Energy Management Strategy based on Reinforcement Learning for Fuel Cell Hybrid Vehicle with A New Reward Function Approach

Fuel cell hybrid vehicle have attracted more attention in recent years to tackle the current encountered environmental issues. Those vehicles raise implementation challenges such as the energy management strategy (EMS) to improve and optimize the energy consumption. This paper takes the learning approach with the reinforcement learning (RL) technique, which requires a proper reward function design to achieve the suited agent behavior in terms of optimization. Studies in this area propose a reward function that penalizes the consumption and uses the charge sustaining method, based on a reference state of charge (SOC). The main contribution and the originality of our research work is to introduce SOC limits to penalize the agent when it reaches the limits, instead of penalizing deviations from a given reference SOC, which constitutes a relevant degree of freedom for the EMS. The proposed approach is trained under a real drive cycle and the results are validated with the WLTP cycle.