Use of machine learning models as surrogate models for finding regions of structural properties of MOFs with high hydrogen storage capacities at room temperature and moderate pressures

Abstract

The development of efficient hydrogen storage materials is essential for the implementation of Fuel Cell Vehicles (FCVs) as a sustainable alternative to fossil fuels. Metal–Organic Frameworks (MOFs) are promising candidates due to their high surface areas and tunable porosity; however, conventional computational methods such as Grand Canonical Monte Carlo (GCMC) and Density Functional Theory (DFT) are computationally demanding, limiting large-scale exploration. In this work, supervised Machine Learning (ML) models were trained on GCMC simulation data to predict usable gravimetric and volumetric hydrogen storage capacities. Unlike previous studies focused mainly on predictive accuracy, this work emphasizes the identification of physically interpretable regions of the MOF design space under realistic operating conditions (298.15 K and 25 MPa). By combining ML models with controlled extrapolation, ranges of structural descriptors associated with high storage performance are identified. The results demonstrate that ML models can act as efficient surrogate tools for guiding the design and screening of high-performance MOFs.