A machine-learning framework for stack voltage prediction in anion exchange membrane water electrolyzers

This study develops a data-driven machine-learning framework for predicting the stack voltage of an anion exchange membrane (AEM) water electrolyzer under varying operatingconditions.Experimental datasets collected over multiple operating days were used to train and evaluate six machine-learning models, namely K-Nearest Neighbors, Gradient Boost-ing, Random Forest, XGBoost, Light-GBM,and CatBoost.Hyperparameter optimization was carried out using Gridsearch combined with a leave-one-day-out cross-validation strategy to ensure robust generalization across different operating days. Model performance was assessed using R2, RMSE, MAE, and MAPE metrics, with CatBoost achieving the highest predictive accuracy, reaching an R2 of approximately 0.95 on the test dataset. Partial dependence plots were employed to interpret the influence of key operational variables on stack voltage.