Evaluating performance of classifiers for supervisory protection using disturbance data from phasor measurement units

Abstract

This paper provides rationale for a supervisory protective system to improve security of power system using classification of PMU data. It evaluates the performance of four major classifiers to classify disturbance events residing within the disturbance data obtained from the Phasor Data Concentrator (PDC) owned by a local utility. These classifiers are Support Vector Machines (SVM), k-Nearest Neighbor Classifier, Naive Bayesian Classifier, and Recursive Partitioning and Regression Trees (RPART). Previous work by authors is used to obtain the targets (classes) for the classifiers. Performance of these classifiers is quantified in terms of accuracy and speed. Their suitability for real time classification to help create the supervisory protection system is discussed.