Machine Learning Models for Predicting Hydrogen Solubility in Saline Aqueous Solutions for Underground Storage

The large-scale adoption of underground hydrogen storage (UHS) is essential to transition to low-carbon energy systems. However, it remains constrained by technical challenges, including the accurate prediction of hydrogen solubility under varying subsurface conditions. This study explores the application of machine learning (ML) techniques to model hydrogen solubility in saline aqueous solutions, aiming to overcome the limitations of traditional thermodynamic models. A comprehensive methodology was developed, incorporating data preprocessing, feature selection, and the evaluation of multiple regression models, including CatBoost, XGBoost, LightGBM, and Support Vector Regressor. Among these, CatBoost demonstrated superior performance with an R² of 0.9939 and RMSE of 0.0188. These findings demonstrate the viability of ML models in providing accurate and scalable predictive tools, contributing to risk assessment in UHS projects. Ultimately, the proposed approach supports safe and efficient hydrogen storage in transitioning to sustainable energy systems.