Reservoir Parameter Prediction Method Based On Support Vector Machine (svm)

Seismic attribute technology widely used in oil and gas exploration and development, in order to obtain high precision of reservoir parameters, it is necessary to find a suitable machine learning algorithm to maximize the seismic and geologic information. Tradition machine learning methods are mostly single variable and linear algorithm. But the geological parameters are the high dimension and nonlinear, traditional machine learning method is difficult to meet the low exploration degree area research needs. Therefore, in lack of drilling data exploration area, in order to obtain the reliable geological information, improve the success rate of well drilling, reduce exploration risk, choosing a kind of support for small sample technique of reservoir prediction is very necessary.This article based on the statistical theory, make full use of support vector machine advantage, and develop its good learning ability to obtain the result of reservoir prediction. The mainly method of support vector machine is to establish a contact mapping relationship between input and output, based on the parameters optimization,using stack kernel function, better to solve the small sample problem. After study the nature of support vector machine, get the principle of the training sample selection. In the selection of the support vector machine, through the practical mathematical model and summarize the enumeration method to optimization the parameters. In the research of kernel function, focusing on local kernel function and a global kernel function, and will attempt to weight stack local kernel function and a global kernel function to get the stack kernel function, finally construct the superposition kernel function support vector machine model.Based on the support vector machine model, with seismic attribute data and drilling data of the actual work area, using two kinds of support vector machine to predict the reservoir parameters of sand and mudstone percentage, oil saturation prediction and porosity. The result display the character of reservoir parameter horizontal distribution has a high agreement degree with the known sample interpolation results; Using the well logging data as the input vector of support vector machine to predict porosity, the prediction result similar with the software Forward interpretation result. The prediction accuracy rate (the relative error is less than10%of the proportion) of two kinds of support vector machine model can reach more than75%.The results of this study show that, the support vector machine is a practical technology based on theoretical, can effectively solve the practical problems in oil and gas exploration, especially to the reservoir prediction in the area of less drilling data. With the methods of optimization parameter become more and more perfect, various types of kernel function superposition is more reasonable, support vector machine for reservoir prediction accuracy will improve further.