Learning-based hybrid control for hydrogen-powered fuel cell bus in entire route with complex traffic dynamics

Hydrogen-powered buses are deployed in many cities and produce no operational pollution. However, their widespread adoption is hindered by the high costs of hydrogen transportation and storage, making energy efficiency improvements crucial. This study analyzes the trajectory control and energy management of a hydrogen-powered fuel cell bus across an entire route, incorporating bus stops and traffic signals. We employ kinematic, powertrain, and energy transfer models to characterize the bus dynamics. A reinforcement learning framework is developed to optimize acceleration, transmission ratio, and power allocation, enhancing operational efficiency and reducing energy consumption. To validate the algorithm’s effectiveness, numerical experiments were conducted based on a real-world scenario, i.e., a bus route in Jiaxing, China. The key contribution of this work is the integrated consideration of vehicle dynamics, powertrain performance, energy transfer, and real-time traffic conditions for hydrogen bus operation. Furthermore, real-time feedback data are used to train the control algorithm, eliminating the need for theoretical model approximations. Results demonstrate that the proposed algorithm outperforms baseline models in energy savings, transmission efficiency, energy management, and vehicle stability.