loads. It is important to identify and repair any invalid data. Currently, most utilities handle invalid
data manually in an ad hoc manner. In practice, this approach does not work, particularly for a considerable amount of time-varying information from smart meters,
as it is impossible to handle a huge data pool using a manual process.
377
Load Data Cleansing and Bus Load Coincidence Factors
This chapter addresses two issues: (1) cleansing of load curve data; and (2) calculation of BLCFs. Load curve data cleansing is the prerequisite for calculating accurate BLCFs, while the BLCF is an application of chronological cleansed load curve
data. Cleansed load curve data also have many other applications in data analytics
of smart grids. Accurate BLCFs are the essential input data for system analyses of
smart grids, including power low studies, stability assessment, network loss calculation, and system reliability evaluation. An actual example is given to demonstrate the
application of BLCFs in system power low studies.
16.2 LOAD CURVE DATA CLEANSING
We can calculate accurate BLCFs by using cleansed load curves. This section discusses the basic concepts and method for cleansing outliers in load curves.
16.2.1 BasIc concePts
We call invalid data in a load curve “outliers,” including missing, corrupted, and unrepresentative data due to various different causes. A load curve is a time series with the y-axis
representing power consumption and the x-axis representing time. Theoretically, outlier
detection in a time series has been a topic in data mining [5,6] and statistics [7]. Outliers
can be classiied into two categories: y-outliers and x-outliers. y-Outliers refer to invalid
y-axis values compared with the behaviors in a local neighborhood [8]. Figure 16.1 shows
two examples of y-outliers, one being a localized spike and the other being a localized
dip. Load curve data exhibit some loose form of periodicity (e.g., daily, weekly, monthly,
seasonal, and yearly). The term loose means that the actual data values can be different, but the trend of the data repeats itself regularly at some interval. x-Outliers refer to
abnormal or unrepresentative data that may occur as a deviation from such periodicities [9]. x-Outliers are caused by random events such as a malfunction of data metering
or transfer systems, outages, unexpected full or partial shutdown of production lines,
unscheduled strikes, temporary weather changes, and so on. Such events are unlikely to
occur again in other periods, and thus are not representative of regular patterns of load
curves. Figure 16.2 shows a load curve with weekly periodicity (high weekdays and low
weekends), in which part of the data in the irst week is an x outlier to this periodicity,
although the values in this part of the data are still within the range of the normal pattern.
Load (kW)
25
20
15
10
0
FIGURE 16.1
15
30
Time (h)
Examples of y-outliers in a load curve.
45
378
Smart Grids
6
Load (MW)
5
4
3
2
1
0
9 Nov
14 Nov
19 Nov
24 Nov
29 Nov
04 Dec
09 Dec
Time (date)
FIGURE 16.2 Examples of x-outliers in a load curve.
16.2.2 method For cleansIng outlIers
The method for cleansing outliers includes the following steps: (1) load curve data
are represented using a smoothing curve, which captures the general trend of the
data; (2) a conidence interval on the smoothing curve is built to identify localized
y-outliers; (3) the smoothing curve is modeled by a sequence of valleys and peaks,
called ∪ shapes and ∩ shapes, which are the potential places where x-outliers could
occur; (4) x-outliers are identiied as the valleys, peaks, or both that do not repeat;
and (5) the detected outliers are approximately repaired [8,9,10].
16.2.2.1 Kernel Function Representation of Load Curve
The irst step in detecting outliers is to model the intrinsic patterns of load curve
data by a smoothing curve. This step can be performed by standard smoothing techniques. Given a time series of load curve data, T = {(ti, yi)} (i = 1, …, n) where yi is the
load value at the time point ti, the regression relationship of the data can be modeled
by a continuous function [10]:
yi = m ( ti ) + εi ( i = 1,..., n )
(16.1)
This function is composed of the regression function m and the observation error
εi. The estimated load value in the regression function at time t is modeled in a form
of nonparametric regression by
1
⌢
m (t ) =
n
n
∑ w (t ) y
i
i
(16.2)
i =1
where wi(t) (i = 1, …, n) denotes a sequence of weights. Equation 16.2 is called the
smoothing curve, in which wi(t) is the weight of the point (ti, yi) at time t in the estimation of the smoothing value. The estimation is neighborhood based in the sense
Load Data Cleansing and Bus Load Coincidence Factors
379
that the further ti is from t, the less weight is given to the point (ti, yi). In the kernel
smoothing, wi(t) is given by
Kernh ( t − ti )
⌢
fh ( t )
wi ( t ) =
(16.3)
where
1
t
Kern  
h
h
Kernh ( t ) =
(16.4)
is the kernel with the scale factor h. By using the Rosenblatt–Parzen kernel density
estimator [11] of the density of t
⌢
fh ( t ) = n −1
n
∑ Kern ( t − t )
h
i
(16.5)
i =1
the Nadaraya–Watson estimator [11] for Equation 16.2 is obtained as follows:
∑
∑
n
⌢
m (t ) =
i =1
n
Kernh ( t − ti ) yi
i =1
Kernh ( t − ti )
(16.6)
The shape of the kernel weights is determined by the function Kern, whereas
the size of weights is parameterized by h, which is called bandwidth and referred
to below as smoothing parameter. A larger h corresponds to a smoother curve. The
kernel function Kern can be chosen to be the normal probability density function
[11], that is,
Kern ( t ) =
1 −(1/ 2 )t 2
e
2π
(16.7)
16.2.2.2 Conidence Interval for Identifying y-Outliers
With a smoothing curve expressed by kernel functions, a localized y-axis outlier
can be easily identiied by building a conidence interval for normal load data on
the smoothing curve. An observation within the conidence interval is considered
normal and an observation outside the conidence interval is considered invalid data.
The error term ɛi in Equation 16.1 is assumed to be normally and independently distributed with the mean of zero and constant variance σ2. Under this assumption, the
conidence interval of a load data point can be computed by its estimated value plus
or minus a multiple of the error. The estimated error si(est) is given by [12]
380
Smart Grids
si ( est ) = MSE + si2 ( yˆi )
(16.8)
where MSE is the error mean square:
1
MSE =
n−d
n
∑ ( y − yˆ )
i
2
i
(16.9)
i =1
and si2 ( y^ i ) is the sampling variance of load at time point ti and is given by
si2 ( yˆi ) = WW T × MSE
(16.10)
where W is called the hat matrix and is calculated by [13]
 W1 ( t1 ) W2 ( t1 )

1 W1 ( t2 ) W2 ( t2 )
W= 
⋮
n ⋮

W1 ( t n ) W2 ( t n )
⋯ Wn ( t1 ) 

⋯ Wn ( t2 ) 
⋮ 
⋯

⋯ Wn ( t n ) 
(16.11)
where Wi(tj) corresponds to the weight of observation at time ti for estimating the load
value at time tj. Note that d in Equation 16.9 is the number of degrees of freedom
of the itted load curve data and can be computed by the trace of the hat matrix W.
For a given signiicance level α, the conidence interval at time ti is estimated by
 yˆi − z1− α / 2 × si ( est ) , yˆi + z1− α / 2 × si ( est ) 
(16.12)
where z1–α/2 is the 100(1 – α/2) percentile of the standard normal distribution. For
example, if the 0.05 signiicance level (α = 0.05) is chosen, the corresponding z1–α/2
is 1.96. This means that a normal observation would fall in the conidence interval
given by Equation 16.12 with a probability of 95%, or, equivalently, that the probability that a data point located outside the interval is invalid is 95%.
16.2.2.3 Candidates for x-Outliers
An x outlier has an unusual trend against the periodicity. With the trend of the load
curve being modeled by the smoothing curve, an x outlier tends to occur at a “valley”
or a “peak” of the smoothing curve, where the smoothing curve has local minimal or
maximal values. To formally deine the location of such valleys and peaks, we irst
introduce some terminologies.
The slope at time ti for a smoothing curve {(ti , m^ i )}in=1 is deined by Δmi/Δti, where
∆mi = mˆ i − mˆ i −1 , Δti = ti − ti–1, for 2 ≤ i ≤ n. A time t in an interval is called a steep
time point if the absolute value of the slope at time t is the maximum in the interval.
In other words, at a steep time point, the load increases or decreases at the maximum rate in the interval. Note that there could be more than one steep time point
381
Load Data Cleansing and Bus Load Coincidence Factors
in an interval. An interval [a, b] is maximal-decreasing if the slope at every time
point in [a, b] is ≤0 and any interval containing [a, b] has at least one point with a
positive slope. For two time points c and cʹ in a maximal-decreasing interval [a, b],
[c, b] is convex-decreasing if c is the rightmost steep time point in [a, b], and [a, cʹ]
is concave-decreasing if cʹ is the leftmost steep time point in [a, b]. In Figure 16.3,
the numbers on the smoothing curve are the slopes of the smoothing curve, and the
horizontal axis is time; [t1, t5] is a maximal-decreasing interval; and t3 and t4 are two
steep time points in [t1, t5]. Since t4 is the rightmost steep time point in [t1, t5], [t4, t5]
is a convex-decreasing interval, but [t3, t5] is not, while [t1, t3] is a concave-decreasing
interval.
Similarly, we can deine maximal-increasing intervals, concave-increasing intervals, and convex-increasing intervals. In Figure 16.3, [t6, t10] is a maximal-increasing
interval, t8 and t9 are steep time points, [t6, t8] is a convex-increasing interval because
t8 is the leftmost steep time point in [t6, t10], and [t9, t10] is a concave-increasing
interval.
The following deinitions formalize the notions of valleys and peaks.
Deinition of ∪ shape: For a smoothing curve, a ∪ shape is a subcurve
T∪ = {(t p , m^ p )}pj = i such that, for some k with i ≤ k ≤ j, [i, k] is a convex-decreasing
interval and [k + 1, j] is a convex-increasing interval.
Deinition of ∩ shape: For a smoothing curve, a ∩ shape is a subcurve
T∩ = {(t p , m^ p )}pj = i such that, for some k with i ≤ k ≤ j, [i, k] is a concave-increasing
interval and [k + 1, j] is a concave-decreasing interval.
In Figure 16.3, the curve [t4, t8] is a ∪ shape formed by a convex-decreasing interval [t4, t5] and a convex-increasing interval [t6, t8]. The curve [t9, t12] is a ∩ shape
formed by a concave-increasing interval [t9, t10] and a concave-decreasing interval
[t11, t12]. Notice that the adjacent ∪ shape and ∩ shape overlap at one point at most.
Let us see the adjacent ∪ shape [t4, t8] and ∩ shape [t9, t12] in Figure 16.3. The ∪ shape
[t4, t8] must end at the leftmost steep time point t8 and the ∩ shape [t9, t12] must start
at the rightmost steep time point t9. Intuitively, ∪ shapes and ∩ shapes capture the
regions on the smoothing curve where the raw load curve has large drops and large
rises. Such regions are the potential places where x-outliers may occur. Therefore, ∪
shapes and ∩ shapes are candidate regions for outliers. All candidate regions can be
found by computing ∪ shapes and ∩ shapes following their deinitions.
m̂
–0.1
1
–1
3
–3
–2
FIGURE 16.3
t2
t3
–1
3
–3
t1
–0.1
t4
t5
0.1
t6
2
t7
t8
t9 t10 t11 t12
t
Explanation of the terminologies in smoothing curve [t1, t12].
382
Smart Grids
16.2.2.4 Identifying x-Outliers
Candidate regions with ∪ shapes and ∩ shapes are extracted based on local neighborhood information (i.e., valleys and peaks). Whether a candidate region contains a
real outlier depends on whether the data in the region deviates from the periodicity.
A candidate is not a real outlier if similar load data occur regularly in other periods
according to the periodicity. The similarity should take into account background
noise and time shifting of periodicity.
To identify all outliers, we consider every candidate region r found in the previous
step. Let C* denote the portion of the raw load curve data contained in r. We want to
emphasize that C* contains raw data in the load curve, not the data in the smoothing
curve. If the data in C* occur approximately in the corresponding regions in different periods, C* is not a real outlier but a part of the periodicity. To differentiate this,
we extract all the subload curves C1, C2, …, Ck, where each Ci is the portion of the
raw load curve in the corresponding region of C* in the ith period. If C* is “similar”
to the majority of C1, C2, …, Ck, C* is considered normal; otherwise, C* is considered a real outlier.
The remaining question is how to measure the similarity between two subcurves
C* and Ci with the same length. There are two considerations in choosing the similarity measure. First, the similarity measure should be less sensitive to background
noise. For example, the two load curves in Figure 16.4 should be considered similar despite some variability at the irst peak due to background noise. Second, the
similarity measure should be less sensitive to time shifting and stretching commonly
observed in load curve data. For example, load curves TA and TB in Figure 16.5
should be considered similar despite minor time shifting and stretching.
The Euclidean distance that is commonly used is not suitable for our purpose
because it is sensitive to time shifting and stretching. If the Euclidean distance were
used, the two curves TA and TB in Figure 16.5 would be recognized as dissimilar.
The dynamic time warping distance for two sequences [14] is not suitable either,
since it would pair up all points of two sequences in comparison, making it impossible to skip noisy points.
What we need is a similarity measure that will examine a small neighborhood in
search of matching points and skip noisy points. For this purpose, the longest common subsequence (LCSS) [15] concept can be adopted. LCSS is a coarse-grained
similarity measure in the sense that it measures similarity in terms of “trends”
y
t
FIGURE 16.4 Two similar load curves with noise.
383
Load Data Cleansing and Bus Load Coincidence Factors
y
TA
TB
t
FIGURE 16.5
Two similar load curves with time shifting and stretching.
instead of exact points. Below, we describe how LCSS is extended to measure the
similarity of load curves.
Given two subload curves A = ⟨a1, a2, …, am⟩ and B = ⟨b1, b2, …, bn⟩, which correspond to C* and Ci, we want to ind the LCSS common to both A and B. The idea
is as follows. To allow time shifting and stretching, ai and bj that are within some
time proximity are examined for matching. If these load points are similar, they are
considered as a match and are kept. Dissimilar values in one or both load curves are
dropped. Mathematically, given an integer δ and a real value ɛ, the cumulative similarity Si,j(A, B) or Si,j is deined as follows:
0,

Si, j = 1 + Si −1, j −1

max ( Si, j −1, Si −1, j )
if i = 0 ∨ j = 0
if ai − b j ≤ ε ∧ i − j ≤ δ
(16.13)
otherwise
In Equation 16.13, the irst if statement does the initialization for the shortest preix. The second if statement builds the similarity recursively: if |ai − bj| ≤ ɛ and if ai
and bj are close enough in time, that is, |i − j| ≤ δ, ai and bj are matched and the similarity is incremented. Note that ɛ represents a tolerance of noise in load value and δ
represents a tolerance of time shifting and stretching.
Let |A| and |B| be the length of A and B, respectively. The LCSS similarity of A
and B is given by
γ ( δ, ε, A, B ) =
S A, B
(
min A , B
)
(16.14)
where S|A|,|B| is the length of the common subsequence to both A and B, which is
cumulatively computed by Equation 16.13. For a user-speciied threshold θ, we say
that A and B are similar if
γ ( δ, ε, A, B ) ≥ θ
(16.15)
384
16.2.3
Smart Grids
rePaIrIng outlIers
Detected outliers should be repaired before the load curve can be used for calculating BLCFs or other applications. Suppose that we want to repair an outlier C*. Some
valid data must be selected to replace the load data of C*. The replacement data can
be derived from the normal data in the corresponding time interval in other periods
with an adjustment for the increasing or decreasing long-term trends over time. This
is expressed as the following multiplicative model [16]:
Y ( ti ) = T ( ti ) × R ( ti )
(16.16)
where Y(ti) represents the value that will replace the abnormal value at a time ti
belonging to an outlier, T(ti) represents the load value in the long-term trend, and
R(ti) represents the periodic index, that is, how much the load curve deviates from
the long-term trend at time ti.
It is possible to estimate T(ti) by the smoothing curve deined in Equation 16.2
with an appropriate smoothness level (bandwidth). However, in the presence of outliers, the smoothing curve may contain invalid data around the outliers to be replaced.
To address this problem, all outliers in the load curve are replaced, irst, by the average of the data at the corresponding time in the previous and following periods. If
load points at these times are also outliers themselves, the corresponding data from
earlier and later periods will be examined until normal data are obtained. After
illing in such average values for all outliers, a new smoothing curve is regenerated
using Equation 16.2 and is used to estimate T(ti).
The periodic index R(ti) for a time ti belonging to an outlier is estimated by the
average of the periodic indexes at the corresponding time of its previous and following periods, that is,
R ( ti ) =
1
2
( R ( t − l ) + R ( t + l ))
i
i
(16.17)
where l is the length of the periodicity. If the data at the previous and next periods
are outliers, earlier and later periods are examined until normal data are obtained.
Note that, for a time ti not belonging to an outlier, the periodic index at ti is computed
by its deinition:
R ( ti ) =
yi
T ( ti )
(16.18)
where yi is the load at time ti. After T(ti) and R(ti) for an outlier are obtained,
Equation 16.16 is used to produce the replacement value Y(ti).
16.2.4
selectIon oF Parameters
The method described above uses several parameters: the smoothing parameter h (see
Equation 16.4), the load stretching threshold ɛ and the time stretching threshold δ (see
385
Load Data Cleansing and Bus Load Coincidence Factors
Smoothing curve
Detected outlier
30
Load (MW)
25
20
15
10
5
0
2005
FIGURE 16.6
2006
2007
2008
Time (year)
2009
2010
An actual load curve, the smoothing curve, and identiied x-outliers.
Equation 16.13), and the LCSS similarity threshold θ (see Equation 16.15). How should
the values of these parameters be set? One approach is to use some statistically “optimal” setting, such as the optimal smoothing parameter [17]. In applications where the
user has background knowledge, the user often desires to have control over a small
number of settings. A practical approach is to provide a mechanism that helps the user
to identify a proper setting of parameters in such a case. We use the smoothing parameter h as an example below, and a similar approach can be applied to other parameters.
A larger h produces a smoother smoothing curve, and thus models the load data in
less detail. In practice, we do not have to make a choice in advance. A software tool
with a user-friendly interface has been developed, which allows the user to slide a
bar for the smoothing parameter and displays the smoothing curve and the identiied
outliers to the user interactively. Based on visual inspection of the raw data, smoothing curve, and identiied outliers, the user can either accept the results or slide the
bar again for a different choice of h and get a display of the new smoothing curve and
outliers based on the new choice of h. This is analogous to sliding the bar at the lower
right of the Word window for setting the font size of a document. With such sliding
bars, a user often quickly converges to what he or she considers to be the best setting.
As an experiment, the method was applied to an actual industrial load curve in
the system of BC Hydro, Canada. This load curve includes hourly MW loads for the
6 years from October 2004 to October 2010, with 24 × 365 × 6 = 52,560 observed
records. Figure 16.6 shows the raw load records, the smoothing curve (by the bold
line) obtained using a speciic smoothing parameter h [8], and the x-outliers (marked
by each rectangle) detected by the proposed method. It can be seen that the smoothing curve models the load curve properly and there is no false-positive or falsenegative error in detecting the x-outliers.
16.3
BUS LOAD COINCIDENCE FACTORS
Once outliers in historical load curves are cleansed, calculating BLCFs using load
curve data is straightforward. This section illustrates the basic concept and procedure
386
Smart Grids
of calculating BLCF, and Section 16.4 will present an example to demonstrate the
application of BLCF to power system analysis.
16.3.1
BasIc concePt oF BlcF
Once load curves at all considered buses are cleansed, it is straightforward to calculate the BLCFs. The BLCF is an important concept in power system analyses,
particularly in power low studies [4]. The load forecast generally provides annual
peak values for the whole system, each region, and individual substations. As mentioned in Section 16.1, loads at substations never reach their peaks at the same point
in time. To perform a power low study at the annual system or regional peak, or at a
percentage level of the annual peak, which may be a seasonal peak or valley, we must
calculate bus loads corresponding to the peak or the level of the peak. The BLCFs at
the buses are used for this purpose.
The concept of BLCF is illustrated using Figure 16.7. For simplicity, only two
substation load curves are considered. The total load curve, which is the sum of the
two substation load curves, can be viewed as the system (or regional) load curve.
Points A, E, and D are the peaks of the system and two substation load curves,
respectively. Apparently, they do not occur at the same time. Points C and B represent the load points of the two substations corresponding to the system (or regional)
peak. The BLCFs of the two substations are deined as
BLCF1 =
LC
LE
(16.19)
BLCF2 =
LB
LD
(16.20)
where LC and LE are the loads at Points C and E in the load curve 1 for Substation 1,
and L B and L D are the loads at Points B and D in the load curve 2 for Substation 2.
90
80
Load 1
Load 2
Total
A
Load (MW)
70
60
50
40
B
D
C
30
E
20
10
0
Time (h)
FIGURE 16.7
Concept of BLCF.
Load Data Cleansing and Bus Load Coincidence Factors
387
Obviously, cleansed load curve data are important to accurately calculate BLCF.
For instance, if the load around Point B, C, D, or E is corrupted, we will get a distorted BLCF value. Even if the load at Point B or C is valid data, a reduced L D or LE
will overestimate the BLCF value and an elevated L D or LE will underestimate the
BLCF value.
Using the example shown in Figure 16.7, we deined the BLCF corresponding
to the system peak. For a given load point in a system or regional load curve, the
corresponding time point on the x-axis is called the reference point. In this sense,
BLCFs can be calculated for any different reference points on a system or regional
load curve (such as the annual peak, winter or summer peak, winter or summer valley). BLCFs can be also calculated for different load categories (such as industrial,
residential, or commercial customers or their combinations) using a similar concept. Apparently, BLCFs vary dynamically from year to year, as the load curve data
change every year. BLCFs should be estimated using multiple years of load data and
updated each year. There are uncertainties around load data points. The system or
regional peak is different in value and moves around in different years, and various
periodicities in a load curve may be shifted. To tackle the uncertainties, instead of
using the load value at each single point (such as A, B, C, D, or E), the average of a
certain number of loads around each point should be used.
Once the BLCFs are estimated, the bus loads corresponding to any reference
point on a system or regional load curve can be calculated using the BLCF values for
that reference point and the forecasted peaks of substation loads.
16.3.2
Procedure For calculatIng BlcF
In the procedure described below, the annual system peak in a 1-year period is used
as a reference point to indicate how to estimate the BLCF values and calculate bus
loads corresponding to the reference point. The procedure for estimating the BLCFs
corresponding to other reference points (a percentage level of the annual system peak
or the maximum/minimum load value in a speciied period, such as winter or summer) in a system or region is similar.
The procedure includes the following steps [18]:
1. The invalid data for yearly load curves at all considered substations are
cleansed using the method given in Section 16.2, including identiication
and repairing of invalid data.
2. All the yearly substation load curves are summed up to obtain the yearly
system load curve. The maximum hourly load value (peak) in the yearly
system load curve is determined and is assumed to be X MW. The corresponding time point on the x-axis is called the reference point.
3. N hourly load points on the system load curve that are closest to the X MW
in the load values are determined, where N (such as 24 or 48) is a proper
number for handling the uncertainty of the system peak. In the case where
the annual system peak is selected as the reference point, the N hourly
load values include the system peak, itself being X MW, with others being
smaller than or equal to X MW. In other cases in which a nonannual system
388
Smart Grids
peak is selected as the reference point, some of the N selected hourly load
values may be larger than X MW but others may be smaller than X MW.
4. M annual largest hourly load values on the yearly load curve of each individual substation are determined, where M is a proper number for handling
the uncertainty of the substation load peak and may or may not be equal to
N, depending on a judgment on the degree of uncertainty of the substation
load peak. Let Lki(P) denote the MW load at the kth point of the M hourly
load points in the ith substation load curve.
5. We locate the N points on each yearly substation load curve that correspond
to the same time points as the N load points on the yearly system load curve,
which have been determined in Step 3. Let Lji denote the MW load at the jth
point of the N selected hourly load points in the ith substation load curve.
The BLCF corresponding to the selected reference point for the ith substation is estimated by
BLCFi =
1
N
N
∑
j =1

L ji
∑
k=M
k =1
Lki ( P )  M

(16.21)
6. There are always some differences in the yearly load curves of a substation
in different years. In many cases, the differences may be relatively small, as
load customers supplied by the substation usually maintain their electricity
consumption behaviors. In other cases, the differences may be relatively large
because the electricity consumption behaviors of some customers change over
the years. To deal with the possible changes, it is better to repeat Steps 1–5 for
more than 1 year’s load curve data. On the other hand, we should not use too
many years’ data, since cumulative changes may be so signiicant that data
from previous years are no longer representative of the most recent pattern.
We suggest considering a time frame of the past 3 years for obtaining three
BLCF values for each substation. There are three possibilities, as follows:
a. Only 1 year’s load curve data are available. In this case, the BLCF for
this year is used.
b. Two years’ load curve data are available. In this case, the difference
between the two BLCF values is calculated as a percentage with respect
to the larger value.
i. If the difference is smaller than or equal to a threshold (such as
5%–10%), the average of the two BLCF values is used as the inal
BLCF.
ii. If the difference is larger than the threshold, the BLCF from the
most recent year’s load curve is used as the inal BLCF.
c. Three years’ load curve data are available. In this case, the differences
between each pair of the three BLCF values are calculated as a percentage with respect to the largest value.
i. If all the three differences are smaller than the threshold, the average of the three BLCF values is used as the inal BLCF.
Load Data Cleansing and Bus Load Coincidence Factors
389
ii. If two of the three differences are smaller than the threshold, the
average of the larger two BLCF values is used as the inal BLCF.
iii. If only one of the three differences is smaller than the threshold, the
average of the two BLCF values corresponding to this difference is
used as the inal BLCF.
iv. If all the three differences are larger than the threshold, the BLCF
from the most recent year’s load curve is used as the inal BLCF.
7. The MW load of each substation corresponding to the selected reference
point (such as the annual system peak, or a percentage level of the annual
system peak, or the maximum or minimum load value in a selected period)
can be calculated using the BLCF value of each substation and the forecast
peak load for each substation in the considered year, by
Li ( MW ) = Li ( F ) × BLCFi
(16.22)
where Li(MW) is the MW load at the ith substation (bus) corresponding to the reference point and Li(F) is the forecasted peak load for the ith substation.
It should be pointed out that the word “system” in the above description can refer
to a whole system, a region or area, a subregion or subarea, or a substation group.
16.4
APPLICATION OF BLCF TO SYSTEM ANALYSIS
A real-life case is used in this section to demonstrate the application of BLCF to
system analysis.
16.4.1
studIed system and load curve data
Figure 16.8 shows a transmission network supplying power to a subarea of BC Hydro in
Canada. The subarea includes three substations named YVR, SEA, and RIM. The power
is supplied from the upstream substation KI2 through two 60 kV transmission circuits
named 60L43 and 60L44. Each circuit includes three sections marked by A, B, and C.
Sections A and B are overhead lines and Section C is an underground cable. The MVA
ratings of the three sections of each circuit in winter and summer are listed in Table 16.1.
The cleansed hourly load curves at the three substations in 2006/2007 are shown in
Figure 16.9. By summing up the loads of the three substations at every hour, the total load
curve of the subarea is created, as shown in Figure 16.10, where the summer and winter
peak periods are framed by the two dashed line rectangles. This subarea has its annual
peak in winter, which appears on November 29, at 5:00 pm, for this particular year.
The maximum loading levels on the two circuits (60L43 and 60L44) occur at the
same time point as the subarea peak. The power lows at the maximum loading levels
on November 29, 2006 at 5:00 p.m. are displayed in Figure 16.11. It can be seen that
the actual maximum loading levels in the normal state are as follows:
In the normal state, 60L43 loading = 56.5/156 = 36.2%
In the normal state, 60L44 loading = 48.2/156 = 30.9%
In N−1 contingency states, the maximum loading level can be calculated by
contingency analysis. It is 105.9/156 = 67.9%.
390
Smart Grids
KI2
60L43C
60L43A
YVR
60L44A
SEA
60L44C
60L43B
60L44B
RIM
FIGURE 16.8 The subarea power supply network.
TABLE 16.1
Rating of the Two Circuits
Circuit Section
60L43A and 60L44A
60L43B and 60L44B
60L43C and 60L44C
16.4.2
Winter Rating
(MVA) (10°C)
156
100
68.5
Summer Rating
(MVA) (30°C)
121
78
63
Bus load coIncIdence Factors at the three suBstatIons
The BLCFs are calculated using the method in Section 16.3 and the historical load
curve data at the three substations. The BLCF values obtained using the subarea winter peak (which is the annual peak in this example) and summer peak as the reference
points are given in Tables 16.2 and 16.3, together with the forecast annual (winter)
peak loads in 2017/2018 and the calculated coincidence bus loads in 2017/2018. It is
important to appreciate that a coincident load for the subarea summer peak as a reference point should be, and is still, calculated using the forecasted annual peak (i.e.,
the winter peak in this case) multiplied by its summer BLCF, since a BLCF is always
deined as a load at the reference point divided by its annual peak. To make a comparison, the winter and summer peak loads of the three substations in 2006/2007 are
391
07/15/06
08/05/06
08/26/06
10/07/06
10/28/06
12/09/06
12/30/06
01/20/07
02/10/07
03/03/07
06/24/06
07/15/06
08/05/06
08/26/06
09/16/06
10/07/06
10/28/06
11/18/06
12/09/06
12/30/06
01/20/07
02/10/07
03/03/07
03/24/07
06/24/06
07/15/06
08/05/06
08/26/06
09/16/06
10/07/06
10/28/06
11/18/06
12/09/06
12/30/06
01/20/07
02/10/07
03/03/07
03/24/07
03/24/07
06/03/06
06/03/06
06/03/06
11/18/06
05/13/06
05/13/06
05/13/06
09/16/06
04/22/06
04/22/06
04/22/06
06/24/06
04/01/06
04/01/06
16
14
12
10
8
6
4
2
0
04/01/06
RIM load (MW)
90
80
70
60
50
40
30
20
10
0
SEA load (MW)
Load Data Cleansing and Bus Load Coincidence Factors
14
YVR load (MW)
12
10
8
6
4
2
0
Time stamp (h)
FIGURE 16.9 The cleansed hourly load curves at the three substations.
also given in the two tables. Note that the load at the RIM substation includes two
components: distribution and industrial customers. The industrial customer at RIM
will not be in service until 2017/2018, and therefore its historical load curve data are
not available for calculating the BLCF. In this case, the BLCF can be assumed to be
1.0. Such an assumption creates a relatively high loading level, which can lead to a
more secure decision in system operation and planning. A small part of the distribution load at RIM will be transferred to another substation by 2017/2018.
392
Smart Grids
Total load (MW)
120
100
80
60
40
20
03/24/07
03/03/07
02/10/07
01/20/07
12/30/06
12/09/06
11/18/06
10/28/06
10/07/06
09/16/06
08/26/06
08/05/06
07/15/06
06/24/06
06/03/06
05/13/06
04/22/06
04/01/06
0
Time stamp (h)
FIGURE 16.10 The load curve of the subarea.
1.061
63.7
T1
9.8
892
SEA 12
0.0
T2
0.0
6.0 15.8
15.8 56.5
T1
1.033
13.0
56.5
6.8
T3
6.8
6.8 48.1
692
SEA 60B1
634
1.061
63.7
RIM 60B3
1.054
63.2
48.2
633
RIM 60B1
1.053
63.2
1.064
63.9
T1
SW
0.0
81.4
16.6
T3
833
RIM 12
FIGURE 16.11
T2
1.061
63.7
40.5 40.5
12.8
2.4
693
SEA 60B2
41.1 41.1
698
YVR 60
9.8
9.8
0.0
1.050
13.2
The power lows at the actual annual maximum loading levels in 2006/2007.
It can be observed that the total load in the subarea is increased from 105.8 MW in
2006/2007 to 123 MW in 2017/2018. This is translated into an average annual growth
rate of 1.5% (1.56 MW/year). The difference between the forecasted (noncoincidental) total load and the coincidental total load is 123−119 = 4.0 MW. This is equivalent
to the load increase percentage in about 2 years. In other words, if the coincidental
loads are used for a planning purpose, this may lead to approximately 2 years’ delay
of the system reinforcement required by the load growth in this subarea. This observation can be veriied by the power low studies given in the next subsection.
393
Load Data Cleansing and Bus Load Coincidence Factors
TABLE 16.2
BLCFs and Coincidence Loads for the Winter Peak at the Three Substations
Substation
RIM—distribution
load
RIM—industrial
load
SEA
YVR
Total
a
Actual Winter
Peak Load in
2006/2007
(MVA)
Forecast Annual
Peak in
2017/2018
(MVA)
Winter BLCF
Coincidental
Load in
2017/2018
(MVA)
83.0
78.2
0.9882
77.3
7.0
N/A (1.0000)
7.0
—a
13.0
9.8
105.8
23.3
14.5
123.0
0.9361
0.8863
—
21.8
12.9
119.0
—, The industrial customer is not yet in service and the load proile is therefore not available.
TABLE 16.3
BLCFs and Coincidence Loads for the Summer Peak at the Three
Substations
Substation
RIM—distribution
load
RIM—industrial
load
SEA
YVR
Total
a
16.4.3
Actual
Summer Peak
Load in
2006/2007
(MVA)
Forecast
Annual Peak
in
2017/2018
(MVA)
Summer
BLCF
Coincidental
Load in
2017/2018
(MVA)
65.4
78.2
0.8331
65.1
—a
7.0
N/A (1.0000)
7.0
12.6
11.6
89.6
23.3
14.5
123.0
0.9359
0.8344
—
21.8
12.1
106.0
—, The industrial customer is not yet in service and the load proile is therefore not
available.
PoWer FloW analysIs usIng the Bus load coIncIdence Factors
The power low studies for the subarea network are performed using the noncoincidental
and coincidental substation loads in 2017/2018. The following four cases are considered:
Case 1: Noncoincidental winter peak loads, which refer to the forecasted
annual peaks at the given three substations.
Case 2: Noncoincidental summer peak loads. For utilities having the annual
peak in winter, although load forecast departments at the utilities may also
394
Smart Grids
provide the forecasted summer peak for the whole system (or areas) in addition to the forecasted winter peaks for the whole system (areas) and all
substations, they generally do not provide the forecasted summer peaks
for individual substations. In the existing practice of utilities, a common
approach is to calculate substation loads at the system summer peak point
by proportionally scal