A Multi-Objective EV Charging Framework Using Reinforcement Learning and NSGA-II for Adaptive TOU Scheduling in Smart Campuses

Abstract

The rapid electrification of transportation systems, especially in academic and institutional campuses, presents both opportunities and operational challenges for local energy infrastructures. Uncoordinated EV charging can impose critical stress on the grid, while static TOU pricing schemes lack the intelligence to adapt to dynamic user behavior and energy profiles. This paper proposes a novel multi-tier hybrid optimization framework that integrates Reinforcement Learning (RL), PSO–Fuzzy TOPSIS-based schedule ranking, and NSGA-II-based multi-objective Pareto optimization for adaptive TOU scheduling in a V2G-enabled smart charging infrastructure. Taking Thiagarajar College of Engineering (TCE), Madurai as a case study location, we simulate realistic EV charging behaviors using brand-specific profiles for two-wheelers, three-wheelers, and four-wheelers, along with a priority-based DSM scheme. The RL agent dynamically learns TOU pricing policies, while the PSO–TOPSIS framework ranks them based on peak load, profit, and discomfort. The final NSGA-II layer identifies optimal trade-offs between economic and grid objectives. Our simulation reveals that the proposed system achieves up to 46% reduction in peak load, enhances energy utilization to over 91%, and minimizes user discomfort by over 85%. These results underline the potential of intelligent, brand-aware pricing control for future-ready energy systems.