Abstract

As one of the classification methods, the Relevance Vector Machine (RVM) can solve the problems of poor sparseness and lack of probability prediction factors in Support Vector Machine (SVM). To overcome the problem of experience-based parameters selection for RVM, a kind of RVM based on particle swarm optimization (PSO) is proposed, which includes the model parameter estimation, training and model optimization based on PSO, and classification. The results of simulation experiment for a typical classification dataset show that its effect is superior to that of classical RVM, and its actual application for gas layer classification in well logging indicates that the classification results are completely consistent with the conclusions of gas trial, and it has the high classification accuracy and the good classification effect.

Paper Details