IEEE Electrification - December 2022 - 96

AD INDEX & SALES OFFICES
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 Electrification Magazine.
COMPANY
Advanced Test Equipment Corp.
RTDS Technologies
PAG E
7
3
URL
www.atecorp.com
rtds.com
PHONE
+1 800 404 ATEC
IEEE Electrification Magazine Representative
Michael C. Ross
Account Director
Naylor Association Solutions
Phone: +1 908 783 2339
mross@naylor.com
Digital Object Identifier 10.1109/MELE.2022.3219905
WHAT'S POPULAR
Successful Applications and Future Challenges of Machine Learning
for Power Systems (continued from page 91)
unsecure event is even worse than
giving a false alarm. While there are
some physics-informed approaches,
such as performance-guaranteed
ML, safe ML, or a hybrid approach of
data-driven and model-based solutions
as a compromise, this area still
needs to be explored by the research
community to achieve better performance
guarantees and explainability
of ML-based algorithms in power
system applications.
Second, there is a need to enhance
the scarce dataset-especially the
abnormal dataset-for power system
operation. An abundant dataset is
one of the boosters of DL success over
the last decade. However, the power
grid is under a normal operation status
for 99.97% of time or even more
(e.g., Loss of Load Expectation is
0.1 days/year as the industrial standard),
so the collected field data
from utilities and independent system
operators may not have sufficient
96
abnormal cases for ML training. To
make matters worse, the field datasets
are often not available to the
broad research community due to
data privacy and security concerns.
As such, it is necessary to further
develop closed-loop testbeds or " digital
twins " as well as ML technologies
like generative adversarial networks
to produce more high-fidelity synthetic
datasets.
Third, the scalability of ML, especially
the DRL techniques, needs to be
studied on real-world, large-scale
power systems. For example, since
most of the DRL algorithms are developed
by the computer science
domain, and their performances are
usually validated on games (e.g.,
board and video games), a big challenge
for RL in power systems is the
scalability issue because the state
and decision spaces increase dramatically
with the number of elements
in the grid, known as the curse of
IEEE Electrification Magazine / DECEMBER 2022
dimensionality. Techniques like multiagent
RL can be a promising direction
for designing scalable algorithms.
In a nutshell, the power system
community is experiencing the bestever
time to explore new revolutionary
ML techniques, which have demonstrated
great potential in power
system operation and planning.
Meanwhile, there are also several
challenges to be addressed by the
research and industrial communities.
Biography
Fangxing Li (li6@utk.edu) is also
known as Fran Li. He is the James W.
McConnell Professor in the Department
of Electrical Engineering and
Computer Science at the University
of Tennessee, Knoxville, TN USA, and
the chair of the IEEE Working Group
on Machine Learning for Power Systems.
He is a Fellow of IEEE and the
Editor-In-Chief of IEEE Open Access
Journal of Power and Energy.

http://www.atecorp.com http://www.rtds.com

IEEE Electrification - December 2022

Table of Contents for the Digital Edition of IEEE Electrification - December 2022