Method for Identifying and Classifying Ultrasonic Image of Partial Discharge in Switchgear Based on Short Time Fourier Transform and Sparse Representation
Ultrasonic partial discharge detection, as a non-contact detection method, has the advantages of strong electromagnetic interference resistance. In this paper, the different types of partial discharge ultrasonic signals in the switchgear are collected, the time frequency graph of the ultrasonic signal is obtained by short time Fourier transform, and the time frequency graph is classified by the sparse representation algorithm. It can quickly and accurately determine whether the discharge occurs and determine which of the following discharge types are: sphere plate discharge, cylinder plate discharge, cone plate discharge, needle plate discharge. In the process of using sparse representation method, two methods of orthogonal matching pursuit and accelerating neighbor gradient method are used to solve the sparse solution. The experiments show that the two methods are suitable for different situations. Compared with the traditional multi classification support vector machine (SVM) method, it has higher accuracy, robustness and stability. It is proved that the sparse representation method is more suitable to solve the problem of partial discharge recognition and classification.