Application Of Support Vector Machine In The Array Of Electromagnetic Propagation Logging Data Processing

Array Electromagnetic Propagation Resistivity Logging(EPRL) data processing is a typical geophysical inverse problem.It is to show the special nature of small-scale, high-precision and real-time. Existing inversion methods can not take into account inversion accuracy and the requirements of real-time application of inversion.The paper seeks to apply Support Vector Machine(SVM) to the inversion of the array of Electromagnetic Propagation Resistivity Logging data to explore a high accuracy and real-time method of processing logging data. In this paper,we do the following work:Firstly,using LIBSVM,a softwear package of Support Vector Machine(SVM),designed by Lin Chih-Jen who is a professor of National Taiwan University, we investigate and analyze the characteristics of the application of SVM in two aspects of data classification and regression,refine the application steps of LIBSVM and make a detailed instructions of the use of LIBSVM.Secondly, based on the analysis of the characteristics of array Electromagnetic Propagation Resistivity Logging(EPRL) inversion problem,we put forward multi-resolution inversion method of artificial intelligence based on Support Vector Machine. The first step,using the classification function of Support Vector Machine,we classify formation models based upon the layer feature;the second step,using the regression function of Support Vector Machine,we further determine the formatiom parameters exact values. The first step is roughly divided, the second step is refined.Finally, using forward modeling method of array Electromagnetic Propagation Resistivity Logging,we generate train data sets and test data sets used in svmtraining and svmprediction of Support Vector Classification and Support Vector Regreession.Using LIBSVM softwear package,we realize multi-resolution inversion method of artificial intelligence put forward in the paper. The training and test results on simulated data sets show that this method is feasible. SVM classification accuracy is above 95%, while SVM regression mean square error is 0.0253455.SVM implements the separation of forward and inversion and the effective application of this method to ensure real-time.