Optimal energy management strategy for multi-stack fuel cell hybrid systems in shunting locomotives based on deep reinforcement learning

This paper proposes a novel hierarchical energy management strategy (EMS) for high-power multi-stack fuel cell hybrid systems (MFCHSs) in shunting locomotive scenarios to achieve optimal power allocation and improve fuel economy. First, the MFCHS model is established, consisting of a Li-ion battery model and a fuel cell system model with integrated efficiency and degradation submodels. Second, in the upper layer, a deep deterministic policy gradient (DDPG)-based EMS is developed for power allocation between the multi-stack fuel cell system (MFCS) and the battery. Specifically, the battery state of charge and MFCS efficiency are incorporated into the DDPG algorithm's reward function, with the agent trained under shunting locomotive operating conditions, improving EMS efficiency and extending MFCS service life during shunting operations. Finally, in the lower layer, an optimal allocation strategy based on efficiency optimization is utilized to allocate the power across individual MFCS stacks. The simulation results demonstrate that, compared to a rule-based EMS combined with an equality-based allocation method, the proposed hierarchical EMS reduces hydrogen consumption by 8.97 % and MFCS degradation by 32.39 %. Hardware-in-the-loop (HIL) experiments further validate the real-time applicability of the method, showing a 0.97 % average relative error between experimental and simulation results.