Research On Selection Method Of Fracturing Oil Wells Based On Multi-kernel Learning

Pressure drive technology has been proved to be effective in oilfield production in recent years.Well selection is an important link in the process of well selection.The combination of machine learning and oilfield development provides a new method for traditional fracturing well selection.In the actual development,the production data is usually limited in sample size,and the data source is complex and nonlinear.In this case,some machine learning methods do not perform well.Multi-kernel learning has good generalization and flexibility when solving problems of different data sources and non-linear data.The research work of this paper focuses on the analysis of the factors affecting the oil increase effect of Pressure flooding Wells and the selection of the kernel function based on multi-kernel learning.The main contents are as follows:1.A multi-kernel combination method based on rank spatial difference is proposed.In this method,the sample data by different kernel function map from original space to feature space,rank of spatial differences in the feature space is calculated,through the different combination of kernel function is the basis of rank space difference comparison,choose the rank space based kernel function combination,one of the largest group of difference as the final multi-kernel learning the kernel function is selected.The random search method is used to replace the grid search method to optimize the kernel function parameters,so as to reduce the model training time on the premise of ensuring the accuracy,and the Simple MKL method is used to learn the kernel function weight.2.Lasso and information gain rate are used to analyze the influencing factors of pressure-drive well selection.Then the analysis results obtained from the two are comprehensively analyzed to determine the degree of influence of pressure flooding factors and order them.3.The prediction model of oil increase effect is established based on multi-kernel learning.The validity of the proposed method is verified on standard data sets and actual production data sets.On this basis,the pressure drive well selection system is established to guide the final pressure drive well selection.