Machine learning-guided multi-parameter performance prediction and screening for anion exchange membrane fuel cell

Anion exchange membrane fuel cells (AEMFCs), recognized as a cost-effective solution for hydrogen energy utilization, have been extensively investigated in thousands of research articles. However, the performance of the anion exchange membranes (AEMs) under operating conditions still falls short of practical requirements. The challenge resides in the current emphasis on analyzing the intrinsic properties of AEMs without explicitly linking these properties to the performance of terminal devices. We developed a comprehensive machine learning model for predicting and optimizing AEMFC power density, encompassing AEM properties, catalyst layer composition, and operating conditions. Having learned from extensive experimental data, the model demonstrates generalization capability, accurately predicting performance for unseen data with an overall error rate below 10%. By implementing the data-driven optimization strategy recommended by the model, the power density of AEMFCs improved by up to 20%. This approach effectively links material-level characteristics to system-level performance, offering a novel perspective on precision-oriented and systematic energy materials innovation.