Feature selection for identification and classification of power quality disturbances

Abstract

Monitoring power quality over extended period of time can result in significant amount of data being gathered. This necessitates automatic processing of the data as the captured phenomena need to be sorted before further analysis can be undertaken. This typically involves a process of identification and classification of various power quality related disturbances. The voltage and/or current waveforms are normally expanded onto some subspaces where there exist some patterns for recognizing them. Advanced signal processing such as Fourier transform and wavelets transform are generally employed to carry out the necessary expansion. This paper explores the features that are commonly selected as the basis for identifying the disturbances. A judicious selection is necessary as it affects the accuracy as well as efficiency of the identification and classification process.