Performance Prediction of PEMFCs Using a Hybrid Approach Integrating BPNN and CFD

Efficient analysis and accurate prediction of both the performance and the multiphysical fields of proton exchange membrane fuel cells (PEMFCs) using machine learning methods are important for design optimization and operation management. In this study, a three‐dimensional multiphase flow PEMFC computational fluid dynamics (CFD) model and a back propagation neural network (BPNN) model are first developed. Next, the effects of various operating conditions (OCs) on the I – V curves of the cell and the multifield distribution features are studied. Then, the BPNN model is used both to predict the performance of the cell and to analyze the global sensitivity of different operating parameters. The main conclusions are as follows: when any of the OCs, that is, temperature ( T ), back pressure ( P ), cathode stoichiometric ratio (Stio_c) and cathode relative humidity (RHc), is varied, the I – V curves of the PEMFC predicted by the BPNN model match those calculated by the CFD model well, with R 2 values greater than 0.98, the mean absolute deviation (MAE) and relative error less than 0.0213% and 2%, respectively. The BPNN model can also accurately predict the voltage of the PEMFC when the OCs are varied simultaneously within a certain range, with R 2 values greater than 0.99. T , P , Stio_c, and RHc are important factors affecting the performance of PEMFCs, but their sensitivity will vary at different current densities. The sensitivity of T is quite high at all current densities, and the sensitivities of P and Stio_c become greater as the current density increases, whereas that of RHc tends to first decrease and then increase with increasing current density. This study provides a basis for further utilization of machine learning methods to simultaneously predict the performance and multiphysical field distribution of PEMFCs with complex structures and even dynamic OCs.