Abstract
Estimating the energy consumption of individual devices and forecasting the total load of end-user apartments have been highly active research areas over the past decades. Many of the proposed approaches and improvements rely on data-driven algorithms, including machine learning, that require large amounts of reliable measurement data for training and evaluation. We identified a severe gap of fully-disaggregated public datasets in research and therefore propose the Device Activity Report with Complete Knowledge (DARCK) dataset, the first dataset monitoring every single appliance in an apartment, including lighting. The dataset was collected in a two-person household in Germany over the span of 6 months and provides power readings of the mains as well as of 51 different appliances at a sampling rate of 1Hz. The dataset aims to complement the existing range of public research data in Non-Intrusive Load Monitoring (NILM) by offering data at commodity hardware precision that can be fully disaggregated. The paper describes the measurement setup and processing steps for data treatment as well as offering several noteworthy insights about the content of DARCK itself and an exemplary benchmark.
Justus Breyer, M.Sc.
Researcher
Kai Gützlaff
Leonardo Pompe, M.Sc.
Researcher
