Neural network-driven genetic algorithm for advanced fuel cell performance enhancement

• Proposed a neural network-genetic algorithm framework enabling continuous global optimization of PEMFC parameters. • Achieved peak power density of 1.0148 W/cm² with optimized operational parameters. • Quantified dominant temperature-pressure interaction (F=26.40) via response surface methodology. • Validated results through CFD simulations and statistical analysis with <1.7% method deviation. This study addresses the multi-parameter optimization challenges in proton exchange membrane fuel cells (PEMFCs) by proposing a hybrid methodology integrating neural networks with genetic algorithms. A five-layer fully connected neural network was developed, establishing an accurate nonlinear mapping between operational parameters and power density (test set R²=0.9845). Subsequent genetic algorithm optimization achieved a power density of 1.0148 W/cm², representing a 2.51% improvement over baseline performance, while identifying the optimal parameter combination. Response surface methodology validation demonstrated the dominant interaction between operating temperature and pressure (F-value=26.40), with an effect magnitude 22.56 times greater than secondary interactions. CFD simulations further confirmed the physical consistency of the optimized parameters at the field-level. This research provides both a reliable collaborative optimization strategy and explicit operational guidance for PEMFC system enhancement.