Prediction Of Reservoir Parameters And Its Application Based On PLS-SVM Algorithms

For the prediction model of traditional reservoir parameter, most is the empirical formula or the general regression formula. Because the internal structure is very complex and the geological conditions affect each other, it leads to the formation parameters multidimensional, nonlinear, correlation great. The oil field development is consuming human and financial resources, so we need to establish a more accurate prediction of reservoir parameters refined to meet the needs of the actual oil production.In this paper, after learning the basic model which calls support vector machine technology, I made a corresponding improvement. We call it partial least squares regression and least squares support vector machine algorithm and apply it on prediction of reservoir parameters. Support vector machine born in the 20th century, the mid-90s by the vapnik et al, it is based on statistical learning theory and machine learning models. Different from neural network algorithms, support vector machine is used as the basis of strict mathematical reasoning, and ultimately converge to the global optimal point, the neural network algorithm to avoid local optimal solution of the shortcomings. And in the solution of nonlinear problems, the support vector machine has the traditional regression methods can not match advantage.In the past, we generally use statistical regression analysis to predict where an explanation. The regression analysis is based on empirical risk minimization principle, such as multiple linear regression, least squares regression, stepwise linear regression. In the case of multi-dimensional variable, since the correlation between the variables would seriously interfere with multiple prediction equation. By least squares multiple regression, principal component analysis and canonical correlation analysis are combined, we get the partial least squares, and the introduction of the correlation coefficient, t test and F test for significance analysis of the model.We combined partial least squares regression and least squares support vector machine. Partial least squares method is used to eliminate the correlation, then the least squares support vector machine is used to put multi-dimensional, nonlinear data to high dimensional space, to get the corresponding fitting formula. The resulting formula is not only better than any alone results, but also avoids the use of neural network algorithm to bring the problem of local optimal solution.Porosity, oil saturation and permeability are very important logging parameters, they are reflected by the well logs such as neutron, density, transit time, etc. When using the empirical formula or even an ordinary regression analysis to predict, because the reservoir parameters parameter relationship with the well logs are very complex and contain multiple correlation between well logs, and high dimension, the predicted results usually have poor precision and generalization. This article put partial least squares regression and least squares support vector machine combined, and successfully applied to prediction of reservoir parameters. After experimental demonstration, we fitted out the formula with high precision and generalization, to achieve the ideal goal.