Power quality disturbance waveform recognition using wavelet-based neural classifier. I. Theoretical foundation

TLDR

Current power quality disturbance waveform recognition relies mainly on visual inspection of waveforms. This paper aims to apply wavelet transforms, artificial neural networks, and evidence theory to automate power quality disturbance waveform recognition. The proposed method performs recognition in the wavelet domain using multiple neural networks whose outputs are fused by voting or Dempster‑Shafer evidence theory. The resulting classifier can assign a degree of belief to each identified disturbance waveform.

Abstract

Existing techniques for recognizing and identifying power quality disturbance waveforms are primarily based on visual inspection of the waveform. It is the purpose of this paper to bring to bear advances, especially in wavelet transforms, artificial neural networks, and the mathematical theory of evidence, to the problem of automatic power quality disturbance waveform recognition. Unlike past attempts to automatically identify disturbance waveforms where the identification is performed in the time domain using an individual artificial neural network, the proposed recognition scheme is carried out in the wavelet domain using a set of multiple neural networks. The outcomes of the networks are then integrated using decision making schemes such as a simple voting scheme or the Dempster-Shafer theory of evidence. With such a configuration, the classifier is capable of providing a degree of belief for the identified disturbance waveform.