Research On Oil And Water Layers Interpretation Method Of Flair Logging Based On Support Vector Machine

Flair logging is a relatively advanced gas logging system at present.Compared with conventional gas logging,it has the advantages of wide detection range,high quantitative degree and more accurate data.However,due to its relatively short time and mainly used in offshore oil and gas fields,the study of its interpretation method has not been carried out in depth,and still stays in the qualitative to semi-quantitative stage,the accuracy rate of interpretation is not high.Support vector machine(SVM)is considered to be the best small sample classification algorithm in many pattern recognition methods.How to combine SVM with Flair logging data to realize quantitative and intelligent interpretation of Flair logging and improve the accuracy rate of Flair logging interpretation needs further study.In this paper,the geological survey of the study area is analyzed,and the theoretical basis and algorithm principle of support vector machine are deeply studied,which provides theoretical support for the follow-up research work.On the basis of analyzing the principle and parameters of Flair mud logging technology,the process of sample data sorting and screening is established,and accurate and representative samples are obtained,which provides the data basis for the establishment of the model.Ttwo different dimension-reduced methods(principal component analysis and Relief F algorithm)are used to reduce the dimension of the data.Using the Libsvm software package,the oil and water layers interpretation model of Flair logging based on support sector machine is established for the dimension-reduced data.According to the validation results,the Relief F algorithm is determined to be most suitable for data dimension reduction of the problem studied in this paper,and the model established from the dimension-reduced data is the optimal model.The interpretation model based on dimension-reduced data of Relief F algorithm is applied in practice,and the prediction accuracy rate is 91.67%,which is higher than the comprehensive interpretation accuracy rate of other Flair logging interpretation method(75%).It further verifies the validity of the model and its strong generalization ability,and also shows that the model can improve the accuracy rate of Flair logging interpretation.To a certain extent,the quantification and intellectualization of Flair logging interpretation are realized.