基于最大熵强化学习的电网自主拓扑切换控制技术  

Autonomous Topology Switching Control Technology for Power Grid Based on SAC

作　　者：马世乾 黄家凯 崇志强 韩枭赟 徐娜 穆朝絮 MA Shiqian;HUANG Jiakai;CHONG Zhiqiang;HAN Xiaoyun;XU Na;MU Chaoxu(Electric Power Research Institute,State Grid Tianjin Electric Power Company,Tianjin 300384,China;School of Electrical and Information Engineering,Tianjin University,Tianjin 300072,China)

机构地区：[1]国网天津市电力公司电力科学研究院,天津300384 [2]天津大学电气自动化与信息工程学院,天津300072

出　　处：《电力系统及其自动化学报》2022年第12期51-60,共10页Proceedings of the CSU-EPSA

基　　金：国网天津市电力公司科技项目(KJ21-2-01)。

摘　　要：随着新能源装机量逐年提高,电网结构日趋复杂,针对新能源并网后导致的运行安全问题,本文提出一种基于最大熵强化学习的电网自主拓扑切换控制算法,为变电站和输电线提供切换策略。首先设计了基于模仿学习的神经网络预训练方案,提出了模仿学习-最大熵算法,解决了电网拓扑切换中面临的爆炸性动作组合空间问题和切换导致的脆弱性问题,智能体可在亚秒级内给出控制指令。基于IEEE 45节点算例对所提算法进行仿真验证,结果表明其具有较强的高效性与鲁棒性。As the installed capacity of renewable energy increases year by year,the structure of power grid is becoming more and more complex.Aimed at the safety problems caused by the grid connection of renewable energy,an autonomous topology switching control algorithm for power grid based on soft actor-critic(SAC)is proposed to provide switching strategies for substations and transmission lines.To solve the problems of explosive action space and vulnerability caused by topology switching,a neural network pre-training scheme based on imitation learning(IL)is designed,and an IL-SAC algorithm is presented.The agent can give control instructions at a sub-second level.The efficiency and robustness of the proposed algorithm are verified based on simulations of an IEEE 45-bus system.

关 键 词：最大熵强化学习算法 模仿学习-最大熵算法 拓扑鲁棒性 电网调控运行 

分 类 号：TM73[电气工程—电力系统及自动化]