Application of Deep Learning into Partial Discharge Source Discrimination

Deep learning Method has shown a very good performance in identifying and categorizing images in recent year, for this reason the application of this method has been evaluated in identifying partial discharge sources discrimination and identification. To do this a database, that consist of time-domain data of 5 different artificial PD models, are created. PD pulses are mapped into 2-dimensional (2-D)space with two methods: first, by application of S-transform and second, by application of continuous wavelet transform. 2-D mapped PD data is fed to convolution neural network (CNN). Feature extraction and classification is done by CNN network. The results show that deep learning method has excellent performance in identifying PD sources, so that using wavelet transform with Morse mother wavelet, its accuracy is more than 99%.

کلیدواژه ها:

convolutional neural network ، partial discharge ، pattern recognition ، deep learning