Steady-state security assessment based on online learning k-nearest neighbor classifier

Abstract

A k-nearest neighbor classifier with online learning procedure for steady-state security assessment is introduced. A dynamic sample set and the related sample editing strategies are proposed. The dynamic samples can keep tracking the operation status of power system to minimize classification error. It is implemented through editing the dynamic samples according to their online performances. The classification result of the real time data is checked with the result of traditional N−1 contingency scan periodically. Misclassified data are appended as a dynamic sample to improve the accuracy of the classifier. A performance value is assigned to each sample. It is updated every time the classifier is used. The sample with the least performance value is removed whenever a new misclassified sample is appended in order to keep the dynamic sample set in a reasonable size. A Case study is performed on IEEE-30 system. The result shows an improvement in the performance of the classifier.