基于联邦强化学习的社区共享储能日前调度  

Day-ahead Scheduling of Community Shared Energy Storage Based on Federated Reinforcement Learning

作　　者：余兴兴 李元诚[1] 王庆乐 郭宜果 杨夯 YU Xingxing;LI Yuancheng;WANG Qingle;GUO Yiguo;YANG Ben(College of Control and Computer Engineering,North China Electric Power University,Changping District,Beijing 102206,China;Economic&Technology Research Institute of State Grid Shandong Electric Power Company,Jinan 250021,Shandong Province,China)

机构地区：[1]华北电力大学控制与计算机工程学院,北京市昌平区102206 [2]国网山东省电力公司经济技术研究院,山东省济南市250021

出　　处：《中国电机工程学报》2024年第20期8103-8112,I0015,共11页PROCEEDINGS OF THE CHINESE SOCIETY FOR ELECTRICAL ENGINEERING

基　　金：国家电网有限公司科技项目:支撑分布式储能网络化运营关键技术研究(5100-202199544A-0-5-ZN)。

摘　　要：社区家庭共享大容量储能设备,实现低储高发套利和家庭需求的时空转移,在分时电价下具有广阔的应用前景。然而,传统储能调度方法难以应对动态变化的储能调度环境。储能调度涉及家庭详细的能源消耗数据,泄露家庭作息习惯、人口数量等隐私信息的风险也不可忽视。为此,采用深度近端策略优化(deep proximal policy optimization,DPPO)强化学习方法,设计了考虑社区实时负荷、分时电价和储能老化的社区共享储能日前调度模型。考虑社区之间的隐私保护需求和数据壁垒问题,进一步提出隐私保护的联邦强化社区共享储能调度模型。该模型通过交互本地与全局DPPO模型梯度等参数信息,保护社区数据隐私。最后,仿真实验表明,所提方法在成本节约以及隐私保护方面具有优势。Community households share large-capacity energy storage equipment to achieve low-storage and high-generation arbitrage and the time-space transfer of family needs,which has broad application prospects under the time-of-use electricity price.However,it is difficult for traditional energy storage scheduling methods to cope with the dynamically changing energy storage scheduling environment.Energy storage scheduling involves detailed energy consumption data of households,and the risk of leaking private information such as household work and rest habits and population size cannot be ignored.To this end,using Deep Proximal Policy Optimization(DPPO)reinforcement learning method,a community-shared energy storage day-ahead scheduling model is designed considering community real-time load,time-of-use electricity price and energy storage aging.Considering the privacy protection requirements and data barriers between communities,a privacy-protected federated enhanced community shared energy storage scheduling model is further proposed.The model protects community data privacy by exchanging parameter information such as local and global DPPO model gradients.Finally,simulation experiments show that the proposed method has advantages in cost saving and privacy protection.

关 键 词：社区共享储能日前调度 联邦学习 强化学习 能源管理 隐私保护 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]