ABSTRACT Operating parameters of Proton Exchange Membrane Fuel Cells (PEMFCs) are important for its output performance and real time control. However, in existing studies, the methods for optimizing its operating parameters either focus on only one aspect, or the multi‐objective optimization process, and is excessively time‐consuming. This study proposes a novel hybrid framework that couples a deep neural network surrogate model with the Non‐dominated Sorting Genetic Algorithm II (NSGA‐II) for multi‐objective optimization of fuel cell operating parameters. Seven key operating parameters—including voltage, temperature, pressure, stoichiometric ratios, and humidity—are selected as decision variables, while power density, system efficiency, and oxygen distribution uniformity serve as optimization objectives. The surrogate model is constructed using an Adam‐optimized backpropagation neural network (Adam‐BP), trained on a dataset generated through orthogonal design and numerical simulations. The model achieves high accuracy in performance prediction and is integrated with NSGA‐II to rapidly generate Pareto‐optimal solutions. The optimal operating parameters are further selected using the CRITIC method, which enables adaptive weighting based on data variability and correlation. Results show that the proposed method improves power density by 27%, system efficiency by 9.7%, and oxygen distribution uniformity by 6.6% compared to baseline conditions. This approach offers a fast, accurate, and systematic tool for PEMFC performance optimization, with strong potential for engineering application.
Zhang et al. (Sun,) studied this question.