Investigation on the effectiveness of classifying the voltage sag using support vector machine

Abstract

Support Vector Machine (SVM), which is based on Statistical Learning theory, is a universal machine learning method. This paper proposes the application of SVM in classifying the causes of voltage sag in power distribution system. Voltage sag is among the major power quality disturbances that can cause substantial loss of product and also can attribute to malfunctions, instabilities and shorter lifetime of the load. Voltage sag can be caused by fault in power system, starting of induction motor and transformer energizing. An IEEE 30 bus system is modeled using the PSCAD software to generate the data for different type of voltage sag namely, caused by fault and starting of induction motor. Feature extraction using the wavelet transformation for the SVM input has been performed prior to the classification of the voltage sag cause. Two kernels functions are used namely Radial Basis Function (RBF) and Polynomial function. The minimum and maximum of the wavelet energy are used as the input to the SVM and analysis on the performance of these two kernels are presented. In this paper, it has been found that the Polynomial kernel performed better as compared to the RBF in classifying the cause of voltage sag in power system.