The Approximation Algorithm Of Kernel Methods On Regularization Path And Its Appliaction On The Prediction Of Oilwell's Production

Kernel method has shown excellent performance in solving the model analysis of nonlinear complex problems.It has become the focus in current machine learning.The regularization path algorithm of kernel learning is an effective method for numerical solution of problems of kernel learning,which can fit the entire path of learning problem's solutions for every value of the regularization parameter with essentially the same computational cost as fitting one model of learning problem's solution.At present,most of the oilfields in China have come to the middle and later development stage.They generally exist problems,the water injection is increasing and the oil production is decreasing.It is necessary to adjust the injection and production constantly to ensure the oil wells and water wells work in a relatively stable situation.Thus,choose the suitable dynamic analysis model of the oil production is very important.The reliability of the model's result is directly related to the medium and long-term planning at oilfield development.By analyzing the trend of oil production in single well,engineers can find problems and find out the causes.Then they present the measures to solve the problems,so as to develop the oil reservoir rationally.The research work of this paper is based on the problem of building the solution path and improving the algorithms' speed,the main work content is embodied in the following aspects:1.Two new approximation algorithms for the regularization path of SVR named SVRRP_MCC,the regularization path of MKL named MKLRP_MCwere proposed in this paper.The algorithms applied Monte Carlo method to randomly sample the kernel matrixs,reduce the size of the matrices then improve the efficiency.2.We determined the relevant factors affecting the production of the oil-wells.Using the Lasso-Lars algorithm to obtain the regression coefficient and regression coefficient variation of each oil production,and the sensitivity of different factors to oil production were determined.3.According to the actual application of irregular data,uneven distribution of samples and large scale of dataset in oilfield,we put the influencing factors of oil production as input values,using kernel method to get the prediction of oil production.