CO2-enhanced shale gas recovery technology (CO2-ESGR) is one of the most potential carbon capture, storage and utilization (CCUS) technologies to mitigate the greenhouse effect and achieve the goal of carbon neutrality. The adsorption characteristics of CO2 and CH4 in shale are the key factors for CO2-ESGR, and fast and accurate prediction of adsorption capacity is still challenging, especially for the mixture of CO2 and CH4. This paper conducted a prediction study of CO2 and CH4 adsorption in shale in a large range of temperatures, pressures and total organic carbon (TOC) by employing the Langmuir model and the back propagation artificial neural network (BP-ANN) method. The key parameters of Langmuir model, saturation adsorption capacity Q0 and Langmuir pressure PL, were optimized for improving the prediction accuracy of CO2 and CH4 adsorption capacity in shale, and the determination coefficient of the improved model was R2=0.9691. In addition, the activation function, number of hidden layer neurons, learning efficiency and other key parameters of the BP-ANN were optimized by using data normalization method. The determination coefficients R2 of the BP-ANN prediction to experimental data for pure CH4 and pure CO2 were 0.9921, 0.9867 , respectively, a little better than the improved Langmuir model. The improved Langmuir model and BP-ANN method proposed in this study can accurately predict the adsorption characteristics of CO2 and CH4 in shale, which can provide theoretical support for CO2-ESGR.
Keywords CO2 and CH4, shale, adsorption, Langmuir model, artificial neural network, CCUS