Integrating Bayesian Optimization with Machine Learning for Predicting Interfacial Tension in Quaternary Aqueous Hydrogen Systems: Implications for Underground Hydrogen Storage

Global decarbonization efforts are positioning hydrogen as a critical energy carrier, with storage in subsurface geological formations standing out as a safe and cost-effective option. Minimizing hydrogen loss in these formations requires an accurate understanding of interfacial tension (IFT), which governs capillary trapping and gas migration. Conventional IFT determination methods are resource-intensive, necessitating advanced predictive techniques. This study presents a machine learning framework for IFT estimation in quaternary hydrogen systems (hydrogen, methane, nitrogen, and brine), which better reflect actual storage scenarios but are rarely addressed in existing models. Bayesian-optimized models, including decision tree, random forest, extremely randomized trees, light gradient boosting machine, and gradient boosting regression, were trained on 70% of the dataset and evaluated on the remaining 30%. The models showed remarkable robustness, achieving a coefficient of determination above 0.95 and a mean absolute error below 1.3 mN/m. Gradient boosting regression (GBR) emerged as the most accurate, outperforming other ensemble learners and published correlations. Sensitivity analysis identified density difference as the most influential parameter. Results also showed that increasing methane fraction lowers IFT, potentially compromising structural trapping and increasing hydrogen loss. GBR’s estimates of capillary entry pressure (10.5–11.7 MPa), column height (1068–1193 m), and storage capacity (266–2379 kg H₂/m²) of a Saudi basaltic formation closely aligned with field reports, supporting model’s practical utility. The proposed framework enhances predictive accuracy and computational efficiency while providing actionable insights for optimizing underground hydrogen storage and reducing loss risks.