Predicting Hydrogen Solubility in Aqueous Solutions via Machine Learning for Optimized Storage in Deep Saline Aquifers, By Ali Ranjbar

Research

Title	Predicting Hydrogen Solubility in Aqueous Solutions via Machine Learning for Optimized Storage in Deep Saline Aquifers
Type	Presentation
Keywords	H2 storage H2 solubility Saline aquifer Machine learning
Researchers	Moien Kafi (First researcher) , Mohammad Rasul Dehghani Firuzabadi (Second researcher) , Yousef Kazemzadeh (Third researcher) , Ali Ranjbar (Fourth researcher)

Abstract

Porous underground structures are increasingly studied for hydrogen gas storage due to their significant capacity. A key challenge in this area is accurately estimating hydrogen solubility in water. This study developed three machine learning models using experimental data, with the LSBoost method emerging as the most precise (R² = 0.9997, RMSE = 4.18E-03), outperforming artificial neural networks and support vector machines. Bayesian optimization was employed for model parameter tuning. Residual error analysis confirmed the LSBoost model's accuracy across all data ranges. Correlation analysis indicated that pressure directly affects hydrogen solubility, while salinity has an inverse relationship; temperature showed minimal impact. The LSBoost method, combined with state equations, offers practical applications for underground hydrogen storage.