电晕Application of the neural network to detecting corona discharge occuring in power cables.pdfVIP

Application of the Neural Network to Detecting Corona Discharge Occurring in Power Cables T. Hara A. Itoh 1i.Yatsuka I. Kishi I.Hirotsn Dept. of Electrical Engineering Suiiiitonio Elect,ric Iiidust.ries,Ltd. Kyoto University Osalia, 554, Japti.11 Kyoto, 606, Japan Abstract waveforms in real systeiii, it L difficult for the iieiiral network to learn all of those Iioise patterns. Thcrti- In this paper, the system of detecting corona dis- fore, we newly made the neural iietwork incliidiiig charges autoniatically with an artificial neural net- only 1neuron in tlie output layer (we call it “1out- work is examined and a network which can distiii- put model”) and made the iietworli learn only coroiia guish between corona and noise patterns occurring in waveforms. We exaniiiied whether this network c.oiiltl power cables is investigated. A feed-forward type of judge corona waveform “coroiia” aiitl juclgr all the a neural network with three layers, i.e., input, hidden other “noise”. and output layers is used. It is found that. the network Tlie methods of inputting frequency spectra of tlic which learns only corona and no noise patterns does waveforms obtained by using FFT (fast foiirier trans- not show a good performance. This iiieans that the form) into the network were also examined. Tlie rea- network should learn both corona aiid noise patterns son why we apply the frequency spectrum da