Development of fault diagnosis system for transformer based on multi-class support vector machines

Abstract

The support vector machine (SVM) is an algorithm based on structure risk minimizing principle and having high generalization ability. It is strong to solve the problem with small sample, nonlinear and high dimension. The fundamental theory of DGA (Dissolved Gas Analysis, DGA) and fault characteristic of transformer is firstly researched in this paper, and then the disadvantages of traditional method of transformer fault diagnosis are analyzed, finally, a new fault diagnosis method using multi-class support vector machines (M-SVMs) based on DGA theory for transformer is put forward. Then the fault diagnosis model based on M-SVMs for transformer is established. At the same time, the fault diagnosis system based on M-SVMs for transformer is developed. The system can realize the acquisition of the dissolving gas in the transformer oil and data timely and low cost transmission by GPRS (General Packet Radio Service, GPRS). And it can identify out the transformer running state according to the acquisition data. The test results show that the method proposed has an excellent performance on correct ratio. And it can overcome the disadvantage of the traditional three-ratio method which lacks of fault coding and no fault types in the existent coding. Combining the wireless communication technology with the monitoring technology, the designed and developed system can greatly improve the real-time and continuity for the transformer' condition monitoring and fault diagnosis.