Lattice Boltzmann prediction of CO2 and CH4 competitive adsorption in shale porous media accelerated by machine learning for CO2 sequestration and enhanced CH4 recovery

CO2 injection has become the effective and potential means to enhance shale gas recovery and realize CO2 geological sequestration because of the CO2-CH4 competitive adsorption. However, the clarification on CO2-CH4 competitive adsorption behaviors is mainly limited to molecular simulations based on single nanoscale pore size, and it is difficult to carry out large-scale calculations based on pore-scale simulations. In this study, a new methodology coupling with molecular simulation, lattice Boltzmann method and machine learning is proposed to accurately simulate and rapidly predict the CO2-CH4 competitive adsorption in kerogen and illite three-dimensional nanoporous media under different CO2 molar fractions. From the features of pore structure and the accurate database of competitive adsorption behaviors from pore-scale simulations which are modified by molecular simulations, the Artificial Neural Network is then trained to be able to predict CO2-CH4 competitive adsorption capacity in arbitrary large-scale porous media. The above method overcomes the limitation of computing resource consumption of molecular simulation and pore-scale simulation, and provides research ideas and basic models for simulation and prediction of fluid adsorption behaviors in large-scale porous media.