| Machine learning prediction of H₂–CO₂/brine interfacial tension for underground hydrogen storage with CO₂ cushion gas |
| Paper ID : 1036-UGESR |
| Authors |
|
Atena Mahmoudzadeh, Mohammad-Reza Mohammadi * Department of Petroleum Engineering, Shahid Bahonar University of Kerman, Kerman, Iran |
| Abstract |
| Accurate prediction of H₂–CO₂/brine interfacial tension (IFT) is important for evaluating multiphase behavior during underground hydrogen storage (UHS) using CO₂ as a cushion gas. In this study, Random Forest and CatBoost models were developed using 301 experimental data points covering a wide range of pressure, temperature, H₂ mole fraction, and brine salinity conditions. The developed models demonstrated high predictive performance, with CatBoost outperforming RF and achieving an overall R² of 0.9951 and an average absolute percent relative error (AAPRE) of 1.40%. Comparison with experimental data showed that the model effectively captured variations in IFT under different pressure and salinity conditions. Furthermore, SHAP analysis identified pressure and H₂ mole fraction as the dominant parameters controlling IFT, while temperature and salinity exhibited comparatively smaller contributions. Overall, the developed model provides an accurate and efficient approach for estimating H₂–CO₂/brine IFT and supporting the assessment of UHS systems utilizing CO₂ as a cushion gas. |
| Keywords |
| Interfacial tension; Machine learning; Underground hydrogen storage; CO₂ cushion gas. |
| Status: Accepted (Poster Presentation) |