Reinforcement learning-based energy management strategies of fuel cell hybrid vehicles with multi-objective control

Along with the rapid development of the artificial intelligence, learning-based energy management strategies (EMSs) for hybrid vehicles have gained increasing attention in recent years, in which reinforcement learning (RL) algorithms are adopted as the mainstream approach. Currently, most of the RL-based EMS research has targeted at general engine-motor hybrid vehicles and focused on emphasizing the effectiveness of one designated type of RL algorithm. Besides, the fuel cell durability has been missed in most of the previous EMS research, which is actually a significant factor for fuel cell hybrid vehicles (FCHVs). In this research, three typical RL algorithms are applied to the energy management problem of an FCHV with the same control framework, which are the Q -learning, the deep Q-network, and the deep deterministic policy gradient algorithms respectively. The EMSs not only consider the fuel economy of the FCHV but also take into account the fuel cell durability based on a fuel cell degradation model. In addition, some techniques are designed for the RL algorithms to improve the performance of the EMSs. The performances of the RL-based EMSs are evaluated and compared in terms of the algorithm convergence ability, the FCHV fuel economy, the fuel cell durability, and the EMS adaptability.