Partial Discharge Feature Extraction And Pattern Recognition Of Cable Terminals

With the progress of society and the continuous improvement of-the people’s living standards,the demand for electrical energy of the country is also increasing,which puts forward higher requirements on the quality of power operations.The operation of China’s urban power grid is currently on the road of comprehensive high-voltage cable.Once a fault occurs in the line,it will lead to serious blackout accident and seriously threaten the lives of the people and property safety.Most of faults of the power system lines are caused by defects of the cable attachments.Therefore,it is necessary to study the fault diagnosis technology of power cable accessories in order to realize the accurate judgment of the faults of the cable accessories,and then take effective measures to avoid insulation accidents.Partial Discharge(PD)is one of the main manifestations of early and sudden failures of cable insulation faults.It is not only the main cause of further degradation of cable insulation,but also the main feature of cable insulation.The research on the pattern recognition and fault diagnosis of cable PD can timely and accurately determine the internal insulation status and defect types of cables,which is of great significance to prevent cable line accidents and ensure the safe and stable operation of the power system.Therefore,on the basis of the existing work at home and abroad,this paper studies the extraction methods of partial discharge pattern recognition feature and the related work of reduction dimension processing of feature through a large number of laboratory partial discharge experiments and data collection and analysis.Through the related study of the method of feature extraction method and feature reduction dimension,it provides a certain basis for partial discharge pattern recognition.The main work and conclusions of this paper are as follows:In this paper,three methods are used to extract the partial discharge characteristics of the partial discharge samples obtained from the experiment.The discharge characteristics of two-dimensional spectral of partial discharge,the moment features of partial discharge grayscale images,and the statistical characteristic parameters of the discharge map are extracted.This paper proposes to "Regionalization" of the spectrum,that is,divide the power frequency cycle into 24 regions according to equal angles,and extract the features of each region for pattern recognition.The purpose is to identify the partial discharge spectra without losing its phase and amplitude information,and it facilitates the "focus on areas" of the spectrum in subsequent work.This not only improves the speed of recognition,but also improves the accuracy.At the same time,the gray scale spectrum and moment features and statistical feature parameters were studied and feature parameters were extracted.Then,based on this,three methods(Principal Component Analysis,Multidimensional Scaling,and Locally Linear Embedding)are used to optimize the feature quantities,and the dimension space is reduced.The feature space was formed by optimized feature quantities,and the training was performed by BP artificial neural network and support vector machine to form an effective classifier,and the test samples were tested.The experimental results show that the proposed method can extract better feature parameters and optimize it to obtain the simplest feature space through dimensionality reduction.The BP neural network and support vector machine are applied to the recognition of the defect type to compare the accuracy of the recognition.The results show that all can achieve better recognition rate.