Identification of interlayers in braided river reservoir based on support vector machine and principal component analysis

Key words:  interlayer identification  braided river  support vector machine  principal component analysis  Lamadian Oilfield

Abstract:

Taking the PI2 braided river sand body of the Lamadian Oilfield in the northern Daqing Placanticline as an example, the support vector machine (SVM) algorithm, combined with principal component analysis (PCA) data dimension reduction was applied to realize automatic identification of braided river interlayer based on four types of logging data. The SVM model was established with twelve characteristic parameters using these four logging curves as input and interlayer types as output, and the optimal parameters (kernel function radius g and penalty factor C) were determined using Gaussian radial basis kernel function and grid search. The results show that the identification accuracy rate of logging feature parameters without dimension reduction was 86.17%, and the accuracy rate of logging feature parameters with PCA dimension reduction was 92.55%, with an increase of 6.38%. The identification accuracy of calcareous interlayers is the highest, and misjudgments occur between muddy and physical interlayers due to insignificant differences in logging response with similar lithology and limitation of logging parameters. However, the SVM algorithm based on PCA has a better reliability for the identification of interlayers and can meet the needs of geological interpretation.