A CFD-driven machine learning approach for predicting PEM electrolyzer performance

Abstract

Proton exchange membrane electrolyzers (PEMELs) powered by renewable energy sources represent a next- generation technology for producing green hydrogen. The performance of PEMEL can be greatly enhanced by optimizing design parameters and operating conditions. This study presents the development of a full-scale, three-dimensional, two-phase computational fluid dynamics (CFD) model aimed at investigating the electrochemical performance of PEMELs under various physical conditions. Numerical simulations were conducted to evaluate the impact of four key physical parameters—temperature, porous transport layer (PTL) thickness, membrane thickness, and cell voltage—on the performance of the PEMEL. Based on the CFD results, four different machine learning (ML) algorithms—Support Vector Machine (SVM), Multilayer Perceptron (MLP), M5P, and Elastic Net—were trained and tested to predict the current density of the PEMEL. The effectiveness of each ML model was assessed using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) metrics for both the training and testing datasets. Among the models, the SVM demonstrated superior predictive accuracy, achieving an MAE of 0.0068, an RMSE of 0.0108 during training, an MAE of 0.0202, and an RMSE of 0.0371 during testing. Moreover, with the SVM method, the R 2 value was found to be 0.9996 for the training set and 0.9953 for the test set, while the a20-index was determined as 93.9 % for the training set and 90.5 % for the test set. ML based on CFD analysis results before fabricating PEM electrolyzers enables rapid performance predictions for new designs, significantly reducing computational time and costs.