A Distributed Training and Scheduling Approach for Real-Time Coordination of Electric Vehicle Fast Charging Stations with Energy Storage

The increasing adoption of electric vehicle (EV) fast charging poses significant challenges to power systems due to its high and intermittent power demands. To alleviate these issues, fast charging stations (FCSs) are often integrated with battery energy storages (BESs) to smooth power fluctuations. However, developing effective energy management strategies for BES-equipped FCSs is nontrivial, involving uncertainties in charging demands, privacy concerns, and substantial computational burdens. In this paper, we introduce an innovative online distributed framework for the real-time coordination of multiple BES-equipped FCSs. The proposed approach begins by decomposing the original complex multi-period, multi-agent optimization problem into a series of simple single-period, single-agent subproblems. A dynamic programming (DP) method is developed to distributively train the value functions for each agent, effectively addressing the uncertainties in EV charging demand. These trained value functions are then combined with a consensus algorithm to enable real-time, distributed decision-making for each BES-equipped FCS, ensuring both system-wide optimization and privacy preservation. Our approach also derives explicit mathematical expressions for both the offline training and online scheduling processes, significantly boosting computational efficiency. Case studies conducted on various test systems validate the effectiveness of the proposed method.