电力系统自动化2025，Vol.49Issue(12)：91-100,10.DOI:10.7500/AEPS20240927004
基于专家知识嵌入强化学习的配电系统灾后恢复决策方法
Decision-making Method for Post-disaster Distribution System Restoration Based on Expert Knowledge-embedded Reinforcement Learning
摘要
Abstract
A safe and efficient distribution system restoration(DSR)decision-making method is of great significance for enhancing the resilience of the distribution system.The traditional mixed-integer programming method relies on precise mathematical models and has long solving time,making it difficult to be applied online.Although deep reinforcement learning has the advantages of not relying on modeling and high decision-making efficiency,it still faces challenges such as huge optimization space and difficulty in security guarantee when dealing with DSR problems.In response to the above problems,a post-disaster DSR decision-making method based on expert knowledge-embedded reinforcement learning is proposed.Expert knowledge includes expert experience knowledge and expert mechanism knowledge.Firstly,a value function pre-training technique based on expert demonstration is proposed,which utilizes the expert experience knowledge to provide better initial optimization point for agents.Secondly,a multi-priority experience replay technology combined with expert demonstration is proposed to prevent agents from forgetting expert knowledge.Finally,an invalid action masking technology based on expert mechanism knowledge is proposed,which effectively reduces the optimization space while ensuring that the generated actions meet safety constraints such as radial operation.The analysis of an modified IEEE 37-bus case and a 362-bus case of the distribution system in a certain area in southern China shows that the proposed method can converge rapidly and generate a power supply restoration decision-making scheme close to the optimal solution that satisfies the safety constraints.Meanwhile,the decision-making efficiency meets the requirements of actual DSR decision-making.关键词
配电系统/灾后恢复/韧性/深度强化学习/专家知识/无效动作屏蔽Key words
distribution system/post-disaster restoration/resilience/deep reinforcement learning/expert knowledge/invalid action masking引用本文复制引用
萧文聪,陈俊斌,余涛,潘振宁,吴毓峰,罗庆全..基于专家知识嵌入强化学习的配电系统灾后恢复决策方法[J].电力系统自动化,2025,49(12):91-100,10.基金项目
国家自然科学基金企业创新发展联合基金资助项目(U24B6010) （U24B6010）
国家自然科学基金资助项目(52207105) （52207105）
广东省基础与应用基础研究基金资助项目(2025A1515010118). 感谢汕头大学科研启动经费资助项目(NTF24030T)对本文的支持! This work is supported by National Natural Science Foundation of China(No.U24B6010,52207105),the Guangdong Basic and Applied Basic Research Foundation(No.2025A1515010118). （2025A1515010118）
