A Two-Stage Framework for Optimal Battery Sizing and Reinforcement Learning-Based Energy Management of EV Charging Stations

With the growing adoption of electric vehicles (EVs), charging stations face increasing power demands, causing significant fluctuations in the utility grid. Battery energy storage systems (BESS) have emerged as effective solutions to mitigate fluctuations by managing power flow and levelling loads. However, the appropriate sizing of BESS and effective energy management are crucial for enhancing system techno-economics. This paper proposes a two-stage framework for optimal BESS sizing and real-time energy management in EV charging stations. In the first stage, a convex optimization algorithm determines the optimal BESS capacity and associated converter size by accounting for the stochastic nature of EV charging demand and detailed component models. The objective integrates investment cost, electricity cost, and penalties for peak-hour power consumption. The optimized BESS size is used for the second stage where a reinforcement learning (RL)-based energy management strategy is developed to achieve real-time decisions on charging prices of EV owners and power allocation between the utility grid and the BESS. To enhance learning efficiency, cyclical feature encoding is introduced to capture periodic patterns in electricity prices and EV arrivals. Simulation results validate the effectiveness of the proposed BESS sizing and energy management strategies, with a 17.6% cost reduction compared to a system without BESS. The RL-based energy management strategy enhances the net income of the charging station by up to 17.5% over a two-week period, outperforming traditional rule-based control methods.