A data-driven approach for multi-dimensional prediction of PEMFC performance using artificial neural networks

The characteristics of internal heat and mass distribution are key determinants of the performance, durability, and efficiency of proton exchange membrane fuel cells (PEMFCs). Owing to the challenges of the high computational cost of traditional CFD simulations and the limited accuracy of data-driven models with large datasets for obtaining the local distribution characteristics, this study introduces a multi-dimensional prediction approach for PEMFC performance analysis by combining physical models with data-driven models. Through geometric partitioning based on rib-priority and relative standard deviation (RSD) convergence evaluation, the data volume can be significantly reduced while preserving the characteristics of the data distribution. Subsequently, a multi-dimensional prediction model is developed using neural network algorithms. The proposed method exhibits excellent performance with a mean absolute percentage error (MAPE) below 3% and a coefficient of determination (R 2 ) exceeding 0.998 for key performance parameters prediction, such as current density, reactant concentrations, water content and temperature. Furthermore, this approach facilitates real-time monitoring and optimal control of PEMFCs through state mapping analysis, which comprehensively explores full-scale performance to guide the optimization of design and control strategies.