Data-Driven Modeling to Estimate Wettability in Hydrogen-Brine-Mineral Systems: Insights for Underground Hydrogen Storage

Abstract Underground geological structures have gained attention for large-scale hydrogen storage. However, this approach faces significant challenges due to the complex subsurface wettability behavior. Experimental techniques are commonly used to understand such complexities; however, they are tedious and expensive. In contrast, machine learning is an effective alternative to predict wettability, but existing models are limited in capturing the nuanced mineralogical interactions. This research proposed a novel model for predicting wettability in Hydrogen-Brine-Mineral Systems. An experimental dataset was collected from the literature, consisting of hydrogen/brine contact angles on pure clean surfaces, under various pressures (10 - 100 bar), temperatures (20 - 80 °C), and salinity conditions (0 - 122347 ppm). Three machine learning models-Support Vector Regression (SVR), Gradient Boosting Regression (GBR), and Extreme Gradient Boosting (XGBoost) were employed to predict the contact angles. The models were trained and validated using KFold cross-validation, and the hyperparameters were optimized through an exhaustive grid-search approach. The model performance was evaluated using the mean absolute percentage error (MAPE), root mean squared error (RMSE), and coefficient of determination (R2). Data analytics illustrate that the contact angle in such systems is highly dependent on the mineral surface, followed by pressure, temperature, and salinity. The comparative performance analysis of the applied models showed that XGBoost outperformed the others, achieving MAPE of 1.13% (training) and 5.67% (testing) with R2 of 98 % on the testing data. The proposed model can predict contact angles for Hydrogen-Brine-Mineral Systems as a function of pressure, temperature, mineral specific gravity, and salinity. Different types of minerals considered in this study include: calcite, dolomite, quartz, granite, anhydrite, gypsum, shale, and basalt. The trained model was tested on an unseen dataset. A group trend analysis was also conducted to confirm the model's integrity. Contact angles generally increase with both pressure and temperature due to changes in hydrogen density and interfacial interactions, with pressure having a more pronounced impact. Mineral composition significantly influences wettability; for instance, gypsum and anhydrite exhibit higher contact angles than basalt and granite under the same conditions. Increased salinity also led to higher contact angles due to ion-induced modification of the surface properties. A novel machine learning model was proposed in this study to predict contact angles for Hydrogen-Brine-Mineral systems. The proposed model may save a significant amount of time when running tedious experiments and molecular simulations, thus adding value to the literature.