A Speedy Reinforcement Learning-Based Energy Management Strategy for Fuel Cell Hybrid Vehicles Considering Fuel Cell System Lifetime

A speedy reinforcement learning (RL)-based energy management strategy (EMS) is proposed for fuel cell hybrid vehicles (FCHVs) in this research, which approaches near-optimal results with a fast convergence rate based on a pre-initialization framework and meanwhile possesses the ability to extend the fuel cell system (FCS) lifetime. In the pre-initialization framework, well-designed power distribution-related rules are used to pre-initialize the Q-table of the RL algorithm to expedite its optimization process. Driving cycles are modeled as Markov processes and the FCS power difference between adjacent moments is used to evaluate the impact on the FCS lifetime in this research. The proposed RL-based EMS is trained on three driving cycles and validated on another driving cycle. Simulation results demonstrate that the average fuel consumption difference between the proposed EMS and the EMS based on dynamic programming is 5.59% on the training driving cycles and the validation driving cycle. Additionally, the power fluctuation on the FCS is reduced by at least 13% using the proposed EMS compared to the conventional RL-based EMS which does not consider the FCS lifetime. This is significantly beneficial for improving the FCS lifetime. Furthermore, compared to the conventional RL-based EMS, the convergence speed of the proposed EMS is increased by 69% with the pre-initialization framework, which presents the potential for real-time applications.