Artificial intelligence based modeling of proton exchange membrane fuel cells

An artificial intelligence (AI) based framework, including artificial neural network (ANN) along with source difference (SD) method, named as SD-ANN, is utilized as a data-driven engineering tool to improve performance prediction of proton exchange membrane fuel cells (PEMFCs) under various operational conditions. In the proposed framework, an ANN is trained on a hybrid dataset composed of both experimental data and theoretical data generated through an analytical model of the PEMFC. Current, temperature, pressure, stoichiometry at both the anode and cathode sides of the PEMFC, and relative humidity are considered as input variables for the ANN. The output of the ANN represents the difference between the experimental and analytical terminal voltages of the PEMFC. The architecture of the ANN was optimized using Gray Wolf Optimization (GWO) algorithm, to obtain maximum accuracy. The proposed model attained a root mean square error (RMSE) of 0.0016 compared to the analytical model’s RMSE of 0.0221, demonstrating a substantially higher level of accuracy. Furthermore, the model is employed to analyze the influence of different operating conditions on the performance of PEMFCs. The findings reveal that the optimal performance, corresponding to a power of 10.154 W for the PEMFC, is attained at a current of 18 A, a temperature of 60°C, anode stoichiometry of 5, cathode stoichiometry of 2, a pressure of 1 bar, and relative humidity levels of 90% for both the anode and cathode.