A novel deep reinforcement learning-based predictive energy management for fuel cell buses integrating speed and passenger prediction

Energy management strategy (EMS) is crucial for the actual performance of fuel cell hybrid electric buses (FCHEB) in complex traffic environments. However, conventional EMS usually performs energy allocation based only on vehicle speed information, neglecting the impact of mass changes on overall vehicle performance. To ensure that FCHEB operates at optimal conditions in real-world scenarios, we propose a predictive EMS (PEMS) with onboard energy systems health management based on the twin delayed deep deterministic policy gradient algorithm (TD3). This strategy considers future driving conditions and integrates a state-of-the-art predictive model for passenger numbers. We aim to reduce the operating costs of FCHEB by comprehensively considering these two key factors. In addition, to enhance the optimization capability and driving condition adaptability of the proposed PEMS, we developed a TD3-based optimizer for the strategy's decision-making process. A dataset based on actual FCHEB operating routes is used to train the proposed PEMS. This dataset includes real driving data and passenger-related information. The experimental results show that the proposed PEMS can reduce FCHEB comprehensive operational costs by 5.92% compared with conventional TD3-based EMS (i.e., a fixed number of passengers).