Fault Classification in Transmission Lines Using Wavelet and CNN

This paper proposes a novel technique using the CNN (Convolution Neural Network) architecture (mainly the VGG-16 and VGG-19) for the fault detection and classification with the implementation of the DWT (db-8) and CWT filter banks. This technique has been implemented for the fault current in three phases detailed decomposition at six levels in a 600 MVA transmission line for 200 km respectively for different distance and fault inception angle. The fault current that is found out using the transmission line in three phases for different time at different distance and fault inception angle is detailed coefficient decomposed using the db-8 wavelet at 12.5kHz sampling frequency for six levels which is then computed by the CWT filter banks to create the time frequency representation which are called as spectrogram. After that we train those images using the VGG-16 and VGG-19 architecture for classifying different faults and comparing the two architecture. This method is quite robust, and validation is done for a large number of data in a minimal amount of time as compared to the ANN architecture.