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

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 fr...

Full description

Saved in:
Bibliographic Details
Published inJournal of control, automation & electrical systems Vol. 29; no. 6; pp. 742 - 755
Main Authors Souza, Wesley A., Marafão, Fernando P., Liberado, Eduardo V., Simões, Marcelo G., Da Silva, Luiz C. P.
Format Journal Article
LanguageEnglish
Published New York Springer US 15.12.2018
Springer Nature B.V
Subjects
Algorithms
Computing time
Control
Control and Systems Theory
Datasets
Decision trees
Electrical Engineering
Electrical measurement
Engineering
Household appliances
Indicators
Measuring instruments
Mechatronics
Multilayer perceptrons
Pattern recognition
Residential development
Residential energy
Robotics
Robotics and Automation
State machines
Support vector machines
Waveforms
Pattern recognition algorithms
Residential appliance recognition dataset
Smart meter
Cognitive meter
Conservative power theory
Online AccessGet full text

Cover

More Information
Summary: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%.
ISSN:2195-3880
2195-3899
DOI:10.1007/s40313-018-0417-4