XLPE Cable Partial Discharge Electromagnetic Coupled Detection And Its Pattern Recognition Research

With the development of power industry and the increment of power installation,XLPE power cable is more and more widely utilized in grid construction andreconstruction. To conform the safety of transmission network, the detection techniqueof XLPE power cable is studied extensively by the experts at home and abroad,especially for XLPE power cable partial discharge (PD) on-line monitoring which is oneof hot hit researches. Based on the present researches, a feasible system for XLPEpower cable PD detection is proposed, and a pattern recognition method on account ofsupport vector machine (SVM) for XLPE power cable PD is put forward.Firstly the influence between the Rogowski type coupler frequency characteristicand its parameters is synthetically analyzed in this paper. According to the calculationand simulation study, a suitable sensor for XLPE power cable PD detection is designed.Then, utilizing the designed current coupler, the detection system of XLPE power cablePD is setup at the platform of LabVIEW. Through the verification of simulation test, itcan be concluded that the detection system designed in this paper could be effectivelyapplied in XLPE power cable PD on-line monitoring.As various interferences existed in the field, especially to periodical narrowinterference and white-noise, comparison to various kinds of de-noising method, anadaptive threshold wavelet packet algorithm is studied to suppress periodical narrowinterference and white-noise, and to improve signal noise ratio (SNR). As the differentdefects have different XLPE power cable PD waveform, four kinds of defects are madein the health XLPE power cable, and the PD signals are sampled in the simulation testthrough the detection system. Based on the defects, five kinds of statistic spectrum aresetup, and 28 statistic characteristic features are calculated for each kind of PD defect.SVM is based on the statistical learning theory and is a new machine learningmethod, and has perfect characteristic in many application domains. As a new-brandpattern recognition technique, on account of practical needs, the suitable classificationalgorithm of SVM is designed in this paper. Through plenty of trainings, a new sampledefect is used for classification. The accurate classification result for the new samplemanifests that PD pattern recognition based on SVM is feasible and effective.