A Deep Reinforcement Learning-Based Energy Management Strategy for Fuel Cell Hybrid Buses

An energy management strategy (EMS) plays an important role for hybrid vehicles, as it is directly related to the power distribution between power sources and further the energy saving of the vehicles. Currently, rule-based EMSs and optimization-based EMSs are faced with the challenge when considering the optimality and the real-time performance of the control at the same time. Along with the rapid development of the artificial intelligence, learning-based EMSs have gained more and more attention recently, which are able to overcome the above challenge. A deep reinforcement learning (DRL)-based EMS is proposed for fuel cell hybrid buses (FCHBs) in this research, in which the fuel cell durability is considered and evaluated based on a fuel cell degradation model. The action space of the DRL algorithm is limited according to the efficiency characteristic of the fuel cell in order to improve the fuel economy and the Prioritized Experience Replay (PER) is adopted for improving the convergence performance of the DRL algorithm. Simulation results of the proposed DRL-based EMS for an FCHB are compared to those of a dynamic programming (DP)-based EMS and a reinforcement learning (RL)-based EMS. Comparison results show that the fuel economy of the proposed DRL-based EMS is improved by an average of 3.63% compared to the RL-based EMS, while the difference to the DP-based EMS is within an average of 5.69%. In addition, the fuel cell degradation rate is decreased by an average of 63.49% using the proposed DRL-based EMS compared to the one without considering the fuel cell durability. Furthermore, the convergence rate of the proposed DRL-based EMS is improved by an average of 30.54% compared to the one without using the PER. Finally, the adaptability of the proposed DRL-based EMS is validated on a new driving cycle, whereas the training of the DRL algorithm is completed on the other three driving cycles.