Lithofacies Discrimination Based On Adaptive Kernel Function of Support Vector Machines

Discrimination of different lithofacies is one of the main objectives of modern quantitative seismic interpretation and the accurate identification of lithofacies is indispensable for reservoir parameter prediction. We propose a method of modifying a kernel function and optimizing selection of parameter to improve the performance of a Support Vector Machine classifier and save much computation time, which also called adaptive support vector machine (AdaptSVM). We introduce the AdaptSVM method to differentiate multiclass lithofacies by our improved separating competence. Examples are given specifically for modifying Gaussian Radial Basis Function kernels. Tests on model indicate that our method can obtain more accurate identification results with less error rate and cost less computation time compared with the conventional SVM. Application to field data represents that lithofacies of target reservoir can be detected clearly.