A Data-Driven Based Approach for Islanding Detection in Large-Scale Power Systems

This paper proposes a novel data-driven approach for detecting islanding in large-scale power systems, offering improved speed, accuracy, and robustness against missing data and measurement errors. The approach utilizes a mathematical combination of system quantities, including voltage, angle, and frequency measurements gathered from phasor measurement units (PMUs), for the identification of coherent areas. The identified groups, characterized using centers of inertia (COIs), are then connected to a distinct point referred to as the center of gravity (COG) through the fictitious reactances. These reactances, interpreted as electrical distances between COIs and COG, serve as valuable indicators for detecting islanding. More precisely, the occurrence of islanding can be identified by comparing the temporal trends of electrical distance variations across different areas. The effectiveness of the proposed methodologies is evaluated using simulated data from the NPCC system, the 73-bus IEEE test system, as well as actual measured data from two historical power system islanding events in a national power grid.