电力系统自动化2025，Vol.49Issue(13)：70-82,13.DOI:10.7500/AEPS20240331009
基于联邦强化学习的主动配电网多主体博弈协同优化策略
Multi-agent Game Collaborative Optimization Strategy for Active Distribution Networks Based on Federated Reinforcement Learning
摘要
Abstract
To address the problems of privacy preservation and trust deficiency in multi-agent collaborative optimal scheduling of active distribution networks(ADNs),this paper proposes a day-ahead and intra-day collaborative optimization strategy based on multi-agent game and federated reinforcement learning(FRL).First,a collaborative optimization framework is established,involving various agents such as distributed generator operators,ADN operators,and energy storage operators.Within this framework,a multi-agent day-ahead and intra-day optimal scheduling model is formulated with dual objectives of maximizing overall revenue and minimizing operational adjustment.In the day-ahead stage,a trust evolution game approach considering bounded rationality is employed to generate preliminary scheduling plans,while an intra-day rolling correction mechanism is implemented using a federated natural policy gradient algorithm.This strategy ensures operation constraint compliance and effectively mitigates privacy leakage risks during information exchange.Finally,the economic feasibility of the proposed model and the effectiveness of the algorithm are verified through simulation analysis.关键词
主动配电网/多主体博弈/优化调度/联邦强化学习/隐私保护Key words
active distribution network/multi-agent game/optimal scheduling/federated reinforcement learning/privacy preservation引用本文复制引用
杨文伟,彭显刚,全欢,褚卓卓,王星华,赵卓立..基于联邦强化学习的主动配电网多主体博弈协同优化策略[J].电力系统自动化,2025,49(13):70-82,13.基金项目
国家自然科学基金资助项目(62273104). This work is supported by National Natural Science Foundation of China(No.62273104). （62273104）
