Longevity-aware energy management for fuel cell hybrid electric bus based on a novel proximal policy optimization deep reinforcement learning framework

Abstract

With the prosperity of artificial intelligence and new energy vehicles, energy-saving technologies for zero-emission fuel cell hybrid electric vehicles through high-efficient deep reinforcement learning algorithms have become a research focus. This article proposes an energy management strategy based on a novel deep reinforcement learning framework to reduce the hydrogen consumption of a fuel cell hybrid electric bus while suppressing the degradation of the fuel cell. To begin, a novel proximal policy optimization framework is designed by taking advantage of multi-thread distributed computation, and then a promising energy management strategy based on this novel framework is proposed. Furthermore, the fuel cell degradation model is established and fuel cell longevity is incorporated into the optimization objective. Finally, the adaptability and computational efficiency of the proposed strategy are verified under the test cycle. Simulation results indicate that the proposed strategy improves the training efficiency effectively, and achieves efficient optimization of hydrogen conservation and fuel cell degradation suppression compared with the strategy based on the proximal policy optimization algorithm. This article contributes to energy conservation and lifespan extension for fuel cell vehicles through deep reinforcement learning methods.