What is it about?
Event-Based NILM uses machine learning models to assign the change of operational states of household devices to a predefined set of devices. We introduce a prior step to remove the data of state changes of devices outside of the given set. This avoids foreign devices being assigned to the monitored set.
Featured Image
Photo by BINGYEN STUDIO on Unsplash
Why is it important?
Since most NILM solutions only monitor a subset of all available household devices, the occurrence of an event produced by a non-monitored device, i.e., a device unknown to the classifier, is inevitable. However, the classifier is usually agnostic to the number and characteristics of unmonitored devices. We introduce a solution to remove such events without prior knowledge of their characteristics, thereby avoiding false positives in the classification process.
Perspectives
The problem of sorting out unknown devices in the classification process is often not addressed in the literature. Instead, during the evaluation, events are manually selected to always fit the available device categories known to the classifier. This obviously does not reflect the real world, and in an attempt to make the concept of NILM more applicable, we address this issue with an elegant, computationally non-intensive solution proposal. Our study includes a broad variety of approaches to remove such events, and the best performing solution shows great promise.
Justus Breyer
Read the Original
This page is a summary of: Advanced Filtering of Unknown Devices in Event-Based NILM, ACM SIGEnergy Energy Informatics Review, December 2024, ACM (Association for Computing Machinery),
DOI: 10.1145/3727200.3727204.
You can read the full text:
Resources
Contributors
The following have contributed to this page
