IEEE Power & Energy Magazine - March/April 2018 - 84

table 1. Comparisons of the features among Go, game-based studies, and search-based studies.
Features

The Game of Go

Game-Based Power System
Studies

Search-Based Power System
Studies

Involving two players

√

Yes, can be 2+ players.

Involving multiple candidate scenarios

√

Possible

Typical

Involving multiple time intervals or stages

√

Typical

MTCS is needed

√

Past experience helps

√

Simulation for RL

√

Required computing time

Minutes

Minutes to tens of minutes

Seconds to minutes

Exponential (e.g., B N $ T )

Exponential (e.g., S M $ T )

State-space complexity*

170

* In the last row, B = the number of bidding options, N = the number of players, T = the number of time intervals (stages), S = the number of scenarios
(e.g., wind scenarios), and M = the number of studied buses.

time, rolling optimization helps overcome the uncertainty of
renewable energy and contributes to higher robustness and
economy of the dispatch schedule.
in rolling optimization, the number of system states can
be tremendous due to a combination of multiple time-intervals and multiple uncertainty scenarios. the objective is to
minimize the total cost under all time intervals, and in each
time interval, multiple uncertainty scenarios are summed up
with probability weights. thus, the decision variables (i.e.,
the output of traditional and renewable generators) are both
time and scenario related. this is a high-dimensional problem, which leads to a massive computational burden for realtime applications.
a dnn can be trained to decide the optimal dispatch
schedule. in this case, the input to the network is the uncertain factors, i.e., wind power and solar radiation, and the output will be the hourly generation of the conventional unit. the
training set includes historical weather data and the associated dispatch schedules. the training set can be expanded
by simulation runs to include all possible real-time operating
conditions. With a well-trained dnn, we may obtain a robust
dispatch schedule without the need of performing an extensive computation.

Summary
Go is an ancient board game that was considered, by far, the
most complex board game for computer software or ai to
solve. alphaGo, developed by the Google deepmind team,
is the first ai application that defeated a human professional
world champion. this article introduced the three critical components, policy network, value network and mctS, in the
alphaGo algorithm. then, several potential ai applications in
electric power systems were categorized into two groups: game
based and search based. table 1 compares the features in the
game of Go, game-based power system studies, and search-

ieee power & energy magazine

based power system studies. Undoubtedly, this will be an interesting area for the electric power community to further explore
such applications and identify potential deep ai applications in
power grid operation and planning.

For Further Reading
d. Silver, a huang, c. J. maddison, a. Guez, l. Sifre, G. van
den driessche, J. Schrittwieser, i. antonoglou, v. panneershelvam, m. lanctot, S. dieleman, d. Grewe, J. nham, n.
Kalchbrenner, i. Sutskever, t. lillicrap, m. leach, K. Kavukcuoglu, t Graepel, and d. hassabis, "mastering the game of
Go with deep neural networks and tree search," Nature, vol.
529, no. 7587, pp. 484-489, 2016.
d. mechner, "all systems Go," Sciences, vol. 38, no. 1,
pp. 32-37, 1988.
J. tan and l. Wang, "enabling reliability-differentiated
service in residential distribution networks with phevs: a
hierarchical game approach," IEEE Trans. Smart Grid, vol.
7, pp. 684-694, mar. 2016.
r. Sunitha, r. Sreerama Kumar, and a. t. mathew, "online
static security assessment module using artificial neural networks,"
IEEE Trans. Power Syst., vol. 28, pp. 4328-4335, nov. 2013.
r. yao, S. huang, K. Sun, F. liu, X. Zhang, S. mei, W.
Wei, and l. ding, "risk assessment of multi-timescale cascading outages based on markovian tree search," IEEE Trans.
Power Syst., vol. 32, pp. 2887-2900, Jul. 2017.
y. Wang, m. liu, and Z. Bao, "deep learning neural network for power system fault diagnosis," in Proc. 35th Chinese
Control Conf., 2016, pp. 6678-6683.

Biographies
Fangxing Li is with the University of tennessee, Knoxville.
Yan Du is with the University of tennessee, Knoxville.
p&e

march/april 2018

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - March/April 2018