Low-temperature energy management for fuel cell buses: A DRL approach integrating powertrain electro-thermal-aging characteristic and cabin thermal comfort

To address the challenges of insufficient fuel cell thermal management, inadequate cabin waste heat utilization, and accelerated powertrain state of health (SOH) degradation in fuel cell buses under low-temperature conditions, this paper proposes a deep reinforcement learning (DRL)-based energy management strategy (EMS) that incorporates low-temperature health awareness and thermal awareness. Specifically, an electrical-aging coupling model of the powertrain, a low-temperature fuel cell thermal management model, and a cabin thermal load model are developed to formulate a multi-objective reward function for the Soft Actor-Critic (SAC) agent. This reward function integrates powertrain SOH, fuel cell temperature, and cabin temperature information, enabling simultaneous optimization of powertrain durability, fuel cell thermal safety, and cabin thermal comfort. Furthermore, for a fixed-route fuel cell bus, a representative driving cycle is constructed to facilitate strategy training, aiming to improve overall energy efficiency and environmental adaptability. Simulation results show that the proposed strategy outperforms the baseline method, reducing key metrics including powertrain SOH degradation (8%–15%), temperature violations (28%–45%), terminal SOC deviation (67%), and driving cost (8%).