Machine Learning-Based Approaches for Transmission Line Fault Detection Using Synchrophasor Measurements in a Smart Grid

Abstract

Modern power technology represents the epitome of engineering achievement, comprising a remarkable interconnected network of elements spanning vast regions. Transmission lines are vital in facilitating power transfer across extensive distances worldwide. The continuous operation and reliability of these transmission lines heavily rely on their effective monitoring and fault mitigation capabilities. Various factors, including natural disasters and other causes, can give rise to faults in transmission lines, which impede the seamless delivery of power. Timely identification and resolution of such faults are paramount to avoid service disruptions and mitigate the risk of cascading blackouts. Phasor Measurement Units (PMUs) have emerged as indispensable devices for monitoring and analyzing transmission lines, offering a dynamic perspective of their behavior due to their high reporting rate. PMUs enable operators to monitor power flow at different locations within the grid, thereby aiding in maintaining system stability and optimizing grid efficiency. These capabilities are essential in realizing the objectives of the smart grid 3.0 paradigm. Swift restoration of transmission line functionality necessitates the rapid detection, classification, and clearance of faults. Digital signal processing algorithms and machine learning techniques have emerged as critical tools in achieving these objectives efficiently. The advent of numerous machine learning algorithms, coupled with their real-time implementation capabilities, has empowered their robust deployment for fault detection and classification in physical transmission lines. This chapter presents the real-time implementation of the machine learning algorithms on a physical laboratory 200 km transmission line. Additionally, it compares the effectiveness of machine learning methods like K-Nearest Neighbour, Support Vector Machine, and Logistic Regression.