Machine learning insights into CO2-brine interfacial tension: effects of salt type, concentration, and temperature

Abstract This study presents a data-driven modeling approach using machine learning techniques to evaluate the impact of various salt concentrations and types on interfacial tension (IFT) under different temperature and pressure conditions. Accurate IFT prediction is critical for optimizing carbon capture and storage (CCS) operations because it controls capillary trapping, plume migration, and injectivity in geological formations A dataset of 1,830 data points was compiled from the literature, encompassing three temperature levels (27 °C, 71 °C, and 100 °C) and pressure ranges from 50 to 250 bar. The salinity types analyzed included NaCl, CaCl₂, and their mixtures, with concentration ranges varying by experimental setup. A stacking ensemble learning model was developed using K-Nearest Neighbors (KNN), Decision Tree (DT), Support Vector Machines (SVM), Naive Bayes, and Logistic Regression. The model achieved a high prediction performance with a R² of 0.960 and RMSE of 1.532 on the test set. Interpretable regression formulas (linear, polynomial, and symbolic) were developed for each brine system, with polynomial regression achieving R² values above 0.97 in individual systems. The model captures known physical trends, such as an increase in IFT with higher salinity and a decrease with increasing temperature. The study demonstrates that divalent ions (Ca²⁺) have a stronger impact than monovalent ions (Na⁺), especially at lower temperatures, which is consistent with recognized interfacial thermodynamics. These findings are consistent with and help quantify salt-type effects on IFT, improving their integration into flow assurance and reservoir simulation frameworks. Furthermore, temperature moderates the salinity impact, resulting in IFT convergence at high temperatures. This study is the first to create a data-driven methodology compares the separate and combined effects of NaCl and CaCl₂ on CO₂–brine IFT. The findings provide a useful method for estimating IFT quickly and can be integrated into reservoir simulation, site screening, and operational optimization in CCS operation. However, additional validation under field-specific conditions is advised to assure accuracy beyond the laboratory scale.