Prediction of Oil and Water Layer by Kernel Local Fisher Discriminant Analysis

The distribution of the oil and water layers in Xinjiang Oilfield is very complex because of the influence by many factors. It is difficult to predict the oil and water layer. In this paper, the oil and water layer of Xinjiang Oilfield was discriminated by kernel local fisher discriminant analysis(KLFDA). The local scatter matrix is defined by a affinity matrix. The original data are projected into the subspace constructed by KLFDA, and the local feature vectors are extracted. Then the prediction (classification) is done in feature subspace by Mahalanobis distance. The results indicate that the performance of KLFDA combining Mahalanobis distance is better than that of LFDA, FDA and ICA; meanwhile, the prediction accuracy of this method is better than that of SVM and ANN.