基于多智能体与改进目标级联法的输配协同优化调度  

Coordinated Optimal Scheduling of Transmission and Distribution Networks Based on Multiple Agents and Improved Target Cascading Method

作　　者：吉兴全 孙辰昊 张玉敏 杨明[2] 叶平峰 韩学山[2] JI Xingquan;SUN Chenhao;ZHANG Yumin;YANG Ming;YE Pingfeng;HAN Xueshan(College of Electrical Engineering and Automation,Shandong University of Science and Technology,Qingdao 266590,China;Key Laboratory of Power System Intelligent Dispatch and Control of Ministry of Education(Shandong University),Jinan 250061,China;College of Energy Storage Technology,Shandong University of Science and Technology,Qingdao 266590,China)

机构地区：[1]山东科技大学电气与自动化工程学院,山东省青岛市266590 [2]电网智能化调度与控制教育部重点实验室(山东大学),山东省济南市250061 [3]山东科技大学储能技术学院,山东省青岛市266590

出　　处：《电力系统自动化》2025年第2期165-174,共10页Automation of Electric Power Systems

基　　金：国家自然科学基金青年基金资助项目(52107111);山东省自然科学基金资助项目(ZR2021QE117)。

摘　　要：输配协同优化调度问题呈现多时间尺度耦合、非凸非线性特征,其本质上是非确定性多项式难问题。随着系统可调节资源的增加,控制变量呈指数增长,“维数灾”现象愈加明显,仅基于物理模型的求解策略难以兼顾计算精度与求解效率。为此,提出一种基于改进多智能体双延迟深度确定性策略梯度(IMATD3)和Peaceman-Rachford(PR)目标级联(ATC)法的输配协同多时间尺度优化调度方法。根据PR分裂方法改进ATC法中主-子系统的信息迭代策略,提出基于PR-ATC的输配协同日前调度方法;通过多智能体与输配网络的离线交互提取连续调度策略的时序特征,并基于物理模型并行计算各系统调度问题,形成基于深度强化学习的离线训练和基于物理模型的在线应用联合调度决策方法。以T6-D7-D7和IEEE 118-D9-D33-D69输配网络为例,验证了所提方法的有效性。The coordinated optimization scheduling problem of transmission and distribution networks presents the characteristics of multi-time scale coupling and non-convex nonlinear,which is essentially a non-deterministic polynomial-hard problem.With the increase of the system adjustable resources,the control variables exponentially increase,and the phenomenon of the curse of dimensionality becomes more and more obvious.It is difficult for the solution strategy based only on physical model to balance the calculation accuracy and solution efficiency.Therefore,a multi-time scale coordinated optimal scheduling method of transmission and distribution networks based on the improved multi-agent twin delayed deep deterministic policy gradient(IMATD3)and Peaceman-Rachford(PR)-analytical target cascading(ATC)method is proposed.Based on the PR splitting method,the information iteration strategy of main-subsystem in ATC method is improved,and a PR-ATC based transmission and distribution coordinated day-ahead scheduling method is proposed.Through the offline interaction between multiple agents and the transmission and distribution networks,the sequential features of the continuous scheduling strategy are extracted,and the scheduling problems of each system are calculated in parallel based on the physical model.A joint scheduling and decision-making method combining offline training based on deep reinforcement learning and online application based on physical model is formed.The T6-D7-D7 and IEEE 118-D9-D33-D69 transmission and distribution networks are taken as examples and the effectiveness of the proposed method is verified.

关 键 词：多智能体系统 协同优化调度 输电网 主动配电网 Peaceman-Rachford目标级联法 深度强化学习 并行计算 

分 类 号：TM73[电气工程—电力系统及自动化]