Analysis and Prediction of Hydrogen Relative Permeability in Underground Storage Systems Using Machine Learning

Underground hydrogen storage (UHS) is emerging as a critical component of future hydrogen infrastructure, offering a reliable solution for energy storage and supply security. A key parameter influencing the efficiency of UHS is the relative permeability of hydrogen (H₂), which governs the flow dynamics of hydrogen in subsurface formations. Accurate prediction of H₂ relative permeability is essential for optimizing storage systems, yet traditional empirical models often fail to capture the complex interactions in hydrogen-water systems. In this study, advanced machine learning (ML) techniques, including Polynomial Regression, Multi-Layer Perceptron (MLP), Gaussian Process Regression (GPR), Kernel Ridge Regression (KRR), Random Forest Regression (RFR), and Gradient Boosting Regression (GBR), were employed to predict H₂ relative permeability under various experimental conditions. The dataset, comprising 130 data points, included variables such as gas saturation, porosity, salinity, and differential pressure. Among the models tested, Gaussian Process Regression (GPR) demonstrated superior performance, achieving an R² of 0.9356, a Root Mean Squared Error (RMSE) of 0.0280, and a Mean Absolute Error (MAE) of 0.0178. These results highlight the potential of machine learning to provide accurate and efficient predictions, significantly outperforming traditional empirical approaches. The findings underscore the importance of integrating machine learning into UHS research, offering a robust framework for predicting hydrogen flow behavior in porous media. Future research should focus on expanding the dataset and incorporating additional factors such as hysteresis and gas mixing effects to further enhance the predictive accuracy of these models. This study contributes to the advancement of sustainable energy solutions by improving the design and operation of underground hydrogen storage systems.