Multi-Agent RL Framework for EV Charging Scheduling Driven by Energy Costs and User Preferences

The increasing popularity of electric vehicles (EVs), calls for more grid-connected charging stations. Managing these stations, poses a complex problem that requires balancing station profitability, user preferences, grid needs, and stability. Finding the ideal charging schedule is difficult because it involves factors such as electricity prices, available renewable resources, the stored energy of other vehicles, as well as the unpredictability of EV arrival and departure times. This paper presents a multi-agent and distributed reinforcement learning framework, achieving high performance under various conditions. The charging spots make their own charging decisions independently, aiming to minimize costs, without exchanging any information. Numerical studies show that the proposed framework improves the scalability and the sample efficiency of DDPG algorithm, offering significantly better results compared to Rule-Based Controllers (RBCs) and other state-of-the-art RL algorithms.