Machine learning-based prediction of high-entropy alloys for hydrogen storage with optimized thermodynamic and kinetic parameters

High-entropy alloys (HEAs), characterized by their multi-principal element compositions, have emerged as promising candidates for solid-state hydrogen storage due to their tunable structures and potential for reversible hydrogen absorption at ambient conditions. However, the vast compositional space of HEAs presents a major challenge in identifying optimal alloys with favourable storage capacity, kinetics, and thermodynamics. In this work, we propose a machine learning-assisted framework for predicting and optimizing HEA systems for hydrogen storage. A Machine Learning based Evolutionary Deep Neural Network (EvoDN2) was trained on experimental data, and tri-objective optimization was performed to maximize hydrogen storage capacity, enhance absorption/desorption kinetics, and minimize activation energy for hydrogen release. Multi-objective algorithms (NSGA-II and cRVEA) were employed to identify promising compositions. Among the predicted systems, the alloy Hf₃₄La₆.₈₁Mg₂₁.₅₃Ta₂₇V₁₀.₆₆ demonstrated the best performance, with a hydrogen storage capacity of 3.05 wt% at ambient conditions, an activation energy of 23.59 kJ·mol⁻¹H₂, and a favourable absorption-to-desorption time ratio of 1.39. Several other alloys, including CoLaTaTi and FeLaNbTiV, also achieved capacities exceeding 3 wt% with competitive kinetics. These results underscore the potential of targeted compositional optimization for achieving a synergistic balance between storage capacity, activation energy, and absorption-to-desorption time ratio, thereby accelerating the discovery of high-performance HEAs for hydrogen storage applications.