Adaptive charging strategies for electric vehicles: A reinforcement learning approach to demand response and resource management

Rising concerns over urban air quality have positioned electric vehicles (EVs) as a sustainable alternative to conventional transport. Large-scale EV adoption, however, depends on the availability of scalable, intelligent charging infrastructure. Effective scheduling of EV charging is vital to ensure reliability, economic efficiency, and user convenience, particularly in the presence of grid constraints and integrated energy storage systems (ESS). This study addresses the optimization of charge scheduling at a multi-lane EV charging station equipped with chargers of varying capacities, grid support, a 500 kW solar photovoltaic (PV) array, and a 200 kW ESS. Six scheduling strategies First-In-First-Out (FIFO), Earliest Deadline First (EDF), Remaining Time (RM), Least Laxity First (LLF), Improved LLF (Imp_LLF), and a Greedy method are evaluated under diverse traffic conditions and realistic operational constraints. A Reinforcement Learning (RL)-based meta-scheduler is further introduced to enable adaptive, real-time selection of the most suitable scheduling policy. By reducing state–action complexity and improving decision robustness, the meta-scheduler enhances both interpretability and adaptability. The agent demonstrates fast, state-dependent decision-making with a maximum action-selection latency of 21.574 ms, and achieves measurable performance gains over all classical baselines, outperforming the best heuristic by up to +101.82 reward points. Evaluation results confirm that the RL-driven framework consistently delivers higher allocation efficiency, reduced deadline violations, and enhanced scheduling stability across dynamic traffic scenarios.