A Reinforcement Learning-Augmented Lyapunov Optimization Approach to DC Fast Charging Station Management

A Lyapunov optimization (LO) approach is proposed in this paper to minimize the operation costs of a DC fast charging station (FCS). The LO method eliminates the need for future forecasts, e.g., EV arrival time or required charging energy, and reduces computation time. The FCS is equipped with a battery energy storage system (ESS) to mitigate the costs and the station’s strain on the grid during peak hours. However, using LO can lead to underutilization of the ESS. To address this issue, a reinforcement learning (RL) agent is trained to change the desired energy level in the ESS such that it is close to the optimal value. Lastly, simulation results demonstrate that the RL-augmented LO method diminishes the costs by 30%.