When a high impedance fault occurs in a distribution network,traditional protection means cannot be effective due to the small fault current.When emerging detection methods perform feature extraction,artificially constructed feature quantities or feature vectors often have difficulty in fully representing the characteristics that distinguish high impedance fault from other events.To this end,this paper first converts zero-sequence currents into time-frequency diagrams using continuous wavelet transform(CWT),and then segments the images into square slices.A convolutional neural network(CNN)is used to identify the characteristic"tooth"feature of high impedance fault.The proposed method is able to visualize high-frequency components of the waveform by means of the time-frequency diagram,and its reliability has been verified in simulations and field sample tests.
关键词
配电网/高阻接地故障/时频谱图/图像分割/深度学习/卷积神经网络
Key words
distribution network/high impedance fault/time-frequency diagram/image segmentation/deep learning/convolutional neural network
维普
万方数据