Classification method of power quality disturbances based on RVM

Abstract

A new method classifying power quality disturbances based on relevance vector machine (RVM) and wavelet packet decomposition is presented. Power quality disturbances are decomposed by wavelet packet, and the energy of the wavelet packet coefficients of each end node are extracted as eigenvectors. The disturbances are classified using RVM based on hierarchical categorization and minimum output coding. Normal voltage and several power quality disturbances (voltage swell, voltage sag, voltage interruption, harmonics, transient oscillation, voltage spikes and voltage notch) are considered. Compared with support vector machine (SVM), RVM classifier achieves higher classification accuracy, requires substantially fewer relevance vectors and shorter test time. It has good ability of generalization for small sample problems and provides predictive distribution, without Mercer's condition restriction on the selection of kernel function. The simulation verifies its validity to classify power quality disturbances.