Prediction of electrochemical properties of La–Y–Ni-based hydrogen storage alloys based on machine learning

Abstract

La–Y–Ni-based hydrogen storage alloys, renowned for their high hydrogen capacity and chemical stability, show significant developmental potential. However, their development is heavily reliant on extensive and costly experimental work. This study established regression models to predict three key properties—electrochemical capacity (0.2C), cyclic stability (80%Age), and high-rate discharge performance (1C)—using Random Forest (RF), Extreme Gradient Boosting (XGB), and Ridge regression (Ridge) algorithms. The RF algorithm outperformed the others, with test set R² values exceeding 0.8 for all properties. Using SHapley Additive exPlanations (SHAP) for model interpretation, this study quantitatively analyzed the optimal models. Under the constraint of the total mass ratio of elements on the A side and B side being fixed at 100, this study analyzed the optimal Mn + Al intervals for replacing Ni on the B side and the optimal Y and Y + Ce intervals for replacing La on the A side.