Coordinating Distributed Energy Storage for Peak Shaving: A Multi-Armed Bandit Approach to Enhance Regulation Resource Synergy

Distributed energy storage (DES) plays a crucial role in enhancing grid flexibility and mitigating peak-to-valley differences caused by large-scale renewable energy integration. However, the effective coordination of DES faces significant challenges due to the uncertainty and variability of user participation in peak shaving. To address this issue, this paper proposes a dynamic scheduling strategy forDES based on an online learning framework. First, a willingness assessment mechanism is introduced to evaluate users' participation tendencies in peak-shaving tasks based on historical data. Then, a two-stage scheduling framework is established, where a candidate resource pool is formed in the day-ahead stage, and real-time dispatch is optimized in response to renewable generation fluctuations. Finally, a Risk-Aware Multi-Armed Bandit (RA-MAB) algorithm is designed to solve the scheduling problem by adaptively balancing exploration and exploitation while considering the risk of uncertain user responses. Simulation results demonstrate that the proposed strategy significantly improves scheduling efficiency, reduces dependence on centralized peak-shaving resources, and enhances the overall stability of DES dispatch.