Multi-Agent Deep Reinforcement Learning for Energy Storage Stations Scheduling in Renewable-Integrated Power Distribution Systems

With the growing emphasis on carbon peaking and carbon neutrality, the installed capacity of renewable energy(RES) has expanded rapidly, resulting in significant load fluctuations. In order to overcome these challenges, it is crucial to implement an efficient scheduling strategy for energy storage stations. Multi-agent deep reinforcement learning (MADRL) has demonstrated strong potential for managing the inherent uncertainties in wind and solar power generation, as well as load variations. Consequently, this paper models the problem using a decentralized partially observable Markov decision process (DMDP) and applies the multi-agent twin delayed deep deterministic policy gradient (MATD3) algorithm to optimize the control of energy storage station networks. The proposed approach enables intelligent, real-time decision-making for dispatching RES resources and regulating energy storage stations. By accommodating the stochastic nature of source-load interactions, the method reduces stress on the upper-level grid and improvesRES integration. The outcomes of the simulation confirm the efficiency of the proposed method.