Evaluating activation strategies and their stability on PEM water electrolyzers using machine learning

Pre-treatment of the proton exchange membrane water electrolyzers is a crucial procedure performed prior to its regular operation. These procedures help in catalyst activation and membrane saturation, thereby, ensuring its optimal performance. In this study, we use machine learning to investigate the impact of three distinct activation procedures on the cell performance and stability. The data set necessary to develop the surrogate models was obtained from a lab scale PEM electrolyzer cell. After evaluating the performance of the three tested models and validating them with experimental data, extreme gradient boosting is selected as the to perform parametric analysis. The modeling predictions reveal that the activation procedures mainly impact the ohmic resistance at the beginning of the cell life. These observations were further corroborated using through sensitivity analysis performed through an explainable artificial intelligence technique. Furthermore, data-driven time-series forecasting analysis to predict cell stability for different activation procedures showed a good comparison between experimental data and model predictions. • Introduced MEA activation as a feature in data-driven modeling of PEMEC. • Combined data-driven models with SHAP for ranking MEA activation procedures. • On the short run, in-situ/water 90 °C activation strategy showcases better performance. • The LSTM model shows a good comparison with experimental data for cell stability .