Influencing factors and prediction of CO2 wettability in coal seams for carbon geo-storage: leveraging data-driven machine learning approaches

Coal seams offer tremendous potential for geological storage of CO 2 and enhanced coal bed methane recovery (ECBM). The efficiency of CO 2 sequestration processes is highly dependent on the wettability of coal when in contact with CO 2 and water. However, coal wettability is a multifaceted function of coal rank, vitrinite reflectance, ash and carbon content, and operating conditions, e.g., pressure, temperature, and brine salinity. Moreover, traditional quantification of coal wettability via dynamic and equilibrium contact angle measurements requires specialized laboratory equipment, and optical measurements may be sensitive to the precise identification of the three-phase contact line. This study, therefore, investigates coal-wetting behavior in the presence of CO 2 using machine learning approaches. The influence of factors such as coal rank, vitrinite reflectance, ash and carbon content, pressure, temperature, and brine salinity on the wetting behavior of coal surfaces is examined. A range of ML techniques, including tree-based methods like random forest and gradient boosting, and deep neural networks, are used to predict contact angles in coal/CO 2 /brine systems. A historical laboratory dataset comprising 470 points was used, with models trained on 80 % of the data and tested on the remaining 20 %. The models accurately predicted contact angles under pressure (0–20 MPa) and temperature (25–75 °C) for coals of various ranks with added uncertainty quantification, with the extreme gradient boosting algorithm outperforming others explored (R 2 = 0.95, MAPE ≈ 6 %). The findings of this study enhance the understanding of fluid interactions and CO 2 wettability of coals under varying coal ranks and operating conditions.