Research Info

As global energy demand rises, the environmental impacts of fossil fuels prompt the search for cleaner energy sources. Hydrogen has emerged as a promising alternative, with efficient underground storage being essential for its large-scale deployment. The sandstone formations are suitable, particularly with cushion gas (i.e. inert gas to maintain pressure and increase pore volume while minimizing water intrusion). In this regard, the gas-rock-brine interactions—governed by wettability and quantified via the contact angle—play a pivotal role in hydrogen trapping and mobility in porous media. This study hypothesizes that machine learning (ML) models can reliably predict contact angles under diverse subsurface conditions, thereby improving the understanding and design of hydrogen storage systems. To test this, a dataset comprising 2,391 experimental data points, collected from a comprehensive review of published literature, was used to train and validate various ML models, including Extreme Learning Machine, Multilayer Perceptron optimized by different algorithm, General Regression Neural Network optimized using the Imperialist Competitive Algorithm (ICA), Least Squares Boosting (LSBoost), Least Squares Support Vector Machine, and K-Nearest Neighbors. Among these, the ICA-LSBoost model outperformed others, achieving a root mean square error of 0.5434 in training and 1.5413 in testing, along with a mean absolute error of 0.3267 and 0.9872 for training and testing, respectively. These results contribute to a better understanding of the simulation and prediction phases of the hydrogen storage process by accurately simulating contact angles and optimizing storage strategies, ultimately facilitating the efficient use of hydrogen as a clean energy source