International Journal of Modern Research in Electrical and Electronic Engineering

This study proposes a reinforcement learning-based multi-objective particle swarm optimization (RL–MOPSO) adaptive control framework to improve hydrogen utilization, enhance component durability, and ensure robust real-time energy management in fuel-cell electric vehicle (FCEV) powertrains under dynamic operating conditions. The proposed framework integrates deep reinforcement learning with multi-objective particle swarm optimization within a hierarchical control architecture. The RL agent learns optimal power management policies through continuous interaction with the vehicle environment, while MOPSO optimizes conflicting objectives including hydrogen consumption, fuel-cell degradation, system efficiency, and battery state-of-charge (SOC) stability. Degradation-aware objective functions penalize rapid load variations, excessive current densities, and operation outside optimal fuel-cell regions. The controller is evaluated using a high-fidelity FCEV model under WLTC and urban driving cycles and compared with rule-based control (RBC), model predictive control (MPC), and standalone RL strategies. Simulation results demonstrate hydrogen consumption reductions of 18–25% relative to MPC and 30–35% relative to RBC. Fuel-cell degradation indicators decrease by 20–28%, while overall powertrain efficiency improves by 12–17% and stable SOC regulation is maintained under varying operating conditions. The RL–MOPSO framework achieves superior energy efficiency, durability, and robustness compared with conventional control methods. The low computational burden of the trained RL policy enables real-time implementation in embedded automotive control systems, supporting the deployment of sustainable and reliable hydrogen-powered transportation.