A collaborative energy management strategy based on multi-agent reinforcement learning for fuel cell hybrid electric vehicles

Deep reinforcement learning (DRL) algorithms have been applied to energy management strategies (EMSs) of hybrid vehicles recently with the development of artificial intelligence. However, the unstable training and inherent lower collaboration ability among agents hinder the application, especially when being faced with complicated control problems. In this research, a novel DRL algorithm, i.e. the multi-agent deep deterministic policy gradient (MADDPG) is applied to a fuel cell hybrid electric vehicle (FCHEV) with the centralized training and decentralized execution (CTDE) framework, where a more detailed reward function is designed to enable a fast and stable convergence. In order to evaluate the effectiveness, the proposed strategy is compared to the dynamic programming (DP)-based, the rule-based and the deep deterministic policy gradient (DDPG)-based EMSs in terms of the energy consumption and SOC maintenance. Results show that the proposed MADDPG-based EMS coordinates the output power of different power sources more effectively and outperforms the DDPG-based EMS in equivalent hydrogen consumption up to 5.8%.