PE JanFeb2023

Volume 21 • Number 1 • January/February 2023
READY
TO
SWITCH? SKM Mobile App for iOS and Android!
Take SKM on the go! The SKM Mobile app provides an intuitive experience that let’s you
quickly and easily collect data from the field and seamlessly import it to your PTW desktop
Receive up to 60% off program! You can also view study results on your phone or tablet - such as what arc flash
PPE to wear or to verify equipment settings!
View one-lines, take pictures, videos, scan QR codes, scribble notes, and more!
when you switch
to SKM.
Benefits MySKM Cloud
You’ll also receive Transfer field collection directly to SKM without having to Upload Mobile app projects to MySKM to use in PTW
write anything down on paper. desktop software and vice-versa!
free service to convert Take pictures, videos, audio, scribble notes and save with Download PTW desktop projects to allow field engineers
each equipment.
competing software to verify equipment settings and study results, such as
available short circuit results, incident energy, required
Scan QR equipment codes to quickly locate the
projects over to SKM! component on the mobile app to view equipment details
PPE selection, and more!
and study results. Assign users and maintain security permissions for each
project.
View entire power system oneline while in the field in the
palm of your hands!
“The main advantage with

Power*Tools for Windows is ARC FLASH EVALUATION
Design safer power systems while ensuring compliance with NEC,
its user-friendliness. PTW is OSHA, BFPA, IEEE, NESC, and CSA standards.
easy to use and the ready- Fully integrated with Short Circuit, Over-Current Coordination,
and Equipment Evaluation modules.
to-use standard models
Evaluate alternatives quickly and easily to establish an optimal design.
enable you to speed up Save time by automatically generating arc flash labels and work permits.
your work. Changing data is Increase flexibility by creating custom labels in any size with
user-defined logos, text, comments, and field placement.
simple and makes evaluating
Compare results from multiple project scenarios in a single table.
different cases very easy.”
ADDITIONAL SOFTWARE FROM SKM

DAPPER® Integrated Electrical Analysis Software. Comprehensive AUTO DESIGN Automatically sizes Cables, Transformers,
Three Phase Short-Circuit Studies, Load Flow Study, Demand Load Buses, and Protective Devices to achieve proper size and
Study, Feeder and Transformer Sizing Study, Impact Motor Starting ratings that meets criteria limits set by the Equipment Evaluation
Study, and Load Schedules. (Unbalanced Studies also available) study. Utilize user-defined Design libraries. Best suited for new system
CAPTOR® Time-Overcurrent Coordination. Graphical Time-Overcurrent designs or to resolve undersized and/or overdutied equipment for
Coordination. Integrated with one-lines, short-circuit modules, Equipment existing systems. (Requires Equipment Evaluation module)
Evaluation, and Arc Flash. Comprehensive protective device library. I*SIM Dynamic Simulation and Transient Stability. Model, Simulate,
AUTO COORDINATION & EVALUATION Instantly identify and Analyze Dynamic Responses to Power System Electro-Mechanical
coordination and protection issues by evaluating each Disturbances. Includes Generator Sizing and Stability, Rotor Angle and
equipment for adherence to the National Electric Code (NEC). Voltage Stability, Flux Level Machine Representation, Identify Fault
Automatically make changes to resolve violations. Clearing Times. User Defined Graphical Models for Exciter, Turbine
Governor, PSS, and other controllers.
CABLE AMPACITY Accurate calculation of ampacity and
temperature rise in various power cable installations. Based GROUND MAT Substation Ground Grid Design and Analysis. Optimizes
on and fully compliant with IEC 60287. grid design using general purpose finite element algorithm for
potential analysis and graphical facilities to validate grounding
DC SYSTEMS ANALYSIS Battery Sizing, Load Flow, & Short Circuit
systems efficiency.
PO Box 12123 Analysis. Evaluate all loading conditions for DC duty cycle loads and
AC emergency loads. Complies with IEEE std. 485, 1115, 399, 946,
Zephyr Cove, NV 89448 Visit skm.com for more products!
and IEC std. 61660.
1-800-500-4SKM
www.skm.com • [email protected]
CHOSEN BY THE TOP 40 ELECTRICAL ENGINEERING FIRMS IN THE WORLD
magazine
Volume 21 • Number 01 • January/February 2023

www.ieee.org/power
on the
cover
73
©SHUTTERSTOCK.COM/URBANS
features
contents
16 Advances in Algorithms for Power 59 Dynamic Estimation-Based
System Static State Estimators Protection and Hidden Failure
By Bei Gou and David Shue Detection and Identification
By Sakis Meliopoulos, George J. Cokkinides,
26 A Powerful Tool for Power System Paul Myrda, Evangelos Farantatos,
Monitoring Ramadan Elmoudi, Bruce Fardanesh,
By Tianshu Bi, Maolin Zhu, and Hao Liu George Stefopoulos, Clifton Black,
and Patrick Panciatici
36 Multifold Insights for Power System
Dynamics From Data Assimilation 73 The Spanish Experience
By Shaobu Wang, Zhenyu Huang, By José A. Rosendo-Macías,
Renke Huang, Ning Zhou, and Junbo Zhao Antonio Gómez-Expósito,
Alfonso Bachiller-Soler,
44 Dynamic Wide Area Situational Miguel Á. González-Cagigal,
Awareness Gabriel Álvarez-Cordero,
By Innocent Kamwa Lucía Mateo-Sánchez,
and Antonio Useros-García
4 columns &
departments
4 Editors’ Voice 83 Book Review
10 Leader’s Corner 84 Calendar
14 Guest Editorial 88 In My View
Digital Object Identifier 10.1109/MPE.2022.3224177
january/february 2023 IEEE power & energy magazine 1

magazine
Editor-in-Chief Assistant Editor IEEE Periodicals/Magazines Department

Innocent Kamwa Sherry Hensley 445 Hoes Lane, Piscataway, NJ 08854 USA
[email protected] Kristin Falco LaFleur, Senior Journals Production Manager
Editorial Board Patrick J. Kempf, Senior Manager, Production
Past Editors-in-Chief L. Anderson, L. Barroso, H. Chen, A. Conejo,
Janet Dudar, Senior Art Director
Steve Widergren M. Correia de Barros, H. Holttinen T. Hong,
S. Imai, C.-H. Kim, D. Lew, Y. Liu, J. Matevosyan, Gail A. Schnitzer, Associate Art Director
Michael Henderson Theresa L. Smith, Production Coordinator
M. Miller, D. Novosel, N. Ochoa, A. Orths,
Founding Editor-in-Chief P. Panciatici, E. Uzunovic Felicia Spagnoli, Advertising Production Manager
Melvin Olken Peter M. Tuohy, Production Director
Spanish Editorial Board Kevin Lisankie, Editorial Services Director
Associate Editors Enrique Tejera, Editor-in-Chief Dawn M. Melley, Senior Director, IEEE Publishing
John Paserba, History Editors: M. Baquedano, J. Cerda, G. Gonzalez Operations
Brian Johnson, Ning Lu, Barry Mather, Issues
Jianhui Wang, Submissions Advertising
Jill Duplessis, Columns and Special Issues Beverly Grimme, Naylor Association Solutions
direct: +1 352 333 3367, cell: +1 904 881 0862, IEEE prohibits discrimination, harassment, and bullying. For more infor-
fax: +1 703 790 9199, [email protected] mation, visit http://www.ieee.org/web/aboutus/whatis/policies/p9-26.html.
IEEE POWER & ENERGY SOCIETY (PES)

The IEEE Power & Energy Society is an organization of IEEE members whose principal interest is the advancement of the science and practice of electric power generation,
transmission, distribution, and utilization. All members of the IEEE are eligible for membership in the Society. Mission Statement: To be the leading provider of scientific and
engineering information on electric power and energy for the betterment of society, and the preferred professional development source for our members.
Governing Board Technical Council IEEE Power Engineering Letters, R. Jabr

J. Bian, President H. Chen, Chair; D. Watkins, Vice Chair IEEE Trans. on Energy Conversion, D. Aliprantis
S. Bahramirad, President-Elect J. McBride, Secretary; V. Vittal, Past-Chair IEEE Trans. on Power Delivery, F. DeLeon
H. Chen, Vice President, Technical Activities IEEE Trans. on Power Systems, J. Milanovic
B. Enayati, Vice President, Education Technical Committee Chairs IEEE Trans. on Smart Grid, C. Canizares
B. Pal, Vice President, Publications K. Cheung, Analytical Methods for Power Systems IEEE Trans. on Sustainable Energy, B. Chowdhury
W. Bishop, Vice President, Meetings J. Lau, Electric Machinery IEEE Open Access Journal of Power and Energy, F. Li
J. Romero Aguero, Vice President Membership R. Thornton-Jones, Energy Development & Power IEEE Power & Energy Magazine, I. Kamwa
& Chapters Generation eNewsletter, TBD
N. Hadjsaid, Vice President, New Initiatives/ B. Chalamala, Energy Storage & Stationary Battery Editor-in-Chief at Large, N. Hatziargyriou
Outreach Y. Wen, Insulated Conductors Marketing, E. Batzelis
Chan Wong, VP Strategic Communications J. White, Nuclear Power Engineering PES Representative to IEEE Press, K. Bhattacharya
& Partnerships C. Preuss, Power System Communications Website, R. Rana
J. Montero, Treasurer & Cybersecurity Meetings
J. Sykes, Secretary L. Lima, Power System Dynamic Performance Committee Chairs
S. Rahman, Past-President E. Hanique, Power System Instrumentation General Meeting Steering, A. St. Leger
M. Chaganti, Region Rep., U.S. & Canada & Measurements Joint Technical Committee Meeting Steering, D. Sabin
H. Albuflasa, Region Rep., Europe, J. Wang, Power System Operation Planning Technically Cosponsored Conferences Steering,
Middle East, & Africa & Economics Daniel Diaz
S. Leon, Region Rep., Latin America M. Yalla, Power System Relaying & Control Transmission & Distribution Conference &
D. Sharafi, Region Rep., Asia & Pacific J. Mathieu, Smart Building, Load & Customer Systems Exposition Committee for North America
M. Chelangat, Member-at-Large Climate Change P. Fitzgerald, Substations Steering, C. Segneri
G. Gonzales, Member-at-Large, Diversity & Inclusion S. Hensley, Surge Protective Devices Innovative Smart Grid Technologies North America,
N. Nair, Member-at-Large, Global Outreach K. Flowers, Switchgear R. Melton
A. Shah, Member-at-Large, Student/Young E. teNyenhuis, Transformers Innovative Smart Grid Technology Conference–
Professionals Representative S. Santoso, Transmission & Distribution Europe, Gianfranco Chicco
Innovative Smart Grid Technology Conference–
IEEE Division VII Director Technical Council Coordinating Committees Asia, D. Sharafi
C. Root H. Sun, Energy Internet PowerAfrica Steering Committee, B. Lequesne
IEEE Division VII Director-Elect J. Follum, Intelligent Grid & Emerging Technology Education
Open D. Alexander, Marine Systems Committee Chairs
Y. Zhang, Renewable Systems Integration Power & Energy Education, S. Kamalasadan
PES Executive Director PES Scholarship Plus, D. Zabetakis
T. Licitra Technical Council Standing Committees
V. Vittal, Awards PES University, Meredith Southergill
Standing Committee Chairs Open, Industry Education Instructional Quality & Initiatives, L. Fan
A. Rotz, Constitution & Bylaws Selection & Quality Control, S. Kamalasadan
J. McBride, Organization & Procedures Website, J. Rickson
J. Mitra, Fellows T. Irwin, Standards Coordination
J. Montero, Finance D. Watkins, Technical Sessions New Initiatives and Outreach
Open, History Open, Webmaster Committee Chairs
S. Bahramirad, Long-Range Planning V. Vittal, Entity Proposal Management Committee Corporate Engagement Program, M. Vaiman
S. Rahman, Nominations & Appointments Diversity, J. McClanahan
Publications Executive Advisory Council, K. Curtis
Region Representatives Publications Board Chair, B. Pal
H. Albuflasa, M. Chaganti, M. Gosalia, IEEE Smart Cities, G. Zissis
R. Kappagantu, J. Khan, S. Leon, M. Papic, Editors-in-Chief IEEE Smart Grid, Wei-Jen Lee
M. Patel, S. Sarkar, X. Wu IEEE Trans. on Energy Markets, Policy, IEEE Smart Village, J. Nelson
and Regulation, D. Kirschen Industry Tech Support Leadership, A. Paaso
Chapter Committee Chairs IEEE Electrification Magazine, L. Fan Media Engagement, D. Kushner
P. Pabst, Chapter Awards and Recognition
TBD, Distinguished Lecturer Program
TBD, Student Chapters Coordinator IEEE Power & Energy Magazine
S. Fattah, Humanitarian Activities IEEE Power & Energy Magazine (ISSN 1540-7977) (IPEMCF) is published bimonthly by the Institute of Electrical and
Electronics Engineers, Inc. Headquarters: 3 Park Avenue, 17th Floor, New York, NY 10016-5997 USA. Responsibility for the
L. Leon, Women in Power contents rests upon the authors and not upon the IEEE, the Society, or its members. IEEE Operations Center (for orders, sub-
S. Akar, Young Professionals scriptions, address changes): 445 Hoes Lane, Piscataway, NJ 08854 USA. Telephone: +1 732 981 0060, +1 800 678 4333.
V. Vittal, Life Members Individual copies: IEEE members US$20.00 (first copy only), nonmembers US$89 per copy. Subscription Rates: Society members
T. Ribeiro, Entrepreneurship included with membership dues. Subscription rates available upon request. Copyright and reprint permissions: Abstracting is permit-
ted with credit to the source. Libraries are permitted to photocopy beyond the limits of U.S. Copyright law for the private use of
Strategic Communications patrons 1) those post-1977 articles that carry a code at the bottom of the first page, provided the per-copy fee indicated in the code
& Partnership is paid through the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923 USA; 2) pre-1978 articles without fee.
L. Paredes Muse, PES Day For other copying, reprint, or republication permission, write Copyrights and Permissions Department, IEEE Operations Center, 445
J. Cepeda, Awards & Recognitio Hoes Lane, Piscataway, NJ 08854 USA. Copyright © 2023 by the Institute of Electrical and Electronics Engineers, Inc. All rights
reserved. Periodicals postage paid at New York, NY, and at additional mailing offices. Postmaster: Send address changes to IEEE
Digital Object Identifier 10.1109/MPE.2022.3224178 Power & Energy Magazine, IEEE Operations Center, 445 Hoes Lane, Piscataway, NJ 08854 USA. Canadian GST #125634188
Printed in U.S.A.
2 IEEE power & energy magazine january/february 2023

POWERFACTORY 2023 IS NOW AVAILABLE!
SELECTED NEW FEATURES
Dedicated Low Voltage Load Flow Calculation function, using coincidence curves to model the
stochastic consumption behaviour of low voltage loads
Developments in Power Quality and Harmonic Analysis, including n-1 contingencies in the Harmonic
Load Flow and Frequency Sweep calculations
Enhancements to Optimal Power Flow, such as redispatch costs for generators, support of external
tap controllers and option to use external solvers for faster optimisation
Flow Decomposition function, for analysing network flows in order to determine the contributions of
loop flows, transit flows, import/export flows etc.
New reporting concept, allowing standard reports to be generated in various output formats such as
PDF, together with a Report Designer that enables users to create their own report templates
Graphical Modelica modelling environment, allowing the graphical definition of complex,
hierarchically structured Modelica simulation models
Check out PowerFactory´s “What´s New in PowerFactory 2023” and learn more about exciting new
tools and features.
For more information visit

www.digsilent.de
Serving more than 170 countries.
editors’ voice
Innocent Kamwa
grid behavior
PMU-enabled dynamic state estimation
T
THE NOTION OF THE STATE OF Understandably, the system operator every location in the network. Most of
health is widely used in asset manage- should work hard to stay as far away as the time, the power grid is indeed in
ment. For instance, electric vehicle bat- possible from this limit so as to always steady-state conditions. Since the 70s,
teries are replaced after they lose ~30% keep the lights on. For that purpose, the the utility industry has developed and
capacity, which means that their state of control room houses big boards report- applied tools known as static state es-
health is not good enough for reliable use ing real-time voltage and breaker status timators to determine the state of grids
in day-to-day travels, even though they on one-line diagrams plus on consoles consisting of hundreds of thousands
still possess 70% capacity. Estimating permitting staff to call up detailed in- of locations connected electrically to-
the state of health of an asset is difficult, formation. When there is a voltage sag, gether and using measurements that are
and for a power grid spanning thou- for example, alarms will sound on ter- noisy. One article in this issue summa-
sands of kilometers and interconnect- minals, and operators will typically get rizes artfully the progress accomplished
ing millions of components, it is barely on the phone to order corrective mea- over that time span and discusses how
imaginable. We use familiar biomedical sures (Figure 1). to enhance the robustness of static state
tests, such as blood tests, to grasp quan- Under steady-state conditions, the estimators to keep them working dur-
titative information about the health of frequency remains constant, and the ing severe power system disturbances
a person, which is then interpreted by power system state of health condition to prevent blackouts.
a doctor. A similar measurement-based is indicated based on voltage and cur- But keeping a steady coupling be-
approach is adopted in power grid rent magnitudes and phase angles at tween generation and consumption
operations. Electronic devices located
in substations acquire and process cur-
rent and voltage signals, breaker status,
meteorological data, etc., which are sent
to control rooms for digestion and inter-
pretation by power system operators.
The state of the power system is nec-
essary for energy market operation and
for the secure operation of the electric
network, to keep every operating vari-
able, such as power through the lines
or voltages at the premises of consum-
ers, within safe bounds. The power grid
state is also relevant in determining the
“safety margin,” i.e., how much room in
generation and transmission capacities
we have left before hitting the maxi-
mum load that can be served reliability.
Digital Object Identifier 10.1109/MPE.2022.3219171 figure 1. Layout of control room wall screens and dispatchers desks in a
Date of current version: 30 December 2022 typical EMS.

Never be
powerless
Hendrix and Kerite: Your partners in overhead

and underground reliability
Hendrix and Kerite offer overhead and • Shorter Lead Times
underground solutions for improved grid • Quicker Responses
reliability - and product that works at every • Customized Solutions
level is just the start. • Reduced Inventory Costs
Stop Outages before

they Stop Service
We keep your lights on (603) 673-2040 • marmonutility.com
exacts a toll. Synchronous generators circuit or other disturbance, one group which rely on phasor measurement units
thousands of kilometers apart must op- of generators could accelerate relative to (PMUs) synchronized through global
erate stably and in synchronism during another group, causing instability and the positioning system (GPS) clocks to col-
infinitely many load and power trans- loss of synchronism. When the frequency lect accurate time-stamped data from all
fer conditions; equipment outages; and moves away from the fixed values, wide- the grids, become necessary. With the
power disturbances. Following a short area monitoring systems (WAMSs), upward trend for the integration of
In This Issue
This issue focusehs on modern applications of state esti- communication system issues, and even cyberattacks.
mation in power system operations. Static state estima- The authors describe the application of DSE for online
tion was first implemented in the power industry in the dynamic model validation and calibration; measurement
late 1960s to provide operators snapshots of steady-state data validation; and data calibration.
system operating points using measurements collected by The author of the second article [A3] describes three
supervisory control and data acquisition systems. These decades of experience at Hydro Quebec toward enhanc-
static state estimators have evolved as the capabilities of ing situation awareness for power system control and
measurement equipment, communication systems, and wide-area protection applications. The end goal is en-
computing systems have improved, allowing system op- hanced EMS with fast-feedback controls supported by
erators more precise control of power systems and more dynamic wide-area situational awareness. The enhanced
complete utilization of assets. EMS enables automation to support decentralized control
Rapid changes to the generation mix with rapid in- in an increasingly digitized, decarbonized, and democra-
creases in inverter-based resources (IBRs), such as wind tized (4D) grid.
and photovoltaic generation, are changing the dynamic The third article [A4] describes a multifaceted synchro-
response of power systems to disturbances. System opera- nized measurement system to provide a full view of the sys-
tors need higher speed dynamic monitoring of the system tem dynamics from load to generation in a power system
operating state as well as more reliable system simulation with a high penetration of IBRs. Four types of synchronized
model data to provide improved situational awareness and measurement devices (SMDs) are described, each with char-
system resilience. This issue will demonstrate how dynamic acteristics to address different classes of monitoring and
state estimation (DSE) techniques can be used for dynamic control. Three of the SMDs provide phasor-based measure-
grid monitoring but can go beyond that to support model ment for specific applications: control; monitoring inverter-
validation, data validation, asset monitoring, stability con- based renewables; and monitoring loads. The fourth type
trol, and protection. of SMD provides waveform data. The authors conclude by
The “In My View” column [A1] discusses the need for describing distributed dynamic state estimators for the dy-
next-generation energy management systems (EMSs) to namic monitoring of systems with a high penetration of IBRs.
truly advance power system operation and control to meet The authors of the fourth article [A5] describe experi-
the challenges to modern power systems with the increase ences applying DSE using measurement data from a va-
in fast-responding IBRs. The author describes how state riety of sources to provide increased ampacity limits for
estimation needs to undergo significant advances to sup- underground cables and overhead through dynamic line
port this paradigm shift to a next-generation EMS. rating. Dynamic line ratings provide operators with the in-
Toward that end, our guest editors have organized creased ability to safely use transmission assets more fully.
five articles in this issue that describe advanced state es- The authors describe methods suitable for underground
timation techniques and present examples where these cables along with methods appropriate for overhead lines.
have been demonstrated in power systems around the The use of DSE improves accuracy by covering gaps in data
world. The first article [A2] provides an introduction to from sensors and weather stations. Application results
DSE using high-speed time-synchronized phasor mea- from Spain show that this method exceeds the season lim-
surement data. Power system operations rely on accurate its normally applied by the transmission system operator.
system models and accurate measurements. Obtaining The fifth article [A6] combines DSE with centralized
accurate dynamic model data has long been a challenge, substation protection approaches based on high-speed
increasingly so with the increased penetration of IBRs. measurements from merging units to create setting-less
Errors enter measurement data through metering errors, (Continued)

In This Issue (Continued )
protection schemes. The resulting protection schemes re- The sixth article [A7] is an unsolicited submission that
duce susceptibility to the common interrelated causes of fits the theme of this issue. The authors outline the chal-
protection misoperations: human errors, complexity, mis- lenges that the solution approach used (commonly applied
coordination, and hidden failures. In addition, the setting- static state estimators) faces in responding to bad measure-
less protection approach is less susceptible to cyberattacks ment data. The authors demonstrate the application of ro-
and supports self-healing in the protection scheme. The bust data analytics to ensure an accurate result using both
authors present results from pilot studies at five substa- synthetic test system data and data provided by utilities.
tions from two utilities. –Brian Johnson
intermittent renewables in the electric of WAMSs and PMUs as technologies estimating power system states un-
grid, it is conceivable that the “con- of choice for controlling the grid behav- der dynamical conditions. While “dy-
stant frequency” assumption enabling ior under varying frequencies that result namic states” have been used in power
static state estimation is a matter of the from load–demand mismatch or distur- system control and protection appli-
past as the new grid will constantly be bances such as wildfires and storms. I cations for a very long time, the cur-
subjected to small fluctuations of the encourage the reader to go back to the rent trend of interest was sparked in
generation and consumption balance, IEEE Power & Energy Magazine Sep- the aftermath of the 2003 blackout in
not to mention the frequent start–stop tember/October 2008 issue, whose con- the Northeast of America, when PMUs
of renewable power plants and inverter- tent is still relevant. The present issue became more widely accepted and
based resources (IBRs). is built on investigations from experts as the power grids were increasingly
This magazine has, on many occa- around the globe into the application of subjected to transients due to inverter-
sions in the past, stressed the importance powerful algorithms to PMU data for based renewables.
DSATools ™ DSAToolsTM CORE MODULES

• PSAT - Powerflow & Short-circuit Analysis Tool • SSAT - Small Signal Analysis Tool
• VSAT - Voltage Security Assessment Tool • DSA MANAGER - Common interface for on-line DSA
Dynamic Security Assessment Software
• TSAT - Transient Security Assessment Tool
Offering a simple, user-friendly interface

with extensive analysis options, DSAToolsTM DSAToolsTM ADD-ON MODULES
is a suite of software tools for power system • UDM EDITOR - Graphical tool to create and examine user defined models for TSAT, SSAT, and VSAT
analysis. The software is designed for • OPF-RA - Identification of remedial actions for improving voltage stability by using optimal power flow approach
applications in both off-line studies and
• SSR - Frequency-domain subsynchronous resonance analysis
on-line dynamic security assessment.
• ST DSA MANAGER - Interface for the study mode in on-line DSA
• TRI - TSAT-RTDS® Interface for Co-Simulation
• TPI - TSAT-PSCADTM Interface for Co-Simulation
• ePMU - Create simulated PMU data (IEEE C37.118)
Contact us: • CDT - Control design toolbox for PSS design and tuning
www.dsatools.com • HARMONICS - Harmonics analysis
[email protected] • CIM IMPORT - Import of powerflow data in CIM/XML format

With this year’s edition, ISGT
Europe retains its status of the
most significant European con-
ference in the domain of smart
grids, with world-leading scien-
tists and researchers from close
to 50 countries all over the globe
and nearly 800 authors/coauthors
in total. Besides the academic
part of the conference, all three
figure 2. The inaugural session. days were accompanied by the
Industrial Exposition and In-
Leader’s Corner tween the board and the IEEE Power dustry Panel sessions, where 10
The IEEE Board of Directors is com- & Energy Society (PES). He joins us in of the world-leading companies
posed of the three IEEE presidents; this issue to demystify the key facets of in the domain of smart grids pre-
the vice presidents and presidents of his role. [A8] He provides a glimpse of sented their latest products and
the six major boards; the secretary; the the various initiatives underway at the advancements to the confer-
treasurer; the 10 Region directors; the board level to enhance the IEEE opera- ence audience.
10 Division directors; and the directors tions, including a modernization of the
emeritus. Among all these officials, the IEEE Fellow elevation process, which Wrap Up
less obvious role is that of the Division will likely benefit PES members from The editors and authors of this issue
directors. Claudio Canizares, the Divi- industry in the long term. have to be congratulated because they
sion VII director, is the interface be- accepted the task of tackling a subject
Book Reviews that was initially felt too technical and
In this issue’s “Book Review” column, unsuitable for a magazine. The concept
[A9] Hannele Holttinen discusses the of state is not taught in the first year of
Wind Energy Handbook, Third Edition, engineering. Some engineers can even
written by Tony Burton, Nick Jenkins, Er- end their training without being ex-
vin Bossanyi, David Sharpe, and Michael posed to this concept, which is required
Graham and published by Wiley in May only for solving advanced problems
2021. The reviewer notes in control engineering but which is a
necessity for large-scale power system
Exposing electrical engineers control and operations. The authors
to the entirety of wind turbine de- did a great job explaining complex
sign might be a good idea—just concepts arising from their research
reading the introduction of each in a tutorial manner. On the algorithm
section and glancing through the side, they discussed the dynamic state
content gives a great overall pic- estimation (DSE) of renewable energy
ture of wind turbine technology. resources; synchronous generators;
and digital substations. On the appli-
Society News and Awards cation side, they highlighted dynamic
In other news, the 10th IEEE PES Inno- situational awareness in control rooms;
vative Smart Grid Technologies (ISGT) model calibration based on wide-area
Europe Conference was held on 10–12 PMU historical data; and control and
October 2022 and organized and hosted protection applications of DSE at sub-
by the Faculty of Technical Sciences, station and system levels.
University of Novi Sad, Serbia (Fig- It is believed that by using the infor-
ure 2). This year’s conference theme mation herein, wide-area power grid
was “Together Towards Digitized, health diagnosis and early warning sys-
Decarbonised, and Distributed Smart tems can be constructed and deployed in
Grids.” Luka Strezoski and Amirhos- control rooms, bringing wide-area con-
sein Sajadi, respectively, the general trols closer to implementation. The sys-
conference chair and the technical pro- tem operator will be better tooled to iden-
gram chair, briefly summarize the main tify and trace the dynamic security level
facts of the event as follows: of the power grid in real time; identify

the different kinds of hidden troubles; ized, digitized, and democratized identification: Inverter-dom-
and bring forward the integrated early electricity grids,” IEEE Power inated power systems,” IEEE
warning report, a secure control strategy E nerg y Mag., vol. 21, no. 1, Power Energy Mag., vol. 21, no.
to eliminate the risk of electric disas- pp. 44–58, Jan./Feb. 2023, doi: 1, pp. 59–72, Jan./Feb. 2023, doi:
ters in the pregnant phase, and hence, 10.1109/MPE.2022.3219179. 10.1109/MPE.2022.3219180.
prevent blackouts from happening. [A4] T. Bi, M. Zhu, and H. Liu, “A [A7] S. Wang, Z. Huang, R. Huang,
powerful tool for power system N. Zhou, and J. Zhao, “Multifold
Appendix: Related Articles monitoring: Distributed dynamic insights for power system dynam-
[A1] B. Fardanesh, “Advances in state state estimation based on a full- ics from data assimilation: Meet-
estimation and the next generation view synchronized measurement ing current challenges,” IEEE
EMS: Finding a killer applica- system,” IEEE Power Energy Power Energy Mag., vol. 21, no.
tion,” IEEE Power Energy Mag., Mag., vol. 21, no. 1, pp. 26–35, 1, pp. 36–43, Jan./Feb. 2023, doi:
vol. 21, no. 1, p. 88, Jan./Feb. 2023, Jan./Feb. 2023, doi: 10.1109/MPE. 10.1109/MPE.2022.3219165.
doi: 10.1109/MPE.2022.3219172. 2022.3219166. [A8] C. Canizares, “Reflecting PES
[A2] B. Gou and D. Shue, “Advances in [A5] J. A. Rosendo-Macías et al., “The views: While representing IEEE
algorithms for power system static Spanish experience: Squeezing interests,” IEEE Power Energy
state estimators: An improved so- line ampacities through dynamic Mag., vol. 21, no. 1, pp. 10–12,
lution for bad data management,” line rating,” IEEE Power Energy Jan./Feb. 2023, doi: 10.1109/MPE.
IEEE Power Energy Mag., vol. 21, Mag., vol. 21, no. 1, pp. 73–82, 2022.3219192.
no. 1, pp. 16–25, Jan./Feb. 2023, Jan./Feb. 2023, doi: 10.1109/MPE. [A9] H. Holttinen, “Wind energy
doi: 10.1109/MPE.2022.3219168. 2022.3219167. handbook,” IEEE Power Ener-
[A3] I. Kamwa, “Dynamic wide area [A6] S. Meliopoulos et al., “Dynamic gy Mag., vol. 21, no. 1, p. 83, Jan./
situational awareness: Propelling estimation-based protection and Feb. 2023, doi: 10.1109/MPE.2022.
future decentralized, decarbon- hidden failure detection and 3219169. p&e
Black box vendor control

simulation with the
RTDS® Simulator
Enabling real-time simulation of detailed controls while protecting manufacturer IP

The RTDS Simulator is the world standard for real-time power system simulation and hardware-in-
the-loop testing of control and protection equipment. The new GTSOC allows for inclusion of OEM
controls in the real-time simulation, facilitating detailed control modelling without requiring physical
hardware to be present - and without compromising the vendor’s intellectual property.
YOUR WORLD IN REAL TIME.

RTDS.COM/GTSOC
leader’s corner
Claudio Canizares
reflecting PES views

while representing IEEE interests
N
NINE MONTHS AGO, I WAS ENT- sidebars allow us to much more effec- ✔ T he Fut u r e of E ngagement
rusted with reflecting IEEE Power & tively carry out the relevant and impor- (Cochairs David Koehler and
Energy Society (PES) views and in- tant discussions and business conduct- John Vig)
terests at IEEE’s Board of Directors ed at IEEE BoD and TAB meetings. ✔ Top-Level Medals and Award
(BoD) and Technical Activities Board My activities as Director this year (Chair Richard Cox)
(TAB) as the incoming Division VII started with a planned in-person BoD ✔ Coordinate IEEE’s Response to
director. I deliberately characterize the retreat and a meeting in January that Climate Change or CRCC (Chair
role’s scope as “reflecting” the PES’s unfortunately became virtual due to Saifur Rahman)
visions and wishes rather than “repre- lingering pandemic circumstances. These committees are focused on ad-
senting” the PES because the director’s Thankfully, the BoD meetings in June dressing a variety of issues and shap-
fiduciary responsibility is to IEEE and and August took place in person, which ing aspects of the present and future
not the PES. In fact, at the PES board does make a difference, as mentioned, IEEE. The committees have since been
meetings and committees, I represent enabling the personal successfully progress-
IEEE’s interests in our Society. I re- interactions that make ing toward their goals
alize that this might be a surprise to the BoD and TAB work At the PES to fulfill their char-
some of you, as it was to me, since PES much more effective, as ters, regularly report-
members elected me. However, this is I know firsthand from board meetings ing back to the BoD at
a relevant distinction to be aware of so
that the role of Division directors can
my work as a Direc-
tor under both formats.
and committees, in-person and virtual
meetings, and will pro-
be properly framed and understood At the retreat and first I represent IEEE’s vide final reports and
within IEEE and the Societies and meeting of the year, recommendations at
councils that elect them. new IEEE President interests in our the last BoD meeting of
I would like to recognize my pre-
decessor, Director Miriam Sanders,
Ray Liu set the tone
and the agenda for the
Society. the year in November,
if not before. Due to the
and her significant contributions to the IEEE year. In particu- limited space available
IEEE BoD, TAB, and PES as well as lar, the following IEEE Ad Hoc Com- in this column to otherwise provide a re-
her commitment, dedication, and hard mittees were formed and approved: view of all these committees, I will com-
work as Division VII Director. I had ✔ Data-Based Business Strategy ment specifically on the two Ad Hoc
the opportunity to observe her in ac- (Chair Rakesh Kumar) Committees that I have been appointed
tion during last year’s board meetings ✔ Fellows Processes (Chair José to and am actively participating in—the
as director-elect, and I can personally M. F. Moura) Fellows Processes and CRCC, which are
attest to the terrific job she did as a ✔ Guiding the IEEE Constitutional probably of particular interest and the
Director. We are all thankful at IEEE Amendment (Chair James Look) most relevant to the PES.
and the PES, especially considering the ✔ IEEE as Your Professional Home The Fellows Processes Commit-
fact that most of her tenure was car- (Chair Barry Tilton) tee is a continuation of similar Ad Hoc
ried out virtually due to the pandemic ✔ IEEE Humanitarian Efforts (Chair Committees established in 2020 and
lockdowns. This certainly presented a Susan K. (Kathy) Land) 2021, which former Director Miriam
challenge as in-person interactions and ✔ IEEE in 2050 (Chair Roger Fujii) Sanders participated in and contributed
✔ IEEE President-Elect Campaign to, with former PES President Damir
Date of current version: 30 December 2022 Pilot Program (Chair David Green) Novosel cochairing the 2021 committee.

Considering that our membership size der, regional, and professional tion diversity that will vet Society/
now makes the PES the second biggest affiliation diversity. I am happy Council Nomination and Eleva-
IEEE Society and close in size to the to report that the PES has just set tion Committees for diversity. This
larger Computer Society, this commit- up this Nomination Committee subcommittee will also review the
tee is of particular relevance, as the under the leadership of President normalized rankings from the
PES has consistently and significantly Jessica Bian, who has ardently divisional committees to flag di-
underperformed with respect to IEEE supported the need to address versity anomalies in the provided
Fellow elevations compared to other diversity, equality, and inclusion rosters for further review.
IEEE Societies and Councils. One of issues in the IEEE Fellow eleva- The Ad Hoc Committee was required by
the main PES concerns, given its sig- tion process. the BoD to fine-tune the approved rec-
nificant membership from industry, ✔ Mandate a composition of the Soci- ommendations to address some issues
is that nominations of industry col- ety/Council Fellow Elevation Com- that were raised by the Board. Further-
leagues seldom result in elevations mittees that guarantees gender, re- more, we are also developing pending
compared to elevations from academia, gional, and professional affiliation recommendations to address diversity
which is compounded by the lower diversity. Both this committee and composition, ranking, and vetting issues
number of nominations of industry col- the Nomination Committee will at the IEEE Fellow Committee level,
leagues. Therefore, the IEEE BoD set be selected by the Society/Council which will be proposed to the IEEE
up this Ad Hoc Committee with the Nomination & Appointment Com- BoD at its September meeting.
following charter: mittee and appointed by its board I am also an active member of the
as well as vetted by the IEEE Fel- CRCC IEEE Ad Hoc Committees, led
Consider “specific improvements low Committee. by PES Past-President and IEEE Pres-
to the processes, mechanisms, ✔ Change the nomination normal- ident-Elect Saifur Rahman. Para-
and evaluation criteria, including ization and ranking process so phrasing, this committee’s charter
proposing new structure and re- that it will now be carried out includes: identifying ongoing climate
taining best practices;” propose at the Divisional level, rather change (CC) and decarbonization efforts
“specific improvements to the than at the IEEE Fellow Com- across IEEE; recommending new CC
composition of the Fellows Com- mittee level, which is the cur- programs and activities for IEEE;
mittees at the Society/Council and rent practice. This change will developing a CC policy statement for
IEEE level, including the selec- give more weight to the Society/ consideration by the BoD; identifying
tion and appointment processes;” Council rankings in the selection key external partners for cooperative ef-
and develop “specific improve- process. forts; and ensuring that IEEE’s response
ments to the Fellows nomination ✔ Provide appropriate access to to CC is substantial, appropriate, and
process to increase the diversity anonymized reference letters and properly engages IEEE, its Members,
of qualified nominees submitted endorsements to and the larger technical
for consideration by the Fellows the Society/Coun- profession. Hence, led
Committee to better reflect the di- cil Fellow Eleva- I would like by the President-Elect,
verse makeup of IEEE (including tion Committees we have been collect-
industry sector, geographic, gen- so that they can
to take this ing information from
der, and underrepresented com- better ascertain opportunity to multiple IEEE organi-
munity diversity).” the significance of zational units (OUs) on
the work and con- kindly request the many CC initia-
After several months of hard work, tributions of the tives currently under-
predicated on the learnings from the nominees.
your active way within IEEE.
two previous Ad Hoc Committees, we ✔ Deliver training for participation I have been par-
made several recommendations that all committees in- ticularly impressed
were approved in principle at the Au- volved in the nomi- in IEEE as PES by the IEEE com-
gust BoD meeting, addressing most of nation and selection mitment and effort
the committee charter assignments. process at all levels,
members. led by its Executive
These recommendations include: as well as provide Director Steve Welby
✔ Create a Fellow Nomination detailed statistics of previous nomi- to reduce the carbon footprint of IEEE
Committee in charge of ac- nation and elevation processes. operations, particularly its buildings.
tively seeking/developing Fel- ✔ Create a Diversity Oversight Sub- I am also pleasantly surprised by the
low nominations at the Society/ committee of the IEEE Fellow large number of OU committees, activ-
Council level, with a prescribed Committee with proper gender, ities, and efforts addressing CC issues.
composition that guarantees gen- regional, and professional affilia- The committee (and particularly the

The PES is cal society in the world with close to
half a million members. Your active
systems. If you are interested in getting
involved, please feel free to reach out to
woefully underr- participation in IEEE will help shape me to discuss how to best become en-
its future and thus contribute to positive gaged in the running and decision mak-
epresented in changes for humanity through the many ing of IEEE, which is in dire need of
many of the activities in which IEEE is involved,
several of them in power and energy
and, hence, seeking new blood.
p&e
multiple OUs
and standing
committees.
President-Elect) has also been engag- E AS Y P O W E R® M O D E L I N G
& A NA LYSI S SO F TWARE
ing multiple institutions, groups, and
organizations outside IEEE, several of
them with strong PES ties, which are
active and visible in the CC space to
establish relevant collaborations, in-
cluding the participation of IEEE at the
Intelligent • Intuitive • Instantaneous
Efficient Modeling
2027 Conference of Parties of United
Nations Framework Convention on
Climate Change (COP’27) in Egypt. · Utility scale and Distributed Generation
In closing, I would like to take this
· Transmission
opportunity to kindly request your
active participation in IEEE as PES · Distribution
members. Currently, the PES has six · Commercial / Industrial
active members in the BoD–President-
Elect Saifur Rahman; Region 7 Direc- Powerful Analysis
tor Rob Anderson; Region 9 Director · ANSI and IEC Short Circuit
Enrique Tejera; TAB VP Bruno Mey- · Protective Device Coordination
er; MGA VP David Koehler; and my-
· Power Flow
· Arc Flash - OSHA Compliant Up to 800 kV
self, which is quite unique for the PES
and a great opportunity for the Soci-
ety’s views and interests to be reflect- · Harmonics
ed within IEEE to impact its relevant
decisions and future. This is especially · Grounding
true in the context of Net-Zero 2050, in ISO 9001:2015 Certiﬁed & NUPIC Audited
which the PES must play a natural and
leading role within IEEE. However,
the PES is woefully underrepresent- As Easy As
ed in many of the multiple OUs and
· Easy to Learn
CURRENT IN AMPERES X 100 AT 480 VOLTS
.5 .6 .8 1 2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 100 2 3 4 5 6 7 8 9 1000 2 3 4 5 6 7 8 9 10000
1000 1000
TX-2
standing committees. Therefore, one

800 800
FLA
600 600
500 500
0.15
400 400
300 TX-2 300

1 / 1.288 MVA
200 13.8 - 0.48 kV 200
6%
18.71
100 BL-1 100 Upstream Upstream Electrode Est Arc Flash Working Incident
GE AKR-50 Arc Fault Arc Fault Electrode
80
TX-2 80
Bus Name Bus kV Trip Device Trip Device Equip Type Configuration Gap Boundary Distance Energy
1000 / 1288 kVA 1600/1760 Name Function (mm) (inches) (inches) (cal/cm2)
· Easy to Use
60 60
50 6% 50
.13
40 40
BUS-3 13.8 R-6 51/50 Open Air VOA 152 32.1 + 26 1.7
24
of my main objectives as BoD director,

30 BL-3 30 BUS-7 13.8 R-7 51/50 Open Air VOA 152 30.5 + 26 1.5
GE AKR-30H
20 800/400 20 M-1 0.48 BL-3 Other HCB 32 31.6 + 18 3.8
1.90
MAIN SWG 3.8 R-18 51/50 Switchgear VCB + HCB 152 189.3 + 18 59.3
BL-1
10 GE MVT
MV T-Plus
MVT-Plus 10 REFINER 2.4 R-7 51/50 Switchgear VCB + HCB 104 261.9 + 18 122.8
17.58
8 Sensor = 1600 8
TIME IN SECONDS
TIME IN SECONDS
Plug = 1600 SWG-4 0.48 R-6 51/50 Switchgear VCB + HCB 32 213.2 + 18 135.1
6 6
Cur Set = 1.1 (1760A)
Cur
.48
5 5 MCC-1 0.48 BL-2 MCC VCB 25 53.6 + 18 6.9

LT
T Band = 1 M-1
19
4 4
R R 3
STPU = 2
STPU
ST Delay
ST D
.5 (4400A)
2.5
y = IInt
elay
ela nt
BL-5
3
MCC-2 0.48 BL-3 MCC VCB 25 53.1 + 18 6.8
C-H HFD
ST
ST Delayy II²t
²t = Out PNL-1 0.208 FS-2 Panel VCB + HBB 25 48.2 + 18 7.1
2 Override
Override = 50000A 225/150 2
PNL-2 0.208 PNL-1 Panel VCB 25 57.8 + 18 7.7
besides actively participating in and

1 1
.8 .8
.6 BL-3 .6
· Fast Results
GE MVT
MV T-9
MVT-9
d
.5 .5
Risk Hazar
.4 Sensor = 800 .4
Plug = 800
and Shock
BL-1
Arc Flash priate PPE Required

.3 C ur Set = 0.5
Cur 0.5
5 (400A) .3
18627A
LT Band = 1 C-6
Arc Flash and Shock
.2 Inst
Inst = 4 (3200A) .2
1 - 400 kcm
kcmil
mil CU
CU
Appro Risk Hazard
.1
TX-2
1000 / 1288 kVA .1
Boundary Appropri
Incident Energy ate PPE Required
.08 INRUSH .08 Flash Hazard inches - Arc Flash or arc rated
coverall
4’ - 0”
contributing to relevant and impactful

BL-5 18
cal/cm at pants
arc rated 4’ - 0”
2
.06
06
C tler Hammer
Cutler H Serie
Series C .06
Flash Hazard Boundary
.05
HFD .05 6.0 shirt and 6.0
BL-3 Arc-rated cal/cm2 at 18 inches
cover is removed
.04 .04
Frame = 225A (150AT) - Arc Flash Incident
29249A
.03 Trip = 150 .03
Hazard when Energy
kV Shock
.02 .02 0.48 Limited Approach 0.48
BL-5 kV Shock Hazard
PANEL PANEL PANEL 20574A 3’ - 6” Restricted
Approach 3’ - 6” when cover is removed
27B) Limited Approach
1’ - 0” (Fed By:1’ - 0” ONLYRestricted
t Name: MCC-23A CONFIGURATION
.01 .01
.5 .6 .8 1 2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 100 2 3 4 5 6 7 8 9 1000 2 3 4 5 6 7 8 9 10000 Approach
CURRENT IN AMPERES X 100 AT 480 VOLTS Equipmen SYSTEM Equipment Name:
NORMAL
VALID FOR VALID FOR NORMAL
SWG-4 (Fed By:
18A)
SYSTEM CONFIGURATION
ONLY
IEEE decisions and committees, is to

enhance the PES members’ participa- Try instantly online or download a free demo copy at:
tion in IEEE.
My closing remarks are thus a www.EasyPower.com/demo
plea to PES members to become ac-
tive IEEE volunteers, which is quite
®
rewarding and impactful, given the
relevance and size of the institute, Power made easy.
the largest and most reputable techni-

guest editorial
Junbo Zhao and A.P. Meliopoulos
parameter visibility
next-generation energy management
I
IN THE AFTERMATH OF THE 1965 slow rates. Many technological advances tem. System operators need a higher
blackout, the need for robust power occurred since then that affected state degree of observability and more reli-
system state estimators became appar- estimation. The introduction of computer able models of the power network to
ent. Edward Teller, a member of the task relaying in 1970, and later, the introduc- improve their situational awareness
force investigating the 1965 blackout, tion of microprocessor (numerical) re- and decision making. This special is-
wrote “the electric power system is a lays (1984) have changed sue intends to share the
dinosaur with a large body and a small SCADA. Specifically, the dynamic state and param-
brain.” The committee recognized the numerical relay has be- It is clear eter estimation develop-
need to have a way of knowing the op- come an integral part of ments for improved grid
erating conditions of the power system SCADA.
that the visibility and demonstrate
in real time. Work following the 1965 Synchronized mea- control and their practical industry
blackout established many engineering surements with microsec- applications in dynamic
principles by which we presently oper- ond precision were intro- operation modeling, monitoring,
ate the electric power system. It is clear duced in 1992 and had a operation, stability con-
that the control and operation of power major impact on power
of power trol, and protection.
systems are primarily model based. State system state estimation, systems are The work “Advances
estimation provides the operating condi- enabling the development in algorithms for power
tions as well as the real-time model of the of commercially avail- primarily system static state esti-
system. Using the model and operating able linear state estima- mators: An improved so-
conditions, we compute control actions, tors as well as distributed
model based. lution for bad data man-
including economic operation; the state estimators. The latest agement,” by Bei Gou and
maximization of operational security; technology is the introduction of merg- David Shue [A1], analyzes the key issues
and other important operational controls. ing units that perform GPS-synchronized of the widely used power system state es-
The first implementation of power sample value measurements at high rates timator that relies on the weighted least
system state estimation in the late 1960s (80 samples per cycle or 256 samples per square algorithm, especially algorithm
was limited by the available technology. cycle). The wide use of phasor measure- convergence in the presence of bad data.
The supervisory control and data acquisi- ment unit measurements makes the dy- Some realistic power systems are used to
tion (SCADA) system was developed us- namic state estimation possible. demonstrate the value of having a robust
ing existing technology based on sensor The technological advances that state estimator. Besides a robust tool to
measurements. The least squares math- enabled the evolution of state estima- capture power system steady-state op-
ematical approach provided a computation for power systems are important erations, a dynamic view provides more
tionally efficient method to perform state for addressing the new challenges faced granular insights for online monitoring.
estimation. Many challenging problems by modern power systems. Significant The article “A powerful tool for power
were addressed, and reasonable solutions challenges have been generated by the system monitoring: Distributed dynamic
have been found. It is clear that the first massive deployment of inverter-based state estimation based on a full-view
implementations of state estimation were generations (IBRs), such as wind farms, synchronized measurement system,” led
limited by the prevailing technology at battery storage, and solar photovolta- by Tianshu Bi [A2], provides a full-view
the time; they were static and operated at ics, changing the dynamic paradigm of synchronized measurement system for the
electric power grids. The future power power electronics-enabled power systems
Date of current version: 30 December 2022 system is an IBR-dominated power sys- that have been developed and installed

in China. Four types of synchronized due to a variety of reasons, such as data from remote weather stations. Field
measurement devices (SMDs) have been miscoordination, complexity, hu- results show that the available ampacity
developed, including SMD for renew- man errors, and hidden failures. The limits determined by this methodology
ables; SMD for control applications; SMD integration of more IBRs makes the exceed the conservative seasonal values
for loads; and SMD for waveforms. These problem even more complicated. To customarily used by the Spanish TSO
SMDs provide the data foundation for the address these issues, in the article most of the time.
development of distributed dynamic state “Dynamic estimation-based protection
estimators for IBR-dominated power sys- and hidden failure detection and iden- Appendix: Related Articles
tem dynamic monitoring. tification: Inverter-dominated power [A1] B. Gou and D. Shue, “Advances in
Dynamic system monitoring relies systems,” Sakis Meliopoulos and his algorithms for power system static
on accurate system models and reliable colleagues [A5] propose dynamic state estimators: An improved so-
measurements. However, with more state estimation-enabled control and lution for bad data management,”
IBRs along with flexible loads, it is chal- protection systems. This dynamic IEEE Power Energy Mag., vol. 21,
lenging to obtain a detailed and accurate state estimation-based approach pro- no. 1, pp. 16–25, Jan./Feb. 2023,
dynamic system model. Furthermore, vides a centralized substation protec- doi: 10.1109/MPE.2022.3219168.
due to communication issues, cyberat- tion system that mitigates the main [A2] T. Bi, M. Zhu, and H. Liu, “A pow-
tacks, and instrument malfunctions, not root causes of relay misoperations by erful tool for power system moni-
all measurements are reliable. The work removing the complexity and coordi- toring: Distributed dynamic state
“Multifold insights for power system dy- nation via the setting-less relay tech- estimation based on a full-view syn-
namics from data assimilation: Meeting nology; facilitating the detection of chronized measurement system,”
current challenges,” led by Shaobu Wang cyberattacks; and enabling self-heal- IEEE Power Energy Mag., vol. 21,
[A3], presents the development and ap- ing against hidden failures. Results no. 1, pp. 26–35, Jan./Feb. 2023,
plication of high-speed phasor measure- from pilot studies at three Southern doi: 10.1109/MPE.2022.3219166.
ment technologies for online dynamic Company substations and two New [A3] S. Wang, Z. Huang, R. Huang, N.
model calibration, data validation, and York Power Authority substations are Zhou, and J. Zhao, “Multifold in-
data calibration. The core algorithm is presented to demonstrate the high val- sights for power system dynamics
based on a two-step prediction-correction ue of the new technologies. from data assimilation: Meeting
process. Pilot project results demonstrate Finally, the work “The Spanish ex- current challenges,” IEEE Pow-
the value of implementing the dynamic perience: Squeezing line ampacities er Energy Mag., vol. 21, no. 1,
state and parameter estimation tools for through dynamic line rating,” by J. A. pp. 36–43, Jan./Feb. 2023, doi:
model and data quality improvement. Rosendo-Macías [A6], presents the recent 10.1109/MPE.2022.3219165.
Enhanced system situational aware- Spanish experience in the application of [A4] I. Kamwa, “Dynamic wide area
ness provides essential information for dynamic state estimation to enhance the situational awareness: Propelling
power system control and protection ap- dynamic line rating technologies for un- future decentralized, decarbon-
plications. In the article “Dynamic wide derground and overhead ac transmission ized, digitized, and democratized
area situational awareness: Propelling lines. Dynamic line rating is becoming a electricity grids,” IEEE Power En-
future decentralized, decarbonized, key component of modern EMS, enabling ergy Mag., vol. 21, no. 1, pp. 44–
digitized, and democratized electric- transmission system operators (TSOs) 58, Jan./Feb. 2023, doi: 10.1109/
ity grids,” Innocent Kamwa [A4] shares or independent system operators to use MPE.2022.3219179.
32 years of experiences at the Hydro- transmission assets more efficiently while [A5] S. Meliopoulos et al., “Dynamic
Quebec Research Institute (IREQ) on maintaining current reliability standards. estimation-based protection and
power system monitoring and control. For underground lines, a method for cal- hidden failure detection and
The long-term vision is to build a new culating the maximum operating current identification: Inverter-dominat-
evolution of the energy management is discussed based on measurement-based ed power systems,” IEEE Power
system (EMS), termed EMS/C or EMS dynamic models describing the tempera- Energy Mag., vol. 21, no. 1, pp.
with faster feedback controls, enabled by ture evolution over time in different sec- 59–72, Jan./Feb. 2023, doi: 10.
dynamic wide-area situational awareness tions of the cable. The problem of param- 1109/MPE.2022.3219180.
and related infrastructure and tools. This eter uncertainty and time variability has [A6] J. A. Rosendo-Macías et al., “The
closed-loop automation contrasts with the also been addressed via dynamic state Spanish experience: Squeezing
current EMS, which, except for the auto- estimation. Field results of a pilot project line ampacities through dynamic
matic generation control and secondary show that the prediction errors of the dy- line rating,” IEEE Power Energy
voltage control, relies mostly on market- namic state estimation-based model are Mag., vol. 21, no. 1, pp. 73–82,
and human-driven feedforward controls. lower than 1.5 °C in all cases. For over- Jan./Feb. 2023, doi: 10.1109/MPE.
Legacy protection and control sys- head lines, a dynamic conductor tempera- 2022.3219167.
tems are vulnerable to misoperations ture estimation tool is developed using p&e

Advances in
Algorithms for
Power System
Static State
Estimators
By Bei Gou and David Shue
M
MORE THAN 50 YEARS HAVE PASSED SINCE THE FIRST STATE
estimator was introduced to power system operations, by Fred Schweppe,
in 1969. Today, state estimation is the most important function in the
energy management system (EMS) because the solution of the state
estimator is the basis of other functions, including contingency analysis,
security assessment, dynamitic security analysis, optimal power flow,
and so on. Numerous research papers and reports have been published
about trying to improve the performance of the power system state esti-
mator since 1969. After the efforts of several decades by researchers and
power engineers, today’s state estimator can be found in almost every
power system control center worldwide. Power system state estimation
was considered a mature area until the occurrence of a 14 August 2003
blackout in the northeastern United States. The state estimator at the con-
trol center crashed due to a single topology error in the system model,
and the system operator could not take appropriate actions to avoid the
occurrence of the blackout.

Date of current version: 30 December 2022
16 IEEE power & energy magazine 1540-7977/22©2022IEEE january/february 2023

The issues that state estimators are not able to guarantee a solution and not able to reject bad data were
well known in the community; researchers and engineers call this a problem of robustness. The August 2003
blackout raised attention for power system engineers and researchers that state estimators still had room
for improvement. Efforts have been made in the past 20 years, trying to solve the problem of robustness;
however, no major breakthroughs were achieved in practice, but incremental improvements were definitely
accomplished. A survey about the loss of EMS performance, published by the North American Electric
Reliability Corporation (NERC) in 2020, showed that 45% of the loss of EMSs is caused by the crashes
of state estimators, which implies that today’s
state estimators are still vulnerable to bad data
An Improved Solution for in the measurements, the system model, and
the parameters.
Bad Data Management and This article analyzes the key issues of the
static power system state estimator, especially
State Estimator Convergence the widely used solver, the weighted least-
squares (WLS) algorithm. This article lists
the main reasons why WLS state estimators
are vulnerable. It also provides some new
understandings of these problems. In addition,
it points out that the current ways of dealing with the weights of zero-injection measurements at power
control centers is questionable, even though it is practiced widely in today’s power industry. Finally, this
article introduces a new state estimator that is completely different from WLS-based state estimators. This
new state estimator is a result of
several National Science Founda-
tion-sponsored projects conducted
by the authors, from 2013 to 2021.
The numerical results of a massive
amount of random testing creating
different types of bad data for a
large test system are delineated in
this article. Further, testing results
using a real case, provided by a
power company in North Amer-
ica, are presented.
Introduction to
Power System Static
State Estimator
Monitoring devices are installed at
substations to provide voltage and
current signals through instrument
transformers. Real and reactive
power flow measurements as well
as voltage and current magnitude
measurements are then generated
from these signals and processed
by the supervisory control and
data acquisition (SCADA) system.
In addition, phasor measurements,
synchronized via the GPS system,
©SHUTTERSTOCK.COM/ILYA LUKICHEV
are supplied by phasor measure-

ment units (PMUs). PMUs pro-
vide direct measurements of phase
angles associated with voltage and
current measurements. All these
different types of measurements

are collected and then processed by the state estimator at the whether it will be possible to estimate the states of the
control center. entire system. If the measurement configuration is not
To perform the state estimation, the studied power sys- able to estimate the entire system, it is often possible to
tem needs to be modeled in a computer with the assumed or instead detect all possible subnetworks (called observable
monitored status of circuit breakers as well as the entire set islands). To make the entire system observable, it is com-
of network parameters, which include the following: mon to supplement the existing measurement configuration
✔ parameters such as the resistance, reactance, and line with sham measurements (called pseudomeasurements),
charging capacitance of all transmission lines which are formed by using scheduled generation informa-
✔ parameters associated with substation components, tion and forecast bus loads. The function of the pseudomea-
including transformers, transformers with off-nomi- surements is to connect and merge all observable islands
nal tap positions, shunt capacitors and reactors, and into a single island. In general, pseudomeasurements carry
series capacitors, as well as the status of power con- big errors, and therefore, to minimize their impact on the
trol devices, such as static var compensators and static state estimate accuracy, only a minimum set of pseudomea-
synchronous compensators. surements is introduced. Pseudomeasurements are able to
The state estimator uses the SCADA measurements from merge multiple observable islands into a single observable
substations and the network model to solve for an estimated island by changing the unobservable branches (marked in
state of the power system for that time window. The resulting red in Figure 2) between observable islands into observable
solution is presented to the system operator and used as an branches. These tasks are conducted by the observability
input for other functions in the EMS (see Figure 1). analysis function.
Before running the state estimator, the first step is to Now the program is ready to solve the state estimation
check whether the measurement configuration is able to problem, based on the given set of measurements, the given
estimate the bus voltage phasors (called states) of the entire system topology, and related parameters. State estima-
system, a step called observability analysis. The number, tion is a function that makes use of the redundancy of the
type, and location of the measurements will determine available measurements to determine the best estimate of
the system state, which is defined as the set of all the bus
voltage magnitudes and phase angles in the system. Among
the large number of possible algorithms applied to solve the
state estimation problem, the WLS and least absolute value
SCADA
are the most commonly used algorithms. Furthermore, the
System Operator WLS algorithm dominates the commercial power system
Grid
State Solution
state estimator market and is utilized at almost every con-
Estimator trol center equipped with a state estimator. Although many
variations of state estimation solvers have been applied in
Other Functions of EMS
today’s power system control centers, they are all based on
Network the WLS algorithm.
Model
The goals of this article are as follows:
✔ to better understand the mechanism of a WLS-based
figure 1. The operating structure of the state estimator in a state estimator
power system control center. ✔ to analyze the inborn flaws of a WLS state estimator
✔ to introduce an alternative
state estimator and demon-
Observable strate performance with test
Island
Pseudomeasurements results
WLS State Estimator

Single
Observable Island History of Least-Squares
State Estimator
The first clear and concise expo-
sition of the least-squares method
Observable Observable was published by Adrien-Marie
Island Island
(a) (b)
Legendre, in 1805. The method
of least squares was chosen as a
figure 2. Merging (a) multiple observable islands into (b) a single observable island standard tool in astronomy and
by using pseudomeasurements. geodesy within 10 years after

The WLS algorithm dominates the commercial power system
state estimator market and is utilized at almost every control
center equipped with a state estimator.
Legendre’s publication, which established an extraordinarily monitor their systems. The necessity of the state estimator to
rapid acceptance of the scientific technique. In 1822, Johann power system operations has been proved through many years
Carl Friedrich Gauss stated that the best linear unbiased esti- of successful operating experience. However, the capability
mator of the coefficients in a linear model, where the errors of bad data detection and identification, one of the original
have a mean of zero, are uncorrelated and have equal vari- important features for state estimation, still has much room for
ances is the least-squares estimator. In summary, a least- improvement, especially for the multiple and mixed instances
squares estimator is optimal under the following conditions: of bad data in the measurements, the system model, and sys-
✔ The model is linear: A linear model is an equation tem parameters. Furthermore, today’s widely used WLS state
that describes a relationship between two quantities estimator is not able to guarantee 100% convergence, which
that show a constant rate of change. makes power system monitoring vulnerable to bad data.
✔ The errors have zero mean: The mean is the average
of a data set. Some Facts About the WLS Static
✔ The errors have equal variances: Variance measures State Estimator
variability from the average or mean.
The second and third conditions imply that the least-squares WLS State Estimators Are Not Optimal
estimator is optimal under the condition that the measure- in Power System Applications
ments contain only white noise. A random signal is consid- A least-squares estimator is optimal only when the model is
ered “white noise” if it is observed to have a flat spectrum linear and the measurements contain nothing but noise with
over a medium’s widest possible bandwidth. zero mean and equal variance. However, it is observed that
Since the date when the state estimator was first intro- bad data are unavoidably present in the measurements and
duced to power systems, a large number of variations have system model in power systems. Furthermore, the power sys-
been proposed for the WLS to try to improve performance. tem model is not a linear model. Therefore, the least-squares
Even though power system state estimators have been state estimator is not optimal, and it is actually a biased state
improved through the efforts of researchers over the past sev- estimator when applied in power systems, due to the presence
eral decades, the robustness problem is not completely solved. of bad data.
In 1982, Tomas Enciso Dy-Liacco summarized the need for
state estimation in power system operations in an article. He WLS State Estimators Are Not Available When
listed the following reasons state estimation is needed. They Are Most Needed
✔ The most practical way of obtaining knowledge of the Since the least-squares estimator uses the sum of the square
system state is through state estimation. of the difference between the samples and their recovered val-
✔ Once the voltages are known through the state estima- ues, it is essentially an average state estimator, from a statistics
tor, all the electrical variables, including those that are point of view. To show the accuracy of different meters, a WLS
not telemetered and which are missing, can be readily state estimator considers a weight for each measurement to
calculated. reflect the degree of its importance in the optimization objec-
✔ State estimation may be characterized either as a tive. It performs very well when measurements contain only
phase angle telemetry system implemented via soft- white noise. If the average value of the samples goes beyond
ware or as a real-time load flow that matches actual the stop criterion, say, the incremental voltage magnitude is
conditions accurately and overcomes the data error less than 0.001, as a result of a huge inconsistency caused by
problem. the bad data, the WLS state estimator will most likely diverge.
✔ The importance of the real-time load flow produced Furthermore, when the system is highly stressed due to
by state estimation lies in its use as a basis for security the heavy load, a least-squares state estimator has difficulty
analysis. reaching a solution. Indeed, test results published by IEEE
✔ An important feature of state estimation is the abil- have shown that the WLS has high probabilities of diver-
ity to detect the presence of bad data and to identify gence under highly stressed conditions. In 2011, it was also
which data are in error. reported in “The Future of the Electric Grid: An Interdisci-
Today, many power systems in the world have success- plinary MIT Study” that “the algorithm (of state estimator)
fully installed and run state estimators in control centers to is not perfect, and state estimators have trouble estimating

The mechanism of a WLS state estimator is not able to recover
the true values of bad measurements because its voltage solution
is biased by bad measurements.
a system state during unusual or emergency conditions— impedance values, or a huge weight for a measurement, etc.
unfortunately, when they are most needed.” In addition, Newton’s method is also sensitive to the initial
value selection requirement that the initial value must be
The WLS Static State Estimator Is Not Able to Precisely close to the true values of the state variables for convergence.
Recover the True Values of Bad Measurements
A WLS state estimator is essentially an average estimator Issue of Leverage Point
that tries to find a voltage solution that can “optimize” the The existence of leverage point measurements is one of the
residuals of all measurements, including the bad measure- main reasons that today’s state estimators are incapable of
ments. In other words, the voltage solution obtained by a rejecting, detecting, and identifying bad data. It has been
WLS state estimator is biased from the true voltage solution demonstrated in the literature that four situations can result
because the voltage solution has to minimize the residuals in leverage points:
of bad measurements. Therefore, the measurement residuals ✔ an injection measurement, indicating net power enter-
are all relatively “small,” but the voltage solution is biased ing at a bus, placed at a bus that is incident to a large
due to the existence of bad data. The root reason is that the number of branches
WLS state estimator does not reject the bad data from the ✔ an injection measurement placed at a bus that is inci-
set of given measurements. To precisely recover the true val- dent to branches of very different impedance values
ues of bad measurements, the voltage solution must be close ✔ flow measurements along branches whose impedanc-
to the true voltage values. The mechanism of a WLS state es are very different from those of the other branches
estimator is not able to recover the true values of bad mea- in the system
surements because its voltage solution is biased by bad measure- ✔ using a very large weight for a specific measurement.
ments. This behavior implies that, to precisely recover the Simply put, leverage point measurements can be under-
true values of bad measurements, we must first effectively stood as those measurements that have much higher weights
reject the bad data. than others in the WLS objective function. Because of their
huge weights, the optimization mechanism will keep their
Issue of Convergence residuals (defined as the difference between the measure-
Since the WLS state estimator includes all the measurements ment and its recovered value) as small as possible to obtain
together in the problem formulation, it is difficult to guaran- an optimal value for the objective function. In this article, we
tee a solution if a gross error or a group of gross errors are far discuss leverage points by using a small example.
from the bulk of the true state. Effectively rejecting bad data
that have large impacts on the solution is a sufficient con- Leverage Points Have a Bigger Impact on the
dition to guarantee a state estimation solution; however, the Solutions of Today’s State Estimators
WLS state estimator is not able to effectively reject bad data. Leverage points are common in power systems because a
The numerical solution algorithm chosen can sometimes high number of short branches exist in power systems and
cause the WLS state estimator to fail to reach a solution. A because of huge weights added on zero-bus injection mea-
common algorithm to solve the WLS state estimator is New- surements. For instance, recent studies have shown that in
ton’s method. Although there are many variations of the some IEEE test systems, leverage points may account for
algorithm, the solution procedure is essentially one where the more than one-third of the set of measurements. A com-
Jacobian matrix (a matrix of the first partial derivatives of mon practice in the power industry is that a zero-bus injec-
equations) is utilized in the iterations. If the Jacobian matrix tion is assigned a huge weight. Unlike the meters that have
is close to singular, the WLS will have difficulty reaching a measurement noise, a bus with no load or generator must
solution, even though the system is fully observable. A num- have strict zero injection according to the power balance
ber of factors could make the Jacobian matrix close to sin- principle. Therefore, it is common and natural to accept
gular. For example, the matrix could be singular due to the the claim that zero-bus injection is accurate. In other
presence of leverage point measurements, or a huge weight words, buses with power injections of zero have a very
assigned to measurements, etc. Leverage point measurements high confidence level. Thus, a huge weight is assigned to
are caused by an injection flow measurement associated with zero-injection measurements by power engineers to show
a large number of branches or branches with very different this high confidence.

This claim is quite true when there is no topology error vulnerable to bad data in the measurements, the topology,
and parameter error on the branches associated to the and the parameters in the system model:
bus. However, when a topology error or a parameter error ✔ Since a huge number of devices (each causing some
occurs on one of the branches connecting to the bus, the degree of measurement noise) are involved in the
zero-injection measurement is effectively no longer zero, process of transmitting samples from the field to the
and assigning a huge weight to it directly creates a leverage control center and because the parameters of transmis-
point and leads to a biased solution. Figure 3 shows the sion branches become inaccurate over time, bad data
negative impacts of the weight applied to a measure- are inevitable in static state estimation. The design of
ment with error. Figure 3(b) and (c) show that increasing the state estimator algorithms needs to consider how to
weight on a single bad measurement increases the bias on deal with bad data instead of obtaining the optimal
the WLS state estimator estimate. When the leverage point solution and then removing the bad data, which is im-
measurement bears bad data, it is impossible to detect using possible if bad data prevent a solution (as discussed
residual analysis. in the preceding). In other words, the main goal of
Inborn Flaws of WLS Static State Estimators

For today’s state estimators, especially WLS state estima- f(x)
tors, a large number of algorithms have been developed
to help detect the bad data in the measurements, the sys- Accurate
Estimate
tem topology, and system parameters. All the algorithms
were developed based on a solution from the state estima-
tor because it is believed that the solution is the key to bad
data detection.
However, if a solution cannot be reached when the bad
data are severe, the algorithms applied have no systematic
strategies to detect and identify and remove the bad data. x
For severe bad data, today’s WLS state estimators encoun- (a)
ter a dilemma: bad data detection needs a solution of the
f (x)
state estimator to detect the bad data, but the state estimator
cannot provide a solution due to the bad data. This situa-
Accurate
tion was exactly what happened in the 2003 blackout. Before Estimate
the occurrence of this blackout, the system operators had to Big Weight
manually fix the divergence problem after a topology error Biased
Estimate
caused the state estimator to crash.
In addition, all the bad data detection algorithms are
based on the residual analysis of an obtained solution. It is
also known that leverage points have smaller residuals, even
though the leverage points bear bad data, which means that x
(b)
bad data in leverage points are impossible to detect using
residual analysis. If the number of leverage points is rela- f (x)
tively small, their negative impact on the quality of a state
estimation solution may be limited. Unfortunately, the num- Accurate
ber of leverage points is quite large in real power systems. Bigger Weight Estimate
Biased
A summary of the inborn flaws of the WLS state estima- Estimate
tor is given as follows:
✔ Today’s state estimators cannot guarantee a solution
in all cases when the system is observable.
✔ A large percentage of leverage point measurements
disables bad data detection.
x
(c)
Static State Estimator Plus
Even though power system static state estimators (SSEs) figure 3. (a) The results from the WLS with an accurate
have been improved through the efforts of researchers and estimate. (b) The WLS static state estimator (SSE) with a
engineers over the past few decades, the robustness problem biased estimate due to a term with a big weight. (c) The
is not yet completely solved. The following considerations WLS SSE result with a bigger weight increasing the bias in
could be part of the reasons that today’s state estimators are the estimate.

the SSE should be to mitigate the impact of the bad Since the mechanism of this new state estimator is quite
data on the voltage solution and to guarantee a solu- different from the existing ones, we call it SSE+. Figure 4
tion with the existence of bad data. shows the difference in the design philosophy between the
✔ Considering that the percentage of bad data is relative- WLS SSE and SSE+. A group of measurements is called
ly high and the measurement redundancy is relatively consistent if the residuals of this group of measurements are
low, a base set formed by the minimum number of er- within the noise level for a given set of bus voltage values.
ror-free measurements to solve the entire system may For any given set of measurements, the minimum set of the
not exist in reality. In other words, if all bad data are most consistent measurement always exists, which implies
effectively removed from the measurement set, the en- that a solution is always available for SSE+. In addition,
tire system is no longer observable and solvable. This because SSE+ selects only the minimum set of the most
reality implies that removing all the bad data from consistent measurements, bad data having bigger impacts
the measurement set to obtain a solution is not a good will be effectively rejected, which guarantees the solution
strategy when designing a state estimation algorithm. with smallest biases if bad data are inevitable. The mini-
✔ We also know that some bad data have a severe nega- mum set of the most consistent measurements is obtained
tive impact on the state estimation solution, while using machine learning techniques. The details of the pro-
some of them have a smaller impact. If bad data can- posed SSE+ are out of the scope of this article, but readers
not be avoided when solving a state estimation prob- interested in the details could directly contact the authors for
lem, it is obvious that we should select those bad data more information.
with a smaller impact to guarantee a solution. In other We tested SSE+ on 25 real cases provided by power compa-
words, we should select those necessary contaminated nies in the United States and two large sample power systems
measurements that are closer to the true values. that are publicly available in open source power system analy-
With the preceding considerations, we developed a state sis software packages. Due to page limits, we present only the
estimator based on a new concept: the minimum set of the results from the test system with 13,659 buses and the results
most consistent measurements. Consistent measurements of a real case provided by a power company in North America.
are basically error-free measurements that stay close to one Our tests mainly focus on the following three features:
another in the residual plot. If we use the true system state 1) the rate of convergence
to recover those measurements, their residuals, defined as 2) the capability of bad data rejection
the difference between the measurements and their recov- 3) the accuracy of bad data recovery.
ered values, should be small, within the noise level. The A massive number of random cases were conducted on
set of most consistent measurements means that if bad data the 13,659-bus system, and 100 random tests were done for
are inevitable in solving the state estimation, only those each case. Three types of bad data are considered: topology
bad data closer to the true values will be included in the errors, parameter errors, and measurement value errors. For
base set. If we use the minimum set of the most consis- the topology error and parameter error testing, lines with
tent measurements to solve the state estimation, we want significant flows, which have larger impacts on the per-
to minimize the negative impact of bad data on the state formance of the state estimator, are selected to set up the
estimation solution. errors. Furthermore, to clearly show the performance of
SSE+ on convergence and bad data rejection, bad data are
added only to noncritical measurements. Please note that
SSE+ has been implemented using a parallel algorithm,
which achieves a calculation time short enough for applica-
tion in commercial EMSs.
Tests on 13,659-Bus System

The system data and measurement configuration for this
test are given in Table 1. Three types of bad data (topology
errors, parameter errors, and measurement value errors)
are tested on this system. To have realistic testing, we set
SSE+ up the measurement redundancy ratio (defined as the ratio
WLS SSE
between the total number of measurements and the total
number of variables) as 2.51, which is not very high. The
SSE Solution SSE Solution
number of bad data instances is from 10 to 500 for each
(a) (b) type of bad data, and 100 random tests were run for each
case. In addition, to test the impacts of large-error bad data
figure 4. A comparison of the WLS SSE and SSE+. (a) The on the performance of SSE+, we set up the topology errors
WLS state estimator. (b) The state estimator plus. and parameter errors on branches with at least 80 MW of

real power flow. The massive random testing shows that
table 1. The measurement configuration of the
SSE+ reached 100% convergence rate, in another word, 13,659-bus system.
all tested case converged to a feasible solution. Figure 5
details the testing results of all types of bad data, includ- Data Configuration
ing topology errors, parameter errors, and measurement Total buses 13,659
value errors. Total branches 20,467
Test on a Real Power System Case Total line flow measurements 35,510
We tested SSE+ on 25 real cases provided by four large com- Total bus injection measurements 27,318
panies in North America. The testing results of one of those
Total bus voltage magnitude measurements 5,772
cases are presented in this section. The system information
and the measurement configuration are listed in Table 2. The Measurement redundancy ratio 2.51
measurement redundancy ratio is relatively low. We tested Total short branches (reactance: <0.001 p.u.) 1,231
the solution quality and capability of bad data recovery for
p.u.: per unit.
SSE+ in the following three aspects.
100 12
Average Rate of Bad Data Rejection (%)
Average CPU Time Per Run (in Seconds)

90
80 10
70
8
60
50 6
40
4
30
20
2
10
0 0
10 100 200 300 400 500 10 100 200 300 400 500
Amount of Bad Data Amount of Bad Data
(a) (b)
1.02
1.019
1.018
Average |v | p.u.
1.017
1.016
1.015
1.014
1.013
10 100 200 300 400 500
Amount of Bad Data
(c)
Topology Errors Parameter Errors

Measurement Value Errors
figure 5. (a) The rejection testing on three typical types of bad data: the average rate of rejection versus the amount of
bad data. (b) The calculation time testing on three typical types of bad data: the average CPU time per run versus the
amount of bad data. The PC had four cores with 1.8 GHz. (c) The average voltage magnitudes for the testing of three
typical types of bad data: the average | v | per unit (p.u.) versus the amount of bad data.

Comparison of Voltage Solution Quality With voltage magnitudes. Smaller voltage measurement residuals
a Standard Commercial SSE indicate that the solved voltage magnitudes are close to the
For this real system case, the power company also provided measured ones. To have an overall comparison, we sorted
us the voltage solution obtained by their standard commercial the voltage residuals of both solutions and drew two curves
SSE. To compare the quality of voltage estimates between in the same plot. Compared to the solution of the standard
SSE+ and the standard commercial SSE, we calculated the commercial SSE, the voltage solution of SSE+ is closer to
residuals of voltage measurements for both solutions. The the voltage measurements.
results are reported in Figure 6 and Table 3. Voltage mea-
surement residuals are the difference between the measured Recovery of Accurate Values Lost Due to
bus voltage magnitudes and the corresponding solved bus Measurement Errors
We know that the WLS state estimator is not able to recover
the true values lost due to bad measurements because it
table 2. The system data and measurement cannot reject the bad measurements. However, the SSE+
configuration of the real utility case.
approach is able to effectively reject bad measurements by
Data Configuration selecting a base set that is formed by the minimum set of
the most consistent measurements, naturally rejecting the
Buses 3,409
bad measurements from the base set. The voltage solution
Transmission branches 4,337 of SSE+ is not influenced by the rejected bad measurements,
and therefore, SSE+ should have a more accurate voltage
Line flow measurements 3,571
solution. To test the capability of bad measurement recovery
Injection measurements 3,022 of SSE+, we performed the following two tests.
Voltage measurements 1,457
Recovery of Bad Data on Injection Measurements
Measurement redundancy ratio 2.15 In this test, we introduced big errors of 1,000 MW to 10
real power injection measurements. Due to the low mea-
surement redundancy ratio, SSE+ selected five real power
0.12 injection measurements among the 10 injection measure-
ments to enter the base set, while the other five entered
SSE+ the redundant set. Four bad measurements were precisely
0.1
Commercial SSE
Sorted Voltage Residuals
recovered, which is shown in Figure 7. One of the bad injec-

0.08 tion measurements could not be precisely recovered, where
the original value was 0 MW and the recovered value was
0.06 3,424 MW. The reason that this measurement could not be
accurately recovered is because its bus voltage was biased
0.04
due to a bad injection measurement at one of its neighbor
buses entering the base set. However, since the other four
0.02
injection measurements were precisely recovered, it is
0 implied that the negative impacts of bad measurements in
0 500 1,000 1,500 the base set are limited and localized.
Bus ID of Voltage Magnitude Measurements
figure 6. The comparison of voltage residuals between

Comparison Between the Original Values
commercial SSE and SSE+ solutions. and Recovered Values by SSE+ (Absolute Values)
60
50
Values (MW)
table 3. A comparison of the voltage solution quality. 40

30
Standard 20
Commercial SSE SSE+ 10
0
Sum of voltage 15.45 p.u. 12.97 p.u. 1 2 3 4
measurement residuals Measurement ID of Four Bad Measurements
Average of voltage 0.0108 p.u. 0.0091 p.u. Original Values
measurement residuals Recovered Values By SSE+
Standard deviation 0.0163 0.0158
figure 7. The bad data recovery test results for bus injec-
Mean voltage solution value 1.0259 p.u. 1.0257 p.u.
tion measurements.

The SSE+ approach is able to effectively reject bad measurements
by selecting a base set that is formed by the minimum set of the
most consistent measurements.
leverage points present in the given set of measurements dis-

Comparison Between the Original Values and ables the capability of bad data detection and identification
Recovered Values From SSE+ (Absolute Values)
450
for the WLS state estimator. An alternative SSE+ was intro-
400 duced in this article, based on the concept of using the mini-
350 mum set of the most consistent measurements. Test results for
Values (MW)
300
250
a case provided by a power company were presented along
200 with results obtained using a large artificial test system. The
150 convergence rate, the capability of rejecting bad data, and the
100
50
capability of recovering bad data were tested, and the results
0 were presented. In summary, the numerous testing results have
1 2 3 4 5 6 7 8 9
shown that the SSE+ approach is a promising offline backup
Measurement ID of Nine Bad Measurements
for today’s vulnerable SSEs and an improved solution for bad
Original Values data management compared to conventional SSE approaches.
Recovered Values by SSE+ The next step could be to further enhance the SSE+ accuracy
and performance by using PMU data more efficiently.
figure 8. The bad data recovery testing for the line flow
measurements. For Further Reading
T. E. Dy Liacco, “The role of state estimation in power
Recovery of Bad Data in Power Line system operation,” IFAC Proc. Volumes, vol. 15, no. 4, pp.
Flow Measurements 1531–1533, Jun. 1982, doi: 10.1016/S1474-6670(17)63216-6.
To test the recovery capability of contaminated line flow A. Abur and A. Gómez-Expósito, Power System State
measurements, we introduced big errors (500 MW) to 10 Estimation: Theory and Implementation. New York, NY,
real power line flow measurements. SSE+ selected one bad USA: Marcel Dekker, 2004.
measurement to enter the base set, and the other nine were A. Monticelli, State Estimation in Electric Power Sys-
rejected and entered the redundant set. To show the recovery tems: A Generalized Approach. Norwell, MA, USA: Klu-
accuracy, we drew Figure 8 to demonstrate the differences wer, 1999.
between the original values and their recovered values. The A. J. Wood, B. F. Wollenberg, and G. B. Sheblé, Power
small differences in Figure 8 imply that the recovery by Generation, Operation, and Control, 3rd ed. Hoboken, NJ,
SSE+ is accurate. USA: Wiley, 2013.
“State estimation algorithms working group, task force
Conclusions on state estimation concepts and terminology,” IEEE Power
It has been more than 50 years since the WLS SSE was Energy Soc., Piscataway, NJ, USA, Tech. Rep. PES-TR20,
first applied in power system operations. Over that time, big May 2016.
accomplishments have been made to improve the robustness J. Zhao, L. Mili, and R. Pires, “Statistical and numeri-
of the state estimator, due to efforts from the power engi- cal robust state estimator for heavily loaded power systems,”
neers and researchers. As a result, SSEs have been success- IEEE Trans. Power Syst., vol. 33, no. 6, pp. 6904–6914, Nov.
fully applied in almost all transmission control centers. The 2018, doi: 10.1109/TPWRS.2018.2849325.
occurrence of the August blackout in 2003 reminded us that “SBIR phase II: A robust state estimator for power grids,”
state estimators still have much room for improvement, espe- Nat. Sci. Found., Alexandria, VA, USA, 2017. [Online].
cially in reducing the vulnerability to bad data and in provid- Available: https://nsf.gov/awardsearch/showAward?AWD_
ing a guarantee of a solution. A survey report published by ID=1660022&HistoricalAwards=false
NERC, in 2020, reminded us again that the vulnerability of
the power system state estimator must be solved for better Biographies
reliability and the security of power system operations. Bei Gou is with Smart Electric Grid, Mansfield, TX 76063
This article gave further understanding of the WLS state USA.
estimator from the point of view of bad data detection and David Shue is with Smart Electric Grid, Mansfield, TX
solution availability. It was shown that a large number of 76063 USA. p&e

A Powerful
Tool for
Power System
Monitoring
By Tianshu Bi ,
Maolin Zhu, and Hao Liu
M
MODERN POWER SYSTEMS ARE WITNESSING
A rapid transition with the increasing penetration
of intermittent nonsynchronous renewable generation
and distributed energy resources. According to the
National Energy Administration, China’s installed
wind and solar energy capacity accounted for 26.7%
of the country’s total power generation capacity at
the end of 2021. In terms of electricity generation,
wind and solar energy accounted for 11.8% percent
of the total electricity use that year. In the renew-
able energy development plan in the 14th Five-Year
Plan (2021–2025), China aims for renewables to meet
33% of national power consumption and for nonhydro
renewables to satisfy 18%. This transition poses chal-
lenges for system operators to grasp the actual status
of power systems accurately and comprehensively.
An energy management system (EMS) is a system of
computer-aided tools used by operators of electric utility
grids to monitor, control, and optimize the performance
of the generation and transmission systems. Hence, an


EMS is commonly applied to pro-
vide critical decision-making sup- Distributed EMS systems. The most commonly
applied SE method in EMSs is
port for system operators. China’s
grid operators, such as State Grid, Dynamic State static SE (SSE). The spatial redun-
dancy of measurements is lever-
have been promoting the use of
EMSs in the dispatch organiza- Estimation Based aged in SSE to estimate a single
snapshot of current system opera-
tions at provincial and higher levels
since 1986, and an EMS platform on a Full-View tion conditions, such as bus voltage
phasors. SSE can provide accept-
named CC-2000 has been devel-
oped, with a set of advanced appli- Synchronized able performance for conventional
power systems dominated by large
cations. The computer-based EMS
represents a great shift from the Measurement synchronous generators (SGs).
However, the measurement data
traditional empirical dispatch to
the modern analytical dispatch of System. used in SSE come from the super-
visory control and data acquisition
China’s power systems. (SCADA) system and are normally
However, measurement data sampled every 2 to 4 s and are not
inevitably contain transducer noise well synchronized. These weak-
and even bad data due to problems, such as communication nesses result in SSE tracking only the quasi-steady-state process
channel noise, electromagnetic interference, and instrument of power systems with electromechanical dynamics.
transformer saturation. State estimation (SE) is used to reduce With the widespread deployment of renewable genera-
the impact of measurement error and provide reliable data suf- tion, smart load controls, energy storage, and plug-in hybrid
ficiently close to real time for EMS applications. Advanced SE vehicles, the behavior of modern power systems is more ran-
techniques become one of the most important parts of advanced dom and dynamic. Thus, the quasi-steady-state assumption
may be invalid. The widespread use
of power electronic equipment has
had a great impact on China’s exist-
ing power grid, leading to a series of
problems, such as subsynchronous
oscillations (SSOs) and jeopardized
system stability. For example, in July
2015, an SSO event occurred in the
wind farms in the Hami area of Xin-
jiang, China, resulting in the trip of
the shaft torsional vibration protec-
tion of three thermal 660-MW power
units 300 km away. In September
2015, a bipolar block fault occurred
on the Jinping–Southern Jiangsu
±800 kV ultrahigh-voltage dc line,
resulting in an instantaneous power
loss of 5,400 MW. Twelve seconds
after the fault occurred, the frequency
of the East China Power Grid dropped
to 49.56 Hz. During events similar to
these emergency situations, the power
system response is very dynamic. SSE
cannot provide timely and accurate
information for system operators,
©SHUTTERSTOCK.COM/THINKHUBSTUDIO
due to its low refresh rate and non-

synchronized measurements. As a
result, the dispatching system cannot
respond to events in time. Therefore,
for future power systems, it is essential
to estimate dynamic states in near real
time in all operating situations.

Phasor measurement units (PMUs) can synchronize the the measurement of a SYMS is developed as an innovative
fast measurements from widely dispersed locations to a GPS dynamic monitoring tool. Finally, the challenges faced by
clock. The wider application of a wide area measurement distributed DSE are discussed.
system (WAMS) based on PMUs allows monitoring electro-
mechanical dynamics, which brings an opportunity for the Full-View SYMS
development of dynamic SE (DSE). As an efficient system SCADA systems with a slow update rate and WAMS based
monitoring tool, DSE can track the evolution of the system on PMUs are not fit for monitoring the complex system
dynamic states accurately. At present, in China’s power grid, dynamic response due to the ever-increasing penetration of
PMUs are installed for transmission lines with a voltage power electronics. Therefore, to bridge the existing research
greater than 500 kV, generator units with a capacity greater gaps, a full-view SYMS is developed. The SYMS contains
than 300 MW, and important 220-kV substations. However, the following four types of synchronized measurement
the frequency components of power signals in areas of dis- devices (SMDs) as well as a data center (see Figure 1):
tributed renewables are becoming increasingly complex. ✔ SMD for renewables (SMD-R)
Existing PMUs designed for the high-voltage transmission ✔ SMD for control applications (SMD-C)
grid have difficulty coping with the fast dynamics of power ✔ SMD for loads (SMD-L)
electronics. In addition, most of the existing DSE stud- ✔ SMD for waveforms (SMD-W).
ies focus on traditional systems dominated by SGs, while
the changed electromechanical characteristics due to the SMDs
high penetration of renewable energy sources (RESs) are The four types of SMDs shown in Figure 1 can estimate the
ignored. The roles and implementation requirements of synchrophasor, frequency, and rate of change of frequency
DSE in EMSs are not well established, and there are still (ROCOF) of power signals. The different types of SMDs can
significant gaps to address in terms of the practicality of also exhibit some differences according to their potential
DSE. Consequently, a full-view synchronized measurement applications. The characteristics of the four types of SMDs
system (SYMS) for power electronics-enabled power sys- are described as follows:
tems is introduced in this article. Then, a distributed DSE ✔ SMD-R: Due to the high penetration of power elec-
scheme considering power systems with RESs and matching tronics and the low inertia of RESs, the frequency
Data Center
of SYMS
Local Measurement Data Local Measurement Data

Concentrator Concentrator
Ethernet
U, I, f, ROCOF U, I, f, ROCOF U, I, f, ROCOF

U, I, f, ROCOF
Wide Frequency Power Quality, dc Power Quality, dc
U, I of ac U, I of ac U, I of ac U, I of ac
U, I of dc U, I of dc U, I of dc
SMD for Renewables SMD for Control SMD for Loads SMD for Waveform
figure 1. The framework and SMDs of the SYMS.

components of power signals in areas of distributed distribution networks with complex loads is still at the
renewables are becoming increasingly complex. Fur- primary stage because of the difficulty of modeling
thermore, the power converters and the control loops the diverse dynamic loads.
have much smaller time constants than those of the ✔ SMD-W: The SMD-W features synchronously mea-
SGs, yielding much faster dynamics. As a result, the suring the waveform with an adaptive sampling rate
measurement system used for the monitoring of RESs from 0.8 to 12.8 kHz. The waveform is then reported
should have a faster sampling speed and broader band- in real time through high-speed internet (such as 5G).
widths for data transmission. The SMD-R is designed It should be noted that the development of synchro-
to have the capability of extracting the accurate syn- nous waveform measurement devices is still in its in-
chrophasor of high-frequency-oscillation signals with fancy, and these settings are the characteristics of the
a higher reporting rate and wider measurement band- SMD-W tested in China. The device can monitor the
width. These enhanced capabilities allow the possi- transient process of disturbances in power electron-
bility of SSO monitoring and control. The SMD-R is ics-enabled power grids, and therefore, it can be used
deployed in the low-voltage level of wind and photo- for security analysis and control. Moreover, it can
voltaic power stations, such as the low-voltage side of be deployed at the key buses that are significant for
the box-type substations of wind turbines. Then, the stability and control applications. The measurements
SMD-R can also facilitate the distributed DSE at each of the SMD-W directly utilize measured values from
wind turbine and photovoltaic array within a wind the waveforms at each sample, with a full description
farm or solar park, which can be used for optimizing of measurement transients. Thus, the SMD-W-driven
the performance of the whole farm or park. DSE that is under development can be used for the
✔ SMD-C: The SMD-C is designed to provide faster mea- applications such as fast data-driven fault cause iden-
surements to the closed-loop control of power electron- tification, harmonics filtering, and so on.
ics-enabled power grids, such as voltage control and
virtual inertial control. The application of the SMD-C Data Center
may not include DSE, but it is an important part of the The data center can be divided into a forward server, a histo-
SYMS. The SMD-C has shorter reporting latency (RL) rian server, and an application server, as shown in Figure 2.
within a narrower measurement bandwidth (48–52 Hz). The forward server is responsible for receiving and parsing a
The measurement requirements and working scenarios tremendous amount of data with different reporting rates. All
of the SMD-C are similar to those of a P-class PMU, historical measurement data are stored in the historian server,
but the former can obtain synchrophasor, frequency, and and various offline and online applications are hosted in the
ROCOF measurements with an RL shorter than 20 ms. application server. Before the measurement data are trans-
✔ SMD-L: The SMD-L is installed in distribution systems mitted to different applications, they must go through SE to
and inside load facilities, which have more harmonics improve data consistency. Hence, SE can be considered one
and a higher noise level compared with transmission of the online applications. Based on the rapid synchronous
grids. The measurement bandwidth of the SMD-L is measurement of the SYMS, DSE can filter out noise, process
set as 45–55 Hz to suppress most of the noise, and bad data, and provide dynamic states, along with the estima-
the SMD-L has a wideband measurement capabil- tion of unknown inputs and other calculations.
ity to estimate the parameters of harmonics and in-
terharmonics. The measurement data can be applied Field Installation and Application Cases
for load modeling, the location of harmonic problems, The SMD-L is easy to install because it is designed to be
distribution disturbance analysis, and so on. DSE for directly connected to the power outlets in residential buildings.
Forward Server Historian Server Application Server

Function:
Data Parser MySQL State Estimation Filter Out Noise
Historian
Database Database Bad Data Process
Measurements Online Application
Data Collection Dynamic States
Offline Application
Unknown Inputs
Data Forwarding Redis
Real-Time ...
Database Database
figure 2. The data center framework. SQL: Structured Query Language.

Therefore, dozens of SMD-Ls have been installed in cities areas. The 20-Hz frequency component of an SSO event
in different regions of China, and various applications are is extracted based on the measurements from the SMD-R
being developed based on the huge amounts of measure- and conventional PMU, respectively, as described in Fig-
ment data. One of the applications is the analysis of the ure 3 (b). The results are then compared with the recorded
frequency spatial–temporal features of large-scale power waveform, which is considered the reference. It is evident
grids. Figure 3(a) gives the measured frequency signals that the oscillation component measured by the SMD-R was
over a period of time from four SMD-Ls installed in differ- close to the actual value, while that measured by the PMU
ent regions of China. It can be seen that the frequency sig- was much less accurate. Therefore, the SMD-R can be used
nals within the same region show strong correlation, while to effectively monitor, locate, and control SSOs. In addition,
the dynamic characteristics of the frequency signals in dif- field tests of the SMD-C and SMD-W have also been carried
ferent regions are different, although the regional power out, although the results are not shown here.
grids are interconnected. In addition, data-driven artificial
intelligence methods can also be incorporated in the grow- Distributed DSE
ing measurement data to extract frequency characteristics Due to sensor error and various sources of interference, the
of power systems. measurements of the SYMS inevitably have random noise
The SMD-R has been installed in some wind farms in and even bad data, which may lead to wrong control actions
China’s Xinjiang and Inner Mongolia, and its measurement and the further expansion of disturbances. Thus, power
data can be used to monitor SSOs in distributed renewable system SE needs to be performed. The purpose of SE is
to estimate system states and maintain the consistency of
the measurement data. The system dynamics are captured
based on the measurement data and the power grid topol-
50.04
ogy coupled with models and parameters of various elec-
50.03 tric equipment. In the electromechanical transient response
Frequency (Hz)
50.02 after a fault, the network topology changes because of

the action of protection devices and automatic reclosing
50.01
devices. Hence, it is difficult for dispatching centers to
50 track topology changes in real time. Thus, from this per-
49.99 spective, centralized SSE based on network topology con-
straints is no longer feasible. In recent years, considering
49.98
0 30 60 the dynamic process of the system and the rapid develop-
Time (s) ment of synchronous measurement technology, distributed
Beijing, North China DSE has attracted much attention. The system dynamic
Zhoukou, North China states provided by DSE constitute the critically important
Nanjing, Eastern China data basis for wide area power system protection and con-
Taining, Eastern China trol applications.
The overall architecture of the distributed DSE is pro-
(a)
vided in Figure 4. Using the widely distributed synchronized
phasor measurement devices, the power system dynamics,
3 including power generation, transmission, and distribu-
tion, can be monitored rapidly and synchronously. Reliable
Amplitude (A)
2 operation data provided by DSE can be applied in various

advanced applications at the dispatch centers. For example,
with the wide area fast estimation of DSE, global optimal
1
control can be carried out considering the compatibility
among controllers. SSE of the transmission network esti-
0 mates only the bus voltage phasors. Linear SE operating at
5 10 15
Time (s) the phasor measurement refresh rate has been developed for
some PMU fully observable networks. However, it does not
SMD-R PMU Recorded Data
track the actual system dynamics and thus is not the focus
(b) of this article. Modeling the diverse dynamic loads in the
distribution network is difficult, and the existing synchro-
figure 3. The application cases. (a) The measured frequen- nous measurement coverage is insufficient, so DSE for dis-
cy signals from SMD-Ls in different regions of China. (b) A tribution networks with complex loads is still in its infancy.
comparison of the 20-Hz frequency component of an SSO Therefore, the discussion is centered on DSE for SGs and
event measured by an SMD-R and a conventional PMU. RESs in the following parts of the article.

Other Areas
Advanced
Applications
Dispatch
DSE Global Center
Control
SMD DSE SMD

Database
Generation SSE
SMD
Distribution
and Load
Transmission
figure 4. The framework of distributed DSE.
Nonlinear Kalman Filter The EKF may work well if implemented correctly, but its
The least-squares method is commonly used in SSE success depends on the local properties of the function to
because only the measurements of a single moment are linearize. Furthermore, the derivatives of nonlinear equa-
used to provide the snapshot-based static view of the sys- tions can be challenging for complicated models and discrete
tem algebraic variables. By contrast, DSE requires inte- switching events.
grating the measurements and the state prediction to estimate To overcome this defect, filtering methods based on non-
the dynamic states, as shown in Figure 5. Thus, the Kalman deterministic sampling filtering and deterministic sampling
filter (KF), which is formulated as a two-step process of pre- are used to approximate the nonlinear probability distribu-
diction and correction, becomes an ideal choice. The KF first tion. The particle filter (PF) is a typical nondeterministic
predicts the system states and also provides the uncertainty sampling filter. The idea of the PF is based on Monte Carlo
of the prediction. Once the measurement is received, the KF methods, which use particle sets to represent probabilities
updates (or corrects) the current states’ prediction and uncer- and can be used in any form of state-space model. In simple
tainty. The uncertainties of the system model and noisy mea- terms, the PF method refers to the process of obtaining the
surements are respectively modeled by process and measure- state minimum variance distribution by finding a set of ran-
ment noise covariance matrixes. Different color lumps are dom samples. The random samples propagate in the state
also used in Figure 5 to represent the propagation of model space to approximate the probability density function, and
and measurement uncertainties, such as the gray lump visu- the solution approach replaces the integral operation with
ally representing the probability distribution of the estimated the sample mean. Moreover, typical deterministic sampling
state vector at time step k – 1.
Power systems are typical
nonlinear systems, of which the dy-
namic variables cannot be directly ent
estimated by a linear KF. Many q u a ti o ns s u rem
ment E Me a
suboptimal approximation methods Measure
Estimated
based on the framework of the KF Estimated System Dynamic Equations State Vector
State Vector
have been proposed. The extended State P
redictio
KF (EKF) is one of the most widely n
used methods for nonlinear sys-
tems. The formulation of the EKF
is based on the linearization of
the system dynamic equations and
measurement equations. Lineariza-
tion enables the state vector and Time Step k – 1 Time Step k
state covariance to be propagated
in an approximately linear format. figure 5. The framework of the Kalman filter updating the state of the system.

filtering methods include the unscented KF and cubature The second-order SG model considers only the mechani-
KF, which propagate the mean and covariance of states cal characteristics of the rotor, and no electromagnetic
using a deterministic sampling approach to pass the sigma characteristics can be reflected. The fourth-order SG model
points through the nonlinear system. They are widely used considers the transient process of the rotor field winding
in the DSE literature because of accurate estimation results and rotor dynamics. The sixth-order model, expanding on
and similar computational efficiency with the EKF. the fourth-order model, takes the dynamics of the damp-
ing windings into account. The subtransient process rapidly
DSE of SGs decays and ends within one to two cycles after a disturbance,
Although RESs have been developing rapidly in recent years, which is difficult for the SMD to track. Thus, the fourth-
SGs are still the core components of the power systems in order model of the SG is commonly used to develop the sys-
China. Their dynamic performance has a great impact on the tem dynamic model used in DSE. In addition, the exciter and
security and stability of the power grid. Due to the large inertia rotor speed governor should also be considered and modeled
of the SG rotor and the constraint of Lenz’s law, the dynamic according to the relevant standards.
states of SGs cannot change abruptly in the electromechanical Using the nonlinear KF methods, the estimated SG
transient process, and the constraints of SG dynamic equa- dynamic variables of the previous time step and the phasor
tions are satisfied. Therefore, the dynamic variables of SGs measurement information are fused to estimate the states
can be estimated by DSE. In addition, the fast phasor mea- of the current time step. The dynamic variables of the SG
surements at the generator outlet are required to decouple SGs fourth-order model include the power angle, rotor speed,
from the external network to facilitate the distributed DSE of and transient EMF vector. The terminal voltage phasor of an
SGs, without considering the network topology. SG is considered the input of DSE to decouple the generator
In China, the installation of a PMU is required at the ter- from the power grid. The measurement vector consists of the
minal of an SG with a large capacity, and the power angle of terminal current phasor and active and reactive power. Fig-
the SG is also required to be measured synchronously. The ure 6(b) provides a comparison between the measured and
power angle for an SG is the angle between the internal elec- estimated values of the power angle of a generator. It can be
tromotive force (EMF) and terminal voltage, and the real seen that DSE can efficiently reduce the measurement noise
power generated by the SG also depends upon this angle. and improve the data accuracy. Besides, it is thought that
Figure 6(a) demonstrates the principle of measuring the the rotor speed measurement from a meter installed on the
power angle based on the key phase pulse signal. First, the shaft of the machine is usually not reliable. The estimated
key phase pulse signal and terminal voltage are input into rotor speed can be used as a reliable input of power system
the SMD. Then, they are compared with the standard time stabilizers (PSSs) to damp out oscillations. Furthermore, the
signal to obtain the rotor position angle and terminal voltage compatibility among PSSs of different SGs can be studied
phase angle. When an SG is in operation with no load, the based on the DSE results to consider global objectives and
power angle is equal to zero, and the constant angle between provide systematic coordination over wide areas.
the EMF and rotor position can be obtained. When the gen-
erator is connected to the grid, the power angle can be calcu- DSE of RESs
lated by subtracting the terminal voltage phase angle and the The scale of the installation of renewable energy power gen-
constant angle from the rotor position angle. eration in China has been rapidly growing in recent years.
Rotor Position
0.72
Internal EMF
Power Angle (rad)
0.7
Constant
Angle Terminal
Voltage
0.68
Rotor
Absolute Time 0.66 Measurement
0 Reference Estimation
Power Angle True Value
0.64
0 2 4 6 8 10
Key Phasor
Time (s)
Pulse
(a) (b)
figure 6. The DSE results for an SG. (a) The measurement of the power angle. (b) A comparison of the measured and
estimated power angles.

China is committed to achieving peak carbon dioxide emis- strategies based on different principles may be used for nor-
sions before 2030 and carbon neutrality before 2060. In this mal conditions, such as vector control and direct power con-
case, RESs will gradually replace the traditional SGs as the trol. Moreover, when a fault occurs, a series of low-voltage
main power source of the power systems. Thus, the DSE of ride-through control strategies with different objectives are
RESs should also be of concern for the critical monitoring, used to keep RESs connected to the grid. Thus, complex con-
control, and protection of future electric power grids. trol strategies pose a challenge for the DSE of RESs. The DSE
Appropriate RES models are the basis of DSE. Wind model for the DFIG, whose rotor field current is controlled by
energy is one of the main types of RESs, where two types power converter, can be reformulated to take the converter
of wind turbines are included: doubly fed induction genera- outputs as the unknown inputs to cope with the diverse con-
tors (DFIGs) and full converter interfaced permanent mag- trol strategies. This approach significantly increases the gen-
net SGs (PMSG). In a DFIG, the stator of the generator is eralizability of DSE methods and reduces the complexity of
directly connected to the grid, while the rotor is connected the DSE model. Furthermore, the unknown inputs can be
to the grid through a back-to-back power converter. The accurately estimated together with the DFIG dynamic states.
DFIG can keep the speed of the rotor rotating the magnetic For RESs that are integrated to the power grid through power
field at the synchronous speed by means of controlling the electronic devices, such as PMSGs and photovoltaic panels,
frequency of the rotor field current through the converter grid-connected converter models should be focused to meet
controller. In a PMSG, the stator is connected to the grid the needs of DSE applications. In particular, the converters
through a power electronic interface, and the rotor excita- should be modeled on a timescale that can be captured by
tion is provided with the permanent magnet instead of the SMDs with limited reporting rates.
external excitation source. A PMSG interfaced through the Similar to SGs, the DSE of RESs also needs the ter-
power electronic converter can operate at low rotor speeds, minal voltage phasors as inputs of distributed DSE. Thus,
and so there is no need of a gearbox. This kind of struc- a measurement device at each generator unit is needed,
ture increases the flexibility in the rotor speed control and as illustrated in Figure 7(a), where an SMD-R installed in
significantly increases the turbine efficiency. Solar energy the tower of a DFIG is shown. However, renewable energy
is another common RES, and photovoltaic panels are the power plants are usually composed of many power genera-
primary form of converting solar energy into electricity. tion units, such as wind turbines and photovoltaic systems.
Like the PMSG, they are also connected to the grid through These RES units form a network with electrical connections,
power electronic converters. as in Figure 7(b). Therefore, it is not necessary to install an
Compared with SGs, the most striking characteristic SMD-R at each generation unit. The voltage phasors of all
of RESs relates to controlling the flow of energy through buses inside renewable energy power plants can be estimated
power electronic equipment. Thus, the converter controller considering the network topology information and reason-
determines the dynamics of RESs. However, their control able measurement configuration. Then a single generation
strategies are usually complex for meeting the requirements unit can be decoupled from the network, and distributed
of modern power system operation specifications. Control DSE can be performed.
Grid
(a) (b)
figure 7. (a) An SMD-R installed on a wind turbine tower. (b) An example wind farm topology.

At present, the configuration of phasor measurement
devices on the ac transmission grid is relatively common,
promoting the development of the DSE of SGs.
The DSE execution rate is usually determined by the interpolation is the best. In addition, a waveform measure-
measurement reporting rate, which is usually set as 50 or ment device, such as an SMD-W, with a higher sampling
100 samples/s. However, the dynamic states of RESs gen- rate can be used for the DSE of RESs. However, it is usually
erally have relatively low time constants due to the small restricted by the higher latency because uploading a large
capacity of RESs and fast control of power electronic equip- amount of recorded data in real time has high requirements
ment. Figure 8(a) presents the variations of the rotor speed for computing power and communication means.
of an SG and a DFIG in the same system in response to a
conductor disconnection fault. It can be found that the rotor Challenges
speed of the DFIG fluctuates more rapidly and stabilizes in a DSE accurately tracks the dynamics of a power system and pro-
shorter time after the disturbance than that of the SG. Thus, vides the evolution of the system state. Although much research
the existing measurement reporting rate may not meet the work has been done for DSE, there is still room for improve-
requirement of the DSE for RESs. ment in terms of theories and practical applications. Here, sev-
A pseudomeasurement obtained by performing a linear eral important aspects are recommended for future work:
interpolation between two consecutive measurement sam- ✔ Modeling of RESs: With more power electronic-based
ples can be used to increase the sampling rate. The interpo- RESs integrated into power systems, DSE for RESs be-
lation parameter should be appropriately chosen to achieve comes an essential tool for monitoring power systems
a good tradeoff between the computing time and estimation that exhibit emerging dynamic characteristics. RESs are
accuracy. The local truncation error can reflect the discreti- often asynchronous, intermittent, and controlled to have
zation error of the predicted states of DSE, so it can be used no inertial behavior. Therefore, it is important to moni-
as an index to adaptively adjust the interpolation parameter. tor their dynamics with DSE to reduce their impact on
Figure 8(b) has the DSE results of an internal model variable system stability. The power plants of RESs are usually
(q-axis rotor flux) of the DFIG rotor winding with no inter- composed of many power generation units. Whether
polation, random interpolation, and adaptive interpolation. to conduct DSE for a power plant with its equivalent
It can be seen that the interpolation can improve the accu- model or each generation unit should be investigated
racy of DSE results, and the performance of the adaptive considering their corresponding application scenarios.
1.004 1.0025
4
1.15 0.39
SG
Rotor Speed
1.001
1.002 0.37
0.9995 3
4.98 5.07 5.16
1 0.85 0.35
q-Axis Rotor Flux
0.33
0.998 2
0 5 10 15 20
0.55
Time (s) 2.6 3.6 4.6 5.6 6.6
0.31
13 13.5 14
(a)
0.8002 1
0.8001
DFIG
Rotor Speed
0.80005
0.8001 0.8
0.79995 0
4.98 5.07 5.16
0.8 No Interpolation Adaptive Interpolation
Random Interpolation True Value
0.7999 –1
0 5 10 15 20 0 5 10 15 20
Time (s) Time (s)
(b) (c)
figure 8. The DSE of a DFIG. (a) The variation of the rotor speed of an SG and a DFIG. (b) The estimation results of the
rotor flux, with different interpolation choices.

The fast dynamics of power electronics in RESs require
a higher measurement reporting rate, where the waveform
measurement device may be more suitable for DSE.
Besides, data-based modeling methods can be considered. Conclusion

By mining the operational characteristics of historical DSE is critical for ensuring the reliable and safe operation
data, artificial intelligence-based tools can refine simple of power systems. With the continuing development of mea-
and accurate models of RESs, which is critical for DSE. surement technology, distributed DSE for dynamic compo-
✔ Measurement for the distributed DSE of RESs: Suf- nents, such as SGs and RESs, becomes possible. The SYMS
ficient measurements are the basis of accurate SE. was introduced, including four types of SMDs discussed for
At present, the configuration of phasor measurement power electronics-dominated systems, which form the basis
devices on the ac transmission grid is relatively com- of distributed DSE. Based on the distributed DSE frame-
mon, promoting the development of the DSE of SGs, work, the DSE of conventional SGs and RESs with nonlinear
but the devices’ application in distribution grids and KFs were then highlighted. Compared with traditional SSE,
RES power plants needs to be improved. The devel- DSE can provide real-time operating states of the system at
oped SYMS will facilitate this expansion. In addi- fast rates. This dynamic data from DSE, in turn, can be uti-
tion, the fast dynamics of power electronics in RESs lized to fulfill the requirements for modern power systems
require a higher measurement reporting rate, where protection and control. However, practical applications and
the waveform measurement device may be more suit- the implementation of DSE for power system with RESs are
able for DSE considering electromagnetic transient still in the development stage and need further investigation.
responses. However, it is difficult to upload the large
amount of recording data to the master station in real For Further Reading
time. Techniques to break through the communication J. B. Zhao et al., “Power system dynamic state and parameter
constraints and solutions performing local distributed estimation-transition to power electronics-dominated clean
DSE need further consideration. Moreover, with the energy systems,” IEEE Power Energy Soc., Piscataway, NJ,
development of prediction technology, short-term USA, Tech. Rep. PES-TR88, Jul. 2021.
power forecasts are becoming more accurate, and the J. B. Zhao and L. Mili, “A theoretical framework of robust
power prediction information of RESs can also be lev- h-infinity unscented Kalman filter and its application to
eraged to improve the performance of DSE. power system dynamic state estimation,” IEEE Trans. Sig-
✔ Filtering algorithms improvement: Power systems nal Process., vol. 67, no. 10, pp. 2734–2746, May 2019, doi:
with RESs are more random and dynamic than tradi- 10.1109/TSP.2019.2908910.
tional power systems, leading to increasing system un- G. Anagnostou, L. P. Kunjumuhammed, and B. C. Pal,
certainties and deteriorated measurement data quality. “Dynamic state estimation for wind turbine models with un-
Thus, the robustness and numerical stability of DSE known wind velocity,” IEEE Trans. Power Syst., vol. 34, no. 5,
algorithms should be further improved. Moreover, be- pp. 3879–3890, Sep. 2019, doi: 10.1109/TPWRS.2019.2909160.
sides the KF, the application of other methods, such as S. S. Yu et al., “An unscented particle filtering approach to
machine learning and artificial intelligence, in DSE decentralized dynamic state estimation for DFIG wind turbines
should also be considered. in multi-area power systems,” IEEE Trans. Power Syst., vol. 35,
✔ Implementation and application of DSE: The level of no. 4, pp. 2670–2682, Jul. 2020, doi: 10.1109/TPWRS.2020.2966443.
the dispatching center where DSE is deployed should be S. Xu, H. Liu, and T. Bi, “A general design method for phasor
considered first to adapt to the hierarchical generation estimation in different applications,” IEEE Trans. Smart Grid, vol.
dispatch organization. The real-time information inter- 12, no. 3, pp. 2307–2319, May 2021, doi: 10.1109/TSG.2020.3040163.
action between DSE and remote measurement devices
should be investigated. Moreover, it is important to devel- Biographies
op the extraction method of the massive information of Tianshu Bi is with North China Electric Power University,
the DSE as the dispatcher’s decision-making basis. Prac- Beijing 102206, China.
tical application demand is an important driving force to Maolin Zhu is with North China Electric Power Univer-
promote the development of DSE. Thus, it is necessary sity, Beijing 102206, China.
to further study the connection between DSE and the ad- Hao Liu is with North China Electric Power University,
vanced applications of the next generation of EMS. Beijing 102206, China. p&e

Multifold Insights
for Power System
Dynamics From
Data Assimilation
Meeting Current Challenges
T
THE WORLD’S ELECTRIC POWER GRIDS HAVE iors and dynamics that the grid has never seen and that were
evolved over the last 120 years from a single power line not considered in its design. Operating such a dynamic grid
to today’s large networks. The evolution will continue at with sufficient reliability and efficiency is a monumental
an accelerated rate with extensive smart grid development challenge. Figure 1 presents three challenges faced by soft-
worldwide. For example, the U.S. government has set a ware tools used in power systems planning and operation
goal of reaching 100% carbon pollution-free electricity by along with solutions.
2035, while the U.S. Department of Energy shared a goal to First, as a basic function of energy management sys-
deploy 30 and 110 GW of offshore wind by 2030 and 2050, tems (EMSs), today’s state estimation largely ignores
respectively. To meet such ambitious goals, a significant the dynamics in the power grid. The electromechanical
percentage of electricity will come from intermittent renew- interaction of generators and dynamic characteristics
able sources and be supplied to a vast number of loads that of loads are not included in commonly applied opera-
will actively respond to grid conditions and incentive sig- tional models. The steady-state-based operational tools
nals. This development is largely driven by environmental do not capture such dynamics in the system and make
and economic factors, such as reducing carbon emissions it difficult to make effective decisions in time. To oper-
and reducing electricity costs for consumers. These energy ate the new power grid more efficiently and reliably, we
resources and responsive loads lead to new uncertain behav- need to explore ways to estimate the dynamic states of
By Shaobu Wang , Zhenyu Huang , Renke Huang,

Ning Zhou, and Junbo Zhao

synchronous generators and inverter-based resources in well. Historically, model validation and parameter calibra-
addition to the static states (the bus voltage magnitudes tion have been implemented using staged testing, which is
and phase angles). very costly and labor intensive. With the recent advance-
The second challenge is model accuracy. Because of uncer- ment of phasor measurement unit (PMU) technologies, it
tainties, models can deviate far from realities. For example, is possible to develop advanced validation and calibration
some of the generator models may not be well validated, result- approaches to adapt power grid models to new dynamics
ing in highly inaccurate model outputs that are inconsistent using measurements captured during operation.
with the measurements. As a case in point, on the 10 August The third challenge is the quality of measurement data.
1996 blackout in the western United States, the model used by Bad data in the measurement may come from large biases
planners and operators predicted system stability while the sys- due to infrequent calibration, instrument failures, impulsive
tem was undergoing numerous cascading failures that resulted communication noise, or even cyberattacks. Adversaries
in a rapid system collapse that occurred within minutes. The may stealthily access the computer networks in power plants,
same phenomenon was observed in the 2011 blackout in San substations, or control centers to disrupt service, manipulate
Diego. In power grid planning and operation, many decisions data, or disable the entire monitoring and control system.
rely heavily on static and dynamic simulation studies derived If left unattended, the bad data can lead to large errors in
from models. As a result, the integrity of power system models estimation and algorithm divergence or even mislead opera-
is very critical to reliable and economical electricity delivery tions. Hence, it is very important to investigate solutions for
to power consumers. data quality issues in power system operations.
Inaccurate models can significantly affect study results, This article summarizes our efforts to meet these chal-
leading to conclusions that deviate from reality, thus jeopar- lenges. The rest of this article is organized as follows. The
dizing system designs and affecting decision making. There- “Core Technologies” section introduces Kalman filter (KF)
fore, the North American Electric Reliability Corporation algorithms and their application in dynamic state estimation
(NERC) requires validating and calibrating large genera- (DSE), model calibration, and data calibration. The “DSE”
tor models every five years, with a main goal of having the section presents examples of application results of DSE. The
model outputs match the recorded measurements reasonably “Power Plant Model Validation and Calibration” section
©SHUTTERSTOCK.COM/METAMORWORKS

summarizes our model calibration algorithms and demon- when the distributions of the state noise have multiple modes.
strates how to integrate results into existing software. The However, computational requirements can be too high for a
“Data Validation and Calibration” section presents sample large system because the required number of samples grows
results of data calibration. Finally, some conclusions are exponentially with the number of states. To determine the
drawn in the final section. best algorithm for a specific application, a user often adopts
a trial-and-error approach to evaluate the estimation accu-
Core Technologies racy and computation time of these algorithms using simula-
Given dynamical models and time series measurements, the tion models.
current states and parameters of a system can be optimally To save computation time and fix the memory size for
estimated in real time using filtering technologies. Many KF real-time estimation, all of these filtering algorithms are
methods have been developed for different applications in often implemented recursively using the two-step structure
the power grid. illustrated in Figure 2. In the first step, the distributions of
the current states are predicted using the previous states.
The KF In the second step, the distribution of the current states is
Assuming linear models and normal distribution noise, corrected when a new snapshot of the measurement data is
Rudolf Kalman developed the KF under a statistical frame- received. The two steps are executed recursively to assimi-
work to estimate the current states of a system. The KF late the measurement data one snapshot at a time, with time
achieved many successes in different applications, starting passing by. The estimated states will converge to the true
with the U.S. space program, because it can provide an accu- states if the system is detectable and, preferably, observ-
rate estimate of the minimum variance and can be recur- able. To ensure the accuracy estimation, the ratio between
sively implemented for real-time applications. covariance of the process noise and measurement noise
To estimate the dynamic states in the power grid, the needs to be properly set up to balance the weights of the
KF method has been extended for nonlinear systems. current measurements and past states in estimating the
These extended methods include the extended KF (EKF), current states. Here, the covariance of the process noise
unscented KF (UKF), ensemble KF (EnKF), and particle describes the uncertainty of the state-transition models,
filter (PF). For power grid applications, each of these algo- which often comes from limited knowledge of how the
rithms has advantages and disadvantages in its estimation states change over time. The covariance of the measure-
accuracy and computation time. The EKF is often consid- ment noise describes the uncertainty of measurements,
ered a standard method for the DSE of nonlinear systems which often comes from the limited accuracy of measure-
because it is easy to implement and has a high computation ment instruments.
efficiency. However, its estimation accuracy degrades rap-
idly when the nonlinearity of the system increases. The UKF Multiple Applications
tends to produce higher estimation accuracy than the EKF Based on the filtering technologies, multiple applications
for a nonlinear system and has similar computation com- have been developed to help a power system engineer under-
plexity. For a large nonlinear system with many states, the stand the status of a system, envision where it is going,
EnKF may overcome the error growth problem of the EKF and ensure the integrity of its measurements. As shown in
while using less computation time. The PF is more flexible Figure 1, the applications of KFs in power systems can be
in modeling noise and can produce accurate estimates, even grouped into the following three categories:
Solution:
Challenges:
Repair Data
How to Track Dynamic States
Calibrate Model How to Enhance Model Accuracy
How to Improve Data Quality

Correct States
Phasor Model States Signal

Data – Mismatch
Simulation Calculation
figure 1. Challenges and solutions.

✔ DSE: KFs have been implemented to estimate the wrongly connected meters, wrong measurement mod-
dynamic states of synchronous machines and wind els, and even cyberattacks. If left unattended, the bad
turbines. The dynamic states are the minimum set data can lead to large errors in estimation, algorithm
of variables that can fully reveal the current status divergence, or even misleading operations. Bad data are
of operations. Not only can they be used as the real- often detected and identified using residual analysis,
time input of a control system to quickly damp out the which studies the statistical features of the difference
unwanted damaging oscillations, but they can also be between forecasted and actual measurements. Once the
used to set a starting point to predict the status of the bad data are detected and identified, their negative im-
power system at the next moment. Because dynamic pacts can be mitigated by adaptively assigning proper
states in power systems often are not directly mea- weights in estimation algorithms. When a bias is ob-
sured, the KFs are applied to accurately and timely served for a particular meter using the KF, the estimat-
estimate them using PMU data. ed bias can be used to calibrate its related measurement
✔ Online model validation and calibration: With more models to improve the measurement accuracy.
PMUs being deployed in power systems, KFs have been With the increasing penetration of modern measurement
used to calibrate the parameters of dynamic models of systems, such as PMUs, more applications of the KF tech-
generators and windfarms. The accuracy of dynamic nology are being developed and implemented to improve the
models is essential to the stability of a power grid be- reliability and efficiency of power grid operations by increas-
cause the models are often used in the planning stage ing operator situational awareness of statistical and dynamic
to reveal how a power grid responds to disturbances. behaviors. To illustrate their applications and impacts, three
In power systems, large model errors often come from examples are detailed in the following sections.
drifting parameters, changing operational conditions,
accumulative model errors, and even human errors, DSE
which can result in a simulation model failing to reveal As a central function in power grid operations, state estima-
a stability problem of an actual system. To avoid mis- tion generates critical inputs for other operational tools. Any
leading information from an inaccurate model, the dy- deficiency in the state estimation process will propagate
namic models of major components, such as generators, through the operational tools and have a direct impact on
are required to be validated and calibrated regularly. how the system is operated. Traditional static state estima-
The KF has been applied to calibrate the parameters of tors receive data from supervisory control and data acqui-
generator models online using PMU data without tak- sition (SCADA) systems, with data updated in the time
ing the equipment offline or injecting large disturbanc- interval of several seconds, which may cause the state esti-
es. The calibration can be performed for both original mator to overlook fast dynamics. Moreover, the static state
dynamic models of a single device or reduced models estimation (SSE) function commonly adopted by the power
of several devices. Calibration tools based on the KF industry is based on the assumption of a quasi-steady-state
have been integrated into commercial software. operation condition. This assumption can be violated by crit-
✔ Data validation and calibration: The KF can be used to ical disturbances, which is more likely in today’s industry
detect and identify bad measurement data and calibrate because of the faster dynamics of inverter-based resource-
measurement devices. Bad data in the measurement dominated systems. It has been reported that the SCADA-
may come from unexpected disturbances, bad sensors, based SSE currently deployed in industry EMSs fails to
Covariances of Process and Measurement at

Measurement Noise the Current Time
Predicted Corrected
1) “Prediction” Current States 2) “Correction” Current States
Initial States Using State- Using Measurement
Transition Models Models
Previous
States
Next Time Instant
figure 2. A recursive implementation of filtering algorithms through general prediction and correction steps.

capture accurate system states after a severe initiating event dynamic states (e.g., rotor speeds or angles) instead of volt-
or during cascading failures. ages. To compare with SSE in the same frame of reference,
Recent developments in PMU technology make estimat- we plot its playback values versus the results of SSE.
ing dynamic states (e.g., generator rotor angles and speeds) From this case, we can see that traditional SSE perfor-
possible with high-speed, time-synchronized measure- mance is sensitive to oscillations. The performance degrades
ment data. The estimation of dynamic states is not possible fast or leads to divergence when oscillations occur. SSE per-
using SCADA measurements due to the low measurement formance degradation is a potential threat to the safe opera-
data update rate. PMUs with a typical data update rate of tion of modern power grids. At the first stage of a cascading
30 frames/s can currently capture the vast majority of high- blackout, the loss of situation awareness means losing the
energy dynamics in power grids and, thus, enable DSE, opportunity to save the system. The high sensitivity of the
which provides a full dynamic view of a power grid and fur- SSE to performance degradation during oscillations blinds
ther enables look-ahead dynamic simulation and dynamic SSE when it is needed most.
contingency analysis.
Through a use case, we demonstrate how the perfor- Power Plant Model Validation
mance of SSE degrades when disturbances occur, and DSE and Calibration
captures the events overlooked by traditional SSE. In this To perform model validation at a power plant level, a sub-
case, we extract a subsystem with 154 buses from the North system model is created using an equivalent generator with
America Western Interconnection. This subsystem has full infinite capacity to represent the remaining power network,
PMU coverage. Since SCADA measurements were unavail- as shown in Figure 4.
able for this case, the time stamp order of PMU data within All of the model details of the power plant to be studied are
the update period was disrupted to mimic the time skew preserved. A PMU installed at the terminal bus can capture
errors of SCADA. Because this is a real case, the true values the fast dynamics of system disturbances, which are reflected
of states are never known. We use the residue between esti- in the recorded voltage magnitude and bus frequency. The
mated values and measurement values as a metric to evalu- outputs (including the active and reactive power) of the power
ate SSE performance. plant obtained via dynamic simulation with play-in signals
As shown in Figure 3, before the disturbance at the time are then compared against the corresponding PMU mea-
instant of t = 88 s (a real event that happened on 13 Septem- surements to identify model deficiency. The play-in signals
ber 2016), the SSE residues (the mismatch between stars and replace the simulated bus voltage magnitude and phase angle
circles) are very small; then, the residue increases sharply (or frequency) with the corresponding PMU at the point of
(i.e., the SSE diverges) as the disturbance occurs. However, connection at each time step during a dynamic simulation.
DSE can still track the system true states, and, finally, the The injected voltage signals will drive the dynamics of the
SSE residue and the oscillation decay return to small val- generator and its controllers before moving to the next step. If
ues when the disturbance is over. Generally, DSE estimates the difference between the simulated and measured response
Estimated Terminal Voltage Magnitude Estimated Terminal Voltage Magnitude

1.5
1.45
1.4 1.4
1.35
V15 Per Unit
V15 Per Unit
1.3 1.3
1.25
1.2 1.2
1.15
1.1 1.1
1.05
1
0 20 40 60 80 100 120 85 90 95 100
Time (s) Time (s)
(a) (b)
Decimated PMU SSE Estimate DSE Playback
figure 3. The (a) SSE and DSE performances following an event and (b) a zoomed-in view around the disturbance.

is sufficiently small, it indicates that the model parameters improves the model quality significantly. (The numbers in
are accurate, so parameter calibration is not needed. Other- red are much smaller than the numbers in black.)
wise, the parameter calibration shown in Figure 4 is used to Model integrity is critical to power grid planning,
find and correct the inaccurate parameters. economic dispatch, and operation, especially with the
The Pacific Northwest National Laboratory team inte- increasingly unpredictable and dynamic behavior of
grated our model validation and calibration algorithms modern power systems. To better comply with industry
into existing commercial software for practical power standards on improving model integrity (e.g., NERC
plant parameter calibration. The architecture design of Standards 26 and 27), the integrated methodology and
the integrated tool is shown in Figure 5. An application software tool suites can systematically validate stabil-
programming interface is developed to connect the model ity models and identify and calibrate problematic model
calibration block and the data preprocessing block. First, parameters using PMU measurements. The tool suite has
we extract the event data and dynamic model for validation been tested on a realistic hydro power plant and dem-
and calibration. Then, we perform validation by examin- onstrates outstanding success in greatly improving the
ing the mismatch between the model’s outputs and the field model performance. As suggested in the NERC guideline
measurements. If the mismatch is big enough (greater than (Standard MOD-026-1, Verification of Models and Data
a threshold), the model is not accurate, and we go to the for Generator Excitation Control System or Plant Volt/
model calibration block to perform validation. Otherwise, Var Control Functions), the PMU-based validation and
if the mismatch is small, the software tool outputs the per- calibration approach can effectively reduce the frequency
formance report. of costly staged generator tests.
In 2016, NERC hosted a workshop about model cali-
bration testing. The workshop held a blind test, with each
participant given the measurement data for 12 events and a
PMU
dynamic model of a hydroelectric power plant. The partici-
Calibration
pants were told that some errors existed in the model (i.e., the Boundary Bus
model’s output could not match the measurement data). The Remainder of
host did not say which parameters were not accurate. Partici- Power System P and Q
Model (External)
pants were asked to calibrate the model using any approach Subsystem Model for Validation
they liked. As one of the participants, the Pacific Northwest Measured 2 Simulated 3
National Laboratory team got the results shown in Figure 6 V, θ, and f 1 P and Q
and Table 1 using the approach described previously. Playback Measured
P and Q
Figure 6 shows the calibration effect for one of the events.
The blue curves represent PMU measurements, and green and figure 4. The model validation with event playback.
red are the model’s output curves
before and after model calibration,
respectively. As can be seen from • EMS Model
Figure 6, the mismatches between • PMU Measurements
the model’s output and measure- • Historical Event Data
and Time (Optional)
ment were reduced significantly API
after model calibration.
Model and Event
Table 1 shows the model cali- Database
Model Validation
bration effect for all 12 events. The Calibration
root-mean-square error (RMSE) Prepare EMS
Cases and Create Subsystem Parameter
is used as a metric to quantify PMU Location and Launch Model Calibration
the mismatch between the output and Retrieve Validation
curve and measurement curve. A PMU Data
low value of the metric means the Data Preprocessing Quality Check Yes
mismatch is small, and the model
Model Deficiency
quality is high; otherwise, the Identified?
mismatch is large, and the model No
quality is low. The numbers in
black are the RMSE values before Performance Model Performance and
Reporting Calibrated Parameters
model calibration, and those in
red are the RMSE values after
model calibration. As can be seen
from these results, calibration figure 5. The architecture design of the integrated tool.

0
82
Reactive Power (MVar)

Real Power (MW)
80 –50
78
–100
76
74
–150
20 25 30 35 40 20 25 30 35 40
Time (s) Time (s)
PMU Measured Playback With Original Parameters Playback With Calibrated Parameters
figure 6. The validation results. Blue is the PMU recording, and red is playback result using the calibrated model.
Data Validation and Calibration Therefore, it is improbable that measurement errors occur
In this section, we discuss error type detection to differ- simultaneously in different measurement devices, but it is
entiate model errors from measurement errors and how to a totally different case for model errors. A model error
handle data errors in measurements, including cyberattacks. will propagate in the whole system and can be detected
in many different measurement devices because system
Error Type Detection states are coupled through state transient equations. Take
In this section, we demonstrate how to differentiate model a 2D case as an example to illustrate the proposed detec-
errors from measurement errors through the mismatches tion approach.
between model outputs and measurements. Because mea- Based on this observation, we use the error type
surement devices operate independently in the power grid, detection method shown in Figure 7. Each small cir-
it is reasonable to assume that measurement errors from cle represents the location of a vector of mismatches.
different devices (or channels) also will be independent.
D2
table 1. The model mismatch quantification by
root-mean-square error. A
RMSE P (MW) Q (MVar)
Original Calibrated Original Calibrated
Events Parameters Parameters Parameters Parameters
B
01 1.61 0.15 17.8 0.76
02 0.52 0.14 3.53 0.28
03 1.08 0.07 13.07 0.5 D1
C
04 0.55 0.13 12.93 1.14
05 0.63 0.12 6.99 0.29
06 0.55 0.14 2.91 0.29
07 1.61 0.24 14.69 0.68
08 0.28 0.06 11.2 0.49
09 1.61 0.22 14.22 0.79
Area for Measurement Errors
10 0.4 0.13 2.52 0.59 Area for Model Errors
11 0.73 0.07 9.24 0.13
figure 7. The error type detection. Green is for model er-
12 0.19 0.08 16.66 0.21
rors, and light blue is for measurement errors.

Normally, without measurement errors and model errors, Conclusions
all of the small circles should be within a boundary circle. Renewable energy and smart grid developments are chang-
When either model errors or measurement errors happen, ing the mix of electricity generation and consumption. Such
the length of the vector defined will increase suddenly a new mix starts to affect the nature of power grid dynamics.
(e.g., points A, B, and C). In accordance with this obser- Data have shown that the power grid is experiencing more fre-
vation, when a measurement error happens, the vector quent and severe frequency deviations during normal opera-
leaves the boundary circle along either coordinate D 1 or tion, which demonstrates that the system is moving further
D 2 (e.g., points A and C). When model errors occur, the away from the quasi-steady-state assumption underlying the
vector leaves the boundary circle along both coordinates traditional SSE. The increased dynamic operational behav-
D 1 and D 2 (e.g., point B). iors call for the estimation of dynamic states in real-time
operation, online model calibration, and data validation and
Treatment of Data Errors calibration. The development and application of high-speed
Testing has demonstrated that the field measurement phasor measurement technologies provide opportunities for
noise/errors do not follow the widely assumed Gaussian implementing DSE, online model calibration, and data vali-
distribution; instead, they follow the thick-tailed distri- dation and calibration. The core algorithm is based on a two-
butions that can lead to a strong bias of the results from step prediction–correction process. Example results were
nonrobust estimators. In addition, with the strong reliance presented to demonstrate the implementation of the two-step
of smart grid functions on communications networks, process in a practical environment to address the challenges
cyberattacks have become a major concern. Typically, facing modern power system operations.
cyberattacks on real-time measurements are classified
as bias-injection attacks, denial-of-service attacks, and For Further Reading
replay attacks. Z. Huang, P. Du, D. Kosterev, and S. Yang, “Generator dy-
Three types of outliers are associated with a given namic model validation and parameter calibration using
dynamical system model, namely, observation outliers, phasor measurements at the point of connection,” IEEE
which affect the metered values; innovation outliers, Trans. Power Syst., vol. 28, no. 2, pp. 1939–1949, May 2013,
which corrupt the predicted state estimates; and struc- doi: 10.1109/TPWRS.2013.2251482.
tural outliers, which affect the system dynamic states and J. B. Zhao et al., “Power system dynamic state estimation:
observation functions. Observation outliers may result Motivations, definitions, methodologies and future work,”
from large biases in PMU measurements due to infrequent IEEE Trans. Power Syst., vol. 34, no. 4, pp. 3188–3198, Jul.
calibration, instrument failures, or impulsive communica- 2019, doi: 10.1109/TPWRS.2019.2894769.
tion noise. As for innovation outliers, they may occur in R. E. Kalman, “A new approach to linear filtering and
several different ways. For example, some of the genera- prediction problems,” J. Basic Eng., vol. 82, no. 1, pp. 35–45,
tor models may not be well calibrated, resulting in highly Mar. 1960, doi: 10.1115/1.3662552.
inaccurate model outputs that are inconsistent with the “WECC generator unit model validation guideline,”
measurements, as discussed in the “Power Plant Model Western Electricity Coordinating Council, Salt Lake City,
Validation and Calibration” section. UT, USA, Apr. 2020. [Online]. Available: https://www.
By contrast, structural outliers are gross errors, such as wecc.org/Reliability/WECC%20Generator%20Unit%20
incorrect circuit breaker statuses, or errors in the param- Model%20Validation%20Guideline.pdf
eters of the transmission lines, synchronous machines, or J. B. Zhao et al., “Roles of dynamic state estimation in
automatic voltage regulators. These gross errors could be power system modeling, monitoring and operation,” IEEE
due to sensor errors, human errors entering model data, Trans. Power Syst., vol. 36, no. 3, pp. 2462–2472, May 2021,
or cyberattacks. The conventional filters, such as the EKF, doi: 10.1109/TPWRS.2020.3028047.
UKF, H-infinity KF, and their variants are not robust to
any type of outliers. To deal with various types of outli- Biographies
ers, robust statistics-powered DSE methods have been Shaobu Wang is with Pacific Northwest National Labora-
developed, including the robust generalized maximum- tory, Richland, WA 99354 USA.
likelihood UKF and its hybrid version using the H-infinity Zhenyu Huang is with Pacific Northwest National Labo-
filtering technique. These robust estimators allow the han- ratory, Richland, WA 99354 USA.
dling of outliers and measurement noise while achieving Renke Huang is with Pacific Northwest National Labo-
good statistical efficiency under a broad range of thick- ratory, Richland, WA 99354 USA.
tailed probability distributions of the process and obser- Ning Zhou is with Binghamton University, Binghamton,
vation noise. In addition, the robustness of DSE naturally NY 13902 USA.
provides the needed power to defend against various types Junbo Zhao is with the University of Connecticut, Storrs,
of cyberattacks. CT 06269 USA. p&e

T
THE MILLIONS OF CONSUMER-OWNED DISTRIB-
uted energy resources (DERs) forecast for the grid by
2050 will trigger a major system change away from
centralized monopolistic utilities to decentralized com-
munity projects exploiting innovative business models.
Such a disruptive change, needed to switch to a carbon-
neutral economy, requires rethinking both the eco-
nomics and dynamics of power systems, knowing that
the electric power sector is going digital. DER-driven
uncertainties will impose costly operational margins
and preventive measures based on solving very com-
plex optimization problems. Keeping human operators
in the loop to supervise actions can limit reaction time
severely by making it difficult to respond in a timely
manner when multiple control systems are required to
stabilize the grid. The operator will hence need to be
assisted by an artificial intelligence system trained at
learning “good” decisions by imitating operators and
assessing the associated operational risk. Hierarchical
monitoring and control systems working in tandem with
decentralized markets and resources will allow the true
secure limits of decentralized, decarbonized, digitized,
and democratized (4D) grids to be identified. This will
be done with preventive/corrective actions executed in
seconds while balancing grid cost versus safety, reliabil-
ity, and stability.
DERs and other digital grid edge technologies
are now at a tipping point. By historical standards,
a 100% adoption rate is expected 20 to 25 years
Dynamic Wide
Area Situational
Awareness
By Innocent Kamwa

from now. A 4D electric grid paradigm will then
emerge, with the ability to provide decarbon-
ized electricity to a decentralized and digitized
economy. The domination of power electronics-
interfaced DERs will challenge the grid status
quo in terms of architecture, planning, and opera-
tions. Preparedness for such a revolution requires
urgent adaption on multiple fronts: 1) enhanced
more-dynamic wide area situational awareness
(D-WASA) tools and platforms; 2) faster, prefer-
ably in a subsecond time frame, decision support
tools and platforms; and 3) more application of
autonomous closed loops to mitigate uncertain-
ties and remove the human from time-critical
loops. The monitoring layer has been the focus
of stakeholders in the past decade, as evidenced
by the deployment of phasor measurement units
(PMUs) and smart meters worldwide. The time
has now come to turn attention to converting data
from these devices into actionable information
and mapping the streaming measurements into
feedback control signals.
Proactive consideration of DER-induced dynam-
ics in energy management system (EMS) applica-
©SHUTTERSTOCK.COM/URBANS
tions and dispatcher tools is needed to take full ben-

efit of the intelligence emerging at the grid edges.
Suddenly, wide area dynamic performance optimi-
zation of the bulk grid becomes possible through
harnessing increasingly smart DERs. However, this
can hardly be achieved by studying distribution and
transmission entities in silos. A more coherent
value chain-oriented view of DERs as assets in an integrated digital grid with millions of smart
devices is essential.
Here are some top priority concerns complicating our march toward an integrated grid highly pen-
etrated by smart DERs. The first is the activation of the network edge in support of new business models
that empower the consumer by taking advantage of the Internet of Things (IoT). Second, “keeping the
lights on” means remaining in control of the stability and reliability of networks dominated by power
electronics by using both decentralized and
centralized digital twin-like controls. The
third priority is the deployment of active
Propelling Future and simultaneous management of both the
technical and the economic performance
Decentralized, Decarbonized, of the electricity network in the presence
of uncertainties. Finally, in support of the
Digitized, and Democratized three previous objectives, it will be nec-
essary to enable the tools and technolo-
Electricity Grids gies required for large-scale integration of
DERs, which include hardware-in-the-loop
(HIL) cosimulation with replicas of supervi-
sory control and data acquisition (SCADA)
and IT systems in the loop.
At the center of this vision, we have a new evolution of the EMS—EMS with faster feedback controls
(EMS/C)—enabled by D-WASA and related infrastructure and tools. This closed-loop automation con-
trasts with the current EMS which, except for the automatic generation control (AGC) and secondary volt-
age control, relies mostly on market and human-driven feedforward controls.

A 4D grid means a more “dynamic” grid, mostly because
of reduced inertia due to the displacement of conventional
generation by power electronics-interfaced devices.
Lessons Learned From a Grid With system stands postcontingency (postdisturbance). Hence, if
Volatile Dynamics the ROCOV is negative and the voltage magnitude is below
A 4D grid means a more “dynamic” grid, mostly because a specified threshold, the system is exposed to a voltage col-
of reduced inertia due to the displacement of conventional lapse after the event is cleared. Similarly, if the ROCOF is
generation by power electronics-interfaced devices. Fig- negative and the frequency is already low, the risk of fre-
ure 1 illustrates the correlation between reduced inertia and quency collapse is higher, and rapid emergency load shed-
degraded frequency response. This recording from an actual ding may be needed to restore the frequency to a safe state.
Latin American interconnection shows poorly damped very By examining many simulations performed offline
low-frequency oscillations triggered by a 10% generation and historical records of actual events, the appropriate
loss. The total embedded generation is 7,350 MW in the threshold, the predictive of power grid instability, can be
first stage and 6,700 MW in the second stage. As inertia set for dynamic states and corresponding WASIs. Artifi-
decreases, the common mode frequency (at 0.05 Hz in Fig- cial intelligence tools are increasingly used to automate
ure 1) becomes more unstable for events of similar sever- this learning task, which is impossible to achieve by trial
ity. Small-signal spontaneous unstable oscillations and a and error when millions of data points are available from
decreased nadir for the same loss of generation are possible PMUs and simulations.
consequences of time-varying inertia. Such credible insta-
bility risks justify a more dynamic situational awareness in Evolution of Wide Area Monitoring
4D grids with reduced and time-varying inertia. at Hydro-Québec
Following a contingency, D-WASA may quickly pro- A historical peak load of 40,325 MW was recorded in Qué-
cess the generators’ states to derive response-based stabil- bec, Canada, on 25 January 2022. The electricity mix was
ity metrics, such as wide area severity indices (WASIs). 94% hydraulic, 4.6% wind, and 1.4% from other renewable
Other metrics often considered are the rate of change of resources, such as biomass. Hydro-Québec hence finds itself
transient energy (RoCoTE), the rate of change of frequency in the enviable position of a 100% renewable utility. However,
(ROCOF), and the rate of change of voltage (ROCOV). The Québec province is far from carbon neutrality since electricity
estimated stability metrics are next conveyed to dispatch- represented only 45% of the final energy used in 2018, due to
ers as forecasts of how far from the edge of instability the reliance on oil and natural gas in transportation and industries.
9 March 2011—5:12 to 5:42 p.m.
60.2
60
Frequency Deviation (Hz)
59.8
59.6
59.4
59.2 Substation 1
Loss of 1,115 MW Substation 2
59 Substation 3
Loss of 645 MW
0 500 1,000 1,500 2,000
Time(s)
figure 1. Low inertia means a higher risk of frequency oscillations and instability in response to disturbances. Because it
is a common-mode frequency oscillation, all signals are superimposed perfectly.

According to the 2022–2029 Hydro-Québec strategic plan, weaken the east–west interties under heavy loading condi-
100 TWh of electricity are needed to achieve the carbon neu- tions, while voltage stability and static loadability are the
trality of Québec in 2050, which means 50% more electricity major concerns in the Montréal area (Figure 2).
than today. A massive ingestion of DERs on the grid edges is With large-angle shifts existing among the 735-kV sub-
deemed necessary for achieving this goal economically, mean- stations separated by long distances, Hydro-Québec was
ing that an accelerated trend toward 4D grids is likely. naturally interested in power grid monitoring using time-
Today’s Québec grid is still a largely centralized and lon- synchronized measurements. Engineers from a unit called
gitudinal system. It was designed to ship about 426,776 MW Grid Behavior, part of the system operations department,
of hydro production from remote powerhouses in Québec and installed two measurements devices designed by Hydro-
Labrador, Canada (5,428 MW at Churchill Falls), to the main Québec Research Institute (IREQ) scientists. They were
cities along the United States border. Part of this electricity is located at the Arnaud substation (near the city of Sept-Îles,
also traded (35.6 TWh of net exports in 2021) with neighbors QC, Canada, in the northeast) and Boucherville (BCV) sub-
in New York and New England, USA. With 34,775 km of station, on the south shore of Montréal, 800 km apart, to
transmission lines and 542 high-voltage substations (includ- track the phase shift between the two locations. Implemented
ing 12,319 km of lines and 41 735-kV substations), Québec prior to digital relays and GPS universal time dissemination,
remains the North American Electric Reliability Council these devices used simple digital counters synchronized
(NERC) interconnection with the largest geographical foot- with the Loran-C network of satellites for navigation.
print. It has a 1,000-km average distance between the main Reporting on this concept, a paper presented at the 1977
power plants and load centers and the same average distance IEEE Power & Energy Society Summer Meeting, in Mexico
between the northeast and northwest power plants. City, Mexico, started as follows: “The bus voltage angle is one
The grid was designed to be basically unstable without of the state variables of a power system. This variable is very
dynamic shunt compensators by minimizing the number closely linked with the behavior of a network.” Although the
of lines for the sake of cost and environmental impact. The characterization of “bus voltage” as a power grid state vari-
1972–1973 power system master plan estimated that for the able was a bit of a stretch (since we know today that it is the
same power transfer capability, roughly four lines between generator rotor angle that should play this role, as discussed
James Bay and Montréal, QC, Canada, were avoided thanks later), it is quite clear from the context that this early work was
to the dynamic reactive power capabilities of the newly instrumental in establishing the need for both PMU technol-
invented thyristor-based static var compensators (SVCs). ogy and its application to dynamic state estimation (DSE).
From 1972 to 1979, 9x250-Mvar synchronous condensers After the primitive monitoring system described previ-
were installed, with 11x300 Mvar of SVCs in the following ously demonstrated its usefulness in forensic investigations
decade. Facing a constant increase in line flows, from an of network events, three generations of wide area monitoring
average of 1,500 MW/line in the 1970s to 2,700-MW/line in systems (WAMSs) were implemented from 1981 until 1998,
the 1990s, 11,000 Mvar of series compensation were further based on inter-range instrumentation group B time synchro-
installed by 1995, in addition to faster transmission line pro- nization with in-house PMU and phasor data concentrator
tection. Power systems stabilizers
(PSSs) also became mandatory in
all power plants of over 10 MW of 1
total capacity in the 1960s. With- Transient
T i and
Dynamic Stability
out the PSSs online, the north–
south power transfer needs to be
curtailed by more than 1,000 MW.
Notwithstanding this careful
design, power transfers in Qué-
bec up to modern times remained
largely stability limited thanks to
the high thermal limits of the 735-kV
transmission equipment, com- Voltage
V
Vo ltag Stability
bined with a pattern of demand 2
Fourteen PMUs
peaking during the coldest days of Eighteen Potential Wide Area PSS Sites
winter, due to the widespread use
One PDC for R&D at IREQ
of electric space heating. While
Fourteen Static Var Compensators
the transient stability limitation is
a characteristic of the power plants Nine Synchronous Compensators
in the north, interarea oscillations figure 2. The enabling conditions for wide area control in Québec (circa 2016).
are triggered by contingencies that IREQ: Hydro-Québec Research Institute; PDC: phasor data concentrator.

(PDC) devices. In 2004, the system switched to Macrodyne on dispatcher screens. Finally, the area control error (ACE),
GPS-based PMUs compliant with IEEE Standard C37-118- at the bottom of Figure 3, shows some spikes, though its
1991, customized to include some algorithmic features spe- standard deviation is small, at v = 80 MW, and consistent
cific to Hydro-Québec, such as harmonics phasors up to rank with NERC requirements.
12 and voltage asymmetry over the three-phase power.
By 2015, the WAMS consisted of 10 PMUs, each associated Why and How SE Became Mainstream
with an electrical coherent area of the bulk system, and a redun- After the frenetic early years of fast-growing interconnec-
dant in-house PDC linked directly to the EMS and located adja- tions to exchange cheaper energy, it was soon realized that
cent to the control center. The WAMS in 2015 covered about operating the resulting giant networks was anything but
25% of the 735-kV buses in the system. The PDC collects the trivial. Widespread blackouts could take place if the actual
PMU data at one sample/cycle and then performs online time conditions of the network were not properly monitored and
reconciliation and transmission/GPS error correction before controlled in real time. A 1965 blackout served as a wake-up
transferring the results to the appropriate applications. call, triggering the use of primitive digital computers and the
Figure 3 illustrates the performance of the Hydro-Québec development of the first SCADA systems and enabling the
WAMS through an eight-year record of frequency and total concept of the control center as we know it today.
harmonic distortion at two typical 735-kV substations. The This platform includes several specialized functions to
frequency has an average error of only 0.5 mHz (which may support operator decision making in addition to the typical
well be related to instrument accuracy rather than an actual monitoring tasks, such as AGC and economic dispatch, for
frequency offset) and a standard deviation of v = 9.2 mHz which the system frequency and generator power outputs
during eight years. The periods of high harmonic distortion, need to be continuously supervised. Initially, operators in the
such as in January 2008, have been correlated with geomag- control room were supplied with operating guides from plan-
netic disturbances (GMDs) induced by solar storms. The ning studies. However, unexpected events often produced
correlation between harmonic distortion and GMDs is the emergency calls to run more cases and modify the operating
reason why this quantity is monitored in real time and shown guides, which was inconvenient to do manually in real time.
Standard Deviation = 10 mHz

600
Deviation (mHz)
400
200
0
–200
–400
1 January 2006 (12 a.m.) 1 January 2008 (12 a.m.) 1 January 2010 (12 a.m.) 1 January 2012 (12 a.m.)
(a)
Mean Value = 0.1%

10
Distortion(%)
0
(b)
Standard Deviation = 80 MW
2,000
1,000
ACE(MW)
0
–1,000
–2,000
(c)
figure 3. The sample recording of the Hydro-Québec WAMS over eight years. The time series plot of the (a) BCV
frequency, (b) Chateauguay substation (CHA) harmonic distortion, and (c) area control error (ACE).

The process was automated a bit by providing the “opera- between the connectivity represented in the EMS model and
tor load flow” software tool in the EMS. The dispatchers the actual status of breakers in the field. Addressing this con-
could manually enter data to check system behavior during cern is also important in the practical implementation of SE.
emergencies. However, it was difficult to match existing SE provides a complete and reliable database, which is
conditions with the load flow results because of insufficient vital for the correct performance of all online application
and erroneous data and network modeling errors. Since the functions that are used in the operation of the system and its
operator’s load flow needed to be accurate so that succeed- secure control. A solution provided by SE is also essential
ing studies were accurate, the process of power system SE for many offline applications that are related to system plan-
was born. SE was cleverly suggested by a famous group ning (load forecasting, reliability, network expansion, and so
of Massachusetts Institute of Technology scientists, led by on) as well as electricity markets.
Fred C. Schweppe, to process available measurements to
estimate the existing state of the system in real time. Dynamic States of the Power Grid
Dy-Liacco once wrote that “control centers are here to stay.” At
Static SE Basics the dawn of a 4D electric industry, paraphrasing Dy-Liacco, we
Let us consider the simple linear and static electrical net- can state that “control centers are here to stay forever but need
work in Figure 4. The network’s parameters are the known to turn dynamics conscious.” Long thought of as security insur-
resistances, R 1, R 2, and R 3, while the current sources I 1 ers of energy flows and market orchestrators, control centers
and I 3 are both unknown and not metered. We have four now need to morph into dynamic asset and network managers
instruments available to monitor this grid: three ampere to materialize the long-due convergence between asset manage-
meters, A 1, A 2, and A 3, and one voltmeter, V. The latter pro- ment and network operations into the same silo.
vides a measurement of the third state, e. This situation is Awareness of “dynamics,” through high-resolution time-
representative of how a real network is seen from the EMS, stamped information, with a time resolution below the
where the measurements are obtained by the SCADA system human reaction time, is facilitated by the increasing IT/oper-
through remote terminal units located in substations, while ational technology convergence. It is a critical condition for
the electric model (the resistance parameters and connectiv- enabling digital twins of cyberphysical assets and networks,
ity among them) are obtained from the operational planning both centralized in EMSs and decentralized in DER man-
model stored in the EMS database. agement systems (DERMSs), substations, and power plant
In statistics, this problem is known as the overdetermined SCADA systems.
least-squares problem because we have more equations than In dynamical systems theory, “states” are a set of vari-
unknown values. The best solution is obtained by minimiz- ables whose initial values, together with a set of differen-
ing the sum of squares of the errors in the estimated val- tial equations, allow one to predict the system future value
ues. The underlying algorithms are used in conjunction with of the “state.” This formulation applies to various kinds of
algebraic power system models that are static and like those systems whose observable behavior evolves with time, such
used in conventional power system load flow analysis. Data as mass–spring systems, electric generators, star systems, and
inputs for a power system case include measurements that chemical reaction processes. Something common to all these
are nonlinear functions of a particularly defined static state, systems is the notion of energy. State variables can describe the
in contrast to this simple resistive network example imple- stored energy at the initial condition. This initial energy is next
mented using linear measurement functions only. propagated over time, with propagation described using dif-
Some key challenges of SE come to mind while exam- ferential equations constraining the state variables. The state
ining Figure 4: What happens when the measurements are
inaccurate? We can still obtain reasonably good results by
using a weighted least-squares solution, where more weight +
V
is put on the errors associated with meters identified as inac- A1 V
curate. When a measurement is “extremely” inaccurate, we A +I
call it “bad data” that we should remove altogether from the R1 e
R2 R3
estimation process. Bad data are detected by identifying out- I1 I3
lier values through analyzing the residual errors of the esti-
mation. The suspicious measurement is discarded, and SE is
A2 A A3 A
repeated until the residual errors are normally distributed.
Another question is, What happens when we lose one
meter, for example, A 1? We can still find a solution in this
example, but if we lose another meter, the network is deemed figure 4. A resistive linear dc network illustrating the static
unobservable because we can no longer find three unknown SE problem. Here, A and V represent ampere meters and
states (I 1, I 3, and e) from two measurements, even if they voltmeters to measure voltage and current, while I and e
are perfect. Another potential concern is having a mismatch represent current and voltage sources.

equations consider both endogenous and exogenous excita- could safely claim that the “state” of a power system was a
tions in addition to internal losses. direct result of the state of the “synchronous generators” in the
With the triumph of the Tesla and Westinghouse ac energy grid. With inverter-based generation becoming widespread and
transmission concept over Edison’s dc distribution, the syn- thus displacing conventional generators, how long this state-
chronous generator became the cornerstone of electric energy ment will hold true remains an open question. But the truth of
systems. Indeed, until recent times, all the electricity consumed the matter is that, for the foreseeable future, the electric system
worldwide was produced only by synchronous generators. We will remain synchronous, meaning that it will still be domi-
nated by synchronous generators
and electronics-interfaced electric-
Reference Voltage ity sources mimicking the behavior
Voltage Generator
Excitation Field Rotor Electrical Electrical of synchronous generators.
System Excitation Four States Coupling A generator model consists of two
One State
Terminal parts, the electrical and the mechani-
Voltage cal models. Each part has its own
Reference
Power Torque
state variables corresponding to
Speed Generator
Mechanical electrical and mechanical differen-
Governing Rotor
System Mechanical Coupling tial equations coupled together by
Network,
Reference One State Speed Two States Loads, and the torque balance equation describ-
Frequency Generator ing the time evolution of generator
Stator interaction with the grid from initial
Generator 2 Network Coupling Equations equilibrium conditions. Timescale
Electrical + Mechanical
decoupling permits considering the
…. electrical states as constant vari-
Other Network Coupling ables equal to their initial values.
Generators Electrical + Mechanical As a result, the generator response
over time is described by the sec-
ond-order swing equation, which
is a mass–spring-equivalent model
Differential Algebraic able to predict the rotor angle and
Equations Equations
electric power oscillations between
(a) the generator and a remote bus to
which it is connected. In this repre-
Terminal Bus sentation, the states are the generator
rotor angle and speed. In 4D grids,
Synchronous Multimachine Power
Generator System the timescale decoupling assump-
tion is no longer reasonable, and
the response of an electrical model
Three-Phase Terminal Known needs to be considered, as well.
Voltages and Generator Figure 5(a) describes a multima-
Currents Parameters Two Mechanical
GPS States
chine power system, with each gen-
Vabc, Iabc
Dynamic erator described by two mechanical
Fundamental State states and four electrical states. Known
Phasors Estimator Four Electrical
Smart Fundamental as the detailed dynamic model in tran-
With States
PMU Frequency sient stability studies, it allows for sys-
Auxiliary Outputs Harmonic Known
Event Flags to SPDC Phasors Inputs tem states to be predicted and con-
Two Unknown
trolled at any future time. Excitation
Inputs
and governing systems introduce
EMS Set Points: Two Transient
Power, Frequency, and Voltage their own states, often approximated
Stability Indices
by the mechanical torque and field
(b) excitation voltage, resulting in a total
of eight states per generator.
figure 5. Decentralized DSE in a multimachine power system. (a) The concept of While the mechanical state vari-
the generator electromechanical model in the multimachine power system simulation ables can be measured, though at
model. (b) The PMU-based DSE algorithm for a single generator, using the subtran- a cost, the other six state variables
sient generator model. SPDC: substation PDC. cannot. We have to “estimate” and

“observe” them from available measurements by using the non- of the round-trip delay between Eaton’s substation PDCs
linear state and measurement equations. This idea first came (SPDCs) and the centralized PDC was less than 16 ms. This
out in the 1970s, when the concept of state observation/estima- latter figure dropped to 5 ms in a newly commissioned tele-
tion emerged in automatic control systems theory. It may be that communication network based on the Multiprotocol Label
the grid was not “dynamic” enough at the time to justify DSE Switching concept.
[Figure 5(b)] working alongside static SE. The sensors needed Around the same time, IREQ licensed its adaptive Kal-
to estimate the terminal values accurately enough for DSE man filter-based PMU algorithms to the grid modernization
were not available, either. Only recently has PMU technology firm Vizimax. The resulting PMU for control (PMU/C) device
changed the landscape enough to meet DSE requirements. received IEEE certification in 2016, demonstrating an abil-
To compute the states, the terminal bus measurements ity to report phasors jitter-free at an update rate of 240 Hz.
Vabc and I abc are used in Figure 5(b), together with power, fre- Its protection class (P class) PMU exceeded all C37.118-2014a
quency, and voltage reference values from the EMS, if avail- requirements, especially for out-of-band attenuation, with a
able. The generator parameters, such as inertia and reactance, fault ride through capability suitable for DSE-based WACSs.
are also necessary to execute the algorithm, which outputs six According to the IEEE certification laboratory, the PMU/C
dynamic states, two unknown inputs (if applicable), and two was compliant with steady-state measurement class accuracy
transient stability indices derived from dynamic states. The requirements while meeting P class latency requirements.
unknown inputs are computed only when the reference values Figure 6 provides two examples of PMU field recordings
of mechanical power and excitation voltage are not available captured during the WACS pilot project at Hydro-Québec.
due to sensing challenges, for instance. The PDC was propelled by Eaton’s Visual T&D software, with
additional features implemented to estimate remote voltage by
PMU Requirements for DSE using a detailed line model. The voltage and frequency of the
DSE is built on top of PMUs, which use an unknown manufac- BCV substation are shown in Figure 6 together with the volt-
turer’s algorithm. When the frequency is close to nominal, the age of the Chenier substation, both near Montréal. Figure 6(a)
measurements look very decent.
When the frequency deviates, dif-
ferent manufacturers’ PMUs will Voltage (p.u.) Frequency (Hz)
have different errors. More work
needs to be done to establish a stan-
dard way for computing phasors
under various conditions in terms
of frequency variations, transients,
16/09/2013 16:21 : 12.1 UTC
and timing uncertainty, which
directly translate to uncertainties
about DSE results derived from
PMUs. However, the latest IEEE
PMU standards, published in 2014,
tightened requirements under tran- (a)
sients, though legacy PMUs that Voltage (p.u.) Frequency (Hz)
do not meet the IEEE compliance
process are here to stay.
In the early 2010s, PMU require-
ments for DSE-based wide area
stabilizing controls were published
by Hydro-Québec after perform-
ing in-house test-based evaluation
of three vendors. Four PMUs from
the batch were installed, as in Fig-
18/09/2013 18:41 : 17.910 UTC
ure 2, in the framework of a wide
area control system (WACS) pilot
project. Over a 12-month period, (b)
no loss of packets was observed.
But with a reporting rate of 60 Hz, figure 6. The dynamic performance of measurement-class PMUs from two vendors
some PMUs suffered from tim- during major grid events. The (a) loss of 1,531-MW exports to New England from
ing jitters of a 0.05-ms mean. the James Bay area and (b) loss of 725 MW of generation in the northeast part of the
The four-times standard deviation grid. p.u.: per unit.

Long thought of as security insurers of energy flows
and market orchestrators, control centers now need
to morph into dynamic asset and network managers.
describes the system response following a heavy loss of exports nous speed of the generators. This is true in a steady state
to the United States. Figure 6(b) reports the system response only. During a speed transition from one equilibrium point to
to the loss of northeastern generation. In both cases, the fre- another, generator speeds and bus frequencies have no reason
quency tracking is deemed accurate enough for a DSE-based to be the same. This fact is implicitly recognized in the DSE
WACS or system integrity protection system (SIPS). algorithm of Figure 5(b), which derives the generators’ rotor
Figure 7 gives the typical architecture of PMU-based cen- speed and angle from terminal phasor measurements.
tralized controls inspired from the International Electrotechni- Interestingly, estimating the rotor speed as the frequency
cal Commission (IEC) 61850-90-5 use case taxonomy. While of the “internal” voltage derived from the relation between the
PMU data from generators can be used directly in these schemes synchronous generator voltage and current phasors was redis-
for actionable decisions, as seen in the next section, the per- covered in the 2010s by the DSE community, although the
formance of the WACS/SIPS might benefit greatly from DSE approach has been in use since PSSs were first applied. Defin-
processing and coordination with static SE in the control center. ing the integral of acceleration power (mechanical – electrical
torques) as a proxy for rotor speed is also a kind of “dynamic
DSE-Informed Power Grid Control: state” estimation principle, already put in practice in PSSs.
Why and How Figure 8 illustrates a highly idealized Hydro-Québec
Since today’s electric grid is synchronous, it might be wrongly network with “smart” PMUs installed at all powerhouses to
inferred that the frequency should be the same at every bus achieve decentralized DSE. The PMUs also provide har-
and that this frequency should equal the common synchro- monics phasors estimation and data reporting. This simulation
Substation or Control Center Selected

Power System
Phasor Values Control
PDC Controller
Equipment
All Phasor Control
Values Commands
Phasor SPDC
Values
PMU PMU PMU PMU
V, I V, I V, I V, I
Substation 1 Substation 2 Substation N
(a)
Actor Name Role Description
PMU Computes Synchrophasors, Frequency, ROCOF, ROCOV, RoCoTE, Harmonics, and so on

SPDC Substation Phasor Data Concentrator—Collects, Aligns, Selects, and Possibly Decimates
Data From Several PMUs in the Substation
PDC Phasor Data Concentrator—Collects, Aligns, Selects, and Possibly Decimates Data From
Several PMUs or Other PDCs
Controller Receives Data From the Local PDC and Executes Algorithms that Implement
Control Functions
Power System Control Elements, Including Breakers, Switches, Flexible ac Transmission
System Control Equipment
Systems, High-Voltage dc, and Similar Equipment
(b)
figure 7. A PMU-based wide area control use case according to International Electrotechnical Commission 61850-90-5.
(a) Two-level wide-area monitoring system. (b) Taxonomy of components in a wide-area monitoring and control system.

LG2 LG3 LG3a LG4 JAMES BAY CHURCHILL CHU
5,600 MVA
5,500 MVA 2,200 MVA 200 MVA 2,700 MVA
LG2_13 LG3_13 LG3a_13 LG4_13 CHU13

1 1 1
1 1
5,600 MVA
5,500 MVA 13.8/735 kV
2 2,200 MVA 200 MVA 2 2,700 MVA 2
13.8/735 kV 2 2 13.8/735 kV
13.8/735 kV 13.8/735 kV CHU735
LG2_735 LG3_735 LG4_735 LdCHU

CHU_ARN 330 MVAR
LG2_LG3
LdLG2 450 km
LdLG3 LdLG4
155 km 660 MVAR LG3_LMO LG4_LMO 660 MVAR
LG2_NEM
660 MVAR 104 km 118 km BaCsARN
279 km BaCsALB
LMO_ALB
218 km CsMTG_ARNN
BaCsNEM LM0735
CsALB ARN735
CsNEM
LdLMO
NEM_ALB735 330 MVAR
CsARNS1 CsARNS2 LdARN
990 MVAR
LdNEM_ALB
NEM_ABI 1,980 MVAR MAN MANIC
236 km
SVC 5,000 MVA
1,200 MVAR
BaCsCHB
ARN_MIC ARN_MAN
CHB_CHB Fault 2 SAG_MIC 180 km 169 km MAN13
CsABI_CHB 1
195 km 247 km
BCsCHMn MIC_MAN
ABI_CHB735 BaCsSAG 5,000 MVA
2
60 km 13.8/735 kV
CsCHMN CsSAG MIC735 MAN735
CHM_SAG
ABI_CHB CHM735 SAG735
271 km CsBER
SVC LdABI_CHB CsPER
600 MVAR
CsCHMS LdSAG
1,320 MVAR BaCsPER BaCsBER
BaCsLVDn 660 MVAR LdMAN_MIC
SVC BaCsCHMs 1,650 MVAR
LdCHM
CsLVDN 600 MVAR 1,320 MVAR
LVD735
MIC_LTD MAN_LEV
409 km 379 km
CsLVDS CHM_JCA
LdLVD 252 km
990 MVAR
BaCsLVDs
SVC
600 MVAR
QUE735
LVD_MTL MTL_QUE
2
219 km 5,000 MVA
260 km 735 kV/25 kV
1
MTL735 QUE25
2 2
5,000 MVA 22,000 MVA Fault 1
735 kV/25 kV 4,000 MW
13.8 kV/735 kV QUEBEC
1 1
LdQUE1
MTL25
MTL13
Discrete,
MTL Ts = 6.51e-0.5 s.
5,000 MVA LdMTL4 LdMTL4x LdMTL1 LdMTL2 LdMTL3
MONTREAL 6,500 MW 2,900 MVAR 4,000 MVAR 12,000 MW 250 MW
figure 8. The simplified simulation model of the Hydro-Québec network in MATLAB/Simulink Simscape Electrical.

In 4D grids, the timescale decoupling assumption is no longer
reasonable, and the response of an electrical model needs
to be considered, as well.
model was implemented in SimPowerSystems and is now Needless to say, other than some early applications in
available as a demo in the Simulink/Simscape Electric pack- local generator controls, no EMS application of DSE has
age from MathWorks. been built from the bottom-up at Hydro-Québec to date. The
Following a contingency (disturbance) applied at the sensor infrastructure barrier is less of a concern today, given
location named “Fault 2,” Figure 9(a) superimposes the the emergence of faster communications and cheaper PMUs
true rotor speed and bus frequency at three power plants, providing accurate phasor and frequency with fast dynamic
while each plot in Figure 9(b) compares phase angle shifts response. However, given the high number of states in a real
between two power plants, with one trace using the mea- system, achieving a DSE solution can be intractable if all the
sured terminal voltage phasors and the other using the rotor generators in a large system are considered simultaneously.
angles. The conclusions from the results shown are that the For example, for a system with 100 machines, we will need to
magnitudes of rotor speed oscillations are bigger than the track 800 dynamic states in real time, using a far larger num-
magnitudes of frequency oscillations during the transient ber of generator measurements. This scalability issue can be
stage, while the two signals are the same in the steady state circumvented by constructing the system model state “locally”
and during very slow common frequency oscillations (Fig- by ascertaining the state of each machine independently.
ure 1). Similarly, the transient angle shifts among remote Although DSE may become suboptimal due to subdividing the
locations are larger in the rotor angle than in the genera- system, this decentralization has several important advantages:
tor terminal bus phase angles. The rotor angles are not the ✔ Each machine state is evaluated individually without
same as the generator bus voltages in the steady state, by any knowledge of the interconnecting network, which
the way. These simulation results answer the “why DSE” is necessary only to predict the total system response.
question: in addition to providing a physics-informed sen- ✔ The DSE output data can then be telemetered to a cen-
sor, DSE enhances observability, which results in crisper tral processor, which filters the individual machine
variables for preemptive decision making and corrective state estimates by using a complete system model to
actions through WACSs and one-shot SIPSs. produce a federated system state estimate.
Rotor Speed Deviation Versus Terminal Bus Frequency Rotor Angle Shift Versus Bus Voltage Phase Angle Shift
0.4 10
LG2 (Hz)
LG2–MTL (°)
0.2
Rotor Speed Deviation
0 Bus Frequency Deviation 0
5 10 15
–10
0.4
MTL (Hz)
5 10 15
0.2 10
0
MAN-MTL (°)
5
5 10 15
0
0.6
MAN (Hz)
0.4 –5
0.2
0 5 10 15
Time (s)
5 10 15
Time (s) Rotor Angle Shift Bus Phase Angle Shift
(a) (b)
figure 9. The DSE of an idealized Hydro-Québec network model: rotor speed and angle deviations are not bus voltage
frequency and phase angles. (a) The La Grande 2 (LG2), Montréal (MTL), and Manicouagan (MAN) rotor-speed versus the
bus frequency. (b) The angle shifts between MTL and LG2 and MAN.

In addition to providing a physics-informed sensor, DSE
enhances observability, which results in crisper variables for
preemptive decision making.
In the central location where generators’ PMU-based data 2) Can we compute the operational limits by using the
are concentrated, we can still estimate the state on a per- actual conditions, focusing on plausible threats and
machine basis so that the computation speed for the system-level possible contingencies over the next operating period?
DSE can be increased through parallel estimation at separate 3) Can we change the way operational reliability is as-
machine locations. Generator PMU data can be processed sessed and used to include probabilistic risk manage-
locally, and the estimated state data transferred to the EMS. ment aspects in operations time frame?
Or, PMU data can be sent unprocessed to the EMS for DSE Once the network rules are adapted, we can think about
on a per-machine basis. But in both cases, we must overcome deploying key technologies leveraging D-WASA in stages.
the limitation of communications channels, which include jit- The following concepts have been discussed since the PMU
ter, colored noise, data dropout, latency, and so on. To some boom in the 2000s:
extent, these issues can be mitigated through filling in the data 1) DSE-informed SIPS algorithms using fuzzy logic
gaps by interpolation and using federated Kalman filters with rule-based systems and other artificial intelligence
adaptive variance. Robust DSE approaches are also promising machines instead of fixed Boolean decision systems
in this regard. 2) integrated probabilistic risk assessment in real-
time operations, rediscussed by stakeholders after
Moving Forward With DSE in EMSs every blackout
3) automatic operators, first discussed by Dy-Liacco,
Technology Gap Analysis in 1996
A robust framework for mastering the dynamic performance 4) dynamic simulators for operator training on D-WASA
challenges of highly penetrated 4D grids can be drafted based on and 4D grids.
technology and research gap analysis. First, the perceived gaps:
1) An integrated region-wide control perspective has not DSE Enables Response-Based WACSs and SIPSs
been formulated. That Improve System Dynamic Performance
2) Widespread deployment of intelligent electronic de- Built on advanced PMU algorithms, DSE can meet the strin-
vices has not occurred, nor has deployment of a uni- gent requirements for automatic feedback control. Suddenly,
versal interface (for instance, in Québec, less than challenging applications long thought intractable and risky,
10% of substations are fully digitized and compliant such as closed-loop WACSs and response-based SIPSs (Fig-
with the IEC 61850 standard today). ure 10), are within reach. These functions will become parts
3) Integrated communications infrastructure is missing of EMS/C, contributing toward a big scheme to improve the
(and may be the missing link in the future despite the dynamic performance of DER-dominated grids. Accord-
increasing adoption of 5G/6G networks). ing to Hydro-Québec findings over the past decade, some
4) The availability of data in the right place at the right time tangible and quantifiable metrics of success attributable to
is still limited by institutional silos and privacy concerns, DSE-based wide area applications are listed in Table 1.
meaning a lot of dormant data in enterprise IT systems. The first application list in Table 1 boosts the reactive
5) Sensor cost remains high, though emerging IEC stan- power at dynamic shunt compensators to prevent voltage col-
dards for power grid IoT applications may change that. lapse, based on using postcontingency remote voltage collapse
6) SE still operates at refresh rates of minutes. Signifi- indicators. The second application modulates reactive power
cantly higher-power computers may need to be em- output from hydropower plants, wind farms, and dynamic
ployed to support subsecond refresh rates and time shunt compensators by using the multifunctional multiband
synchronization with DSE. PSS (MF-MBPSS) fed with an average speed from DSE, as
To close these gaps, we need to think outside the box shown in Figure 11. The MF-MBPSS can be enhanced with
to unlock new business opportunities and new reliability/ center-of-inertia and ROCOF information. The third applica-
security models and create an environment favorable to 4D tion, studied since the mid-1990s and demonstrated at a pilot
grids. Here are some hypotheses to consider: scale in the China Southern grid, damps interarea oscilla-
1) Can we capture a modest share of the safety margin (say, tions in ac–dc systems by using rotor angle shifts to modulate
5% of the total) while counting on, as a last resort, a DSE- generator excitation systems and modulate real and reactive
informed intelligent SIPS as an automatic safety net? power set points for high-voltage-dc and DER installations.

Given the high number of states in a real system, achieving a
DSE solution can be intractable if all the generators in a large
system are considered simultaneously.
Getting Ready for DSE Proof-of-Concept This practical experience validated, in our mind, the ben-
and Adoption efits of interfacing HIL simulation with EMS replicas by
Considering the tight coupling required between DSE using SE-based power system snapshots to enhance WACS
and EMSs at the submillisecond timescale, real-time simu- testing and validation. We concluded that a necessary prior
lation of EMS/C with a 4D grid model, including the IT stage in the development of 4D grid technologies leverag-
and telecommunication layers, is unavoidable. During the ing DSE was a simulation environment that could accurately
WACS project in Figure 2, which involved only three sub- represent the temporal dynamics of the real-world systems,
stations, IREQ’s Hypersim real-time simulator was used including IT and telecommunication layers. The vision was
extensively in conjunction with the IREQ PDC connected to integrate real-time simulation of electrical, telecoms, and
to the SPDCs through actual communication links. Moni- IT layers together while adding real automation control-
toring screens were installed side by side with simulator lers and protection for central and DER systems, replicas of
screens and a one-line diagram of the field experiment to EMS, and SCADA, all linked by sophisticated middleware.
facilitate systemwide debugging and to mimic dispatch- Figure 12 sketches IREQ’s cosimulation platform, initi-
driven actions from the EMS. The system under test ated in 2016. By including detailed representations of DERs,
was simultaneously being simulated to confirm the field microgrids, bulk grids, and the telecom networks connecting
results and explore more complex failure mechanisms under them together, a high-fidelity simulation platform becomes a
actual conditions. genuine “digital twin” of the smart grid. Credible offline and
Wide Area
D-WASA Brain
Data Uplink
Substation
Phasor Analytics SPDC Supervisory Phasor Analytics EMS/C
i 1 Wide Area Snapshot Model
Severity Data to
Dynamic Center-of- Decision From SE
C37.118-
I,V
States Inertia- Indices Making
Calibration Variables Transient
2014a F
v Engine SPDC Energy
Class P/M SPDC Pilot
... 2 Indices Analytics for
…. ROCOF
PDC Variables Data to
DSE …. Digital Automatic
Harmonic ….
Engine Std PMU SPDC Tracking Twin Operator
IED Phasors ROCOV ROCOF,
Data ... Modal
ROCOV Indices
Harmonics IED
Features Flags SPDC Wide Area DSE
N Data to Features D-WASA to
Action to Action Control Actions
SPDC
MF-MBPSS
Response-Based SIPSs,
DERs
WACSs, and Dispatcher Screens
DERMS/C Wide Area
Data Downlink
figure 10. From smart PMUs to D-WASA: in the era of the IoT, smarter synchrophasors and intelligent electronic
devices (IEDs) will accelerate the convergence of EMSs and WACSs to enable an autonomous EMS/C with fast feedback
control loops to mitigate uncertainties. Std: standard; DERMS/C: DERMS control.

DSE technology is rapidly gaining in maturity thanks to recent
progress in algorithms and enhancements of PMU devices and
communication networks.
real-time studies can then be performed of complicated sce- digital twin, masking the domain expert from the complex-
narios, such as cyberattacks targeting network assets (sub- ity inherent in a system of cyberphysical systems.
stations and power plants), to determine potential impacts on
power system performance and control and to evaluate risks Conclusions
to the reliability of electricity service to consumers. Because DSE technology is rapidly gaining in maturity thanks
The originality of this concept, which has at its kernel to recent progress in algorithms and enhancements of PMU
InSystemLab middleware, developed by E-Sim Solutions, is devices and communication networks, a wide variety of
to make it possible to study a system of systems by allowing applications in monitoring and control are emerging. While
each domain expert to analyze the behaviors in his or her the long-term vision is clearly to build a new kind of EMS,
own environment, with familiar tools specific to his or her EMS/C, that is more aware of systems dynamics and able to
field of expertise. However, in the background, the system is control the changing load–generation–frequency patterns
represented, without compromise, as a dynamically accurate autonomously in 4D grids, we remain very far from that goal,
table 1. The low-hanging fruits for DSE in a stability-constrained power grid:

the implementation of WACSs and response-based SIPSs.
Application Findings From Hydro-Québec Case Studies Grid Benefit
Voltage stability Modulating 11 compensators increased peak load transfer Deferral of investments in expensive
limits limits by 400 to 2,000 MW, depending on system conditions, dynamic shunt compensators near load
compared to only 400 MW when using purely local control. centers
Primary frequency This improved the eastern interconnection frequency (nadir) by Increased hosting capacity of renewable
response more than 10%–20% by modulating the load voltage and using resources through lessening the frequency
less than 1% of the reactive var resources. regulation constraints at low grid inertia
Oscillatory A wide-area PSS was five to 20 times more effective than a local Enhanced reliability at heavy loading
stability limits PSS in damping interarea modes, providing the same damping and more resiliency against multiple PSS
as two to three PSSs with the same reactive power range. outages
Local Substation
With MF-MBPSS Local
Remote Wide Area Information
Substation IEDs Information Processing
Processing
Local Local
Variables: Substation IEDs
Remote
Variables: Local Metrics
Global Metrics of Proximity to
of Proximity Instability
to Instability
Less than 200 ms
figure 11. The wide area response-based and closed-loop control applications to enhance the dynamic performance
of 4D grids.

DERMS
IEC 61850 Keysight Emulators,
EMS/SCADA PDC/SPDC
Devices Omicron Sources, and Control
LAN
Shared Memory v and Protection Devices
Hypersim
EMTP Targets PMU/C
PSS/E InSystemLab
Middleware
OPNet and OMNET+

Power System (Telecoms Simulation )
Analysis Tools Panning, Training
Analysis, and Testing
Data Access
and Streaming Human Simulink
Interfaces to Real Time
Unified Co-Simulation
Historian
Tools
Relay Functions
Relay
SEL-787 Relay Simulink
Servers and Protection Functions
Data Analysis:
DER Models
Osisoft PI
and so on
Electrical Automation/IEC Telecoms/
Experts 61850 Expert Security
Expert
figure 12. The HIL cosimulation platform developed at IREQ for proof-of-concept DSE-based applications in 4D grids.
LAN: local area network. EMTP: Electromagnetic Transient Program; PSS/E: Power System Simulator for Engineering.
and the path to get there is fuzzy. In the intermediate stages, we I. Kamwa et al., “Synchrophasors data analytics framework
are using the ability to obtain a real-time view, though imper- for power grid control and dynamic stability monitoring,” Eng.
fect, of the dynamic state of the system over a wide area for Technol. Reference, early access, May 2016, doi: 10.1049/
transitioning the old generation of topology-based WACSs and etr.2015.0049.
SIPSs to more adaptive stability controls. Our findings suggest C. F. Mugombozi et al., “Collaborative simulation of het-
that these response-based controls can assist in ensuring, eco- erogeneous components as a means toward a more comprehen-
nomically, the good reliability of future 4D grids characterized sive analysis of smart grids,” in Proc. 2019 7th IEEE Workshop
by a more volatile dynamic operational environment. Model. Simul. Cyber Phys. Energy Syst. (MSCPES), pp. 1–6,
doi: 10.1109/MSCPES.2019.8738794.
Acknowledgment D. Rimorov et al., “Dynamic performance improvement
This work was performed, in part, at IREQ, where the author of New York state power grid with multi-functional multi-
spent more than 32 years as a research engineer. The support band power system stabiliser-based wide-area control,” IET
of the Natural Sciences and Engineering Research Council, Gener., Transmiss. Distrib., vol. 11, no. 18, pp. 4537–4545,
through an Alliance grant for the “development and proto- Dec. 2017, doi: 10.1049/iet-gtd.2017.0288.
typing of decentralized and carbon-free electrical systems,” “Power system dynamic state and parameter estimation-
is gratefully acknowledged. transition to power electronics-dominated clean energy sys-
tems,” IEEE Power & Energy Society, Piscataway, NJ, USA,
For Further Reading Tech. Rep. PES-TR88, Jul. 2021.
I. Kamwa et al., “Wide-area monitoring and control at T. E. Dy-Liacco, “Enhancing power system security con-
Hydro-Quebec: Past, present and future,” in Proc. 2006 trol,” IEEE Comput. Appl. Power, vol. 10, no. 3, pp. 38–41,
IEEE Power Eng. Soc. General Meeting, Montréal, QC, Jul. 1997, doi: 10.1109/67.595291.
Canada, pp. 1–12, doi: 10.1109/PES.2006.1709097.
C. Basu et al., “Situational awareness for the electrical Biography
power grid,” IBM J. Res. Develop., vol. 60, no. 1, pp. 10:1– Innocent Kamwa is with Laval University, Québec City, QC
10:11, Jan./Feb. 2016, doi: 10.1147/JRD.2015.2498818. G1V0A6 Canada. p&e

Inverter-Dominated
Power Systems
Dynamic
Estimation-
LOV
S US
REY
/ AND
K.C OM
© SH U TT E RSTOC
Based Protection
and Hidden
Failure Detection
and Identification
By Sakis Meliopoulos , George J. Cokkinides,
Paul Myrda, Evangelos Farantatos, Ramadan Elmoudi,
Bruce Fardanesh, George Stefopoulos,
Clifton Black, and Patrick Panciatici

january/february 2023 1540-7977/22©2022IEEE IEEE power & energy magazine 59

L
LEGACY PROTECTION AND CONTROL (P&C) SYSTEMS capability of the system will not be discussed in this article
are vulnerable to misoperations due to a variety of rea- due to space limitations.
sons: miscoordination, complexity, human error, and hid- The rCSP technology applies DSE at the protective
den failures. Adding to these, the characteristics of the relay level at rates of 40 times per cycle, as well as at the
power system are changing due to the addition of inverter- substation level at rates of once per cycle. It simplifies the
interfaced generation. Inverters have different character- P&C functions by eliminating the need for coordination,
istics than synchronous machines when responding to eliminating relay misoperations, facilitating the detection
faults or disturbances, and the logic of commonly applied of cyberattacks, and enabling self-healing against hidden
P&C systems may fail to operate as expected. Some recent failures and cyberattacks.
relay misoperations have been attributed to inverter-inter-
faced resources. Overall Description
Advanced state estimation methods form the building block
Introduction for addressing the issues mentioned previously. A DSE-
Present P&C systems, which are based on numerical, mul- based protection approach (also known as, setting-less pro-
tifunctional, multidimensional, communications-enabled tection) has been introduced that does not require coordina-
relays are complex protection systems requiring coordinated tion with other protection functions. The DSE-based method
settings. North American Electric Reliability Corporation supervises and continuously assesses the correctness of the
statistics and other sources around the globe indicate about data going into the relays. If the data are corrupted for any
10% protection misoperations; where many misoperations reason, i.e., instrumentation channel failures, hidden fail-
result in major system disturbances, making them the num- ures, or cyberattacks, the system identifies the root cause
ber one cause of grid unreliability. Present approaches lack and identifies the compromised data and replaces them
effectiveness in dealing with the root causes discussed with estimated data. DSE enables immediate detection of
previously. As an example, detection of hidden failures such anomalies. Identification of the root cause and of the
is limited and when a hidden failure is detected, affected compromised data are performed via a series of hypothesis
relaying functions are inhibited and protection is entrusted testing. Subsequently, the developed system takes automatic
to back up protective relays. If hidden failures occur simulta- corrective action to replace corrupted data with estimated
neously with faults, no known detection method exists. values, thus allowing the P&C system to operate normally;
Dynamic state estimation (DSE) is introduced as a and at the same time sends messages to operators for repairs.
new approach to P&C with many advantages: 1) it reduces The proposed technology, rCSP, mitigates the main root
the complexity of P&C by eliminating the need for coor- causes of relay misoperations: 1) complexity and the need for
dination among protection functions, 2) it provides a coordination is removed by the use of the state estimation-
method to validate the measurements/inputs to the relay, based relay technology, and 2) it self-heals in the presence of
and 3) in case of anomalies, it identifies the root cause of hidden failures by identifying compromised data and replac-
anomalies (which can be faults, hidden failures, cyberat- ing them with estimated values. It also sends a message to
tacks, and others) via hypothesis testing. These features operators with the exact location of the hidden failure for
provide actionable information, which is used to drasti- repair. It is recognized that cyberattacks may compromise
cally increase the reliability of the protection system, as data and/or insert malicious commands; both may result in
described in this article. protection misoperations. The rCSP offers defenses against
The application of DSE for P&C has led to the develop- cyberattacks by leveraging the knowledge of the system state
ment of the resilient and secure centralized substation pro- to detect and identify data attacks and/or malicious com-
tection (rCSP). rCSP mitigates the main root causes of relay mands. In summary, rCSP eliminates the need for coordina-
misoperations with the following features: 1) complexity and tion among protection functions, facilitates the detection of
coordination is removed by the use of setting-less relay tech- cyberattacks, and enables self-healing against hidden fail-
nology, 2) it detects hidden failures and self-heals by identi- ures and cyberattacks, thus eliminating misoperations from
fying compromised data and replacing them with estimated most common root causes.
values, and 3) if necessary, it sends a message to opera- The rCSP is illustrated in Figure 1. It is a two-level rCSP
tors with the exact location of the hidden failure for repair. system with self-healing capability against abnormali-
Cyberattacks may compromise data and/or insert malicious ties: 1) level one provides reliable and fast protection for
commands; both may result in protection misoperations. The each protection zone of a substation/plant using the recently
technology offers defenses against cyberattacks by leverag- introduced DSE-based protection (also known as setting-less
ing the knowledge of the system state to detect and identify relay) operating on sample values, SV (left-hand side of Figure
data attacks and/or malicious commands. DSE provides the 1), and 2) level two ensures the validity and trustworthiness
full state of the system at fast rates (once per cycle) enabling of the data via a substation-wide DSE using phasor data (see
full state feedback optimal control at speeds not possible right-hand side of Figure 1). In case of abnormalities, the
before for a geographically dispersed power system. This substation-level centralized system detects the abnormality

immediately. Subsequently, an identification process is initi- the physical substation. Merging units (MUs) are installed
ated to determine the root cause of the abnormality (hidden in the substation switchyard. They are shown symbolically
failures, cyberattacks) via hypothesis testing. Once the root on the left-hand side of Figure 1 as “MU.” The MUs stream
cause is known, corrupted data are replaced with corrected sample values to the process bus. There the sample values
data so that the setting-less relays will operate with cor- are ordered by time tag and accumulated in circular buf-
rected data. This action eliminates relay misoperations due fers. This process is performed by the sample value data
to corrupted data. At the same time, operators are alerted concentrator (SVDC), to be discussed later in the section
to the cause of the problems so that appropriate repairs can “Process and Station Bus Management.” The protection
be implemented. In summary, the continuous tracking of functions are implemented in a multicore computer. Each
the system state enables data validation and, most impor- protection function runs in a dedicated core of the computer
tantly, detection of hidden failures. Typical hidden failures and receives data from the process bus; only the data needed
are blown fuses, incorrect settings, incorrect current transfor the specific protection zone. This block is designated as
former or voltage transformer ratios (human error), faults in “Zone 1 Setting-Less Relay,” “Zone 2 Setting-Less Relay,”
the instrumentation, and so on. Finally, self-healing is selec- and so on. If the system is free from internal failures (hid-
tively applied to remove the abnormalities. den failures, communication failures, instrumentation chan-
The recommended implementation of the system is shown nel shorts, blown fuses) the setting-less relays will operate
in Figure 1. Note the upper left corner of Figure 1 shows normally and will perform as designed to protect the system.
ebCSP
Convert Time Domain Model into Dynamic Phasor Model
Zone 1 Zone 2 Zone N
AQCF Model AQCF Model AQCF Model
Convert sample Values into Phasors
Zone 1 Zone 2 Zone N

Measurement Measurement Measurement
Definitions Definitions Definitions
Substation Substation
AQFC Model Measurement Model
MU Zone 1 rCSP
Setting-Less
Relay DSE
Process Bus (SVDC)
Station Bus (PDC)
MU (Detection Process)
Zone 2
Setting-Less Phasors
Relay Hypothesis Analysis
MU
(Identification Process)
Zone N Compromised Data Correction

Setting-Less
Relay
MU
Cyber
Validate Intrusion
Real Time Model
Optimal Substation Detection

Subsystem
Control Model
Real To Control System
Time and State Center
Control
From
Other
Substations
figure 1. General organization of resilient and rCSP. PDC: phasor data concentrator; SVDC: sample value data
concentrator.

The rCSP offers defenses against cyberattacks by leveraging the
knowledge of the system state to detect and identify data attacks
and/or malicious commands.
Specifically, if there is a fault, only the setting-less relay of actual power system fault, rCSP does not take any action and
the faulted protection zone will operate and disconnect the allows the relays to perform their function(s). In any other
faulted protection zone. Considering the possibility of fail- case, the compromised data are identified. Next, using the
ures in the P&C system, a supervising system is operating best estimate of the substation model, the best estimate of the
in parallel and automatically. This system is shown on the physical quantities corresponding to the compromised data
right-hand side of Figure 1. are computed and are fed back to the process bus replacing
The supervising system works as follows: All the data the compromised data. The replacement of the compromised
collected at the process bus are converted into phasors at a data with estimated data are shown in Figure 1 with the feed-
rate of one phasor per power frequency cycle. These pha- back line “estimated sample values.” The action results in
sors and the models of the protection zones are transmitted corrected data in the process bus and the subsequent correct
to the substation-wide dynamic state estimator. Note that 1) operation of the relays. The rCSP project set and achieved the
the models of all the protection zones of the substation are performance goal that the analytics be completed in less than
used to construct the substation model indicated as “substa- two fundamental frequency cycles. This means that relays
tion AQCF model” in Figure 1 (AQCF stands for algebraic will experience compromised data for a period of less than
quadratic companion form), and 2) the totality of phasors two cycles. Once the compromised data are replaced with
creates a high redundancy measurement set indicated as estimated (good) data, the relays will automatically reset.
“substation measurement model.” Redundancy is defined as This performance has been achieved. The constituent parts
the number of measurements divided by the number of states of the rCSP are described next in greater detail.
of the substation model. For a typical substation, the redun-
dancy is above 2,000%. The substation measurement model State EBP (Setting-Less Protection)
is used to run the substation-wide dynamic state estimator. The use of DSE to determine the health of power compo-
There are two possible outcomes of the substation-wide state nents has emerged as an approach that is immune to the new
estimator (standard estimator output): 1) all measurements characteristics of the power system. First, setting-less protec-
are consistent with the substation model; in this case the sys- tion does not require coordination with any other protection
tem does not need to do anything else, or 2) the measure- functions, resulting in great simplification of P&C schemes.
ments do not fit the model of the substation well. In this case, Second, it is immune to fault current levels and fault current
the state estimator detected an abnormality; further analysis components (negative, zero sequence), enabling successful
is needed to determine the root cause of the abnormality. protection in systems where inverter-based generation has
For this purpose, a series of hypothesis analysis is initiated. drastically altered the fault current levels and characteristics.
The following hypotheses are tested: 1) an anomaly is an It is emphasized, setting-less protection is immune to sup-
actual fault in the power system, 2) a hidden failure or a fail- pressed or inconsistent negative and zero sequence quanti-
ure in the instrumentation, or 3) a cyberattack that altered ties during faults due to inverter controls. Note that many
data (identification process). The procedure is described legacy protection functions depend on the presence of nega-
in a subsequent paragraph in greater detail. In case of an tive and zero sequence fault quantities consistent with linear
circuit behavior for their logic.
The DSE-based protection is referred to as estimation-
I2 T, ω based protection (EBP). It is important to note that present
I1 I3 I, V Temp day technology of MUs seeing increased application has
made the application of EBP seamless. The basic idea of the
EBP has been inspired by the Kirchhoff’s current law (KCL)
Protection Zone Protection Zone
differential protection function (no coordination needed),
Differential Function DSE Based Protection but it is much different than conventional differential pro-
Internal tection, as depicted in Figure 2.
Measurements
In differential protection, the electric currents at all ter-
KCL Only All Physical Laws minals of a protection zone are measured, and their weighted
sum must be equal to zero (generalized KCL). As long as
figure 2. The EBP or setting-less relay concept. the sum is zero or near zero, no action is taken. In EBP, all

It is emphasized, setting-less protection is immune to
suppressed or inconsistent negative and zero sequence
quantities during faults due to inverter controls.
existing measurements in the protection zone are utilized. of magnitude higher than the first-order integration methods
The measurements may include all or part of: 1) currents and that are typically used. The increased accuracy of the inte-
voltages at the terminals of the protection zone, 2) speed, gration method results in better performance of the DSE.
acceleration, torque, temperature, in rotating machinery, 3) The DSE is executed using sample values (point on wave-
voltages, and currents at other locations inside the protection form or time domain samples). The relatively new technol-
zone (as in capacitor protection), 4) temperatures in case of ogy of MUs has introduced two standard sampling rates: 1)
oil-filled transformers or any other internal measurements, 80 samples per cycle—recommended for protection, and 2)
and other. Assuming that all the quantities of interest are 256 samples per cycle—recommended for power quality.
measured, then one can monitor whether all the physical For protection and assuming a 50-Hz system, this means
laws, such as KCL, Kirchhoff’s voltage law, laws of motion, that the sampling period is 250 μs. Since a relay must operate
laws of thermodynamics, and others, are satisfied by the in real time, this means that the analytics of the DSE must be
power apparatus under protection. If all the physical laws are completed within 250 μs. In our case, using quadratic inte-
satisfied, this indicates that the component is healthy, and gration, the DSE is executed once per two successive sets of
no action is required. When an internal fault or abnormality samples. In this case, the analytics of the state estimation
occurs, one or more physical laws are violated triggering the must be completed within 500 μs. This time reduces to
detection of the anomaly or fault. It is important to note that 416 μs for 60-Hz systems. Today this is possible, assuming
the EBP replaces all the known protection functions with that the code has been optimized and a high-end micropro-
one single relay that protects the protection zone against all cessor is used. As an example, Figure 3 provides a snapshot
the types of faults covered by all other protective relays (dis- of the execution of the DSE and shows that the analytics are
tance, overcurrent, directional, and so on). performed within 25% of the available time.
A systematic way to monitor the satisfaction of all physical It is also important to note that depending on the topol-
laws is provided by the dynamic state estimator. Specifically, ogy of the protection zone, in addition to the actual mea-
one can develop the detailed dynamic model of the component surements, virtual, derived, and pseudomeasurements can
that expresses all the physical laws obeyed by the component. be added to augment the measurement set and improve the
Then each measurement can be expressed as a function of the performance of the DSE. In legacy P&C systems it is cus-
state of the component defined in the dynamical model, which tomary to use primary and backup protection systems for
in general may include algebraic and differential terms. The increased reliability. The same practice can be applied to the
formulation of the DSE simply needs the mathematical model ESB. It is recommended that redundant parallel systems on
for each measurement as a function of the state. separate computing platforms be used for reliability.
The DSE is formulated as an optimization problem
whose solution determines the best estimate of the state that Instrumentation Channel Model
will minimize an objective function defined as the distance An important issue for the performance of the ESB is the accu-
between the measurements and the mathematical model racy of the data. Specifically, MUs are, in general, higher preci-
expressing the physical quantity associated with the mea- sion equipment as compared to values communicated through
surements. The optimization problem can be solved with typical supervisory control and data acquisition (SCADA)
a number of well-tested methods, such as 1) constrained systems or protective relays. MUs provide measurements
optimization, 2) least squares approach, 3) least absolute that are time synchronized with precision better than 1 μs and
value, 4) minimax approach, and 5) any variation of Kalman magnitude accuracy better than 0.1% of the input waveform.
filtering. It is important to note that the formulation of the This performance is not achieved in an actual field installation
problem includes differential terms. Kalman filtering uses because MUs receive inputs from instrument transformers,
first-order integration methods to perform the solution itera- over control cables, and other subsystems, which introduce
tions, typically through a predictor/corrector algorithm. To errors that are much greater than the accuracy of a typical MU.
improve the performance of the solution method we have These errors are referred to as instrumentation channel error.
introduced a second-order integration method (quadratic The end result is that “raw” data are laced with errors.
integration), which converts the algebraic/differential opti- Conceptually, the overall precision issue can be resolved
mization problem into an optimization problem in terms of with sophisticated calibration methods. Calibration pro-
an algebraic model. The accuracy of this model is two orders cedures are quite expensive and face difficult technical

problems. An alternative approach is to utilize appropriate This distinction between local and remote is necessary
error correction methods for the purpose of correcting the because the remote data (telemetered data) have substan-
sample values, assuming that the characteristics of the vari- tial higher time latencies. The validity of their time tag is
ous GPS-synchronized pieces of equipment are known and assessed, and their physical location (i.e., the physical quan-
the instrumentation feeding this equipment is also known. tity they represent) is objectified into a mathematical model
A viable and practical approach is the utilization of DSE so that it can be interpreted correctly by upstream users
to correct errors from instrumentation. Specifically, the (setting-less relays, conversion to phasors, substation state
models of the MUs and the associated instrumentation chan- estimation, and so on). The overall organization of the pro-
nel model are all integrated into a single model that provides cess bus and the SVDC makes the utilization of the sample
the transfer function from the primary side to the output values by the various protection functions (relays) and other
of MUs. Using this model, the error correction problem is applications much simpler, as shown in Figure 4.
formulated as a DSE problem in which one of the states is The first level of upstream functions is the protection
the primary quantity. DSE provides the best estimate of the function [designated as “EBP (Time Domain”)]. In this
primary quantity. Numerical experiments have verified that function, the user defines any number of protection func-
the accuracy of the best estimate of the primary quantity is tions as required. The number of protection functions
comparable to the accuracy of the MUs. depends on how many protection zones exist in the substa-
tion. For each protection zone, the user must define what
Process and Station Bus Management measurements are used by the particular protection function
The number of sample values collected by the MUs in a sub- by simply selecting from the available measurements at the
station is very large. In a typical substation that will be sev- process bus. The user interface is via a simple graphical user
eral tens of millions of sample values per second. Recently, interface. Subsequently, each EBP automatically generates
telemetered sample values are also available from remote virtual, derived, and pseudomeasurements to supplement
MUs. To manage this data, we have developed the SVDC that the actual measurements from the process bus. The math-
is part of the process bus. The SVDC time-aligns the sample ematical model of each measurement is also automatically
values as they are received from various MUs with variable generated. The automation has been achieved by adopting
latency. The sample values (measurements) are stored in cir- an object-oriented approach for each element of the process,
cular buffers as shown in Figure 4. Note that there are two i.e., device model, measurement model, and so on. The pro-
sets of redundant buffers, one for the sample values from the cess is facilitated with a user interface that makes the selec-
local MUs and another set for the data from remote MUs. tion of the measurements for each protection zone as simple
Sampled Values From MU at a Sampling Rate of 4,000 s/s
250 µs 250 µs 250 µs 250 µs

t–h t+h t + 2h t + 3h t + 4h
375 µs 375 µs
125 µs 125 µs 125 µs
(Idle Time) (Idle Time)
500 µs 500 µs
DSE Execution DSE Execution DSE Execution

at Time t at Time t + 2h at Time t + 4h
DSE Execution Once per 2h (or 500 µs)
Computation Time %
0 20 40 60 80 100
125 µs
Computation Time = × 100 = 25 %
500 µs
figure 3. The execution times of an EBP relay.

Using this model, the error correction problem is
formulated as a DSE problem in which one of the
states is the primary quantity.
as clicking on a list. The second level of upstream functions uses all the measurement data from the relays converted into
is the substation-wide state estimation, which is discussed in phasors. This is an automated process.
the next section. The information flow from the EBP relays to the sub-
station-wide state estimator is illustrated in Figure 4, see
Protection Supervision Via Substation the block on the right of the figure designated as “SQDSC
Level State Estimation (phasors).” SQDSC stands for substation quadratic device
Numerical and EBP relays, as well as electromechanical and state concentrator. Note the substation-wide DSE is
relays, are vulnerable to any false input data, independently characterized with high redundancy and computes the best
of the source of errors, i.e., faults in instrumentation chan- estimate of the substation dynamic state. Subsequently, the
nels, failures in instrumentation circuits, cyberattacks, hidden goodness of fit between the measurements and the substation
failures, and others. It follows that faulty input data to relays is model is computed via well-known analytics in state estima-
detrimental and undermines protection reliability. Methods to tors. The goodness of fit index expresses the probability that
detect faulty input data are essential for reliable P&C systems. the differences between the measurement and the substation
High-fidelity state estimation provides a systematic method to model are within the metering error. This probability is also
supervise the input data to relays in a substation to secure their referred to as confidence level. A high probability of good-
operation against faulty input data. In the case of detected ness of fit, i.e., more than 0.90 results in declaring all the
faulty data, state estimation can provide corrected data by uti- data in the substation valid. In this case, no further action
lizing the real-time model of the system and the best estimate is required. Otherwise, existence of faulty data is declared.
of the system state. For this purpose, the substation-wide DSE Subsequently, the identification process starts via hypothesis
is constructed from all the models of the protection zones and testing to determine the root cause of the faulty data. Once
Process Bus 2 WinXFM 2

EBP (Time Domain) WinIGS 1 SQDSC (Phasors) WinIGS 2
Redundant SVDC PDC
Process Bus 1 WinXFM 1

Local Measurements User Interface Substation Level State Estimation
Measurement Values ∗ Define Logical Node Automatically Copy all
in Circular Buffers ∗ Select Measurements Measurement Definitions and
(Local + Remote) Models into SQDSC
Domain
Time
Measurement Automatic Generation of

Definitions Automatic Generation of: Additional:
• Virtual Measurements • Virtual Measurements
Phasors
• Derived Measurements • Derived Measurements

• Pseudo-Measurements • Pseudo-Measurements
Remote Measurements
Measurement Values
(Latency) in Circular Buffers Measurement Measurement Measurement Measurement
Definitions Models Definitions Models
Domain
Time
Measurement
Definitions
Phasors
LAN Sockets
Actual Measurements Only
figure 4. Process bus organization, the SVDC, and relationship to upstream functions. LAN: local area network.

the root cause and faulty data has been identified, the esti- representing the voltages of the three phases and the neu-
mated state and real-time model is used to create estimates tral. The state is extended to include internal states of all
of the faulty data and correct/replace the faulty data. devices, for example, the torque angle and the rotor speed
The substation-wide DSE is formulated in the dynamic of a generator or the magnetic flux linkage of a transformer.
phasor domain. Specifically, the mechanical and elec- Furthermore, the system state also includes the states of the
tromagnetic dynamics are included while the electrical buses at the other ends of the lines/circuits connected to the
dynamics are neglected. The electrical quantities such as substation. The conceptual illustration of the dynamic state
voltage and current are represented in phasor form. The estimator and the state definition is shown in Figure 5. The
substation model is a three-phase, breaker-oriented, and best estimate of the state variables is obtained by utilizing
instrumentation-inclusive model. The available data at the measurements and corresponding models only at the local
substation are numerous and provide high redundancy, substation to avoid telemetering requirements.
defined as the number of measurements over the state esti-
mates to be computed. The overall approach allows sharp Measurement Set, Quadratization, Quadratic
faulty data detection, identification, and rejection, as well Integration, and Measurement Model
as alarm analysis and root cause identification. The advan- The measurement set for the substation DSE comprises 1)
tage comes from the fact that the substation state estimator GPS-synchronized voltage and current phasor measure-
problem is much smaller in size (as compared to a central- ments and 2) traditional, nonsynchronized measurements
ized state estimator) and, therefore, powerful hypothesis (voltage magnitude, active and reactive line flows, and
testing methods are applied for both faulty data and topol- bus injections, and other standa rd SCA DA data). In
ogy errors identification without substantial deterioration addition to the actual measurements the substation DSE
of the computational efficiency. In the following sections, measurement set is augmented with a number of virtual,
the definition of the states, the mathematical model of the derived, and pseudomeasurements.
devices, the generic model of the measurements, and the Each measurement is expressed as a function of the sub-
solution algorithm are provided. station states. For this purpose, each power system compo-
nent is modeled as a physically based, three-phase dynamic
Definition of States model that is neutral/ground inclusive. Components include
The state of the DSE is defined as the minimum informa- transmission lines, transformers, generators, and others. Any
tion that completely describes the operating conditions of measurement on any power system component uses this
the substation. DSE uses dynamic voltage phasor states as model to derive the mathematical model of the measurement
well as other quantities such as speed of the generator rotor, as a function of the state variables. This approach results in a
magnetic flux linkages, mechanical power, and other infor- high-fidelity representation of any measurement in the system
mation to represent the operating conditions of the system. and the ability to estimate the dynamic and transient behavior
Specifically, all states are functions of time, and they sat- of the power system. The DSE formulation includes the fol-
isfy the model equations, which may be a combination of lowing steps:
algebraic and differential equations. The voltage states for ✔ All the measurements are expressed as an object
every three-phase bus are defined with a set of four elements (mathematical expression of a measurement as a func-
tion of the state of the substa-
tion). This object is a nonlin-
V, δ ear function with algebraic
Substation of Interest and differential terms.
1
(DSE Application) ne
L i ✔ The object is converted into a
n
dωg1 s io nonlinear function with non-
δg1, ωg1, is
dt V, δ V, δ m linearities not greater than qua-
s
an
Tr dratic (for example, squares of
variables) by introduction of ad-
yg1
ditional state variables if neces-
Tr sary. Subsequently, a numerical
an
sm
iss integration method is applied
io to convert each measurement
n
yg2 Li
ne model into a companion alge-
dωg2 V, δ 2
δg2, ωg2,
dt braic form. The quadratic in-
tegration method is used and
V, δ
specifically the Lobatto three
co-location point method. The
figure 5. Substation DSE-definition of states. end result is that the object of

each measurement is in a companion algebraic form of the DSE in the form of a graph of the variable k versus
with nonlinearities not greater than quadratic. the confidence level (probability). This is the equivalent of
✔ Complete the formulation of the DSE problem by se- providing the expected error (which equals the variable k
lecting an objective function. The weighted sum of the times the standard deviation of the meter accuracy) versus
residuals squared is a typical selection. This process probability (confidence level). Note that at 80% confidence
makes the substation DSE a constrained optimization level the value of k is 0.6. This means that the state estimate
problem with a quadratic objective function and con- is on average 60% more accurate than the accuracy of the
straints with non-linearities not higher than quadratic. raw measurements.
The state estimation problem is then solved with well-
known methods. Hypothesis Testing
When DSE detects an anomaly via the chi-square test, it is
Performance Metrics necessary to determine the root cause of the anomaly. Three
The accuracy of the state estimate is characterized using possibilities exist:
the standard chi-square test and the statistical properties
of the state estimator. The chi-square test provides the
State Estimation Quality
probability that the distribution of the residuals, defined as
1k
the differences between the measurements and the model
values, are within the expected bounds. Because there are
Parameter k
100
many data acquisition devices in any substation with dif-
ferent degrees of accuracy, a normalization constant k is 10
introduced. The variable k is defined as follows: If it is 1.0,
then the standard deviation of each measurement error is 1
equal to the accuracy of the meter with which this measure-
ment was obtained. If different than 1.0, then the standard 100 m
deviation of the measurement error equals the accuracy of 0 20 40 60 80 100
Confidence Level
the meter times k. The introduction of the variable k allows
the characterization of the accuracy of the state estimate figure 6. A chi-square test in terms of normalization
with only one variable. Figure 6 provides a typical report variable k versus confidence level.
IRIG-B
GPS Clock Process Bus

Process Bus
Computer
16:03:17 Ethernet Switch 1
Dual Optical Fibers DFRs Ethernet

(SV + PTP)
MUs
USI 9000 USI 9000
Station Bus
Ethernet
Computer
Analog Voltage and Current Signals
Omicron Amplifiers Ethernet

Ethernet Switch 2
Station Bus
Power System
Simulation
Computer
Ethernet
D/A Converter Bank
figure 7. Laboratory equipment configuration. SV: sample value; PTP: precision time protocol.

✔ One or more power faults
exist in the system
Protection Zone 4 ✔ One or more protection fail-
Transmission Line Protection Zone 5
ures exist (hidden failures,
TLBUS1 TLBUSFUTU cyber-attack, instrumenta-
A,B
C
,C
tion failure, and so on)
B,
LB Future
A,
LB0
MU2
MU3 ✔ Simultaneous occurrence
AN
,BN
,CN of faults and protection fail-
ures exist.
BUS115kV BUS115KV Protection Initially, one should consider
LB1
Zone 2 the following possible events:
2TXBUS1 ✔ occurrence of a protection
N,CN
AN,B ,C failure
Protection Zone 1
A,B
✔ occurrence of power fault(s)
A,
MU5
B ,C
∆
1
MU4 in a protection zone

A
Transformer AN,B,C ✔ simultaneous occurrence
2
Y
,BN
C
,CN
B,
of a faulted protection zone

A,
and a protection failure.

2TYBUS1 Hypothesis analysis starts with
these three hypotheses, tested in
YB1
sequence; if none is tested posi-

tive, then other possible events
AN,BN,CN
BUS13KV AN,BN,C must be considered. The hypoth-
N
MU9 MU6
esis testing is guided by the fol-
FB1
BUS13kV
FB2
lowing observations:
,C
A,
A,B
B,
MU8 MU7 ✔ At the substation level, the

C
C A,B,C
A,B,
Protection Zone 3 redundancy is high (over
FDRBUS1 FDRBUS2 2000%). This implies that
Distribution Line 1 Distribution Line 2 the possibility of leveraging
points is remote and there-
figure 8. A single line diagram of the test system running used to test the rCSP in the fore normalized residuals
laboratory with HIL from the state estimator
can be used as guidance of
where the problem is lo-
cated. Leveraging points are
measurements that are bad,
but they show a very small
residual, making their detec-
tion much more complex.
✔ The system is continuously
running. This implies that
any abnormal event will be
captured in real time. Note
that the probability of si-
multaneous failure events
occurring at exactly the
same time is low. For this
reason, the hypotheses are
initially limited to first-or-
der (single) events.
The mechanics of hypothesis
(a) (b) testing are as follows: First, sus-
pect measurements are identified
figure 9. An EBP relay of MSU1 line user interface. ESSD: estimated state from normalized residuals that
standard deviation. are provided by the DSE. Second,

suspect measurements are grouped using the source of the system. For all other root causes, the operator is alerted to
data (MUs) and also by protection zone. Third, protection initiate repairs. At the same time, automatic self-healing
zones are classified into “possibly faulted” or “not-faulted” can be performed, as described next.
depending on the EBP relay report. An expert type of sys-
tem generates hypotheses to be tested using the aforemen- Data Self-Healing
tioned classification of the suspect measurements. Subse- Anomalies other than power faults, i.e., hidden failures,
quently, the hypothesis testing is performed sequentially. cyberattacks, and others, can be repaired by software until a
The hypothesis testing terminates when a hypothesis pro- technician provides repairs and/or disinfection of the com-
vides an affirmative result (it is tested positive). If hypoth- promised cyber asset. Specifically, these anomalies result
esis testing identifies the root cause as a power fault, the in some data being compromised. For each compromised
relays are allowed to perform their task of protecting the datum identified, the real-time model of the substation and
MUs
Siemens 7SC805
Alstom/Reason
MU
(a)
Massena Reactor Currents MSU1 Line Currents Marcy Reactor Currents MSU1 Line Currents
AA
SVData
SVData
IRIG-B IRIG-B
Power
Power
Voltages Voltages
Massena MU Marcy MU
(b)
figure 10. The Installation of rCSP at (a) Southern Company and (b) NYPA.

the best state estimate of the substation is used to compute buses via optical fibers (multimodal 1300 nm). Monitoring
what this datum should be. The real-time model and the best and configuration of the MUs is handled via the process bus.
estimate of the state is provided by the substation-wide DSE Digital fault recorders (DFRs) with processor management
in the phasor domain. The phasor data are converted into time unit capability (shown as USI9000) and numerical relays (not
domain data by using the basic correspondence between pha- shown) are connected to the substation bus. The system has
sors and time domain waveforms. The time domain wave- been designed to interface with MUs as well as with numerical
forms are used to compute the estimated samples at the same relays, DFRs and other intelligent electronic devices. We refer
sampling rate as the actual MUs. Subsequently, the estimated to this system as the hybrid P&C system. A multicore personal
sample values are streamed into the process bus, replacing computer (with Windows 10 operating system) accesses the
the compromised data. The process enables all applications process and substation buses, respectively. Another computer
that depend on the sampled data in the process bus, i.e., zone (power system simulation computer) drives a bank of digital-
protection, to operate on corrected and validated data. to-analog (D/A) converters providing voltage and current sig-
nals to the inputs of the amplifiers; the output of the amplifiers
Laboratory Testing With Hardware in the Loop is connected to MUs, DFRs, and relays. Note that the voltages
The overall approach of using DSE to perform protection and and currents are amplified to match MUs, relays, and DFRs
supervise the correctness of input data to relays has been orga- analog input range via Omicron amplifiers.
nized into the prototype rCSP platform. The system has been The laboratory setup allows the testing of all the func-
tested in the laboratory with hardware in the loop and has also tions of the rCSP system in the comfort of the laboratory
been installed in several substations at partner utilities (three but with a system that duplicates the P&C system of a sub-
substations of the Southern Company system and two substa- station. Figure 8 shows a diagram of an example substation
tions of the New York Power Authority (NYPA)). Prior to the implemented in the laboratory. Figure 9 shows a snapshot
installations the system has been “factory tested” at the Geor- of the operation of the rCSP. Specifically, Figure 9 shows
gia Tech laboratory, with the configuration shown in Figure 7. the user interface of one EBP relay for a transmission line.
The laboratory configuration consists of a process and sub- It is important to note all EBP relays are running simulta-
station bus implemented using managed Ethernet switches. A neously. Space limitations prohibit showing snapshots of
set of 13 MUs are connected to both process and substation all EBP relays running simultaneously.
90 90
120 60 120 60
150 30 150 30
Vc
Vc
Ic
Ia Ib
180 0 180 0
Va Ia
Va
Vb
Vb
Ib
Ic
210 330 210 330
240 300 240 300
15 Sept., 2021–14:50:00.014619 15 Sept., 2021–14:50:00.014619
V2_A_MASSENA 440.3 kV –15.54° I3_A_MASSENA 1,222.3 A –10.53°

V2_A_MARCY 439.1 kV –28.2° I3_A_MARCY 1,228.6 A 154.71°
V2_B_MASSENA 434.3 kV –137.55° I3_B_MASSENA 1,272 A –133.44°
V2_B_MARCY 439 kV –148.67° I3_B_MARCY 1,248.3 A 30.56°
V2_C_MASSENA 431.6 kV 103.4° I3_C_MASSENA 1,228.4 A 104.3°
V2_C_MARCY 437.3 kV 91.52° I3_C_MARCY 1,173.9 A –90.43°
figure 11. A snapshot of data reported by the EBP relay for a 765-kV, 133-mile line.

january/february 2023
(a) (b)
figure 12. Analytics of EBP relay of MSU1 line: modified data simulating a fault. (a) The computations of the EBP relay during a 0.1-s interval. (b) The results during the
simulated fault.
IEEE power & energy magazine

71
Typical performance metrics from the laboratory test- system with high penetration of inverter-based resources.
ing are as follows: Each of the EBP relays executes a DSE Second, substation-wide state estimation enables validation
once every 416 μs (60-Hz system). The maximum execution of the measurement data streaming into relays, thus ensuring
time of each DSE for any one of the five protection zones of the reliability of the P&C system. Third, the rCSP system
this substation is 45 μs. The substation-wide state estimation is capable of detecting hidden failures, cyberattacks on the
executes once per 60-Hz cycle and takes less than 7.5 ms to data, and other disturbances while at the same time utilizing
execute. This performance enables real time operation with estimated quantities, enabling the continuous and reliable
plenty of margin. operation of the P&C system.
Field Installations and Experience For Further Reading

As mentioned, the rCSP system has been installed at three Protection System Misoperation Task Force, “Misoperations
Southern Company substations and two NYPA substations. report,” NERC Planning Committee, Atlanta, GA, USA,
Part of the installation for one of the Southern Company sub- 2014.
stations and the installation of one MU at the Massena substa- S. Meliopoulos, G. Cokkinides, J. Xie, and Y. Kong,
tion at NYPA are shown in Figure 10(a) and (b), respectively. “Instrumentation error correction within merging units,” in
The NYPA installation of the EBP relays has been operational Proc. Georgia Tech Fault Disturbance Anal. Conf., Atlanta,
since July 2021. A visualization of some of the data from one GA, USA, Apr. 30/May 1, 2018, pp. 1–7.
protection zone, the MSU1 line (a 765 kV, 133-mile-long R. K. Hunt, “Hidden failure in protective relays: Supervi-
line), is shown in Figure 11. The user interface and the per- sion and control,” M.S. thesis, Virginia Tech, Blacksburg,
formance report of the relay are shown in Figure 9. The per- VA, USA, 1998.
formance report for the DSE shown in Figure 6 is taken from D. Pal, B. Mallikarjuna, M. J. B. Reddy, and D. Mohanta,
this system. As reported in the discussion of Figure 6, the state “Analysis and modeling of protection system hidden failures
estimate is on average 60% more accurate than the accuracy and its impact on power system cascading events,” J. Con-
of the raw measurements. No fault has occurred in the system trol, Automat. Elect. Syst., vol. 30, no. 2, pp. 277–291, Apr.
during the short period of the field installations, and as a result 2019, doi: 10.1007/s40313-019-00444-3.
we do not have actual fault data to report. Z. Jiao, H. Gong, and Y. Wang, “A D-S evidence
Additional detailed computations of the EBP relay are theory-based relay protection system hidden failures de-
shown in Figure 12. Specifically, Figure 12(a) shows the tection method in smart grid,” IEEE Trans. Smart Grid,
computations of the EBP relay during a 0.1-s interval. The vol. 9, no. 3, pp. 2118–2126, May 2018, doi: 10.1109/
top two traces show the voltages and currents at both ends TSG.2016.2607318.
of the line. The third trace shows the chi-square value (very H. F. Albinali and A. S. Meliopoulos, “Resilient protec-
small value), and the fourth trace shows the confidence level tion system through centralized substation protection,” IEEE
that the data fit the model of the protection zone (the values Trans. Power Del., vol. 33, no. 3, pp. 1418–1427, Jun. 2018,
are practically 100%). The bottom trace shows the execution doi: 10.1109/TPWRD.2017.2789318.
time of the DSE. Note it is below 40 μs. Because there were
no faults at the time of the writing of this article, the data Biographies
were altered to simulate a fault. The results during the simu- Sakis Meliopoulos is with Georgia Institute of Technology,
lated fault are shown in Figure 12(b). Note that the bottom Atlanta, GA 30332 USA.
three traces show the chi-square, confidence level, and trip George J. Cokkinides is with the Georgia Institute of
decision, respectively. Technology, Atlanta, GA 30332 USA.
Paul Myrda is with the Electric Power Research Insti-
Conclusions tute, Chicago, IL 60126 USA.
The rapidly changing scenario of the electric power grid Evangelos Farantatos is with the Electric Power Re-
into an inverter-dominated dynamical system incorporat- search Institute, Los Angeles, CA 90280 USA.
ing smart grid technologies necessitates faster tracking of Ramadan Elmoudi is with New York Power Authority,
power system operations, faster analytics, fast response Buffalo, NY 14225 USA.
times, protection immune to the changing characteristics of Bruce Fardanesh is with New York Power Authority,
fault currents, and protection against P&C system failures White Plains, NY 10601 USA.
and cyberattacks. The application of DSE plays a critical George Stefopoulos is with the Solar Energy Technolo-
role toward meeting the new requirements. This article has gies Office, Washington, DC 20003 USA.
described the rCSP system as an example of an intelligent Clifton Black is with Southern Company, Birmingham,
approach toward practical solutions to meet the new chal- AL 35203 USA.
lenges. The rCSP system is possible with today’s technology Patrick Panciatici is with Réseau de Transport
and offers many advantages. First, it provides an approach d’Electricité (RTE), 92073 Paris, France.
p&e
to P&C which is immune to the characteristics of a power

The Spanish
Experience
D
DYNAMIC LINE RATING (DLR), ALSO KNOWN IN
some instances as real-time thermal rating, is becoming Squeezing Line
a key component of modern energy management systems
for transmission system operators (TSOs) or independent Ampacities Through
system operators (ISOs) to use transmission assets more
efficiently while maintaining current reliability stan- Dynamic Line Rating
dards. However, the need for this tool will be even more
overwhelming in a decarbonized future, given the fore-
seeable line congestion originating from the coincidence
of several factors: 1) significant growth of the share of
intermittent renewables; 2) the further electrification of the the impact of those actions, TSOs/ISOs are focused on mak-
residential, transportation, and industrial energy demand; ing better use of the available infrastructure, which calls for
and 3) increasing difficulties and costs for the construction more sophisticated tools capable of permanently monitoring
of new transmission assets. the thermal status of transmission lines and transformers.
Network congestion may force system operators to The urgency for better utilization of the existing assets has
implement some preventive or corrective measures, such as prompted specific actions from regulators. For example, the
generation rescheduling, with undesirable economic conse- U.S. Federal Energy Regulatory Commission is evaluating
quences for the system as a whole. Therefore, to minimize the economic benefits and challenges of applying DLR com-
pared to the traditional static line rating (SLR).
Digital Object Identifier 10.1109/MPE.2022.3219167 Broadly speaking, the capacity of a transmission system can
Date of current version: 30 December 2022 be directly associated with the transfer capacity of its individual
By José A. Rosendo-Macías , Antonio Gómez-Expósito ,

Alfonso Bachiller-Soler , Miguel Á. González-Cagigal ,
Gabriel Álvarez-Cordero, Lucía Mateo-Sánchez,
and Antonio Useros-García
IMAGE LICENSED BY INGRAM PUBLISHING
january/february 2023 1540-7977/22©2022IEEE IEEE power & energy magazine 73

The U.S. Federal Energy Regulatory Commission is evaluating
the economic benefits and challenges of applying DLR compared
to the traditional static line rating.
components, whether they are underground lines (UGLs) for OHLs, measurement sampling rates are typically low
or overhead lines (OHLs). In the simplest case, the operat- when compared to the thermal time constant. Therefore,
ing procedures adopted by TSO/ISOs assume that the rated in this case, steady-state conditions are assumed for the
capacity of transmission lines is determined by a single SLR. DLR calculation. Finally, the conclusions derived from the
Also common is the use of several SLRs throughout the year, obtained results and suggested future lines of research related
normally chosen on a seasonal basis. For UGLs, the Inter- to DLR are outlined.
national Electrotechnical Commission (IEC) standard IEC-
60287-3-1-2017 includes a list with the ambient temperature Underground Transmission Lines
and soil properties to be used for the calculation of these ther- In the particular case of Spain, the risk of steady-state ther-
mal ratings in several countries. As these procedures do not mal overload for UGLs is quite low under normal operating
consider the state of the system and weather conditions at any conditions. This low risk is due to the relatively low average
given time, over the last decades, researchers have explored typical load (below 30% of the rated capacity), which leaves
more innovative approaches, collectively known as DLR, spare capacity for N-1 outages, especially when facing peak
aimed at obtaining the thermal capacity of transmission assets load conditions.
in real time to squeeze out their real capacity. However, apply-
ing DLR involves the installation of new and costly monitor- DLR for UGLs
ing equipment, along with the development of suitable data The rated ampacity of UGLs is usually a conservative value
analytics algorithms. Therefore, several research lines are because it does not take into account the high thermal inertia of
currently focused on minimizing the required investment the materials involved. Indeed, a buried conductor subjected to
without jeopardizing the safe system operation or the integ- a load step can take years to reach the maximum steady-state
rity of assets. operating temperature. Among the factors affecting both the
A key concept related to UGLs is the current carrying steady-state and transient thermal behavior of UGLs, the fol-
capacity of the conductor (also known as the ampacity). This lowing stand out: the burial depth, ambient temperature, and
concept is defined in IEC 60287 as the maximum permanent soil thermal properties. Note that the latter two are time vary-
current that asymptotically heats the conductor insulation to its ing, which should be considered by the DLR methodology.
maximum temperature. For example, for low-density degassed Several strategies are used to duly consider the thermal
XLPE-insulated cables, temperatures of more than 90 °C might inertia of UGLs for the estimation of their ampacity for a
jeopardize the service life of the cable. The ampacity concept few hours, days, or even months ahead. In the case of cyclic
can also be used for OHLs. In this case, the maximum steady- loads, a first strategy consists of calculating a so-called
state operating temperature is established to prevent metallur- “increased ampacity,” either using a load factor or a cyclic
gical damage or violation of critical distances. In the case of rating factor, as described in IEC 60853. This standard is
Europe, temperatures up to 90 °C are adopted for aluminum- dedicated to the calculation of the cyclic and emergency cur-
conductor steel-reinforced conductors. rent rating of cables.
This work presents the recent Spanish experience with the A second strategy can only be applied if soil parameters
application of DLR for underground and overhead ac trans- are known or assumed (from actual measurements or con-
mission lines. For each type of asset, brief descriptions of the sidering the same parameters used to calculate the conven-
methodology adopted and of the main difficulties encoun- tional ampacity). The following load scenarios are consid-
tered in the practical implementation are first provided. Next, ered, where different information related to the conductor
practical experience gained in the pilot projects undertaken temperature can be numerically calculated:
by the Spanish TSO Red Eléctrica de España (REE) is pre- ✔ For a load step, it is possible to calculate the time it
sented, showing that the proposed techniques improve the takes for the cable to reach the steady-state operating
utilization of existing assets as compared to traditional solu- temperature from its previous value. This time constant
tions. In the UGL case, dynamic state estimation (DSE) is can be used when analyzing network contingencies.
used to obtain an accurate model of the temperature evolu- ✔ A representative cyclic load curve can be designed to make
tion, where the involved parameters are simultaneously esti- the cable reach its maximum operating temperature. This
mated. Although similar methods could be used, in theory, value can be used as the base ampacity for further studies.

✔ It is possible to combine the two previous scenarios so causing the heating and the ambient temperature. The
that a balance is achieved between the ampacity increase model also depends on the geometry of the cable and the
and the operational security. soil thermal properties (the resistivity and capacity). In the
✔ The temperature of the conductor can be updated based applications we discuss here, much more complex ladder-
on real-time current measurements. Given the slow evo- type models have been considered, including the effects of
lution of the temperature, this provides a reaction mar- convective cooling for cables with protective pipes and the
gin of hours. mutual heating for three-phase configurations. In practice,
A third strategy, more accurate and flexible, consists the innermost insulator temperature, which can be approx-
of monitoring the cable temperature in real time by using imated by the conductor temperature, is the limiting factor
a distributed temperature sensing (DTS) system based on for ampacity.
optical fiber (OF), which is embedded in the sheath or the To assess the accuracy of this analytic model, the results
jacket of the cable. The measurements provided by this obtained for the conductor temperature were compared with
device, updated every 5 min, include the average tem- those provided by the FEM-based simulation. This compari-
perature for every meter of the OF. Both spatial and time son was conducted under different load scenarios and chang-
steps are tunable. These measurements allow underground ing values for the ambient temperature, soil thermal resistivity,
cables to operate according to the actual temperature evo- and volumetric thermal capacity. The maximum estimation
lution. The operating principle of the DTS system is based error was around 0.8 °C, as shown in Figure 2 for an example
on Raman or Brillouin effects, which estimate the conduc- scenario covering one week.
tor temperature through changes in the frequency bands Hence, for a given current profile, the ladder-type model
produced by light scattering. has proven to be sufficiently accurate for calculating the
An alternative for temperature monitoring in the case evolution of the temperature in each section of the cable,
of thermally limiting and accessible sections of cables allowing a simpler way to assess if the maximum operating
without OF is the use of a temperature probe in the jacket temperature is exceeded.
of those accessible cable sections, which are considered
limiting in terms of the overall ampacity. This strategy has Parameter Estimation
been applied by REE on more than 20 220-kV and three A potential limitation of the ladder-type discrete model is
66-kV UGLs. that it involves a set of parameters that are not necessarily
known. To address this issue, a DSE model together with a
DTS-Based Methodology for DLR nonlinear Kalman filter (KF) has been used for joint state
To properly use the DTS measurements for the DLR calcula- and parameter estimation. In the DLR application, the DSE
tion, a thermal model of the cable is required. In this context, model can be used for the estimation of the temperatures and
several models have been developed for underground cables the unknown parameters in the presence of measurement and
using the finite-element method (FEM), which can be model noises. For this purpose, an augmented state vector is
numerically solved using simulation tools, such as COM- defined, including the state variables of the model and the
SOL (Figure 1). whole set of parameters to be estimated. This way, the KF
However, these simulations
are not appropriate for real-time
applications due to the calcula-
tion time required and the diffi- –0.75
cult integration into the utility’s –0.8 60
–0.85
software. To overcome this prob- –0.9
lem, simpler models can be used –0.95 50
–1
to describe the thermal dynamics –1.05
y-Position (m)
Temperature (°C)
of the cable. Particularly suitable –1.1

–1.15 40
for this purpose is the ladder- –1.2
–1.25
type discrete dynamic model of –1.3 30
the cable and the surrounding –1.35
–1.4
soil, leading to the well-known –1.45 20
state–space representation char- –1.5
–1.55
acterizing linear systems. The –1.6 10
state vector includes the cable –1.65
temperatures (the conductor –0.6 –0.4 –0.2 0 0.2 0.4 0.6
and sheath) and those of several x-Position (m)
concentric slices of the soil. The
input is composed of the currents figure 1. An example of a COMSOL thermal simulation.

Several research lines are currently focused on minimizing
the required investment without jeopardizing the safe system
operation or the integrity of assets.
process updates the whole state vector with each new sample section of the underground cable and to check whether
provided by the DTS. the maximum operating temperature is exceeded.
The overall process of the DTS-based method, including The accuracy of the KF-based technique has been evalu-
the estimation of parameters, is illustrated in the flowchart ated in a set of simulated scenarios, where an FEM-based
of Figure 3, and summarized here: tool is run with different parameter values to emulate the
✔ Initial estimations of the state variables and parameters DTS measurements included in the KF algorithm. The max-
are introduced in the KF algorithm, along with the cor- imum estimation error for each parameter can be taken as a
responding covariance of the initial estimation error. metric of the overall method performance.
✔ Then, the DTS measurements are processed by the
KF to obtain estimates of the parameters involved Field Application
in the model. The DTS technology presented previously has been used
✔ Finally, given the expected current profile, the esti- in a pilot project in the Balearic Islands, Spain. The DTS
mated model is used to predict the temperature of each information is complemented with the real-time estimation
55 0.8
Analytic
COMSOL 0.6
Error in Conductor Temperature (°C)
50
Conductor Temperature (°C)
0.4
45
0.2
40
0
35
–0.2
30
–0.4
25 –0.6
20 –0.8
0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168
Time (h) Time (h)
(a) (b)
figure 2. The conductor temperature with (a) a ladder-type model and FEM and (b) the resulting error.
DTS Measurements Expected Current Profile
Estimated
State and Predicted
Initial Estimation and Parameters Ladder-Type State
Overheating
Its Error Covariance KF Algorithm Estimated
Check
Model
figure 3. A flowchart of the DTS-based method using KF.

The rated ampacity of UGLs is usually a conservative value
because it does not take into account the high thermal inertia
of the materials involved.
of three parameters external to the cable, leading to a more Figure 6 shows a sample with the spatial distribution of
accurate prediction of the conductor temperature for future the real-time estimation of the conductor temperature
load curves. This more sophisticated estimation technique using the DTS measurements.
has been tested on three operative UGLs, namely, two 220-kV So far, the proposed technique has been tested on sev-
circuits in Barcelona, a 66-kV short cable in Mallorca, and eral 132- and 220-kV UGLs. Future research efforts will
a 132-kV circuit in the Mallorca–Ibiza interconnection. The be oriented to fully assess the DSE-based estimation tech-
goal was to make 48-h predictions of conductor temperature nique with higher conductor currents and different layouts
with an error of less than 3 °C. of cables, such as those including galleries and overhead-to-
For the current profile presented in Figure 4(a), the underground transmission towers.
sheath temperature measurements provided by the DTS are In summary, the main innovations currently being inves-
those shown in Figure 4(b), which is the only information tigated in Spain are the simultaneous, real-time estimation
used for the parameter estimation carried out by the KF- of the external parameters arising in the thermal cable model
based DSE. considered and the improvement in the operation of UGLs
The time evolution of the estimates for the three param- by means of DLR.
eters to be estimated is represented in Figure 5(a)–(c). Notice
that the estimated values remain within an acceptable range, Overhead Transmission Lines
except for the ambient temperature Ta. Finally, the DSE-based This section presents the application of DLR techniques to
model was used to predict the sheath temperature for the next OHLs, paying special attention to distinguishing features
48-h period, which can be compared with the actual DTS that are not found in the case of UGLs.
measurements to test the accuracy of the model. The differ-
ence between both values (estimation error) is represented in DLR for Overhead Transmission Lines
Figure 5(d), where it can be observed that the resulting errors In Spain, owing to the relatively short length of OHLs, rarely
remain below 1.4 °C in all cases. This confirms the suitabil- exceeding 200 km, the main restriction in the transfer capac-
ity of both the ladder-type discrete model and the KF-based ity is related to their thermal limit, i.e., to the conductor tem-
DSE for DLR of UGLs. perature at a given time, which needs to be compared to the
A pilot implementation of the DLR methodology maximum operating temperature of the corresponding cir-
described previously has been thoroughly tested. To give cuit. Conductor overheating can be the cause of the follow-
an idea of the information the user can access in real time, ing two undesirable phenomena:
450 35
400
DTS Temperature (°C)
350 30
300
Current (A)
250
25
200
150
20
100
50
0 15
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Time (h) ×104 Time (h) ×104
(a) (b)
figure 4. The (a) current profile considered and (b) DTS measurements in the presented application.

✔ An excessive increase in the maximum sag of any Several techniques make it possible to increase the maxi-
line section may result in violations of the mandatory mum operating current for a given line, such as increasing
minimum electrical clearances. the height of the towers, substituting the conductors with
✔ There may be deterioration of the conductor by an- a different type of material with higher operating tem-
nealing of the aluminum wires, eventually reducing peratures (for some options, more than 100 °C), or adding
the breaking load of the material. more conductors per phase. However, all of these strategies
The operational impact of these phenomena depends largely require circuit disconnection for relatively prolonged peri-
on the type and properties of the conductor. ods and the corresponding administrative authorization. The
2.5 1.4
1.2
2
cv (MJ/K . cm3)
ρs (m.K/W)
1.5
0.8
1
0.6
0.5
0.4
0 0.2
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Time (h) ×104 Time (h) ×104
(a) (b)
70 1.4
60 1.2
Prediction Error (°C)
50 1
Ta (°C)
40 0.8
30 0.6
20 0.4
10 0.2
0 0.5 1 1.5 2 2.5 3 0 6 12 18 24 30 36 42 48
Time (h) ×104 Time (h)
(c) (d)
figure 5. (a)–(c) The estimated parameters using Kalman filtering and (d) the sheath temperature prediction error in the
48-h period.
28
26
Estimated Conductor
Temperature (°C)
24
22
20
18
16
14
0 1,010 2,020 3,030 4,040 5,050 6,060 7,070 8,080 9,090

Length (m)
figure 6. The real-time temperature estimation throughout the conductor length.

real-time application of DLR to
OHLs can largely overcome these
Current Through Maximum
problems since it does not require the Line Temperature
line disconnections.
The following classification Real-Time
Measurement: Conductor Equivalent
can be made regarding the differ- Temperature, Tension, Ampacity
Temperature Wind Speed
ent methods used to approach the Sag, etc.
DLR, depending on the informa- Ambient
tion considered: Temperature
✔ Weather data: The meteoro-
logical conditions (tempera- Solar
ture as well as wind speed Radiation
and direction) are obtained
from weather stations, al-
figure 7. A flowchart of the CTE-based methods.
lowing the maximum cur-
rent to be calculated given
a limiting conductor temperature. This method, along study. These magnitudes are generally very different in terms
with field results, is described in the next sections. In the of time scales (minutes and hours) and dependence on posi-
context of weather data, the conductor temperature can tion and time, which makes the update rate of external condi-
be updated with real-time current measurements, as in tions a key factor in the DLR of OHLs.
the case of UGLs, increasing the reaction margin in The weather information can be used either to calculate
the operation of the monitored asset. This strategy is the conductor temperature, given the value of the line cur-
currently being applied in northern Spain to a subtrans- rent, or to obtain the current that makes the conductor reach
mission corridor connecting wind farms to the grid. a certain temperature, with the latter being directly related
✔ Conductor temperature evaluation (CTE): This method to the calculation of ampacity in OHLs.
is motivated by the high variability of both wind If the sampling rate of the measurements provided by the
speed and wind direction along the OHL. To overcome installed weather stations were sufficiently high, then the
this problem, the equivalent conductor temperature is thermal differential equation could be the basis for the appli-
estimated online using different measurements of the cation of DSE to dynamically estimate the evolution of con-
monitored asset, such as the direct monitoring of tem- ductor temperature, much like in the UGL case. However, in
perature, conductor voltage, span sag measurement, the OHL case, the rated conductor temperature in the steady
etc. Once the equivalent temperature of the conduc- state is customarily the only factor considered for DLR calcu-
tor is estimated, the ambient temperature and the so- lations, allowing a static thermal balance equation to be used.
lar radiation are used to obtain an equivalent wind
speed, which is finally used to calculate the actual
DLR. The overall process of this DLR methodology,
currently being investigated, is summarized in the
flowchart in Figure 7. Joule and Magnetic
✔ Distance monitoring: In this case, either the line Heating
sag or the clearance to ground are measured to pro-
Ps Pj
vide real-time information on the thermal state of Solar
the OHL. Heating
Convective Cooling
DLR Calculation Based on Weather Data
In this case, the thermal equilibrium between the solar, Joule,
Pc
and magnetic heat sources, on the one hand, and the convective
and radiative cooling mechanisms, on the other, as described
by the corresponding IEEE standard 738-2012 or the Confer-
ence Internationale des Grands Reseaux Electriques (CIGRE)
Pr
guide for thermal rating calculations of OHLs (working group
B2.43), is illustrated in Figure 8.
Unlike the DSE-based technique presented earlier for Radiative Cooling
UGLs, in the case of OHLs, the weather conditions—namely,
the ambient temperature, solar irradiance, wind speed, and figure 8. A representation of the thermal equilibrium in
wind direction—play a critical role in the problem under OHLs (adapted from CIGRE guide).

To assess the accuracy of this analytic model, the results obtained
for the conductor temperature were compared with those provided
by the FEM-based simulation.
In this work, the CIGRE guide is used to calculate the dif- since the ambient temperature and the solar radiation
ferent terms involved in the static thermal equation. The terms are less dependent on position.
comprise not only meteorological variables but also a series of ✔ Local monitoring: In this case, the weather stations are in-
parameters related to the mechanical (the material, diameter, stalled on one or more towers of the OHL (see Figure 9),
number of wires, etc.) and thermal (the emissivity, absorptivity, and the DLR is only calculated for the respective seg-
etc.) properties of the conductor. In this regard, several studies ments. As in the previous case, the overall ampacity of the
have been carried out to determine which of these variables has circuit is approximated by the minimum of these values.
the greatest influence on the calculated DLR value. Experimen- Although more accurate results of weather conditions are
tal results obtained in these studies show that wind speed and provided by local stations, there is still a possibility that
wind direction have the greatest effect on the calculation of the the critical segment (i.e., the one with the lowest ampac-
maximum operating current since the convective effect is pri- ity) is not monitored, leading to excessive DLR values.
marily responsible for the cooling of the conductor. To overcome this problem, a security coefficient ranging
However, despite its relevance, locally characterizing between zero and one, which can be obtained by using
the wind based on weather stations is challenging. In this historical data, can be applied to the resulting DLR value.
regard, two different locations for the weather stations
can be considered to obtain the meteorological condi- Field Application
tions, as follows: The Spanish regulation includes recommended ampacity
✔ Remote measurements: For each section of the OHL, values for different types of conductors and OHL configura-
weather information is taken from the nearest meteo- tions, allowing the use of alternative models, such as that
rological station, and the maximum operating current considered by the Spanish TSO to calculate the seasonal
is calculated for that particular section. Then, the DLR ampacities from historical data. With this model, the par-
for the whole circuit is obtained as the minimum of ticular conditions of each season are taken into account to
these values. In this approach, the distance between obtain the maximum current (e.g., the winter ampacity is
each line segment and the corresponding weather sta- higher than in the rest of the year since air temperature and
tion may cause an excessive error in the local meteo- irradiation are typically lower during this season).
rological variables, particularly the wind speed and Although these seasonal values improve the overall utilization
direction. In case of doubt, a conservative value of the of the existing assets, there has been a growing interest related to
DLR can be obtained by taking a reduced value of the the incorporation of DLR into the operation of the network. As
wind speed, such as the commonly accepted 0.6 m/s, a result, several research lines have emerged to develop a more
or even considering natural convection exclusively, accurate methodology for the calculation of the ampacity of OHLs
using meteorological data from weather stations without compro-
mising the safe operation of the corresponding lines.
The first case study refers to a 220-kV OHL located in
Zaragoza. The goal was to compare several methods for
line monitoring to obtain the conductor representative or
equivalent temperature. The methods considered were using
the information provided by local and remote weather sta-
tions, DTS, inclinometers and lidar devices. For this pur-
pose, a set of those devices was installed along the circuit,
providing the corresponding local measurements every
10 min. The current through the line was also considered to
calculate the conductor temperature from the thermal equi-
librium equation. As an example of the results obtained in
this comparative study, Figure 10 represents the conductor
equivalent temperature calculated using local weather sta-
tion data, along with that obtained using the geometrical
figure 9. A weather station installed on a tower. information provided by a lidar device. It can be observed

that both values are quite similar during the whole period
considered, which demonstrates the accuracy of the infor- 100
Weather Station
mation provided by the weather station. Lidar
Conductor Temperature (°C)

To show the potential advantages of the DLR calculated 80
from weather station measurements (received every 10 min)
with respect to the seasonal ampacities customarily used, a 60
case study corresponding to a 220-kV OHL located in La
Rioja is presented. 40
The increase in ampacity provided by the DLR is illus-
trated in Figure 11, where a load–duration curve is separately
20
represented for each season. It is observed that the DLR cal-
culated from local weather station data for this particular
circuit is higher in at least 97% of the cases, even after apply- 0
6 a.m. 9 a.m. 12 p.m. 3 p.m. 6 p.m. 9 p.m.
ing the security coefficients accounting for partial monitor- (h)
ing. These results are encouraging and demonstrate that the
application of DLR techniques allows better utilization of figure 10. A comparison of the conductor temperature
transmission assets compared to seasonal ampacities. calculated using local weather station and lidar methods.
1,200
1,200
1,000
1,000
800
800
DLR (MVA)
DLR (MVA)
600
600
400 400
200 200
0 0
0 1,000 2,000 3,000 4,000 0 2,000 4,000 6,000
Cases Cases
(a) (b)
1,000
1,000
800
800
DLR (MVA)
DLR (MVA)
600 600
400 400
200 200
0 0
0 2,000 4,000 6,000 8,000 0 2,000 4,000 6,000 8,000
Cases Cases
(c) (d)
Weather Station Seasonal
figure 11. The load–duration curves for each season: (a) winter, (b) spring, (c) summer, and (d) autumn.

circuit through the thermal equilibrium equation. In an attempt
to improve the accuracy of the estimated conductor tempera-
ture when data from remote weather stations are used, further
tests have been carried out combining meteorological data with
information collected by other local sensors, such as inclinom-
eters. Both methodologies have been validated by comparing
the results with those calculated on the basis of information
provided by a lidar device. For the tested circuits, field results
show that the ampacity provided by the proposed methodology
exceeds the seasonal values customarily used by the Spanish
TSO most of the time (up to 97% in an exceptional case).
The experience gained from these pilot projects has
proven essential in the progressive evolution of DLR-related
figure 12. An inclinometer installed on a phase conduc- know-how. In the case of OHLs, which are by far the most
tor. (Source: Relogable; used with permission.) common, one of the main lessons learned lies in the key role
of monitoring resources (the latency, availability, accuracy,
As mentioned before, alternative DLR schemes are also location, communication coverage, etc.) for a proper calcula-
being currently tested, including CTE. Particularly, con- tion of the conductor ampacity. In light of this preliminary
sidering the results presented before for the 220-kV OHL experience, rating forecasts are being explored, where the
in Zaragoza, the Spanish TSO is committed to adopting a monitored meteorological variables are replaced by pre-
double monitoring system, using weather station data jointly dicted values hours or even days ahead. The application of
with measurements of the conductor sag in critical sections. rating forecasts might contribute to reducing the costs asso-
The latter are provided by locally installed sensors, such as ciated with the removal of technical constraints in power
the one shown in Figure 12. The variation of the maximum systems. Another lesson learned is that the potential benefit
sag, compared to that calculated in reference conditions, can of using DLR must be analyzed on a case-by-case basis.
be directly associated with the temperature of the conductor.
When combined with the measurements from the weather For Further Reading
station, this obtains the maximum operating current. “Guide for thermal rating calculations of overhead lines,”
CIGRE, Paris, France, CIGRE WG B2.43, 2014.
Conclusions A. Gómez-Expósito et al., “City-friendly smart network
This article presents the recent experience of the Spanish technologies and infrastructures: The Spanish experience,”
TSO in the application of DLR, for both underground trans- Proc. IEEE, vol. 106, no. 4, pp. 626–660, Apr. 2018, doi:
mission cables and OHLs. 10.1109/JPROC.2018.2793461.
First, a method for the calculation of the maximum oper- M. W. Davis, “A new thermal rating approach: The real
ating current in UGLs based on dynamic models describing time thermal rating system for strategic overhead conductor
the temperature evolution in different sections of the cable transmission lines—Part I: General description and justifi-
was discussed. The method considers a ladder-type discrete cation of the real time thermal rating system,” IEEE Trans.
thermal model, which has proven to be accurate enough in Power App. Syst.* (through 1985), vol. 96, no. 3, pp. 803–
simulated scenarios when compared to the results provided 809, May 1977, doi: 10.1109/T-PAS.1977.32393.
by FEM. The problems of parameter uncertainty and time
variability were also addressed, as some of the parameters Biographies
included in the thermal model may be unknown or inaccu- José A. Rosendo-Macías is with the University of Seville,
rate. For this purpose, a DSE based on the Kalman filter- Seville 41092, Spain.
ing approach has been adopted to jointly estimate the tem- Antonio Gómez-Expósito is with the University of Se-
peratures and parameters. The field results of a pilot project ville, Seville 41092, Spain. He is a Life Fellow of IEEE.
show that the prediction errors of the DSE-based model are Alfonso Bachiller-Soler is with the University of Seville,
less than 1.5 °C in all cases. Future research in this context Seville 41092, Spain.
is oriented to applying this same technique to more complex Miguel Á. González-Cagigal is with the University of
circuit configurations and other voltage levels. Seville, Seville 41092, Spain.
Regarding OHLs, different possibilities for calculating Gabriel Álvarez-Cordero is with Redinter (Redeia), Alc-
the DLR are currently being studied and tested. Among these obendas, Madrid 28108, Spain.
techniques, the one based on real-time information provided Lucía Mateo-Sánchez is with Red Eléctrica, (Redeia),
by weather stations has so far been the most successful, pro- Alcobendas, Madrid 28109, Spain.
vided the stations are located closely enough to the critical Antonio Useros-García is with Red Eléctrica, (Redeia),
circuit sections. The approach calculates the ampacity of the Alcobendas, Madrid 28108, Spain. p&e

book review
Wind Energy Handbook

of value for education and reference
T
THIS ISSUE’S “BOOK REVIEW” relevant issues such as turbulence and
column discusses Wind Energy Hand- extreme winds.
book, Third Edition, written by Tony The key area of aerodynamics is
Burton, Nick Jenkins, Ervin Bossanyi, covered in detail with two large sec-
David Sharpe, and Michael Graham. tions. The mechanical design load sec-
The reviewer writes, “Exposing electri- tion is also thorough. Wind turbine de-
cal engineers to the entirety of wind tur- sign is first described on a conceptual
bine design might be a good idea—just level, ranging from rotor diameter; ma-
reading the introduction of each section chine rating; rotational speed; and the
and glancing through the content gives number of blades to teetering; power
a great overall picture of wind turbine control; fixed/two speed; variable slip
technology.” and variable speed operation; braking
systems; and drivetrain options. Down/
Wind Energy Handbook, upwind rotors, augmented flow, and
Third Edition multiple rotor structure turbines are
–By Tony Burton et al. discussed as well. The second design
Wind Energy Handbook, Third Edi- chapter goes through the 10 main com-
tion, is an extensive book covering ponents of wind turbines.
the theory of operation behind wind The controller section of the book
turbine technology. The first edition, explains techniques to design control
published 20 years ago, has had two algorithms for pitch, generator torque,
updates over 10-year intervals, reflect- (visiting professor at the University and yaw. Wake effects and wind farm
ing the evolutions of design rules and of Strathclyde); and Michael Graham control include active wake control
principal innovations in the technol- (professor of unsteady aerodynamics at methods as well as controls for grid
ogy. Offshore wind was added to the Imperial College). support (for example, curtailment and
second edition, and a section on wake The Wind Energy Handbook, as fast frequency control).
effects and wind farm control is added suggested by the title, provides compre- A brief section on wind power proj-
in this third edition as well as updates hensive coverage of wind turbine tech- ect development includes visual impact
across all chapters. The writers are a nology. The focus is on wind turbine and ecological assessments; noise; and
group of renowned senior profession- design. A vast established knowledge electromagnetic interference impact
als in wind power technology from the base is covered in the book, suitable on communication systems. The off-
United Kingdom: Tony Burton (wind for use by students (final year under- shore section covers wind resources;
energy consultant after retiring from graduate or postgraduate) and people design loads; and fixed and floating
DNV GL); Nick Jenkins (professor of involved in the design, manufacture, or support structures.
renewable energy at Cardiff Univer- operation of wind turbines. Wind energy integration with the
sity); Ervin Bossanyi (senior principal The introduction gives a com- power system has its own section, from
researcher at DNV GL); David Sharpe prehensive overview of how turbine wind power plant electrical systems
technology has evolved into modern to both distribution and transmission
wind turbines. The wind resource sec-
Date of current version: 30 December 2022 tion focuses on mechanical design- (continued on p. 86)

calendar
PES meetings
for more information, www.ieee-pes.org
T
THE IEEE P OW E R & E N E RGY May 2023 September 2023
Society’s (PES’s) website (http://www. IEEE International Conference on IEEE International Smart Cities
ieee-pes.org) features a meetings section, Electrical Machines and Drives Conference (ICS2 2023), 24–27
which includes calls for papers and (IEMDC 2023), 14–17 May, San Francisco, September, Bucharest, Romania,
additional information about each of CA, USA, contact Leila Parsa, leila@ contact George Cristian Lazaroiu,
the PES-sponsored meetings. Please ieee.org clazaroiu@yahoo
check the conference website for the
most current information. IEEE PES International Confer- October 2023
ence and Exposition (GT&D Tur- IEEE PES/IAS PowerAfrica Con-
January 2023 key), 22–25 May, Istanbul, Turkey, ference (PowerAfrica 2023), 18–20
IEEE PES Joint Technical Committee contact Omer Usta, [email protected], October, Marrakech, Morocco, con-
Meeting (JTCM 2023), 8–12 January, https://ieee-gtd.org/ tact Essaaid Mohammad, essaaidia@
Jacksonville, FL, USA, contact Dan Sabin, gmail.com, https://ieee-powerafrica.
[email protected], https://pestechnical.org/ June 2023 org/
IEEE Transportation Electrification
IEEE PES Innovative Smart Grid Conference and Expo (ITEC 2023), IEEE PES Innovative Smart Grid
Technologies (ISGT 2023), 16–19 21–23 June, Detroit, MI, USA, contact Technologies Europe (ISGT Eu-
January, Washington, DC, USA, contact Rebecca Krishnamurthy, rebecca.k@ rope 2023), 23–26 October, Grenoble,
Kathy Heilman, kathy.heilman@ieee. rna-associated.com, https://itec-conf.com France, contact Bertrand Raison, ber-
org, https://ieee-isgt.org/ [email protected],
IEEE PowerTech Serbia (PowerTech https://ieee-isgt-europe.org
March 2023 2023), 25–29 June, Belgrade, Serbia, contact
IEEE PES Innovative Smart Grid Predrag Stefanov, [email protected], IEEE PES Innovative Smart Grid
Technologies Middle East (ISGT https://attend.ieee.org/powertech-2023/ Technologies Conference Asia (ISGT
ME 2023), 12–15 March, Abu Dha- Asia 2023), 27–29 October, Wuhan,
bi, UAE, contact Ehab Fahmy El- July 2023 China, contact Shangzhi Pan, shangzhi.
Saadany, [email protected], IEEE PES General Meeting (GM [email protected]
https://ieee-isgt-me.org/ 2023), 16–20 July, Orlando, FL, USA,
contact Roseanne Jones, roseanne. For more information on additional
April 2023 [email protected], https://pes-gm.org/ technical committee meetings, webi-
IEEE PES Grid Edge Technologies nars, and events, please visit our IEEE
Conference and Expo (Grid Edge), August 2023 PES calendar: https://www.ieee-pes.org/
10–13 April, San Diego, CA, USA, IEEE Electric Ship Technologies meetings-and-conferences/conference
contact Kathy Heilman, kathy.heilman Symposium (ESTS 2023), 1–4 Au- -calendar.
@ieee.org, pes-gridedge.org gust, Arlington, VA, USA, contact
Julie Chalfant, [email protected],
p&e
Date of current version: 30 December 2022 https://ests21.mit.edu/

in my view (continued from p. 88)
intervention of the system operators to a opportunities for more automation surements via a model-based approach
more desirable operating state. For real- and improved operation and control and identify and reject bad data before
power/frequency control, this is becom- of power systems for increased reli- any data are transmitted to the control
ing more challenging due to reduced ability and efficiency. center. This local distributed SE will
inertia from inverter-based resources One might argue whether nimbler be essentially producing the “state” of
(IBRs), and guaranteeing system sta- EMS systems and faster more granular the substation in real time and has been
bility is going to become more difficult. SEs are really needed since we have demonstrated to provide a versatile and
Even with grid-forming inverters, due operated with this slower system for effective SE-based protection system.
to their very fast response and intermit- many years now. But with less inertia The validated data are then transmit-
tency, nimbler grid control systems will and faster responding IBRs, there is ted to the EMS at the control center via
be required. For reactive-power/volt- a need for more agile controls in the redundant dedicated fiber-optic links
age control, the ability of the advanced power grid. Currently, we rely on 2–6 s with low well-defined latencies. Other
IBRs in supplying fast reactive power of SCADA data and solve the SE in the high-bandwidth communication means
system can be exploited to actively con- minutes time frame. However, to enact of comparable low latencies should
trol the entire system in a more system- more advanced or perhaps automated also be considered for special situa-
wide and coordinated way. controls, this frequency of data acqui- tions and/or as a means for backup and
To truly advance power systems op- sition needs to be greatly increased. diversification.
erations and control, a paradigm shift In a 60-cycle system, 2–6 s will allow Some utilities are implementing
in the current EMS is needed—a next- huge gaps where the EMS is blind to self-owned and -operated fiber along
generation EMS. Since SE can be con- system events. Even though the faster all transmission lines for backbone
sidered to be the backbone of the EMS, speeds may be above the response ca- communications to primarily address
advances in SE will be most important pabilities of the system operators, our this need and enhance their protection
to achieving higher fidelity EMS. It is objective should be to move toward systems. Note that this local estimator
great that this issue of PEM is dedicat- much higher situational awareness and
ed to the topic. more automated and faster response to
The goal of the next-generation EMS system events in the future.
is to implement a grid control system To provide more accurate time-syn-
based on fully synchronized measure- chronized wide-area measurements at
ments that provides improved grid ob- the control center, the vision is to have a
servability and situational awareness GPS-synchronized RTU or a PMU with
as well as more automated real-time a considerably larger than the typical
control action or high-quality advice number of channels at each substation
and more flexibility to the operators of that collects voltage and current phasor
the grid. Such a system will help opti- measurements as well as breaker (and s Meets FAA Specifications!
Color – Size – Shape!
mize power flows, minimize production possibly motor-operated disconnect/ – International Orange
costs, and reduce congestion and will al- disconnect switch) status information. s Tested and approved by
low faster and more precise reactions to In the interim, and until there are many major power companies!
s Thousands still in service
system changes, among other benefits. more synchronized measurements, the after 40 years
In a next-generation EMS, the ex- combined use of both SCADA and s Universal attaching! Fits any
wire .1" to 1"!
isting substation remote terminal units PMU data can be contemplated to fill in s Installs in 5 minutes!
(RTUs) should be replaced with PMU- for the lack of complete synchronized s Withstands hail!
type devices producing both precisely data. Preferably, instead of concentrat- s No maintenance! Does not
slip, oscillate, chafe, cause
time-tagged phasor data and breaker ing a large number of measurements in electrolysis or harmonic
status information. Dedicated, re- one or a few GPS-synchronized IEDs, vibration.
s Ships in halves nested.
dundant, and minimum-latency com- one can utilize the synchronized data
9, 12, 20, 24, 30, and 36" balls
munications links from all (or at least streams out of digital fault recorders
& special sizes available
all above a certain voltage class) sub- and protective relays.
stations to the control center are key To create a hierarchy for a more Call now 573-796-3812
requirements. Advanced synchro- manageable system and at the same ext. 2001
phasor-based algorithms to estimate time implement local SE-based pro- Fax 573-796-3770
or directly solve for the state of the tection systems these measurements www.tanawiremarker.com
TANA WIRE MARKERS
power system in cycles rather than should comprise inputs to a local SE at
P.O. Box 370, California, MO 65018
minutes will provide breakthrough the substation level to validate all mea-
january/february 2023 634276_Tana.indd IEEE

1 power & energy magazine
23/03/13 10:3885
AM
may be able to provide an accurate come possible. A number of monitor- frequent repeated solutions or dynamic
breaker/disconnect status without a lot of ing applications and control algorithms SEs where the actual system dynamic
hardwiring of the binary breaker contact or methodologies may be deployed, state (as in the differential equations, in
information. A hybrid approach may which should utilize the SE outputs the control systems sense) is estimated.
also be taken for accurate identification as input data. A feed-forward model- The term state is used in two differ-
of system topology. Implementation of based approach is proposed here, as ent senses as it relates to SE: repeated
this local SE would also be greatly fa- an example, to continually drive the snapshot SE solutions versus instanta-
cilitated by substation power grid along a pre- neous time-function three-phase estima-
automation systems defined optimal trajec- tion at the synchronized sample level.
or even more by IEC With less tory. It is assumed that To develop superior power system
61850-based substation a validated accurate dy- operation and control tools and capa-
designs transferring inertia namic (rotor angle dy- bilities, advanced state estimation will
data and commands at
faster rates.
and faster namics only) model of
the power system exists
help pave the way for advancements
and movement toward more robust and
Data errors due to responding and can be utilized. The nearer-to-real-time analysis and deci-
communications prob- time-synchronized mea- sion-making tools for power systems
lems should be identi- IBRs, there surements are of great operation and control. The current trend
fied and corrupt data
rejected at this stage,
is a need for value in improving and
validating the system
in the deployment of synchronized
measurement devices and the required
assuming adequate re- more agile dynamic model as well. dedicated low-latency communication
dundancy in the mea- Ultimately, the goal of networks to transmit these measure-
surements. In addition, controls in the the controller will be to ments to control centers are most fitting
data reconstruction al-
gorithms can be used to
power grid. minimize the frequency
deviation from the nom-
and timely for more advanced state esti-
mation and the next generation of EMS.
“fill in” missing or cor- inal 60 Hz and maintain Distributed computing approaches, as
rupt data. Such data should be properly voltages and power flows within ac- mentioned previously, i.e., at the sub-
identified for subsequent utilization by ceptable bounds while keeping up with station level and then at the global level,
other applications. It is envisioned that load fluctuations and rejecting distur- are important in obtaining the system
with high-performance computing, SE bances such as line and generator trips. state and model at very fast rates. The
solution times in the order of several The most important application or EMS vendors have not been actively
power cycles can be ultimately achieved. the “killer application” for synchro- developing and offering more advanced
With the state of the entire grid phasor or PMU data is their use in the SE tools and the next-generation EMS.
known at this very fast rate, closer- local substation level and global con- It is time for the utilities, ISOs, and
to-real-time grid monitoring can be trol center SE to obtain the dynamic RTOs to demand superior SE and EMS
achieved, and closer-to-real-time com- state of the grid. A more dynamic state capabilities in light of upcoming grid
putations of control actions will be- in SE is achieved through either more operation and control challenges. p&e
book review (continued from p. 83)

grid connections; grid codes; and power sign; controls; transformers; power plant be a good idea—just reading the intro-
quality. Power system operation and re- internal grids; and grid connection and duction of each section and glancing
source adequacy are briefly described. protection. The offshore wind section through the content gives a great overall
Although the book is not focused on includes the specificities of submarine picture of wind turbine technology.
electrical engineering, it covers most cables. Exposing electrical engineers to –Hannele Holttinen
areas of electrical design: generator de- the entirety of wind turbine design might p&e

advertisers index
The Advertisers Index contained in this issue is compiled as a service to our readers and advertisers: The
publisher is not liable for errors or omissions although every effort is made to ensure its accuracy. Be sure
to let our advertisers know you found them through IEEE Power & Energy Magazine.
Company page# URL Phone

ASPEN, Inc. CVR4 www.aspeninc.com +1 650 347 3997
DIgSILENT GmbH 3 www.digsilent.de/
EasyPower LLC 13 www.EasyPower.com/demo
Marmon Utility LLC 5 www.marmonutility.com +1 603 673 2040
PowerSys/EMTP 8 www.emtp.com/renewable
Powertech Labs, Inc. 7 www.dsatools.com
PSCAD 11 pscad.com
RTDS Technologies, Inc. 9 www.rtds.com/gtsoc
SKM Systems Analysis, Inc. CVR2 www.skm.com +1 800 500 4SKM
Tana Wire Markers 85 www.tanawiremarker.com +1 573 796 3812
445 Hoes Lane, Piscataway, NJ 08854
IEEE power & energy magazine representative

Beverly Grimme
Naylor Association Solutions
direct: +1 352 333 3367, cell: +1 904 881 0862, fax: +1 703 790 9199
[email protected]

in my view
Bruce Fardanesh
advances in state
estimation
finding a killer application
E
ELECTRICAL POWER SYSTEMS’ tools are collectively referred to as an some of the key challenges that aug-
operation and control functions rely on energy management system (EMS). In ment the need for reforming the way
analytical tools for setting safe and se- the online analysis, the collected asyn- the electricity grid is controlled and
cure operating limits and for maintain- chronous telemetry data are provided operated. Luckily, with the acceptance
ing reliable service. These tools utilize to a tool referred to as a state estima- and proliferation of new smart sensors
mathematical models of the power tor (SE). and other intelligent electronic devices
system components and equipment in The SE, as the backbone of the EMS, (IEDs), such as phasor measure-
the form of differential and algebraic estimates the state (voltage magnitude ment units (PMUs), and the availability
equations. These mathematical models and phase angle at all of the necessary mod-
are formulated and coded in the form the buses) of the power er n com munications
of computer programs used to deter- system as well as the re-
Today, power infrastructure to trans-
mine the power system operating state sulting power flow val- systems are mit such high-quality
or its response to various changes and ues using conventional system measurements
disturbances, such as generation and best-fit procedures and typically to control centers, it is
load variations; switching operations; iterative Newton-based very timely that the in-
and faults. The results of these analy- solutions. The SE out-
analyzed dustry starts to develop
ses are heavily relied upon for the op- put is also used to deter- in both an and adopt superior tools
eration and control of power systems mine system quantities and techniques for nim-
and for managing and monitoring that are not directly “offline” and bler operation and con-
their performance. measured or measur- trol of power systems.
Today, power systems are typically able. Other functions,
an “online” In an article pub-
analyzed in both an “offline” and an such as contingency (near real-time) lished in 2002 in IEEE
“online” (near real-time) manner. Of- ranking and security Computer Applications
fline analysis tools utilize power flow analysis; economic dis- manner. in Power magazine, the
and stability solutions for the purpose patch; etc., may also be IEEE Power & Energy
of system planning, expansion, and implemented using the SE output. Magazine (PEM) predecessor, I made
limit setting. Online analysis tools, on For more than 50 years, since the the case for a hierarchical coordinated
the other hand, are utilized for real- inception of the EMS, there has been closed-loop control of power grids with
time operation and control of power gradual evolutionary progress in the active steering of the grid operating
systems at utility or independent sys- makeup of the EMS and SCADA sys- point to both a stable and optimal op-
tem operator (ISO) or regional trans- tems, but no foundational or transfor- erating state. This new approach was
mission operator (RTO) energy control mational changes have materialized. suggested in lieu of, as opposed to the
centers. These online tools typically Currently, however, the electric power current practice of relying on the sys-
require telemetered data collected via industry is undergoing a major trans- tem inertia and costly preset operating
supervisory control and data acquisi- formation. The increased penetration margins, for the system to coast to a
tion (SCADA) systems. SCADA sys- of nondispatchable renewable energy stable but perhaps nonoptimal operat-
tems and the associated computational sources; wide deployment of distrib- ing point upon, as a result of a distur-
uted resources at the distribution sys- bance, and then be driven by manual
tem level; and increased customer
Date of current version: 30 December 2022 empowerment and engagement are (continued on p. 85)

Now Accepting!
Submissions
For The New IEEE Transactions on
Energy Markets, Policy, and Regulation
The rigorously peer-reviewed journal will be

disseminating relevant knowledge and
developments in the organization and
structure of rapidly evolving multi-energy
supply and demand systems.
Learn More & Submit

ieee-pes.org/publications/transactions

PE JanFeb2023

Uploaded by

Copyright:

Available Formats

PE JanFeb2023

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PE JanFeb2023

Uploaded by

Copyright:

Available Formats

Volume 21 • Number 1 • January/February 2023

“The main advantage with

ADDITIONAL SOFTWARE FROM SKM

Volume 21 • Number 01 • January/February 2023

Digital Object Identifier 10.1109/MPE.2022.3224177

january/february 2023 IEEE power & energy magazine 1

Editor-in-Chief Assistant Editor IEEE Periodicals/Magazines Department

IEEE POWER & ENERGY SOCIETY (PES)

Governing Board Technical Council IEEE Power Engineering Letters, R. Jabr

2 IEEE power & energy magazine january/february 2023

For more information visit

4 IEEE power & energy magazine january/february 2023

Hendrix and Kerite: Your partners in overhead

Stop Outages before

6 IEEE power & energy magazine january/february 2023

DSATools ™ DSAToolsTM CORE MODULES

Offering a simple, user-friendly interface

january/february 2023 IEEE power & energy magazine 7

8 IEEE power & energy magazine january/february 2023

Black box vendor control

Enabling real-time simulation of detailed controls while protecting manufacturer IP

YOUR WORLD IN REAL TIME.

reflecting PES views

10 IEEE power & energy magazine january/february 2023

12 IEEE power & energy magazine january/february 2023

standing committees. Therefore, one

300 TX-2 300

of my main objectives as BoD director,

5 5 MCC-1 0.48 BL-2 MCC VCB 25 53.6 + 18 6.9

besides actively participating in and

Arc Flash priate PPE Required

contributing to relevant and impactful

IEEE decisions and committees, is to

january/february 2023 IEEE power & energy magazine 13

14 IEEE power & energy magazine january/february 2023

january/february 2023 IEEE power & energy magazine 15

Digital Object Identifier 10.1109/MPE.2022.3219168

16 IEEE power & energy magazine 1540-7977/22©2022IEEE january/february 2023

are supplied by phasor measure-

january/february 2023 IEEE power & energy magazine 17

WLS State Estimator

18 IEEE power & energy magazine january/february 2023

january/february 2023 IEEE power & energy magazine 19

20 IEEE power & energy magazine january/february 2023

Inborn Flaws of WLS Static State Estimators

january/february 2023 IEEE power & energy magazine 21

Tests on 13,659-Bus System

22 IEEE power & energy magazine january/february 2023

Average CPU Time Per Run (in Seconds)

Topology Errors Parameter Errors

january/february 2023 IEEE power & energy magazine 23

recovered, which is shown in Figure 7. One of the bad injec-

figure 6. The comparison of voltage residuals between

table 3. A comparison of the voltage solution quality. 40

24 IEEE power & energy magazine january/february 2023

leverage points present in the given set of measurements dis-

january/february 2023 IEEE power & energy magazine 25

Digital Object Identifier 10.1109/MPE.2022.3219166

26 IEEE power & energy magazine 1540-7977/22©2022IEEE january/february 2023