A Distributed Multi-Agent Reinforcement Learning Approach for Efficient Charging Station Recommendation in Large-Scale Environments

Abstract

The rapid growth of electric vehicles (EVs) and the increasing deployment of charging infrastructure have introduced new challenges in optimizing charging station recommendations. Current methods often fail to consider the dynamic and large-scale nature of charging environments, leading to increased waiting times and suboptimal user experiences. To address these limitations, we propose a distributed multi-agent reinforcement learning framework for EV charging station recommendations, with the objective of minimizing overall driving and queuing time. Our model leverages mean field theory to address the variable number of agents and uses a distributed decision-making approach, allowing each agent to select charging stations based on local observations while coordinating with others. Simulation results demonstrate that our proposed CSMF algorithm significantly outperforms conventional methods, such as Nearest, DQN, and MADDPG, by achieving lower mean waiting times. Future research will focus on incorporating personalized user preferences to further enhance recommendation accuracy.