shale caprock wettability modeling via machine learning for safe underground hydrogen storage

Underground hydrogen storage is one of the most effective solutions for long-term, large-scale energy storage, with its safety directly dependent on caprock performance. among various caprock types, shales are the most common and critical sealing formations due to their widespread occurrence and extremely low porosity and permeability. the sealing efficiency of shale caprocks is primarily governed by rock–fluid wettability, which is quantified by the contact angle (ca) and plays a key role in controlling capillary pressure, hydrogen migration, and long-term storage stability. however, experimental measurement of ca in shale–brine–hydrogen systems is challenging, costly, and limited by severe technical and operational constraints. in this study, machine learning (ml) models based on support vector regression (svr) and random forest (rf) were developed to predict the ca using pressure, temperature, brine salinity, and total organic carbon (toc) as input parameters. the results demonstrate that svr outperforms rf, achieving higher predictive accuracy (r² ≈ 0.93 for the testing dataset) and lower prediction errors, highlighting its superior capability in capturing complex nonlinear relationships. pearson correlation analysis further identifies pressure and toc as the most influential factors controlling ca behavior. overall, this study demonstrates the strong potential of ml—particularly svr—as a reliable and efficient tool for predicting shale wettability and assessing caprock integrity, thereby supporting the safe and optimized design of underground hydrogen storage systems.