Font Size: a A A

Research On Partial Discharge Fault Diagnosis In GIS Based On UHF

Posted on:2014-01-13Degree:DoctorType:Dissertation
Country:ChinaCandidate:C X WangFull Text:PDF
GTID:1222330467484809Subject:High Voltage and Insulation Technology
Abstract/Summary:PDF Full Text Request
Partial discharge(PD) ultra high frequency(UHF) on-line monitoring technique is an important technique to evaluate the insulation degradation state of the GIS equipment. However, the presence of a large number of on-site interference affects the detection sensitivity and reliability, interference suppression is the key to online monitoring; the purpose of online monitoring is for equipment condition assessment, determining the running state of the equipment and detecting the fault and early warning is another key issue. Therefore, the anti-interference and multi-source discharge separation method, the fault diagnosis and early warning in GIS PD UHF detection technique are studied in this thesis, the main contents are as follows:In order to solve the problem of the superposition of impulse type interference and multi-source discharge, a multi-source discharge separation technology based on amplitude ratio clustering is proposed. For the difference of the propagation path and attenuation characteristics of UHF partial discharge signals,the ratio of the amplitude of sync pulse collected from two UHF sensors is extracted to separate the multi-source discharge signals by clustering method, laboratory and field test results demonstrate that this method can solve the superposition of multiple-pulse discharge. In order to further improve the accuracy of source separation and make up for the deficiency that the amplitude ratio clustering technology can not reflect the frequency domain features, a new UHF partial discharge signal conditioner with high sensitivity and wide dynamic range is developed based on dynamic frequency selection and envelope demodulation technology, on the basis of the conditioner, an intelligent technology for multi-source discharge separation based on frequency characteristic, and then an interactive dynamic clustering (IDC) algorithm is designed to separate the signals. Comparative test shows that the proposed method is much better than the traditional fuzzy clustering method(FCM), the separation accuracy is maintain more than90%. Using a single sensor can achieve the intelligent automatic separation of multi-source discharge signals and discharge pattern spectrum. Finally, the validity and practicability of the method is verified by the laboratory test.PD development process of five kinds of typical discharge failure in the GIS is studied by the test with boosting voltage method and the constant voltage method. The PRPD spectrogram, PRPS spectrogram and PD fingerprint information of different discharge stages are obtained. According to the different discharge type, this paper extracts the PD characteristics fingerprint which can effectively present the severity of partial discharge, forming a PD severity diagnostic fingerprint repository. On the basis of PD characteristic fingerprint monotone mutation change tendency, the K-means clustering and minimum distance principle is used to establish a classification rule of PD development stage, and explores the evolution mechanism of different PD discharge’s development process. Experimental result shows that, when discharge type is different, the probability of breakdown or flashover is different, and the fingerprint feature represented PD severity is also different. For metal protruson defect, the fingerprint characterization the severity are mainly discharge phase width (φw), the ratio of the discharge number of positive and negative half cycle(N+/N-), the entropy of the PD amplitude distribution (En(V)), the box dimension of⊿ui distribution spectra(⊿ui(DB)); For the floating electrode defect, the fingerprint characterization the severity are mainly the average discharge interval(⊿Tave), the box dimension and information dimension of positive and negative half cycle of PRPD; For the free metal particles defect, the fingerprint characterization the severity are mainly discharge phase width (φw), the correlation coefficient of the N-cp spectra (N-φcc), the correlation coefficient of the Vmax-cp spectra; For the metal electrode defect on GIS insulator, the fingerprint characterization the severity are mainly discharge phase width ((pw), the variance of dicharge frequency distribution(σ(N)), the entropy of dicharge frequency distribution (En(N)), average discharge voltage gradient sequence(μ.⊿u); For the surface discharge defect on insulator, he fingerprint characterization the severity are mainly discharge phase width (φw), the entropy of the largest PD amplitude distribution(En(Vmax)), the variance of dicharge amplitude distributionσ(Nv).The PD patterns library and characteristic fingerprints library are established based on a large number of laboratory test data and scene data. The characteristic parameters related to the phase and unrelated to the phase that can represent the types of PD are optimization extracred using criteria function based on the distance. A hierarchical PD identification method is proposed in this thesis, the test results show that the hierarchical identification method using the extracted characteristic parameters is superior to conventional methods, the recognition accuracy rate is increased from81.9%to98.3%, and solve the PD fault identification problem in the scene when the phase information can not be obtained.For PD fault early warning problem, using ARMA (autoregressive moving average) model predictive theory to predict the PD development trend of short-term is proposed in this thesis. Based on the test data of PD development process, modeling and predicting the development trend of characteristic of linear change, step change, nonlinear changes using ARMA model, prediction results show that:For the characteristic parameter of linear change, short-term prediction using ARMA model is feasible; for the characteristic parameter of step change, the prediction error can be kept within10%when the prediction time is less than30minutes; for the characteristic parameter of nonlinear change, the substantially change trend of PD can be basically predicted using ARMA model, but it is difficult to gaive an accurately prediction.Finally, based on the PD identification confidence, PD severity diagnosis confidence, PD risk degree, the level of failure probability, the consequences of failure and maintenance outage costs, the GIS PD fault diagnosis and early warning rules and mathematical model is established, and GIS PD fault intelligent diagnosis and early warning system is developed, test results show that the system is running stable, and the diagnosis and early warning results is accurate.
Keywords/Search Tags:GIS, Partial discharge, Ultra high frequency, Signal separation, Faultdiagnosis, Early warning
PDF Full Text Request
Related items