Power System Disturbance Classification Using CWT Guided Customized AlexNet CNN

This letter aims at designing a convolution neural network (CNN)-based classifier that detects and classifies 13 different types of power system events causing poor power quality. Early and accurate detection of those classes of disturbances helps power engineers take quick and suitable measures to protect power system components and to supply uninterrupted power to the consumers. The whole process comprises of the sensing of disturbance creating current signals from experimental setup, filtration by unsupervised
$k$
-mean machine learning to remove noisy data, Big Data generation, architectural modification of AlexNet CNN, time–frequency informative image creation by continuous wavelet transform followed by convolution using kernel filters, and classification and performance evaluation. The proposed customized CNN has been compared with other benchmark CNN architectures. The method has shown 98.46
${\%}$
accuracy with minimum computational time.