Optimization of multi-vehicle charging and discharging efficiency under time constraints based on reinforcement learning

In the Vehicle-to-Grid (V2G) scenario, a multitude of coordinated electric vehicles (EVs) equipped with high-capacity batteries actively participate in power grid dispatching as energy carriers, aiming to achieve a tripartite objective encompassing peak load reduction and valley filling, enhanced utilization of renewable energy sources, and added benefits for electric vehicle owners. To address the existing limitations in the charging–discharging decision-making process for electric vehicles based on V2G, such as the lack of consideration for charging pile constraints, EV profitability, EV transportation timeliness, and high costs associated with central servers, we proposed a reinforcement learning-based Multi-vehicle Joint Routing and Charging–Discharging Decision algorithm (MJRCDD). Firstly, the Markov decision process (MDP) was established to describe the problem, and the route selection and charging–discharging behavior of the vehicle were innovatively integrated in the vehicle action space. Secondly, the multi-vehicle joint route planning and charging–discharging decision problem was solved by multi-agent reinforcement learning. Finally, the effectiveness of MJRCDD was verified by simulation and comparison experiments based on PeMS.