基于深度强化学习的有源配电网实时电压控制策略  

Strategy of Real-time Voltage Control in Active Distribution Network Based on Deep Reinforcement Learning

作　　者：陈潇潇 周云海[1] 张泰源 郑培城 CHEN Xiaoxiao;ZHOU Yunhai;ZHANG Taiyuan;ZHENG Peicheng(College of Electrical Engineering and New Energy,China Three Gorges University,Yichang 443002,China)

机构地区：[1]三峡大学电气与新能源学院,湖北宜昌443002

出　　处：《三峡大学学报（自然科学版）》2024年第1期76-84,共9页Journal of China Three Gorges University：Natural Sciences

基　　金：国网冀北电力有限公司科技项目(SGJBDK00DZJS2310065)。

摘　　要：大规模分布式光伏的接入给配电网电压控制带来了挑战.针对下垂控制中各光伏逆变器无法实现协同控制的问题,本文提出了一种基于多智能体深度强化学习的有源配电网实时电压控制策略.该控制策略将有源配电网电压控制物理模型转变为分散部分可观测的马尔科夫决策过程,并通过多智能体双延迟深度确定性策略梯度算法训练各智能体,在集中式训练-分散式执行的框架下实现光伏逆变器的协同电压控制.该策略无需精确的配电网物理模型,将各光伏逆变器作为强化学习环境中的智能体,与环境交互的过程中学习最优控制策略,能够应对有源配电网中源荷的随机变化,实时开展电压控制.改进的IEEE-33节点算例验证结果表明,本文所提策略具备良好的稳压减损性能.The access of large-scale distributed photovoltaic(PV)brings the challenge to the voltage control in distribution network.Aiming to the problem that all inverters of PV cannot realize the cooperative control in sag control,a novel strategy of real-time voltage control in active distribution network based on multi-agent deep reinforcement learning is proposed in this paper,which transforms the physical model of voltage control into a decentralized partially observable Markov decision process.The gradient algorithm of the multi-agent double delay depth deterministic strategy is adopted to train each agent,and the collaborative voltage control of the PV inverter is realized under the framework of centralized training-distributed execution.The strategy does not require the accurate physical model of distribution network,and each PV inverter is taken as an agent in the reinforcement learning environment to learn the optimal control strategy in the process of interacting with the environment.It can cope with the random changes of the source load in active distribution network and also can carry out the voltage control in the real time.The results of the improved IEEE-33 node example show that the proposed strategy has better voltage stability and the performance of loss reduction.

关 键 词：配电网电压控制 分布式光伏 多智能体深度强化学习 数据驱动 马尔科夫决策过程 

分 类 号：TM761[电气工程—电力系统及自动化]