FAULT CLASSIFICATION AND LOCATION IN SERIES COMPENSATED LINE USING SVM

Abstract

It has been a difficult task to classify and locate fault in series compensated lines because of nonlinear operation of protective equipments installed with it such as Metal Oxide Varistor (MOV). Therefore new protection approaches are required. This dissertation presents a novel approach for the protection of Thyristor-Controlled Series Compensator (TCSC) line using support vector machine (SVM). The presented work deals with the problem of Fault Classification and Fault Location (Section Identification and Distance Protection) using SVM. Two SVMs are trained to provide fault classification, and section identification respectively. And two Support Vector Regression (SVR) models, one for each section, are also trained for the purpose of distance protection. The proposed method takes the advantage of a more informative feature set for classification, resulting in better speed and accuracy. Time taken for testing is significantly reduced and results are perfectly accurate which makes it suitable for real time application for protection of the transmission line with TCSC. Moreover, it also relieves the system from extra components such as- zero-sequence analyzer. The proposed scheme was tested on 10000 data instances with a very wide variation in system conditions such as- compensation level, source impedance, location of fault, fault inception angle, load angle at source bus and fault resistance. The model gives an accuracy of almost 100% for fault classification which is not reported in the literature so far (precisely 1 misclassification out of 10000 testing data instances). Experiments are also conducted and results have shown the excellent performance improvement of the proposed methodology over state-of art methods even with a very small amount of training data. Overall the proposed methodology provides a robust, highly accurate, economical and fast method for. Fault Classification and Section Identification in TCSC compensated EHV transmission lines. The proposed distance protection model has never been tried before and it also shows superior results than previous attempts. A correlation score of 0.97112 was achieved using the proposed regression model.