Metal hydride composition-derived parameters as machine learning features for material design and H2 storage

Abstract

Though hydrogen is a promising energy carrier for a green future, many challenges persist. One is the difficulty in engineering storage solutions, with metal hydrides being a leading contender among solid-state strategies. To facilitate efficient searching of candidate materials, ridge regression, simple decision trees, random forest ensembles, and gradient boosting ensembles were employed to predict the energy of formation, with the random forest ensemble resulting in the lowest test set error. First, two public databases, Materials Project and HydPark, were searched for metal hydrides. Feature engineering was performed before the models were developed, resulting in electronegativity, density, atomic density, d-character, f-character, band gap, hydrogen weight fraction, magnetization, temperature, and pressure being retained. The models were then benchmarked by the lowest test error before a random forest ensemble was used to populate entries missing energy of formation. All were then scored by hydrogen storage capacity and energy of formation suitability. Readily available features, including several derived from only the chemical formula, were found to be highly predictive and so are promising for high-throughput screening of arbitrary novel hydride formulations and blends for thermodynamic feasibility.