What is it about?
Non-Intrusive Load Monitoring (NILM) aims to infer the energy consumption of singular devices in a household from the metering data of a central smart meter. This is achieved using machine learning techniques, which in turn require training and test data for evaluation. We provide a dataset collected over 6 months in a real household with submeters for every single device, enabling broad evaluation on the scalability and accuracy of new algorithms.
Featured Image
Photo by Solen Feyissa on Unsplash
Why is it important?
Since NILM algorithms are typically based on machine learning approaches, large quantities of data are required. Especially in terms of transferability of the models, it is always useful to have more options at hand. When creating our own dataset, we considered the critique of several reviews of recent years: The current variety of datasets lacks a setting where every single device is submetered. This is why we created the Device Activity Report with Complete Knowledge (DARCK), where for every existing appliance in a real-world household, the power consumption has been monitored, providing ground-truth and the capability to completely disaggregate the consumption measured at the smart meter.
Perspectives
Although it is not always necessary to train NILM models on the complete variety of existing devices in the household, for certain evaluations (e.g., event detection), it is still invaluable to know what is going on in the household. With DARCK, we now provide a definite way for researchers to check their algorithms, e.g., whether the event detection produced a False Positive.
Justus Breyer
Read the Original
This page is a summary of: Dataset: Device Activity Report with Complete Knowledge (DARCK) for NILM, November 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3736425.3771959.
You can read the full text:
Resources
Contributors
The following have contributed to this page
