Application Research On Partial Discharge Recognition Of High Voltage Equipment Based On Wavelet Sparse Representation And Deep Learning

GIS is one of the most important equipment in power equipment,and its partial discharge phenomenon is a common internal fault,which is an early sign of insulation deterioration.By monitoring the operation status of GIS equipment and collecting the partial discharge signal,image and position,the measured signal can be classified,analyzed,feature extracted,judged,identified and processed,and the type,position and development of insulation defects can be estimated,so as to effectively prevent the recurrence of partial discharge.Therefore,the purpose of partial discharge identification is to determine the location,type and damage degree of defects,which has a very key application value in practice.The traditional PD identification method usually uses the practical engineering experience and theoretical knowledge of power experts for judgment,which has great limitations.With the rapid progress of network technology and signal processing technology,more and more new recognition methods are used in partial discharge judgment,which have their own advantages and disadvantages.For example,sparse representation algorithm only needs a few basis functions to represent the decomposed signal,which makes the information easier to process;Deep learning technology shows different application advantages from traditional methods through the rapid classification of data.Because the image recognition technology is similar in nature,this paper mainly studies the introduction of sparse representation and deep learning method for partial discharge pattern recognition.The partial discharge of GIS equipment is simulated,and the defect model and training sample library are established for recognition research.The image data of four kinds of GIS discharge defects are collected respectively,and the typical sample library is formed as the basis for subsequent training,testing and research.Depending on the sample library,firstly,the sparse representation algorithm is used to identify the test samples of various discharge types,and the recognition rate is obtained.Then,the deep learning technology is used to identify and judge.Finally,the performance differences of sparse representation,deep learning and sparse representation algorithm based on deep learning feature are compared.The results show that the effect of joint recognition is better than that of single recognition technology,the overall recognition rate is higher and the error rate is lower.Therefore,the sparse representation algorithm based on deep learning features has better recognition effect than other recognition methods because of its improved recognition mechanism.This paper has 35 figures,8 tables and 139 references.