Pattern Recognition Of Transformer Partial Discharge Based On Fuzzy ART Neural Network Research

As the most important part of the power system, and the protection of the reliability ofthe power grid, once the transformer failure, usually caused by the collapse of the powersystem and serious economic losses. The data indicate that the main reason of the transformerfault is the deterioration of the insulating properties, so in order to improve the reliability ofthe insulation, all countries in the development of state of repair that can real-time onlinemonitoring and diagnosis In the operating state of the transformer, partial discharge patternrecognition is one of the current research focus. However, because of the complexity of thetransformer structure,the uncertainty of the partial discharge, a lot of electromagneticinterference in the running scene,and immaturity of the recognition algorithm, these all gavepartial discharge caused a lot of difficulties at detect, locate and identify classification. Thisarticle on the basis of a lot of domestic and foreign literatures about transformer partialdischarge, pattern recognition, artificial neural networks, do some research as follows:This article introduced the basic concepts and the structure of the pattern recognitionsystem, and analysis of several important issues in pattern recognition system from theclassifier design specially, included the design of the feature space, the classifier designstandards and basic methods, the selection of the discriminant function and the determinationof the parameters, and system of training and learning, these are the theoretical basis of thisarticle.This article analyzed the various sources of interference signal included in thetransformer partial discharge signals, summarizes the various methods of interferencesuppression commonly used. Then according to the relevant literature, discussesed four typesof discharge characteristics commonly used in partial discharge pattern recognition:3D listdata characteristics, statistical characteristics of the partial discharge, the image gray momentfeature and time-frequency characteristics, discussed the characteristic quantities separabilitycriteria.This article discussesed the structure and working principle of the the ATR1neuralnetwork, introduced the origin, development and application of the fuzzy theory, and analyzedthe of structure and algorithm of the fuzzy ART neural network.The fuzzy ART neuralnetwork combined the characteristics and operation characteristics of the fuzzy sets to theART1neural network, make the the input of the ART1network fuzzified, so can computeanalog, make up for defects of the input of the ART1network only for binary. This is aself-organizing neural network, can real-time unsupervised learning, and suitable for thepattern recognition. This article drawed on five different experiments of partial discharge model from theprevious literature, and got the three-dimensional spectra of each discharge model, and makethe list data characteristic quantities from the spectrum as input of the neural network, andsimulated with the fuzzy ART neural network and BP neural network in MATLAB.Experimental results show that the ART Neural network has obvious advantages inrecognition rate, recognition speed and stability relative to BP neural network, it has goodpractical value and application prospects in the field of partial discharge pattern recognition.