摘要
针对传统优化方法及集中式联邦强化学习在隐私保护和计算效率方面存在的局限性,提出一种基于分布式联邦强化学习的多区域综合能源系统优化调度方法。每个区域综合能源系统由单独智能体管理,各智能体通过双延迟确定性策略梯度算法优化本地Critic网络的参数,并与邻域智能体进行参数信息交互,无需额外的中央服务器即可高效管理能量调度。为了保证全局最优,参数交互的权重系数由双随机矩阵元素确定。算例分析结果表明,所提方法能在增强对综合能源系统隐私保护的同时,展现出良好的收敛性能,有效降低了运营成本。
Aiming at the limitations of traditional optimization methods and centralized federated reinforcement learning in terms of privacy protection and computational efficiency,a distributed federated reinforcement learning-based optimal scheduling method of multi-regional integrated energy system is proposed.Each regional integrated energy system is managed by an individual agent.Each agent optimizes the parameters of its local Critic network using the twin delayed deep deterministic policy gradient algorithm and exchanges parameter information with the neighboring agents,thereby enabling efficient energy scheduling management without reliance on an additional centralized server.To ensure global optimization,the weight coefficients of parameter interaction are determined by the elements of a doubly stochastic matrix.The results of case study analysis show that the proposed method can enhance the privacy protection of the integrated energy system while demonstrating excellent convergence performance and effectively reducing operation costs.
作者
朱新文
王家奇
李生炜
林文杰
吴祥
郭方洪
ZHU Xinwen;WANG Jiaqi;LI Shengwei;LIN Wenjie;WU Xiang;GUO Fanghong(College of Information Engineering,Zhejiang University of Technology,Hangzhou 310023,China)
出处
《电力自动化设备》
北大核心
2026年第4期94-102,共9页
Electric Power Automation Equipment
基金
国家自然科学基金资助项目(62373328)。
关键词
综合能源系统
优化调度
分布式联邦强化学习
智能体
双延迟确定性策略梯度算法
integrated energy system
optimal scheduling
distributed federated reinforcement learning
agent
twin delayed deep deterministic policy gradient algorithm
