An autonomous control technology based on deep reinforcement learning for optimal active power dispatch. (February 2023)

• Record Type: Journal Article
• Authors: Han, Xiaoyun; Mu, Chaoxu; Yan, Jun; Niu, Zeyuan
• Is Part Of: International journal of electrical power & energy systems. Volume 145(2023)
• Publication Date: 2023-02
• Subjects: Active power dispatch -- Renewable energy penetration -- Soft actor–critic (SAC) -- Imitation learning (IL) -- Lagrange multiplier method -- Robustness; Electrical engineering -- Periodicals; Electric power systems -- Periodicals; Électrotechnique -- Périodiques; Réseaux électriques (Énergie) -- Périodiques; Electric power systems; Electrical engineering; Periodicals; 621.3
• DOI: 10.1016/j.ijepes.2022.108686
• Languages: English
• ISSNs: 0142-0615

Abstract

Abstract: The large-scale renewable energy integration has brought challenges to energy management in modern power systems. Due to the strong randomness and volatility of renewable energy, traditional model-based methods may become insufficient for optimal active power dispatch. To tackle the challenge, this paper proposes an autonomous control method based on soft actor–critic (SAC), a deep-reinforcement learning (DRL) strategy recently developed, which provides an optimal solution for active power dispatch without a mathematical model while improving the renewable energy consumption rate under stable operation. A Lagrange multiplier is introduced to the SAC (LM-SAC) to promote algorithm performance in optimal active power dispatch. A pre-trained scheme based on imitation learning (IL-SAC) is also designed to further improve the training efficiency and robustness of the DRL agent. Simulations on the IEEE 118-bus system with the open platform Grid2Op verify that the proposed algorithm effectively achieves better renewable energy consumption rate and robustness compared with existing DRL algorithms. Highlights: SAC is used in power systems, which realizes real-time optimal dispatch of power. LM-SAC based on the Lagrange multiplier method is proposed to improve SAC. A pre-trained scheme based on imitation learning is designed to propose IL-SAC.