Feature Extraction of UHF PD Signals Based on Diode Envelope Detection and Linear Discriminant Analysis

Ultra-high frequency (UHF) partial discharge (PD) detection method is an effective method for insulation defect diagnosis in Gas Insulated Switchgear (GIS). The time-resolved partial discharge (TRPD) method has the advantage of small amount of data compared the phase-resolved partial discharge (PRPD) method. However, when applied to on line monitoring, there are still disadvantage of slow processing and lack of methods for real-time analysis of signals. In this paper, a time domain feature extraction method of UHF PD signals based on diode envelope detection and linear discriminant analysis is proposed, which can reduce the data volume and improve the processing speed while ensuring the diagnostic accuracy. First, the envelope of the UHF time domain signal is extracted by the diode envelope detection method, and the smooth envelope signal is obtained by the low pass filter. Then, six dimensionless eigenvalues of envelope signal are extracted and dimensionality reduction is made by linear discriminant analysis (LDA). Finally, classifier is used to classify the features after dimensionality reduction. The results show that compared with the traditional Hilbert algorithm extraction envelope, the computational complexity of this method is fast, and the six dimensionless eigenvalues combined with LDA can achieve better diagnostic results.