Multi-Agent Deep Reinforcement Learning for Efficient EV Routing and Intelligent Charger Allocation in Smart Cities

The road power system networks are getting strained due to the quickly growing electrification of city transportation especially electric vehicle-based transportation. This research presents an integrated framework utilizing multiagent deep reinforcement learning (MADRL) to concurrently optimize electric vehicle route planning and charging station selection. Each electric vehicle (EV) is shown as an autonomous agent proficient in acquiring real-time decision-making about route selection, charger selection, and charging level. We use a centralized-training and decentralized-execution (CTDE) architecture with Multi-Agent Proximal Planning Optimization (MAPPO). The investigation outcomes are shows that the recommended strategy decrease average travel time by 21.8%, average charging costs by 32.5%, and peak transformer loads by 18.5 percent when compared to conventional routing and nearest-charger strategies. The results outcomes are helping to work with smart charging and vehicle-to-grid (V2G) systems in the future.