Publicação: A NILM Dataset for Cognitive Meters Based on Conservative Power Theory and Pattern Recognition Techniques
Carregando...
Data
Orientador
Coorientador
Pós-graduação
Curso de graduação
Título da Revista
ISSN da Revista
Título de Volume
Editor
Tipo
Artigo
Direito de acesso
Acesso 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
Palavras-chave
Cognitive meter, Conservative power theory, Pattern recognition algorithms, Residential appliance recognition dataset, Smart meter
Idioma
Inglês
Como citar
Journal of Control, Automation and Electrical Systems, v. 29, n. 6, p. 742-755, 2018.
