| The rapid development of strong smart grid put forward higher requirements onthe security and reliability of power transformers. Partial Discharge (PD) is the maincause of deterioration of insulation materials and the manifestation of early, onlinemonitoring of PD is important for transformer fault diagnosis and real-time conditionassessment. Digital de-noising and feature extraction are the keys of onlinemonitoring of PD. This paper focuses on methods of digital de-noising and featureextraction of PD, and the main works are as follows:First of all, according to CIGRE MethodⅡand the common defects of powertransformer insulation structure, three kinds of defect models and correspondingelectrode system was devised to stimulate corona discharge, surface discharge, cavitydischarge. Four types of PD experiments are performed in laboratory, the use of pulsecurrent method to measurement the PD signal, and get a large number of experimentaldata.Secondly, on the basis of detailed analysis the EMD and autocorrelationfunctions. Improve the spatial and temporal de-noising method using the significantdifferences of PD’s and white noise’s autocorrelation function. Aiming at the inherentdisadvantages of spatial and temporal de-noising method, improved de-noisingmethod based on EMD are raised, which combined with improve spatial and temporalde-noising method and threshold de-noising method. The results show that, theimproved de-noising approach based on EMD is capable of obtaining low waveformdistortion and higher signal-to-noise ratio of de-noised signals than waveletde-noising approach.Finally, based on the analysis of the Singular Value Decomposition (SVD),feature extraction method of EMD-SVD was presented. Aiming at four typical PDsingals, singular value and tabulated data are extracted respectively. BP artificalnetwork is used to classify diffrernt kinds of discharge, result show that the singularvalue has high pattern recognition accuracy with a small number of features and lesstime. |
PDF Full Text Request