Diagnosis of the single phase-to-ground fault in distribution network based on feature extraction and transformation from the waveforms

作     者：Shi, Fang Zhang, Linlin Zhang, Hengxu Xu, Kai Vladimir, Terzija 

作者机构：Shandong Univ Minist Educ Key Lab Power Syst Intelligent Dispatch & Control 17923 Jingshi Rd Jinan Shandong Peoples R China State Grid Jinan Power Supply Co Jinan Shandong Peoples R China Univ Manchester Sch Elect & Elect Engn Manchester Lancs England 

出 版 物：《IET GENERATION TRANSMISSION & DISTRIBUTION》 (IET发电，输电与配电)

年 卷 期：2020年第14卷第25期

页      面：6079-6086页

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 

基　　金：National Natural Science Foundation of China National Key R&D Program of China [2017YFB0902800] 

主　　题：feature extraction power distribution faults fault diagnosis earthing distributed power generation fault location eigenvalues and eigenfunctions pattern clustering fault currents time series feature extraction distributed generators power system fault diagnosis Peterson coil fault current single phase-to-ground fault section identification method eigenvalues time-sequenced features fault features synchronous current waveforms fault recorders topology related fault feature matrix time-series features random matrix theory distribution characteristics fault cases multifeeder distribution network improved K-means clustering algorithm 

摘      要：The increasing permeation of the distributed generators in the power system brings great challenges for fault diagnosis, especially for the distribution networks with ungrounded neutral or grounded by Peterson coil as the fault current is limited and easily affected by the noises and interferences. A single phase-to-ground fault section identification method is proposed based on feature extraction of the synchronous waveforms and the calculation of the eigenvalues for the time-sequenced features. First, several fault features are defined and extracted from the synchronous current waveforms obtained by the fault recorders. Then, the topology related fault feature matrix is constructed using the time-series features obtained from different measurement sites, and the eigenvalues of the matrix are calculated based on the random matrix theory. Lastly, using the distribution characteristics of the eigenvalues, improved K-means clustering algorithm is utilised in classifying the fault cases and identifying the faulty sections. The effectiveness of the proposed scheme is verified by IEEE 34 nodes test system and a multi-feeder distribution network.