| 引用本文: | 王仁浚,高红均,罗龙波,陈明辉,徐杰,刘俊勇.基于深度强化学习的新型配电系统优化运行研究综述[J].电力自动化设备,2025,45(9):152-164. |
| WANG Renjun,GAO Hongjun,LUO Longbo,CHEN Minghui,XU Jie,LIU Junyong.Review of research on new distribution system optimization operation based on deep reinforcement learning[J].Electric Power Automation Equipment,2025,45(9):152-164. |
|
| 摘要: |
| 高比例新能源接入背景下，新型配电系统运行优化问题不断凸显。考虑新型配电系统多元化分布式资源广泛接入、多类型主体互动协同与系统规模不断扩大的发展特征，综述由此带来的运行高度不确定性、复杂交互关系难以建模以及大规模系统求解困难等关键挑战。针对运行优化问题在集中/分散式决策、安全性约束及多主体互动性等方面的差异化特性，系统梳理并归纳了基于深度强化学习的关键运行优化技术，比较了不同算法的性能优劣及适用场景。总结得出了深度强化学习在新型配电系统优化运行中面临的主要技术瓶颈，展望了其未来在多时间尺度协同、模型泛化与智能交互等方向的发展前景。 |
| 关键词: 新型配电系统 多元化分布式资源 多类型主体 规模化发展 深度强化学习 |
| DOI：10.16081/j.epae.202505008 |
| 分类号: |
| 基金项目:国家自然科学基金资助项目(52077146) |
|
| Review of research on new distribution system optimization operation based on deep reinforcement learning |
|
WANG Renjun1, GAO Hongjun1, LUO Longbo2, CHEN Minghui2, XU Jie1, LIU Junyong1
|
|
1.School of Electrical Engineering, Sichuan University, Sichuan 610065, China;2.Guangzhou Power Supply Bureau of Guangdong Power Grid Co.,Ltd.,Guangzhou 510620, China
|
| Abstract: |
| On the background of high penetration of renewable energy, the problems of operation optimization for new distribution system are emerging. Considering the development characteristics of new distribution system, such as extensively access of diversified distributed resources, the interactive collaboration of multi-type entities, and the continuous expansion of system scale, significant challenges such as the high uncertainty of operation, difficulty in modelling complex interaction and difficulty in solving large-scale systems. Aiming at the differentiated characteristics of the operation optimization problem in centralized/decentralized decision-making, safety constraints and multi-agent interaction, key operation optimization technologies based on deep reinforcement learning are systematically sorted out and summarized, the performance advantages and disadvantages of different algorithms and their applicable scenarios are compared. The major technical bottlenecks faced by deep reinforcement learning in optimization operation of new distribution system are summarized, and its future development prospects in multi-time scale coordination, model generalization and intelligent interaction are prospected. |
| Key words: new distribution system diversified distributed resources multi-type entities large-scale development deep reinforcement learning |
