The server is under maintenance between 08:00 to 12:00 (GMT+08:00), and please visit later.
We apologize for any inconvenience caused
Login  | Sign Up  |  Oriprobe Inc. Feed
China/Asia On Demand
Journal Articles
Bookmark and Share
Time-Frequency Features Extraction and Clustering Analysis of Partial Discharge UHF Pulses
Pages: 211-219
Year: Issue:  2
Journal: Transactions of China Electrotechnical Society

Keyword:  Partial dischargepulses clusteringS transformnon-negative matrix factorizationfuzzy C means;
Abstract: In this paper, four types of artificial defect models are used to generate typical PD sources, and laboratory PD experiments are performed to measure the UHF pulses. S transform(ST) is introduced to present a time-frequency analysis of partial discharge(PD) UHF pulses to explore PD sources separation. In the proposed method, ST is firstly applied to the UHF pulses of PD, and non-negative matrix factorization(NMF) is utilized to decompose the S transform amplitude matrix into a number of base vectors in frequency domain and location vectors in time domain. Then, a few parameters such as sharpness, sum of derivative, information entropy and sparsity are extracted from various base and location vectors, forming a PD feature space which can fully reflect the timefrequency information of the UHF Pulses. Finally, a fuzzy C means(FCM) algorithm is used to perform the clustering of extracted feature vectors. Clustering results of the experimental samples show that the UHF pulses from different sources can be effectively separated by the extracted time-frequency feature vectors. The highest successful clustering rate of 90.33% is obtained by the 10-dimensional feature vectors when NMF parameter r =2. In addition, ST can make better clustering accuracies than traditional Wigner-Ville distribution(WVD). The clustering performances of the proposed timefrequency features would be dramatically influenced when multiple refraction and reflection exists, and further investigations should be presented.
Related Articles
No related articles found