Statistical mechanic and machine learning approach for competitive adsorption of CO2/CH4 on coals and shales for CO2-enhanced methane recovery

Understanding the adsorption behavior of CO2, CH4, and their mixture on coal at high pressures is necessary to achieve enhanced CH4 recovery and the simultaneous sequestration of CO2. The adsorption of CO2 and CH4 on dry coal is known to exhibit complex behavior as a function of temperature, pressure, and composition. In this study, a model for CO2 adsorption under supercritical conditions was proposed based on a combination of surface adsorption and dissolution in the coal matrix. However, the corresponding CH4 adsorption can only be presented by a surface mechanism. Although the theoretical model provides an idealized description of the heterogeneous nature of coal, it retains the ability to capture the qualitative features of the experimental isotherms. The results indicate reasonable adsorption on the coal surface and dissolution into the coal matrix in the model mechanisms. The model also provides binding energies, surface areas, absolute adsorption isotherms, and isotherms in terms of the fractional occupancy. Considering the complex behavior of mixture adsorption, a machine learning (ML) approach was applied to the adsorption data of various coals. The ML model was reliable for predicting the competitive adsorption of CO2 and CH4 regardless of the coal type (R2 = 0.9950 and 0.9923 for CO2 and CH4, respectively). According to the analytical results obtained from the theoretical model and the ML approach, the volatile matter content, fixed carbon content, and vitrinite reflectance of coal were determined to be important properties for predicting the competitive adsorption of CO2 and CH4 on coal.