Construction of gas-bearing characteristics model based on KPCA-SVR for southern Sichuan shale gas

Abstract The gas content characteristics in shale gas reservoirs, encompassing both the volume and dispersion of gas, are vital in augmenting the efficiency of extraction and resource utilization of shale gas. The correlation between gas content (Vg) and well logging parameters, for instance, porosity (POR), density (DEN), natural gamma (GR), along with geochemical parameters such as total organic carbon content (TOC), potassium-uranium-thorium ratio (U), organic matter maturity (Ro), and high-precision resistivity (ρ), continues to be ambiguous. Moreover, a gas content characteristic model apt for the southern Sichuan region is still in the process of being developed. As a result, this research introduces a method, grounded on Kernel Principal Component (KPCA) and Support Vector Regression (SVR), for the quantitative prediction of Vg. Initially, we performed a cross-analysis of diverse parameters to pinpoint the sensitive parameters for Vg, and the investigation revealed that POR, TOC, U, ρ, PERM, DEN are sensitive parameters for Vg. Subsequently, we utilized these sensitive parameters to formulate a shale gas content characteristic model based on KPCA-SVR. Ultimately, we implemented this method in the shale gas field in the Changning area of southern Sichuan, and by juxtaposing the predicted values and actual values of three verification wells, we discerned that the discrepancy between the two was minimal, validating the applicability of this model. The principal contribution of this research lies in the successful development of a high-precision model, employing machine learning methods, for the quantitative prediction of shale gas content.