The Study of PSO-SVM and PSO-GRNN Algorithm Used in the Fault Pattern Classification of Transformer

Abstract

Support Vector Machine and GR neural network are widely used in the fault pattern classification. For large data samples and fault types, these methods have high classification accuracy. In order to further improve the practical range and operational efficiency of numerical classifier, the Particle Swarm Optimization Algorithm is used to the optimize penalty and kernel function parameters of SVM, weight and threshold parameter of GRNN. By numerical verification, it is found that the classifier is applied to multi-peak complex curved surface function, has nice classification ability. The classifier is used in classification of DGA data of actual oil immersed, and it is found: running time of PSO-SVM classifier increases with the increase of the fault mode, and the recognition accuracy shows the trend of the overall decline. The classification accuracy of the PSO-GRNN classifier is relatively stable, but the fault classification is more time-consuming. Through the research and analysis of this paper, validity of optimization classification algorithm is verified by the simulation, compared performance of the PSO-GRNN and PSO-SVM classification algorithms, which provides ideas and methods for the fault classification method of oil immersed transformer.