Enhancing generalizability of energy management strategies for fuel cell buses: A cross-domain adaptive energy saving control approach

Existing eco-driving and energy-management strategies may suffer performance degradation when applied to fuel-cell hybrid buses with different powertrain parameters. To address this issue, a cross-parameter-domain adaptive cooperative control framework is proposed. Three archetypal domains are constructed, covering fuel-cell rated powers of 60–100 kW and battery capacities of 150–210 Ah. The upper-level controller integrates Soft Actor-Critic with a Domain-Adversarial Neural Network to reduce policy sensitivity to parameter-domain variations, while the lower-level Deep Deterministic Policy Gradient controller performs real-time power allocation. An information bridge incorporates lower-level powertrain states and Critic-value feedback into the upper-level decision process to coordinate eco-driving and energy management. Simulation results show that the proposed model maintains consistent performance across the investigated nominal domains and independently perturbed parameter configurations. Across 15 parameter-perturbation episodes, it reduces mean equivalent hydrogen consumption by 4.28% relative to SUMO and by 3.60%–9.65% relative to the learning-based comparison models. The mean terminal SOC increases from 0.449 for SUMO to 0.475, corresponding to a relative increase of 5.79%. The proposed energy-management strategy also achieves lower equivalent hydrogen consumption than A-ECMS and MPC while remaining close to the offline DP benchmark. Hardware-in-the-loop tests further demonstrate satisfactory closed-loop responsiveness under different parameter configurations. These results confirm the improvements in energy efficiency, driving stability, and cross-parameter-domain adaptability.