Abstract
Enhancing household energy efficiency is crucial, and Non-intrusive Load Monitoring (NILM) offers a valuable solution by giving consumers insights into their energy use without individual device monitoring. However, the deployment of NILM models in new settings is challenging due to their training on domain-specific data. To effectively use public data for training NILM models for identifying individual appliances, understanding the challenges of model transfer is crucial. This study explores several factors that could hinder successful model transfer and highlights the challenges in broader NILM system deployment. We developed and tested various NILM models, both event-based and eventless, across multiple household domains and found that domain bias, e.g., noise and line frequency, does not significantly impact model performance.
Justus Breyer, M.Sc.
Researcher
Sparsh Jauhari
René Glebke, M.Sc.
Researcher
Asst. Prof. Dr. Muhammad Hamad Alizai
LUMS, Lahore, Pakistan
Markus Stroot
