Line Voltage Based Distance Relay Using a Multistage Convolutional Neural Network Classifier

Abstract

This paper presents a novel proposed method for the fault diagnosis in distance protection of transmission line, by which the real time voltage signal is the only required relay input. Unlike conventional protection schemes, the current signal is excepted without influencing the basic functions of the relay as a protecting and a monitoring device to detect and locate the fault. The new method is based on a pre-trained Convolutional Neural Network (CNN) with a combination of the higher order spectral estimations, which performs a deep learning classification with a very high accuracy. This research has succeeded in proposing an efficient 2D CNN model that takes the Short-Time Fourier Transform (STFT) of the signal for high accuracy fault detecting, locating, and classifying. The performance of the proposed models is tested using a new large dataset prepared using Simulink/ Matlab. The results show a high numerical performance evaluation that validates the consistency of the proposed methods.