Multi-Agent RL Optimizes Large EV Fleet Charging.

The widespread adoption of electric vehicles (EVs) presents significant challenges for power grid management, including increased peak demand, voltage fluctuations, and the need to integrate variable renewable energy sources efficiently. To address these issues, particularly for large EV fleets, implicit coordination among vehicles is essential to minimize user costs and prevent network overloads. This study investigates and compares two independent multi-agent reinforcement learning (MARL) approaches for optimizing decentralized EV charging: contextual combinatorial bandits and policy gradient algorithms. These methods aim to enable individual EV agents to make smart charging decisions based on local information. The evaluation was conducted within a realistic simulation environment where autonomous agents made decisions considering factors like price signals, state-of-charge, and temporal constraints. The study assessed the performance of these MARL approaches across different congestion levels and mixed-strategy configurations, incorporating heterogeneous agent groups and dynamic electricity pricing derived from real photovoltaic production data. The findings contribute to understanding how AI can facilitate the efficient and sustainable integration of large EV fleets into existing power grids.