Armazenamento de hidrogênio em zeólitas por adsorção: uma abordagem computacional baseada em simulação molecular e aprendizado de máquina

Abstract

Efficient hydrogen storage is one of the main challenges in transitioning to a more sustainable energy matrix. This study investigates hydrogen adsorption in zeolites as an alternative storage method, employing molecular simulation and machine learning. Grand Canonical Monte Carlo simulations were performed to obtain adsorption isotherms at 298 K, followed by the development of supervised learning models to correlate zeolite structural properties with their adsorption capacity. The results indicate that both small- and large-pore zeolites can exhibit good performance, depending on factors such as structural density and pore accessibility. The XGBoost model demonstrated high accuracy (R2 = 0.993) and was used to predict hydrogen adsorption across a wide range of structures. Predictions revealed that approximately 50% of the analyzed zeolites adsorb between 3 and 4 mol/kg at 35,000 kPa, with the IRR zeolite standing out, reaching 7.04 mol/kg. Similarity analysis showed that structural density is a key factor in optimizing adsorption. The findings reinforce the potential of zeolites as promising materials for hydrogen storage and highlight the effectiveness of computational modeling as a tool to accelerate the selection of viable materials for energy applications.