Classification of UHF Partial Discharge in GIS Using Optimal Complex Wavelet-based Features

Abstract

In the field of electrical power engineering, pattern recognition is commonly used for classification of defects caused of Partial Discharge (PD) in Gas Insulated Substation (GIS) in order to assess the insulated system in time. As phase information can't be acquired effectively using real wavelet transform for extracting features because of only real coefficients being existed, in this paper, the method of complex wavelet transform (CWT) for decomposing ultra-high frequency (UHF) PD signals is proposed; complex coefficient is constructed by combining the real part and imaginary part of complex wavelet coefficients which can describe PD signal integrally; and the features can be extracted from complex coefficients of each scale; additionally, optimal complex coefficients are selected by comparing Jsum criterion; Finally, PD samples are acquired through large number of experiments, and BP neural network is used to identify the PD sourced by four typical insulated defects effectively. The results show that the features extracted from the optimal compound coefficient corresponding to the best complex wavelet have wonderful results. I INTRODUCTIONGIS has developed very quickly and has been in use all over the world since the early seventies [1]. It is well known that insulation breakdown is often preceded by PD activities, so the detection of PD is very important to early detection of insulation defects and increases the reliability of the power supply [2][3][4]. Wide band PD measurement in the UHF range offers a better Signal to Noise Radio (SNR) than the conventional method, which has been in use widely.In the past, there were three different categories of PD pulse data patterns such as phase-resolved data, time-resolved data, and frequency-resolved data [5]. With the fast development of the computer technology, wavelet transform have attracted much attention in PD pattern studies. As phase information can't be acquired effectively using real wavelet transform for extracting features because of only real coefficients being existed, in this paper, the method of CWT for decomposing UHF PD signals is proposed, and complex coefficient n I R is constructed by combining the real part and imaginary part of complex wavelet coefficients which can describe PD signal integrally; and the features can be extracted from complex coefficients of each scale using cluster analysis; additionally, optimal complex coefficients are selected by comparing J sum criterion; Finally, PD samples are acquired through large number of experiments, and BP neural network based on chaos and tabu search optimization algorithm is used to identify the PD sourced by four typical insulated defects effectively. The results show that the features extracted from the optimal compound coefficient corresponding to the best complex wavelet have wonderful results. II PD FEATURES BASED ON COMPLEX COEFFICIENT A. Construction of Complex CoefficientsAt present, the method for extracting features from wavelet coefficients mainly includes parts of wavelet coefficient...

Intelligent Monitoring System Based on Optical Fiber Acoustic Emission Sensor and Its Application in Partial Discharge Diagnosis of Gas-insulated Switchgear

“…Then, seven time domain features and 18 frequency domain features were extracted and respectively standardized in the time and frequency domains as shown in Table 1. (17)(18)(19)(20)(21) Finally, the selected features were used to train a three-layer BP-ANN to construct a defect recognition model as shown in Fig. 5.…”