基于鲁棒强化学习的配电网电压实时控制方法  

Real-time voltage regulation technique for distributionnetworks utilizing robust reinforcement learning

作　　者：宫建锋 齐屹 韩一鸣 张斌 张诗豪 韩照洋 尹孜阳 GONG Jianfeng;QI Yi;HAN Yiming;ZHANG Bin;ZHANG Shihao;HAN Zhaoyang;YIN Ziyang(Economic and Technical Research Institute,State Grid Ningxia Electric Power Co.,Ltd.,Yinchuan Ningxia 750011,China;Tianjin University,Tianjin 300072,China)

机构地区：[1]国网宁夏电力有限公司经济技术研究院,宁夏银川750011 [2]天津大学,天津300072

出　　处：《宁夏电力》2025年第1期43-49,共7页Ningxia Electric Power

基　　金：宁夏回族自治区自然科学基金项目(2023AAC03845)。

摘　　要：分布式光伏的大规模接入对配电网的电压控制提出了严峻挑战,为此,提出了一种基于鲁棒强化学习的配电网电压实时控制方法,以应对量测不确定性带来的影响。首先,建立以电压偏差最小为目标的配电网电压控制模型,并将其转化为马尔可夫决策问题;其次,通过设置攻击智能体对观测状态施加扰动,模拟量测误差;最后,设计了一种同步训练、异步学习的求解机制,提升智能体的鲁棒性。在IEEE 33节点系统上的仿真结果表明,所提方法具有良好的有效性和可靠性。The large-scale integration of distributed photovoltaic(PV)systems poses considerable challenges to voltage regulation in distribution networks.A real-time voltage control approach utilizing robust reinforcement learning is proposed to alleviate the effects of measurement uncertainty.First,a voltage control model for distribution networks is established to minimize voltage deviation,subsequently reformulated as a Markov decision process.Secondly,a distur-bance agent is introduced to induce perturbations in the observation state,so imitating measurement inaccuracies.Finally,a framework combining synchronous training and asynchronous learning is developed to improve the agent′s robustness.The simulation findings for the IEEE 33-bus system indicate that the proposed solution is both effective and reliable.

关 键 词：分布式光伏 电压控制 不确定性 鲁棒强化学习 

分 类 号：TM73[电气工程—电力系统及自动化]