Dynamic Time Series Data Reduction for NILM Appliance Identification

Abstract

Advancements in Internet of Things capabilities along with cheap & easy-to-use sensors have led to the development of many new domains, Non-Intrusive Load Monitoring being one of them. A crucial element of these technologies is appliance identification based on disaggregated power consumption signatures. The length of said signatures depends on the data collection frequency, with higher frequencies corresponding to longer time series. A dynamic time series data reduction method is introduced which can effectively extract a region of interest from very long time series. Appliance classification accuracy with these sub-ranges is then tested using Matrix Profile. Plug-Load Appliance Identification Dataset was used to carry out the experiments.