Power quality event detection using joint 2D-wavelet subspaces

Abstract

In this work, we present a novel 2D representation of power system waveforms for the automatic analysis of transients. The representation is composed of a matrix whose rows are formed by time segments of digital waveforms. By the appropriate selection of the time segment length, the 2D data exhibits wave-like image shapes. The general shape is immediately disturbed whenever a quality event occurs. We propose the use of two dimensional discrete wavelet transforms (2D-DWT) to detect these shape disturbances. It has been observed that, after omitting the approximation space signals of the wavelet transform and denoising the detail space signals, the inverse 2D-DWT provides good detection and localization results, even for cases where conventional methods fail. Examples are presented.