Real-Time Two-Stage Scheduling of Electric Vehicle Charging Stations Using a SMPD Framework

Real-time scheduling of large-scale electric vehicle (EV) charging stations is essential for improving service efficiency, operational profitability, and grid coordination. However, most existing studies formulate charging scheduling as a single-stage unified decision problem that jointly handles discrete service access and continuous power control. Such a formulation fails to capture the inherently hierarchical operating mechanism of large-scale charging stations and often suffers from limited interpretability, enlarged action spaces, and reduced scalability under stochastic arrivals, dynamic departures, and time-varying resource constraints. To address these issues, this study reformulates the real-time charging scheduling problem as a two-stage collaborative decision process and proposes a Supervised Service Matching and Reinforcement Power Dispatch (SMPD) framework. In the first stage, a supervised bipartite matching network is developed to determine the service access relationships between waiting EVs and available chargers. In the second stage, a Soft Actor-Critic (SAC)-based continuous control strategy is employed to optimize charging power allocation for connected EVs under charger-level and station-level constraints. Experimental results demonstrate that the proposed framework effectively reduces waiting time while improving charger utilization, charging-demand satisfaction, and economic performance. Comparative and robustness analyses further verify its superior scheduling effectiveness, training stability, and adaptability under different infrastructure scales and random disturbance scenarios.