Log Interpretation Methodofsupport Vector Machine And Its Application

In oil and gas exploration and development activities and deepening the exploration targets is getting more complicated,the traditional logging interpretation method to calculate the porosity and permeability of the coincidence rate has been difficult to meet the needs of production in lithology recognition.Through the introduction of support vector machine method has unique advantages in high dimension,nonlinear and small sample,and cross validation method used for optimizing the key parameters,the predicted model of lithology,porosity and permeability.This paper firstly introduces the basic statistical theory of support vector machine,expounds the principle of support vector machine and the concept of kernel function,the kernel function with different properties were compared.This paper put area data into SVM with different kernel functions,the highest accuracy of kernel function is radial basis function(RBF),K fold cross validation method is used to find the key parameters of error cost coefficient C and kernel parameter optimization for G to establish the model of support vector machine.In the prediction of lithology identification,firstly,we analyses thegeological conditions of the study area toclassificate lithology and collect the logging and thin section identification and coring data.The lithology of Sulige area is classified into four types: quartz sandstone,lithic quartz sandstone and lithic quartz and mudstone;The lithology of Qianshan area is divided into nine types: lithologic mixed granite gneiss,mixed,mixed into the rocks,leucoleptite,shallow grain migmatite and granulite gneiss,amphibolite,and mafic intrusive rocks;The lithology ofX well is divided into five types : carbonate reservoir lithology sandstone,mudstone,dolomite,gypsum and salt rock,The lithology of Y well is divided into four types: lithologic limestone,mudstone and gypsum and rock salt.Analysis of logging and lithology sensitivity in perplexing logging response characteristics,selection response characteristics can be applied to the area of lithology identification and lithologic logging response using sensitivity analysis to determine the characteristics of lithology sensitive response,lithology sensitive response characteristics of Pe and GR of Su Dong and Su Xi area are selected as the input features of support vector machine,lithology sensitive DEN,CNLand GR of Qianshan area are selected as the input features of support vector machine,ithology sensitive response characteristicsRHOZ,DT,and GR ofX well are selected as the input features of support vector machine,thology sensitive response characteristicsRD,AC and GR of Y well are selected as the input features of support vector machine.Determine the input characteristics of support vector machine,according to the selection of training samples using 70 percent off cross validation of key parameters for optimization of each area of support vector machine modeling,get Su Dong area support vector machine C=1.7411 penalty factor and kernel function parameter g=9.1896,Su Xi area support vector machine C=3.0314 penalty factor and kernel function parameter g=3.4822.Penalty factor C=36.7583 and kernel function parameter of support vector machine in Qianshan g=6.9644.X well support vector machine penalty factor C=222.8609 and kernel function parameter g=0.3789,Y well support vector machine penalty factor C=2.2974 and kernel function parameter g=18.3792.After finding the key parameters,the support vector machine model is established,and the same training samples are input by different methods.The results of the 35 groups of data in the area of Su Dong were obtained by the method of cross validation.The accuracy of the support vector machine method of cross validation was 94.3%,the accuracy rate of the traditional SVM was about 91.4%,and the accuracy of the neural network method was 88.6%.The accuracy of cross validation SVM is 2.9% higher than that of traditional SVM,the accuracy of cross validation SVM is higher than that of neural network was 5.7%.The results of the 40 groups of data in the area of the Su Xi were predicted,and the accuracy of the support vector machine optimized by cross validation was up to 92.5%.The accuracy rate of the traditional SVM was 90%,and the accuracy of the neural network method was 87.5%.The accuracy of cross validation SVM is 2.5% higher than that of traditional SVM,the accuracy ofcross validation SVM is higher than that of neural network method SVM was 5%.In the Qianshan area,the 201 groups of data were predicted,and the results showed that the accuracy of thecross validation SVM was 90%,the accuracy rate of the traditional SVM was about 86.6%,and the accuracy of the neural network method was 82.6%.Compared with the traditional SVM and neural network,the accuracy of cross validation optimized SVM is 3.4% and 7.4%,respectively.The data of 50 groups of X well and Y well in a certain area are predicted.The accuracy of the support vector machine is 86%,the traditional support vector machine(SVM)is,the accuracy rate of the neural network is 82% and the accuracy of the SVM is X.The accuracy of the support vector machine optimized by cross validation is 88%,and the accuracy rate of the traditional SVM is about 84%,and the accuracy of the neural network method is,which is about Y.The accuracy of cross validation optimization of SVM is higher than that of traditional SVM and neural network method,respectively.In the prediction of porosity,the Sulige area 43 coring data,3808 groups of physical analysis data,the data of the 400 groups preferred the training samples,mainly uses the sonic logging response characteristics AC,CNL and DEN as input features using a support vector machine the regression model parameter optimization algorithm,the result is the penalty factor and kernel function parameter g=0.5637 C=8.2621.At the same time,the establishment of statistical model,prediction results of 9 wells in 40 groups of data of different wells,SVM average absolute error value of the average relative error is less than 0.87 of the value of the statistical model is 0.97,less than a porosity unit in accordance with the industry standard.The permeability prediction,the Sulige area 41 cored wells,3425 groups of data,the data of the 300 groups were selected to be the training samples,mainly uses the sonic logging response characteristics AC,CNL,DEN,GR,RT and Pe as input features using support vector machine model regression algorithm,the parameter optimization is the penalty factor and kernel function parameter g=1.5326 C=12.0325.At the same time,the establishment of statistical model,prediction results of 7 wells in the 58 sets of data,the average absolute error SVM mean absolute error is less than 0.15 of the value of the statistical model is 0.24,the average absolute error is at a low level.According to the comparison of the results achieved,that support vector machine optimization based on cross validation compared with other methods have certain advantages in lithology identification and porosity prediction and permeability prediction,and has a good application effect.