基于深度强化学习的馈线-台区两阶段电压优化  

Two Stage Voltage Optimization of Feeder-Station Area Based on Deep Reinforcement Learning

作　　者：徐晓春[1] 李佑伟 戴欣 袁洲茂 田恩东 姚顺 窦晓波[2] XU Xiaochun;LI Youwei;DAI Xin;YUAN Zhoumao;TIAN Endong;YAO Shun;DOU Xiaobo(State Grid Jiangsu Huaian Power Supply Company,Huaian 223001,Jiangsu,China;Southeast University,Nanjing 210096,Jiangsu,China)

机构地区：[1]国网江苏省淮安供电公司,江苏淮安223001 [2]东南大学,江苏南京210096

出　　处：《电网与清洁能源》2023年第3期63-73,共11页Power System and Clean Energy

基　　金：国网江苏省电力有限公司科技项目(J2021036)。

摘　　要：分布式电源(distributed generation,DG)在10 kV和400 V配电网中大量接入,给配电网安全运行带来了巨大挑战。由于DG不确定性以及400 V台区实时量测数据不全的问题,基于最优潮流的优化方法难以解决馈线与台区的协同优化问题。为此,该文提出了一种基于电压越限风险和深度强化学习(deep reinforcement learning,DRL)的馈线-台区两阶段优化方法。首先,基于概率最优潮流计算得到10 k V馈线系统的最低电压越限风险下的调控策略,以及节点电压期望值并下发至台区。接着,利用台区调控资源,基于深度强化学习实现台区电压与光伏消纳的多目标优化。最后基于改进的IEEE33节点系统验证了该文方法的有效性。Distributed Generation(DG)is widely connected in 10 kV and 400 V distribution networks,which brings great challenges to the safe operation of distribution networks.Due to the uncertainty of DG and incomplete real-time measurement data in the 400V station area,it is difficult to use the optimization method based on optimal power flow to solve the collaborative optimization problem of feeders and stations.Therefore,this paper proposes a feeder-station optimization method based on voltage overlimit risk and Deep Reinforcement Learning(DRL).Firstly,based on the probabilistic optimal power flow,the regulation strategy under the minimum voltage overlimit risk of the 10 kV feeder system and the expected voltage of the node are calculated and sent to the platform area.Secondly,the multi-objective optimization of the voltage and PV absorption of the platform area is realized based on deep reinforcement learning by utilizing the platform area control resources.Finally,based on the improved IEEE33 node system,the effectiveness of the proposed method is verified.

关 键 词：电压优化 概率最优潮流 非全观测配电网 深度强化学习 分布式电源 

分 类 号：TM76[电气工程—电力系统及自动化]