Abstract
Medium Voltage Distribution systems are susceptible to high impedance faults (HIFs) and incipient faults (IFs). The HIFs are permanent, not detectable by Conventional relays, and caused by an energized conductor that in contact with ground surfaces. The IFs occur to underground cables, which are self-cleared, and may develop to a permanent fault if they kept unchecked. This paper proposes an automated detection and classification technique based on the Morlet Wavelet and the transfer learning technique via AlexNet Convolutional Neural Network (CNN). At first, these faults, as well as load switching and capacitor energization, were created on the IEEE-13 bus system. These events were detected using the proposed technique via the Morlet Wavelet. Then, the detected events are converted to scalogram RGB images with Continuous Wavelet Transform (CWT). These images were used as a set of training data for the transferred AlexNet CNN. After that, to have our testing images for the transferred AlexNet, a new set of events are created on the IEEE-34 bus system where they are detected using the same proposed technique and converted to RGB images. The discoveries were very promising where the proposed technique had a 100% detection accuracy, and the classification process had a 100% classification accuracy for HIFs, IFs and load switching. The classification accuracy for the capacitor energization were 94% which can be improved dramatically if the proposed technique was linked to a simple logic algorithm with the timing schedule of capacitor energization.