Study On Classification And Recognition Of Partial Discharge Defects For GIS Based On Ultra High Frequency Method

Gas insulated switchgear(GIS)has gradually become one of the important equipments in modern power systems due to its compact structure,high reliability and easy maintenance.However,The insulation performance of GIS devices will deteriorate during production and long-term operation and the most typical feature is Partial Discharge(PD).The partial discharge detection method based on ultra-high frequency(UHF)in the partial discharge detection technology of GIS device has the characteristics of good anti-interference performance and high sensitivity.This thesis mainly studies the problems related to PD detection and pattern recognition by UHF method.The specific work of this thesis is as follows:Firstly,this thesis mainly investigates the types and generation principles of PD in GIS devices,sets the internal defects of the four GIS devices to be tested and introduces the partial discharge experimental platform.And this thesis analyze the UHF PD signal characteristics corresponding to the four insulation defects,establish basic data for subsequent denoising and pattern recognition studies.Secondly,in order to ensure the reliability of the UHF sensor in the partial discharge test,this thesis evaluates the uncertainty of the UHF sensor.The IEEE recommended electric field strength calibration scheme is the main idea,and the UHF sensor calibration system with the electric field strength as the detection amount is adopted.And this thesis writes software programs to control related instruments.This thesis evaluates the uncertainty of UHF sensor by Monte Carlo method,and will provides some reference value for subsequent related work.Then this thesis introduces the empirical mode and the empirical wavelet algorithm,and sets the contrast signal of complex frequency components to carry out the signal decomposition experiments of the two.According to the decomposition and reconstruction signal results,the EWT algorithm can greatly reduce the decomposition of complex frequency signals.Avoid modal aliasing.The UHF PD signal denoising study based on the EWT-wavelet transform algorithm is studied.Finally,the UHF PD signal is easily interfered and the recognition accuracy is reduced.The improved empirical wavelet-wavelet denoising algorithm is proposed and the defect recognition model is established.The improved algorithm is applied to the denoising experiment of simulating UHF PD signals,and effective noise suppression is realized.The improved de-drying algorithm and wavelet denoising are applied to the measured signals at the same time.It is found that the paper algorithm has more advantages in noise suppression ratio.This thesis decomposes the above denoised signal into EWT and extracts the 13-dimensional features of high-order discrete quantities of the signal as the input layer of the random forest algorithm.Through the OOB out-of-bag error and training time in the random forest algorithm as the reference reference quantity,the number of random forest decision trees and random feature quantity are found.Finally,the best parameters are selected to identify the defect classification of UHF PD signals.The measured data shows that the classification recognition algorithm proposed in this thesis can achieve the correct classification rate of 97.5%,and compared with other algorithms to improve the defect classification recognition rate.