A Deep Reinforcement Learning-Based Energy Management Optimization for Fuel Cell Hybrid Electric Vehicle Considering Recent Experience

This study emphasizes a recent experience sampling method in conjunction with Deep Deterministic Policy Gradient (DDPG) to enhance the training speed and improve the training outcomes. Firstly, to ensure the safe operation of the battery and energy storage system under peak power, a power demand decoupling method based on frequency domain is proposed to achieve power stratification. Subsequently, a multi-objective equivalent consumption minimization strategy model is established based on the data types of the experimental platform, and the improved DDPG algorithm is employed to solve it. Finally, simulation results demonstrate that compared to conventional DDPG algorithms, the improved DDPG algorithm can enhance efficiency by an average of 2.02%.