Synchrophasor based data driven approach for fault identification using multi-class support vector machine

Abstract

Traditional supervisory control and data acquisition (SCADA) system measurements suffer from low resolution (2-4 samples/sec). The proliferation of Phasor Measurement Units (PMUs) in power grid have enabled high resolution (25-120 samples/sec) monitoring. This has resulted in the explosion of data characterized by large volume, velocity and variety. Diversity of intra-class events may lead to bad performance by deterministic event detection schemes. To this, we propose a data driven technique based on multi-class Support Vector Machine (SVM) for fault classification. The proposed method identifies both symmetrical and unsymmetrical ground faults in the system utilizing only three input features based on voltage of phases A, B and C. A One Against One (OAO) and One Against All (OAA) SVM formulation is used for multi-class classification. The user-defined parameters C and γ are tuned using k-fold cross-validation and grid search technique. Comparison between deterministic and data driven fault classification scheme is also discussed in this work. © 2020 IEEE