Analysis Of Transformer Winding Fault Diagnosis Based On BP Neural Network

Electricity is an essential life and production of energy, Is the driving forcefor social progress and economic development. Long distance transmission must usetransformer way, so in order to ensure grid safety, must make the transformer operating in asafe state. Previous studies have confirmed that the main position of transformerfault is transformer winding. If can not timely detect transformer winding deformation, itmay be short circuited winding burn, resulting in widespread power outages and brings theloss and inconvenience to the development of the national economy and people’s life. It hasimportant meaning to research the transformer winding deformation diagnosis method.At present there are three main kinds of transformer winding deformation detectionmethods: frequency response method, low voltage impulse method and short circuitimpedance method. Low voltage pulse method is very vulnerable to externalelectromagnetic interference, short circuit impedance method is a method for detectingbased on small current and low voltage, weak deformation is not easy to be detected, andthe detection time is longer. Compared with the above two, FRA detection has strongability of repeatability, anti-jamming, and detection equipment is light and simple, andfrequency response curve can describe the characteristics of the winding and reflect thesmall change.The object of this paper is the analysis of transformer winding fault diagnosis basedon BP neural network. It introduces the research background and significance oftransformer winding deformation, and various methods of detecting winding deformationwere compared, frequency response method has strong advantages. Research the influencethat common power system fault have on transformer winding deformation and stress inthese situations, the frequency response curve and the change of transformer windingdeformation are analyzed, according to the frequency response of the equivalent circuit, afrequency response simulation model is created by PSPICE software, and analyzes thechange of the parameters of equivalent model in the several typical faults, and finallyproposes a transformer winding fault diagnosis system based on neural network. Thesystem mainly consists of feature extraction and neural network training, firstly introducesthe feature extraction, mainly using the change of frequency response curve ofdisplacement amplitude and resonant point to find the characteristic value, extractingfeatures by wavelet transform; and then introduces the basic principle and the structure ofBP neural network, and BP neural network is built according to the characteristics oftransformer winding deformation diagnosis system, the feature vectors extracted fromfrequency response curve are used by the training and verification in BP neural network;finally, it introduces the software development environment, gives the softwarearchitecture of diagnosis system of transformer winding deformation, and theman-machine interface, database management module are introduced in this paper. Through the simulation experiment of transformer winding diagnosis system based on BPneural network, it can better contribute to the diagnosis of transformer windingdeformation type.