Variable-Length Event Classification using PMU Data with Naïve Bayes

Abstract

Increasing levels of non-synchronous generation prompted by global emissions targets has resulted in power systems with low inertia. This has led to changing system dynamics and evolving trends in system events which are difficult to classify through traditional means. Many countries have invested in Phasor Measurement Units (PMUs) to monitor these systems over large geographical areas which form Wide Area Monitoring Systems. Due to the increased use and improved technology of PMUs this has generated vast quantities of data for system operators to process. Automatic methods for event diagnosis are required due to the complexity of system events, including variable event lengths. This paper demonstrates an approach for the widearea classification of a number of power system events. Event sequencing is used to solve the variability of event lengths. Sequential feature selection is adopted on wide-area synchronized frequency, phase angle and voltage measurements to extract the optimal features. Successful event classification is obtained by employing a Naïve Bayes classifier on the features. The reliability of this method is evaluated using simulated case studies and benchmarked against various sequence lengths.