Machine Learning-Driven Prediction of Density and H2-Brine Interfacial Tension: Implications for Underground Hydrogen Storage

Abstract

Underground hydrogen (H2) storage presents a viable way towards energy transition and meeting the growing energy demand. To achieve the net-zero target and mitigate anthropogenic greenhouse gas emissions, the contribution of H2 as a clean energy source has proved to be an efficient alternative for future use. Interfacial tension (IFT) is a paramount parameter that influences the displacement of H2 and its storage capacity in geological conditions. This paper aims to accentuate the storage of clean H2 at a large scale and leverage robust machine learning algorithms to predict the IFT of H2-brine systems at various thermophysical and salinity conditions. In this study, predictions of brine density and IFT of H2-brine systems are performed using input features, including, pressures (0.1, 5, 10, 15, and 20 MPa), temperatures (25, 50, and 70°C), and salinities including deionized water, seawater and brines of 1 molar and 3 molar concentrations for various salts (NaCl, KCl, MgCl2, CaCl2, and Na2SO4) from our experimental data. Extensive experimental measurements are recorded and a dataset consisting of 180 data points is utilized. Three robust machine learning techniques, viz., gradient boosting regressor (GBR), extreme gradient boosting (XGBoost), and random forest regressor (RF) are employed to predict the brine density and IFT of H2-brine systems. Various statistical techniques are employed to initiate this process, and histograms of the input and output parameters are generated. The correlation heat maps are populated with Pearson's correlation coefficient for IFT prediction. The relative importance of input parameters with IFT data of H2-brine systems is determined by Pearson, Spearman, and Kendall models. Evaluation of the machine learning results indicates a high prediction accuracy with RF and XGBoost algorithm with a coefficient of determination (R2) value of 99.9% for training and 99.5% for testing data. Root mean square error (RMSE) and mean absolute error (MAE) are subsequently determined. The findings from this study demonstrate the remarkable accuracy of machine learning algorithms in predicting the density and IFT of H2-brine systems. This study can be effectively utilized further to evaluate the displacement of H2 and storage capacity, reducing the time and cost associated with determining IFT with traditional methods.