Research of Feature Extraction and Recognition of Partial Discharge Based on the Probabilistic Neural Network

Wei Lin,Xin Wang,Yihui Zheng,Lixue Li
DOI: https://doi.org/10.3969/j.issn.1000-3886.2018.03.018
2018-01-01
Abstract:To improve the accuracy and training speed of classification recognition of partial discharge (PD ) signal of the gas insulated switchgear (GIS),on the basis of probabilistic neural network (PNN ),a method was proposed for classification of GIS partial discharge signal as well as unified characteristic parameters expressing signal shape and statistical characteristics.Firstly,the ultra high frequency (UHF)method was used to obtain PD signal generated by GIS PD signal simulator as data set to plot a scatter diagram,and characteristic parameters describing image properties completely and accurately were chosen to improve data discrimination.Then,characteristic parameters were extracted and dimension reduction was completed to avoid low efficiency due to excessive dimensionality.Data sets were divided into training set and test set at random.Finally,PNN was used to complete data training and test,thus overcoming low recognition training speed,likelihood to fall into local minimum and other shortcomings of traditional modes such as back propagation (BP)neural network and support vector machine (SVM).Results of experiments and comparison showed that the proposed approach had high accuracy and the accuracy of the proposed unified parameter classification was much better than that of the traditional method.
What problem does this paper attempt to address?