Surrogate-enhanced multi-objective optimization of on-board hydrogen production device for carbon-free heavy-duty vehicles

The design of on-board ammonia decomposition units (ADUs) and its integration with ammonia-hydrogen hybrid powertrains present a critical challenge in the development of carbon-free heavy-duty vehicles. This study addresses this challenge through a novel surrogate-enhanced optimization framework for ADU design, introducing a dual-phase hybrid optimization framework combining non-dominated sorting genetic algorithm for partitioned exploration and Bayesian optimization for local refinement. The framework employs sequential domain decomposition using genetic algorithm-driven Pareto sampling integrated with surrogate training data accumulation, followed by Gaussian process-guided refinement that fuses adjacent optimal regions through covariance-based surrogate merging. Experimental validation demonstrates the effectiveness of the framework in achieving balanced system performance in key metrics. The results show that the powertrain equipped with the optimized ADU achieves a system efficiency of 31.24 % and an ADU efficiency of 76 % at minimal system costs, with dynamic validation more than 3.5 %.