GIS Insulation Defect Diagnosis Method Based on Improved MFCC and PCA-SVM Model

In order to improve the accuracy of GIS partial discharge pattern recognition, this paper proposes an improved MFCC and PCA-SVM voiceprint recognition model based on ultrasonic signals. Firstly, pre-process the ultrasonic signal according to the characteristics of GIS partial discharge; secondly, calculate the standard MFCC parameters, and calculate the first and second difference accordingly, and combine the dynamic and static parameters as the voiceprint characteristic quantity; finally establish the PCA-SVM classifier performs dimensionality reduction and recognition on the feature quantity. The test was conducted on a 110kV GIS, and the GIS partial discharge ultrasonic signals of four insulation defects of metal particles, floating potential, creeping discharge and corona discharge were collected for testing. The test results show that the MFCC feature vector after adding the first-order and second-order difference features can accurately and effectively reflect the ultrasonic signal characteristics under different insulation defects of the GIS, and the PCA-SVM algorithm has a high recognition and diagnosis success rate and a high recognition speed. The test proves that the diagnosis model proposed in this paper can effectively improve the efficiency of GIS defect recognition.