Optimization of Multi-Objective Electric Vehicle Charging and Discharging Scheduling Based on Multi-Agent Reinforcement Learning

The rapid expansion of electric vehicles (EVs) challenges charging and discharging scheduling. This paper introduces a multi-objective EV scheduling method using multi-agent reinforcement learning (MARL). We build an optimization model considering user satisfaction, grid load fluctuation, and renewable energy integration. Using Q-learning and Deep Q-Network (DQN) algorithms, simulations based on Japanese travel data show that when renewable energy integration is prioritized, EVs tend to swap batteries more, increasing renewable energy matching by over 200% with DQN. Compared to Q-learning, DQN offers better renewable energy matching and lower user costs. This study offers valuable insights into multi-objective EV scheduling, aiding renewable energy integration and grid stability.