Feature Extraction for Peak-time Series of Partial Discharge at HVDC
Author(s): SI Wen-rong, FU Chen-zhao, HUANG Hua, LI Jun-hao, LI Yan-ming, Electric Power Research Institute, SMEPC, State Key Laboratory of Electrical Insulation and Power Equipment, Xi’an Jiaotong University
Pages: 17-24
Year: 2013 Issue: 11
Journal: High Voltage Apparatus
Keyword: HVDC; partial discharge; peak-time series; feature extraction; GA-BP;
Abstract: Three basic defect models of oil-paper insulation including corona, bounded cavity and surface discharge,and a typical discharge model of corona in air are described after the DC PD test and measurement system had been shown. The PD measurement system has a special anti-noise technique which resorts to the characteristic of pulse waveshape signals in time and frequency domain to achieve fast separation of random pulsed noise signals. Based on those works done above, the parameter Delta(t)is introduced for DCPD to form a new analysis approach, which perform the same function as the popular method named phase resolved partial discharge(PRPD)histograms of PD at AC voltage. Thirty-six fingerprints of DCPD are gained resorting to application of the standard statistic operators to some selected time and amplitude resolved partial discharge(TARPD)histograms. Then, feature selection are achieved by using a BP Neural Network optimized by genetic algorithm(GA-BP), while the result is that only twenty optimized fingerprints are reserved.
Pages: 17-24
Year: 2013 Issue: 11
Journal: High Voltage Apparatus
Keyword: HVDC; partial discharge; peak-time series; feature extraction; GA-BP;
Abstract: Three basic defect models of oil-paper insulation including corona, bounded cavity and surface discharge,and a typical discharge model of corona in air are described after the DC PD test and measurement system had been shown. The PD measurement system has a special anti-noise technique which resorts to the characteristic of pulse waveshape signals in time and frequency domain to achieve fast separation of random pulsed noise signals. Based on those works done above, the parameter Delta(t)is introduced for DCPD to form a new analysis approach, which perform the same function as the popular method named phase resolved partial discharge(PRPD)histograms of PD at AC voltage. Thirty-six fingerprints of DCPD are gained resorting to application of the standard statistic operators to some selected time and amplitude resolved partial discharge(TARPD)histograms. Then, feature selection are achieved by using a BP Neural Network optimized by genetic algorithm(GA-BP), while the result is that only twenty optimized fingerprints are reserved.
Related Articles