Safe and tariff-aware reinforcement learning for EV charging using real OCPP data

The rapid growth of electric vehicle (EV) adoption presents significant challenges for distribution networks, particularly in managing peak demand, operational cost, and system safety. This paper proposes a constrained multi-objective reinforcement learning framework (CMOR-PPO) for grid-safe and tariff-aware EV charging using real-world Open Charge Point Protocol (OCPP) data. The proposed approach integrates a lightweight forecasting module, a tariff-aware policy optimization scheme, and a projection-based safety mechanism to ensure strict adherence to feeder constraints while maintaining high service reliability. The framework incorporates multiple objectives, including cost minimization, peak demand reduction, risk mitigation, and energy delivery performance, within a unified learning architecture. A Lagrangian-based constraint update is employed to enforce operational limits, while robustness is evaluated under demand scaling and forecast uncertainty scenarios. Experimental results demonstrate that the proposed method achieves a reduction in peak demand (approximately 5%) and operational cost compared to uncontrolled and rule-based strategies, while maintaining 100% charging completion and near-zero constraint violations. Furthermore, the approach exhibits stable performance under varying noise levels and demand conditions, confirming its suitability for real-world deployment. The results highlight the effectiveness of combining real OCPP data, tariff-aware reinforcement learning, and safety-constrained optimization to enable practical and scalable EV charging management in modern smart grid environments.