Cost-Aware Charging Scheduling for Large-Scale Electric Vehicles Based on Hybrid Reward Multi-Agent Reinforcement Learning

The charging cost of large-scale plug-in electric vehicles (EVs) is primarily affected by charging behaviors. In this article, a cost-aware charging scheduling strategy is proposed based on a hybrid reward multi-agent reinforcement learning (HR-MARL) algorithm for multifunctional charging stations that integrate charging, storage, and discharging functions. First, the multicharging constraints are constructed, and the cost-aware charging scheduling problem is formulated. Then, the HR-MARL algorithm is designed to schedule the charging behaviors of large-scale plug-in EVs, considering electricity price fluctuations, users' requirements, and constraints from EVs, charging piles, and charging stations. The main novelties of the HR-MARL include a hybrid reward learning strategy (HRLS) and a local-global information-based replay buffer (LGI-RB). The HRLS helps ensure the target consistency of all agents, while the LGI-RB enhances the learning efficiency of the multiagent system. Finally, the proposed HR-MARL-based charging scheduling strategy is validated through a real-world charging station. The validation results demonstrate that the proposed HR-MARL outperforms several state-of-the-art (SOTA) methods in reducing charging costs for large-scale plug-in EVs.