A comparative study on artificial intelligence-based methods for fault detection, classification, and localization in distribution lines

Abstract

In recent years, supervised machine learning (SML) has demonstrated its effectiveness in pattern recognition and outcome prediction within datasets. The objective of this research is to develop an algorithm that utilizes SML classification and regression equation capabilities to accurately classify and locate faults occurring in electricity distribution lines. The proposed algorithm takes the measured values of electrical current and voltage at one end of the distribution line as input data and outputs the type of fault. The algorithm evaluates its performance by simulating the IEEE 14-bus power system using MATLAB and generating various types of faults at different locations and with different fault resistances to create a comprehensive fault database. The algorithm can employ various types of SML techniques and approaches, including Linear Discriminant Analysis (LDA), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and the Least Mean Squares (LMS) regression method, to compare their abilities in classifying fault types and identifying fault locations. Additionally, the study investigates the system's vulnerability to variables such as uncertainty in transformer instrument measurements and the presence of generator or transmission line outages in the power system.