Prediction of self-sealing efficiency of caprock in long term geological CO2 storage based on GA-BP neural network and numerical method

The sealing efficiency of the overlying caprock constitutes one of the crucial indicators in the safety assessment of CO2 geological storage engineering. During the long-term CO2 geological storage process, the minerals in the caprock will undergo precipitation and dissolution reactions, which changes porosity of the rock and thereby influence the sealing performance of the rock. The self-sealing efficiency of caprock is regarded as safety- enhancement of GCS projects. However, Previous studies rarely applied artificial neural network to predict the caprock self-sealing efficiency caused by different host mineral compositions. And the gray correlation analysis (GRA) has been implemented to evaluate the strength of caprock self-sealing efficiency caused by each mineral. In this paper, a GA-BP neural network and a numerical simulation model have been established to predict the degree of self-sealing effect brought about by different mineral compositions. The volume fraction of each mineral is employed as an input parameter to prognosticate the variation of the rock pore at the bottom of the caprock. To verify the accuracy of the prediction model, representative caprock samples in China were gathered, and their fundamental mineral compositions were ascertained by X-ray diffraction and SEM methods. The results show that calcite significantly decreases self-sealing efficiency while ankerite, albite and Ca-smectite are key minerals triggering self-sealing enhancement. The output results of artificial neural networks demonstrate that the root mean square error (RMSE) of the GA-BP neural network prediction model is 1.1596.