Day-ahead Scheduling of Community Shared Energy Storage Based on Federated Reinforcement Learning
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.
community energy storage day-ahead schedulingfederated learningreinforcement learningenergy managementprivacy protection
万方数据
维普
