基于MDE-PPO算法的配电网多时间尺度无功优化  

Multi Time Scale Reactive Power Optimization of Distribution Network Based on the MDE-PPO Algorithm

作　　者：尹力 盛俊杰 杨帆 袁杰 朱陶之 冯燕钧 Yin Li;Sheng Junjie;Yang Fan;Yuan Jie;Zhu Taozhi;Feng Yanjun(Wuhan Power Supply Branch,State Grid Hubei Electric Power Co.,Ltd.,Wuhan Hubei 430000,China;School of Electrical Engineering,Southeast University,Nanjing Jiangsu 210096,China)

机构地区：[1]国网湖北省电力有限公司武汉供电公司,湖北武汉430000 [2]东南大学电气工程学院,江苏南京210096

出　　处：《电气自动化》2025年第1期79-81,85,共4页Electrical Automation

基　　金：国家电网有限公司科技项目(5400-202322566A-3-2-ZN)。

摘　　要：考虑配电网不同设备动作特性,提出了一种多时间尺度无功优化策略。首先,建立计及分布式能源的配电网无功优化模型以降低网络损耗与电压偏移;其次,将短时间尺度决策问题建立为马尔科夫决策过程以便深度强化学习算法的优化训练;然后,提出一种基于电压偏差的改进差分进化算法,并构建了改进差分进化-近端策略优化算法协同训练架构以满足不同时间尺度动作设备的调度需求;最后,在修改的IEEE 33节点系统上进行仿真。结果表明,所提方法能够有效抑制电压波动并降低网络损耗,具有实际的应用价值。A multi time scale reactive power optimization strategy was proposed based on the MDE-PPO algorithm,which considered the operational characteristics of different equipment in the distribution network.Firstly,a reactive power optimization model was established for the distribution network that took into account distributed energy to reduce network losses and voltage offsets;secondly,the short-term decision problem was established as a Markov decision process for the optimization training of deep reinforcement learning algorithms;then,an improved differential evolution algorithm based on voltage deviation was proposed,and a collaborative training architecture of improved differential evolution proximal strategy optimization algorithm was constructed to meet the scheduling needs of action devices at different time scales;finally,the simulation was carried out on the modified IEEE 33 node system.The results indicate that the proposed method can effectively suppress voltage fluctuations and reduce network losses,which has practical application value.

关 键 词：配电网 深度强化学习 改进差分进化 近端策略优化 多时间尺度无功优化 

分 类 号：TM73[电气工程—电力系统及自动化]