Optimization of cathode catalyst layer composition for PEMFC based on an integrated approach of numerical simulation, surrogate model, multi-objective genetic algorithm and evaluation strategy

The cathode catalyst layer (CCL) provides the core site for the electrochemical reactions that occur in the proton exchange membrane fuel cell (PEMFC). Its composition directly determines the electrochemical and mass transfer performance of the fuel cell. Current experimental and modeling methodologies still have shortcomings in the comprehensive optimization of CCL performance. Therefore, this work proposes a machine learning framework combining a data-driven surrogate model and multi-objective optimization to effectively evaluate the influence of CCL compositions on three novel performance indexes, including voltage (V_cell), average value of CCL oxygen concentration (C_O2), and water saturation (CCL_sat). Firstly, CCL agglomerate model is integrated with the multi-physics PEMFC model to achieve an accurate characterization of performance. Numerical simulations provide a reliable database for training the surrogate model based on back propagation neural network (BPNN). Subsequently, the surrogate model is integrated with non-dominated sorting genetic algorithm-Ⅱ (NSGA-II) to expedite the assessment of fitness value for optimizing three performance indexes. Finally, the optimal CCL composition scheme located on the Pareto frontier is prioritized through the technique for order preference by similarity to an ideal solution (TOPSIS). The results show that the decision-optimal model demonstrates varying degrees of improvement compared to the base model, with a 6.17% improvement in V_cell, increases of 5.84% in C_O2, and reductions of 3.77% in CCL_sat at 1 A cm⁻². Therefore, a novel integrated optimization framework considering multi-factors and multi-objectives proposed in this study is of great significance in seeking the optimal CCL composition parameters and ultimately enhancing the comprehensive performance of PEMFC to accelerate its commercial application.