Optimization Inversion of Pseudo-NMR Echo Derived from Imaging Logging and New Porosity Calculation Using Support Vector Machine

Abstract

Due to the complex pore structure and strong heterogeneity of carbonate reservoirs, accurately characterizing reservoir properties remains a major challenge. Electrical imaging logs and nuclear magnetic resonance (NMR) logging are the essential techniques for analyzing reservoir pore structures and assessing fluid distribution. However, the high cost of NMR logging often limits its field application, resulting in sparse data coverage. To overcome this limitation, a pore spectrum model was constructed from electrical imaging logs, and corresponding pseudo-echo signals were generated. Through an optimized inversion algorithm, key parameters such as pseudo-NMR total porosity, fracture porosity, and other pore-structure-related parameters were extracted. These parameters, together with acoustic and density logs, as well as other derived logging features, were integrated to develop a comprehensive porosity prediction model based on support vector machines (SVMs). The model was applied to carbonate reservoirs in the Tarim Basin. For Well AT01, the average relative errors of the four-parameter SVM model and the full-parameter SVM model on the independent test set were 43.49% and 9.70%, respectively. The relative error of the full-parameter SVM model ranged from 4.40% to 15.30%, indicating improved prediction accuracy compared with direct use of pseudo-NMR-derived porosity. In addition, the full-parameter SVM model provided porosity estimates that were generally consistent with core-measured porosity in the locally calibrated test interval. These results suggest that integrating pseudo-NMR-derived pore-structure parameters with conventional logging features can improve porosity evaluation in complex carbonate reservoirs, especially when measured NMR logging data are limited.