Aplicação das trajetórias V-I para métodos de extração de características e classificação de cargas elétricas

Abstract

Due to the growing demand for electricity, energy efficiency programs have started to include energy management and conservation solutions. In order to collaborate in the awareness and management of consumers’ energy use, non-intrusive load monitoring techniques allow the individual consumption of devices in an installation to be informed to the user. To implement this approach, there are four main steps: data collection, event detection, feature extraction, and load identification. The performance of the last step depends on how the previous steps are performed, especially the feature extraction, which involves the quantitative description of the behavior of an electrical appliance, also defined as load signature. In this context, this work focuses on the development of two approaches based on load signatures called V-I trajectories. In the traditional extraction approach, new characteristics are proposed, for use in steady-state and transitory state of the load signature. The second approach consists in developing a method that describes the load signature itself, through the two-dimensional Fourier series (2D) descriptors. With those proposed approaches, the results in load identification are evaluated with different classifiers (k-nearest neighbors, Ensemble, decision tree, support vector machine, and linear discriminant analysis) for three distinct and publicly available databases. Besides the evaluation of computational resources of the proposed methods, such as time and memory in an embedded system, the robustness of the methods to noise insertion is evaluated. The results obtained for both methods, in most cases, present classification accuracy higher than 90%. Also, the use of computational resources provides competitive results when compared to another method in the literature (state-of-the-art), based on the Wavelet Transform.