Interfacial Tension Modeling of Aqueous Ionic Liquids via Machine and Deep Learning

Surface phenomena play a crucial role in the success of enhanced oil recovery (EOR) operations in petroleum engineering. Recent studies have demonstrated that ionic liquids (ILs) can effectively reduce the surface tension (SFT) and interfacial tension (IFT) of aqueous phases against air and crude oil, respectively. This study employed two deep learning models, including gradient boosting neural networks (GrowNet) and deep neural networks (DNN), and two machine learning models, including categorical boosting (CatBoost) and Gaussian process regression (GPR), to predict these important parameters. These models utilized input parameters of temperature (T), ionic liquid concentration (IL Conc.), IL molecular weight (MW), and critical properties of ILs, including critical temperature (Tc), critical pressure (Pc), and acentric factor (ω). Comprehensive statistical analysis revealed CatBoost's superior performance in SFT prediction (AAPRE = 0.19, RMSE = 0.21, R2 = 0.998) and IFT estimation (AAPRE = 0.35, RMSE = 0.11, R2 = 0.998). The model consistently outperformed alternatives across all parameter ranges. Notably, the CatBoost predictions captured the expected physical behavior, showing inverse relationships between both temperature/IL concentration and SFT/IFT values. SHAP analysis identified IL temperature and concentration as the most influential parameters. In the final phase of this study, leverage analysis validated the data set quality, confirming 94.97% of SFT and 94.74% of IFT measurements as statistically reliable. The results of this study established CatBoost as a robust, data-driven alternative to conventional empirical methods for predicting SFT/IFT of IL-containing aqueous solutions in EOR applications.