TY - JOUR
T1 - Self-Dispatch of Wind-Storage Integrated System
T2 - A Deep Reinforcement Learning Approach
AU - Wei, Xiangyu
AU - Xiang, Yue
AU - Li, Junlong
AU - Zhang, Xin
N1 - Funding Information:
This work was supported by the National Natural Science Foundation of China under Grants U2166211 and 52177103. Paper no. PESL-00158-2021.
Publisher Copyright:
© 2010-2012 IEEE.
PY - 2022/7/31
Y1 - 2022/7/31
N2 - The uncertainty of wind power and electricity price restrict the profitability of wind-storage integrated system (WSS) participating in real-time market (RTM). This paper presents a self-dispatch model for WSS based on deep reinforcement learning (DRL). The designed model is able to learn the integrated bidding and charging policy of WSS from the historical data. Besides, the maximum entropy and distributed prioritized experience replay frame, known as Ape-X, is used in this model. The Ape-X decouples the acting and learning in training by a central shared replay memory to enhance the efficiency and performance of the DRL procedures. Besides, the maximum entropy framework enables the designed agent to explore various optimal possibilities, thus the learned policy is more stable considering the uncertainty of wind power and electricity price. Compared with traditional methods, this model brings more benefits to wind farms while ensuring robustness.
AB - The uncertainty of wind power and electricity price restrict the profitability of wind-storage integrated system (WSS) participating in real-time market (RTM). This paper presents a self-dispatch model for WSS based on deep reinforcement learning (DRL). The designed model is able to learn the integrated bidding and charging policy of WSS from the historical data. Besides, the maximum entropy and distributed prioritized experience replay frame, known as Ape-X, is used in this model. The Ape-X decouples the acting and learning in training by a central shared replay memory to enhance the efficiency and performance of the DRL procedures. Besides, the maximum entropy framework enables the designed agent to explore various optimal possibilities, thus the learned policy is more stable considering the uncertainty of wind power and electricity price. Compared with traditional methods, this model brings more benefits to wind farms while ensuring robustness.
KW - Wind farm
KW - deep reinforcement learning
KW - distributed prioritized experience replay
KW - electricity market
KW - energy storage system
KW - maximum entropy
UR - http://www.scopus.com/inward/record.url?scp=85126325945&partnerID=8YFLogxK
U2 - 10.1109/TSTE.2022.3156426
DO - 10.1109/TSTE.2022.3156426
M3 - Article
AN - SCOPUS:85126325945
VL - 13
SP - 1861
EP - 1864
JO - IEEE Transactions on Sustainable Energy
JF - IEEE Transactions on Sustainable Energy
SN - 1949-3029
IS - 3
ER -