Citation

Hossen, Md Sabbir and Ramasamy, Gobbi and Ngu, Eng Eng (2026) AI-OCPP Framework for Smart and Safe EV Charging: Real-World Performance Evaluation. IEEE Access, 14. pp. 34755-34768. ISSN 2169-3536

Text AI-OCPP_Framework_for_Smart_and_Safe_EV_Charging_Real-World_Performance_Evaluation.pdf - Published Version Restricted to Repository staff only Download (6MB)

Abstract

Electric vehicles (EVs) are changing how people use energy for transportation, and their growing adoption poses new challenges for safe and cost-effective charging operations. When multiple EVs charge simultaneously, aggregate demand can rise sharply, increasing the risk of grid overloads and high electricity costs. The Open Charge Point Protocol (OCPP) enables interoperability between charging stations and backend systems, but it does not provide built-in intelligence for forecasting or charging control. To address this gap, this paper presents a practical AI-driven OCPP framework that integrates short-term demand forecasting, tariff-aware reinforcement learning, and runtime safety enforcement into a unified control loop. The framework is evaluated using real OCPP logs and employs a tariff-aware, safety-constrained Proximal Policy Optimization (PPO) scheduler that issues charging setpoints while respecting feeder and connector limits at every control step through a projection-based safety layer. A lightweight forecasting module provides short-horizon operational support aligned with the 15-minute control resolution, without being treated as a standalone predictive contribution. The proposed approach is evaluated under multiple tariff regimes, including flat rate, time-of-use (ToU), and real-time pricing (RTP), and compared against proportional fairness (PF) and rule-based baselines. Results show that when configured with explicit energyparity constraints, the PPO controller maintains near-baseline energy delivery while achieving substantial reductions in peak demand and total energy cost. Across evaluated scenarios, peak demand is reduced by up to 30% and energy cost by approximately 25–33%, without violating grid safety constraints. The controller also demonstrates stable behavior across different pricing and loading conditions without extensive manual tuning. Overall, this study demonstrates that combining established learning methods with rigorous runtime safety enforcement and real-world validation can yield deployable and reliable AI-based charging control within OCPP-operated infrastructure. The framework emphasizes practical integration and deployment realism rather than algorithmic novelty, providing a viable pathway toward safer, more efficient, and costaware EV charging networks.

Item Type:	Article
Uncontrolled Keywords:	Tariff-aware reinforcement learning, grid-safe charging, electric vehicles
Subjects:	T Technology > TL Motor vehicles. Aeronautics. Astronautics > TL1-484 Motor vehicles. Cycles
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	02 Apr 2026 02:55
Last Modified:	02 Apr 2026 02:55
URII:	http://shdl.mmu.edu.my/id/eprint/15622

Downloads

Edit (login required)