Real-World Tariff-Aware Safe Reinforcement Learning for Grid-Stable OCPP EV Charging Networks

The rapid growth of electric-vehicle (EV) charging networks introduces coupled challenges in grid safety, tariff responsiveness, and protocol interoperability. This paper presents a real-world, tariff-aware reinforcement learning (RL) framework for grid-safe load management in Open Charge Point Protocol (OCPP) networks. The approach combines a projection-based safety layer with a tariff-sensitive policy that optimizes operating cost, peak demand, fairness, and tail delay under flat, time-of-use (ToU), and real-time pricing (RTP) regimes. Using operational OCPP logs from a Malaysian multi-site deployment, we instantiate a multi-scenario evaluation in which the proposed proximal policy optimization (PPO) controller reduces aggregate peak demand by about 25% versus proportional fairness and nearly 40% versus a rule-based baseline, while maintaining comparable or intentionally reduced energy throughput when required by cost–peak trade-offs, without compromising grid safety. Under ToU and RTP, the controller achieves cost reductions up to 30% without violating feeder or station caps; improvements in fairness (lower Gini index) and queueing performance (lower <tex-math notation="LaTeX">$w95$ </tex-math> delay) confirm that safety projection and tariff awareness jointly enhance equity and grid compliance. The methodology includes a formal projection operator that enforces hard constraints at dispatch, a multi-objective reward with tariff terms, and statistical validation based on site-level medians, confidence intervals, and Wilcoxon tests. To our knowledge, this is the first demonstration of a tariff-aware, safety-constrained RL controller validated on real OCPP data across multiple tariff scenarios, providing a practical path to scalable, economically adaptive, and grid-stable smart charging.