Machine Learning-Driven Wettability Prediction for Underground Hydrogen Storage in Gas Reservoirs

Wettability of geological formations is a critical factor in underground hydrogen storage (UHS) efficiency, commonly evaluated through experimental contact angle (CA) measurements. However, experimental CA measurements under reservoir-relevant conditions are challenging due to hydrogen’s high reactivity and embrittlement potential. This study demonstrated the successful application of machine learning models for predicting contact angle (CA) in H_2/CH4–brine–rock systems under conditions relevant to underground hydrogen storage (UHS). Using 150 experimentally measured data points, four ML algorithms were employed to predict the CA of the H_{2–CH4/brine/rock} system: Extreme Gradient Boosting (XGB), Random Forest (RF), AdaBoost Regression (AdaBoost), and Gaussian Process Regression (GPR). Among them, XGB exhibited the highest overall performance, with strong predictive accuracy and generalization, followed by RF. The close agreement between predicted and actual CA values, supported by both statistical metrics and cross-plot analysis, confirms the effectiveness of XGB and RF models for reliable CA prediction in UHS contexts.