Cooperative Optimization Strategy for Multi Virtual Power Plants Based on Multi Agent Soft Actor Critic Reinforcement Learning

Xiaofeng <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\operatorname{Ren}^{3,4, e} { }^{3}$</tex> State Grid Jiangsu Electric Power Co., LTD., Jiangsu, China <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">${ }^{4}$</tex> Suqian Power Supply Branch, Jiangsu, China e-mail: [email protected] Abstract-With the rapid distributed resources (DERs), virtual power plant (VPP) relying on advanced digitization techniques to aggregate massive heterogeneous DERs are increasingly becoming a research hotspot. Therefore, how to achieve a win-win situation for multiple subjects without sharing the private information of VPP is a major challenge for cooperative optimization scheduling of multi VPP(MVPP). To this end, this paper constructs a VPOMVPP bi-level optimization model to determine their scheduling strategies over 96 time periods in a day. In addition, this paper introduces a multi agent soft-actor-critic (MASAC) reinforcement learning algorithm to solve the bi-level optimization model, which protects the privacy security of the MVPP through the feature of distributed execution. Finally, the effectiveness of the model and proposed in this paper is verified by arithmetic examples.