On the applicability of 2D local binary patterns for identifying electrical appliances in non-intrusive load monitoring

Abstract

In recent years, the automatic identification of electrical devices through their power consumption signals finds a variety of applications in smart home monitoring and non-intrusive load monitoring (NILM). This work proposes a novel appliance identification scheme and introduces a new feature extraction method that represents power signals in a 2D space, similar to images and then extracts their properties. In this context, the local binary pattern (LBP) and other variants are investigated on their ability to extract histograms of 2D binary patterns of power signals. Specifically, by moving to a 2D representation space, each power sample is surrounded by eight neighbors at least. This can help extracting pertinent characteristics and providing more possibilities to encode power signals robustly. Moreover, the proposed identification technique has the main advantage of accurately recognizing the electrical devices independently of their states and on/off events, unlike existing models. Three public databases including real household power consumption measurements at the appliance-level are employed to assess the performance of the proposed system while considering various machine learning classifiers. The promising performance obtained in terms of accuracy and F-score proves the successful application of the 2D LBP in recognizing electrical devices and creates new possibilities for energy efficiency based on NILM models. Springer Nature Switzerland AG 2021.