An Application Research Of Reservoir History Matching Based On Support Vector Machine

Reservoir history matching is an important part of reservoir simulation. The traditional history matching has some problems such as high cost and poor results. Support Vector Machine (SVM) is a new method of machine learning. It has some advantages such as generalization ability, nonlinear and high dimensions. Whereas the relationship between SVM and reservoir history matching, the SVM is adopted to predict reservoir history matching parameters.In this paper, we start with an overview of statistical learning theory and the principle of Support Vector Regression (SVR) and kernel function methods are explained in detail. Secondly, the principles, methods and steps of reservoir history matching are introduced. Then, the SVR model of reservoir history matching is established through theoretical analysis and practical testing on this basis. Finally, the SVR model is used to predict multiple sets of data. And the predicted parameters values are loaded to reservoir model and are compared with actual historical production data.To predict reservoir characteristic parameters of four regions exactly, a method based on the attribute reduction by the rough set and SVR is presented. Firstly, the rough set theory is used to reduce the attributes of sampling data in order to select the decision-making attributes for constituting a new simply dataset. Secondly, the theory of SVR is used for training data and establishes the predicting model. Then, the test data will be predicted.Finally, the predicted history matching parameters obtained from the reservoir model are then loaded to the simulation model and are compared with actual historical production data to ensure that the guidance suggested by the system is accurate. The experimental results show that the method proposed in this paper can get a better fitting result and can also reduce the computational complexity of SVR in training data and improve the accuracy of reservoir physical parameters. The implementation of the method can provide a foundation for decision making for reservoir development.