基于图感知强化学习方法的配电网在线无功电压控制策略研究  

Research on Online Reactive Power and Voltage Control Strategy for Distribution Network Based on Graph Aware Reinforcement Learning Method

作　　者：吴浩 杨虎 韩禹 杨金明 李季 WU Hao;YANG Hu;HAN Yu;YANG Jinming;LI Ji

机构地区：[1]江苏祥泰电力实业有限公司,江苏泰州225300 [2]泰州开泰电力设计有限公司,江苏泰州225300 [3]江苏安泰输变电工程有限公司,江苏泰州225300

出　　处：《电力系统装备》2025年第1期60-62,共3页Electric Power System Equipment

摘　　要：随着分布式可再生能源的接入规模逐渐增加,配电网电压越限和网损增大等问题日益严重.文章提出了一种基于图神经网络和双延迟深度确定性策略梯度算法的配电网无功电压控制策略.先将配电网的无功电压控制问题建模为马尔科夫决策过程,并采用图神经网络提取配电网拓扑结构特征,增强了智能体对系统状态的感知能力,实现对光伏逆变器和静止无功补偿器的无功输出优化.最后结合改进IEEE33节点验证了该方法在电压稳定性和能量优化方面的有效性,为高渗透率分布式能源下的配电网优化提供了一种可行的解决方案.As the scale of distributed renewable energy access gradually increases,problems such as voltage exceeding limits and increased network losses in the distribution network are becoming increasingly serious.The article proposes a reactive power and voltage control strategy for distribution networks based on graph neural networks and dual delay depth deterministic gradient algorithm.Firstly,the reactive power and voltage control problem of the distribution network is modeled as a Markov decision process,and a graph neural network is used to extract the topological structure features of the distribution network,enhancing the intelligent agent's perception ability of the system state and achieving the optimization of reactive power output for photovoltaic inverters and static reactive power compensators.Finally,the effectiveness of this method in voltage stability and energy optimization was verified by improving the IEEE33 node,providing a feasible solution for distribution network optimization under high penetration distributed energy.

关 键 词：配电网 马尔可夫决策 图神经网络 无功电压控制 

分 类 号：TM761.1[电气工程—电力系统及自动化]