New-generation machine learning models as prediction tools for modeling interfacial tension of hydrogen-brine system

Recently, hydrogen (H₂) gas has gained prodigious attention as a sustainable energy carrier to reduce acute dependence on fossil fuels due to its fascinating properties. To ensure it continuous availability, hydrogen storage in underground geologic formations has been proffered. Nonetheless, H₂ storage in underground formations is dependent on fluid-fluid interfacial tension (IFT). Herein, new-generation machine learning models namely Gaussian Process Regression (GPR), the Elman Neural Network (ENN), and the Logistic Regression (LR) were used to predict the IFT of the H₂-brine system. For this purpose, the includes temperature (T), pressure (p), and density difference (Δρ), with the surface tension (γ) as the output variable. The effectiveness of each model was assessed through a variety of metrics including the Nash-Sutcliffe efficiency (NSE), the Mean Absolute Error (MAE), the Mean Absolute Percentage Error (MAPE), the Correlation Coefficient (PCC), the Root Mean Square Error (RMSE), and BIAS. Moreover, the limitations of traditional chemometrics feature extraction was overcome by utilizing an original linear matrix input-output (M1-M3) feature extraction approach. The result generated demonstrates that the suggested models and correlation offer sterling IFT estimations. The Gaussian Process Regression (GPR) model outperformed the other evaluated machine learning methods. Particularly, the GPR-M2 model combination showed extraordinary effectiveness, outperforming the BTA-M1 model, which had the lowest performance by 22%. Numerical comparison indicated that GPR-M2 with MAPE = 0.0512, and MAE = 0.002 emerged as the best reliable model. This study extends the frontier of knowledge in achieving carbon-free and sustainable energy society via accurate IFT prediction of H₂-brine system.