A NILM Dataset for Cognitive Meters Based on Conservative Power Theory and Pattern Recognition Techniques

Nenhuma Miniatura disponível

Data

2018-12-15

Orientador

Coorientador

Pós-graduação

Curso de graduação

Título da Revista

ISSN da Revista

Título de Volume

Editor

Tipo

Artigo

Direito de acesso

Acesso abertoAcesso Aberto

Resumo

This paper presents a novel dataset capable of classifying and disaggregating residential appliances for the development of smart or cognitive power meters. This novel dataset uses power indicators (also denoted as conformity factors) from the conservative power theory (CPT), which are calculated from measured voltage and current waveforms during the operation of residential loads. The association of CPT power indicators with suitable pattern recognition algorithms (PRA) and a power signature state machine provides proper identification of each appliance. So, the paper also presents a detailed evaluation of possible PRA for this application, especially the SVM—support vector machine, OPF—optimum-path forest, MLP—multilayer perceptron, KNN—K-nearest neighbor and DT—decision tree. All these algorithms have been compared regarding accuracy and computational time. Validation results point out that KNN would be the best choice for dealing with the proposed dataset, leading to an accuracy higher than 98%.

Descrição

Idioma

Inglês

Como citar

Journal of Control, Automation and Electrical Systems, v. 29, n. 6, p. 742-755, 2018.

Itens relacionados