EMBED

Abstract

Electricity disaggregation, the task of inferring appliances' energy consumption in a building from a few sensing points, has received attention in the energy community. In this paper, we introduce EMBED, a publically available dataset for Energy Monitoring through Building Electricity Disaggregation. EMBED is the most comprehensive fully labeled disaggregation dataset to date containing the aggregate power measurement as well as plug load data of different appliances for three residential units, each was collected for at least two weeks. Therefore, this dataset could be used for comparing the performance for both event-based and non-even-based methodologies. In this dataset, all the events were uniquely labeled by identifying separate labels for different transition states for each appliance. This feature enables comprehensive evaluation of classification or clustering algorithms for disaggregation. The paper discusses the dataset, hardware setup, data collection procedure, and unique characteristics compared to the similar works. In addition, it presents preliminary analyses on the dataset as a benchmark to encourage future research in the disaggregation domain.