The Application of Support Vector Machine to Estimate Synthetic Shear Sonic Log

Abstract

Abstract – Rock physics modelling is commonly applied to characterize the subsurface. Sonic log provides the elastic properties in advanced petrophysics modelling or rock physics modelling. Although it is very important, to obtain shear sonic measurement results is very expensive. Therefore, empirical and artificial intelligence allow some solutions to estimate synthetic shear sonic log. This study applicate PCA as feature selection and SVM as the regressor with TAF as the target interval for well NEGF1P. The results of feature selection are GR, DTC, and MSF log as selected features. GS optimizes the SVM kernel parameter using selected features. The best parameters for each kernel (linear and rbf) and selected feature are the input to estimate synthetic shear sonic log. The estimation result using linear kernel has R2 0.845 and root mean square error (RMSE) 15.132 and using rbf kernel has R2 0.886 and RMSE 12.989. The estimation results construe that rbf kernel estimates the synthetic sonic log with more precision than the linear kernel and indicates the linear relation between the estimated and origin log. The three other wells apply SMV with rbf kernel best parameters and selected features to estimation the synthetic shear sonic in similar interval and younger interval (GUF).