A non-intrusive approach to classify electrical appliances based on higher-order statistics and genetic algorithm: a smart grid perspective

Abstract

Electrical appliance monitoring systems have received a lot of attention in recent years. These systems can provide users with valuable information for energy saving. In this article, a non-intrusive approach to classify electrical appliances based on higher-order statistics (HOS) is proposed. Aiming at reducing the computational cost of the proposed method, Fisher's Discriminant Ration and Genetic Algorithms (GA) were used for selecting a finite set of representative features among those obtained by HOS. The core idea of using GA was to simultaneously reduce the data dimension and optimize the classifier performance. The method was carried out over experimental signals, collected from the main power service entry of a house. Eleven electrical appliances were studied and fifty current signals of each of these loads were acquired; only the transient state of these signals was analyzed. The final classification was performed by multilayer perceptron (MLP) and decision tree (DT) classifiers, reaching an overall validation efficiency of 100% and 99.5%, respectively. The proposed classifiers used only 6 extracted features (second and fourth-order cumulants) and are suitable for real-time application.