Automatic classification of Power Quality disturbances via higher-order cumulants and self-organizing networks

Abstract

This work accomplishes the classification of Power Quality (PQ) disturbances using fourth-order sliding cumulants' maxima as the key feature. These statistics estimators are calculated over high-pass filtered real-life signals, to avoid the low-frequency 50-Hz sinusoid. Four types of electrical AC supply anomalies constitute the starting grid of a competitive layer performance, which manages to classify 90 signals within a 2D-space (whose coordinates are the minima and the maxima of the sliding cumulants calculated over each register). Four clusters have been clearly identified via the competitive network, each of which corresponds to a type of anomaly. Then, a Self-Organizing Network is conceived in order to guess additional classes in the feature space. Results suggest the idea of two additional sets of signals, which are more related to the degree of signals' degeneration than to real new groups of anomalies. We collaterally conclude the need of additional features to face the problem of subclass division. The experience sets the foundations of an automatic procedure for PQ event classification.

An Embedded System in Smart Inverters for Power Quality and Safety Functionality

“…Higher-order statistics (HOS) have shown their potential to deal with impulsive and non-linear time-series in the PQ analysis frame [18,24]. In fact, HOS take into account an extended set of features that enhance detection and make the decision-making stage feasible.…”