Physics-informed surrogate-assisted co-optimization of fuel cell electric vehicle powertrain and energy management for long-haul heavy-duty applications

Achieving accurate and efficient co-optimization of the powertrain configuration and energy management strategy (EMS) of the fuel cell electric vehicle (FCEV) is crucial to enhancing the performance. To address current challenges such as insufficient model accuracy, the difficulty of driving cycles in characterizing practical operating characteristics, and suboptimal performance of optimization algorithms, this study proposes a hierarchical co-optimization toolchain applied to the multi-objective co-optimization of the heavy-duty FCEV. Firstly, the physics-informed neural network (PINN) is designed for the high-fidelity model (HFM) of the FC system and a driving cycle synthesis method that can characterize the operating features of commercial vehicles. In the optimization framework, the upper layer conducts a large-scale preliminary search for Pareto solutions by combining HFM with an improved non-dominated sorting genetic algorithm-III (NSGA-III), while the bottom layer constructs the surrogate model (SM) based on the solutions from the upper layer and performs Bayesian optimization (BO) to obtain more precise solutions. The results show that the HFM of the FC system achieves both high accuracy and interpretability and the synthesized driving cycle is consistent with the characteristics of the original dataset. After optimization using the toolchain, the heavy-duty FCEV, in comparison with the baseline, realizes a 5.36% reduction in hydrogen consumption, a 6.93% decrease in state of health (SOH) loss, and a 4.16% lowering of manufacturing cost. This hierarchical co-optimization toolchain integrates the advantages of models, driving cycles, and optimization algorithms, providing a flexible and efficient solution for the co-design of FCEVs.