Publikation
NILM domain adaptation: When does it work?
Muhammad Muaz; Ingo Zinnikus; Mahnoor Shahid
In: 2024 10th International Conference on Smart Computing and Communication (ICSCC 2024). International Conference on Smart Computing and Communication (ICSCC-2024), July 25-27, Bali, Indonesia, Pages 524-528, ISBN 979-8-3503-6311-1, IEEE, 2024.
Zusammenfassung
This study investigates the transferability of Non- Intrusive Load Monitoring (NILM) models. Our methodology incrementally increases source domain data for transferability on the target domain. Testing 641 models, successful adaptation typically involves data from the same country or continent, achieving baseline performance (F1 score 0.7) with data from 5– 6 houses. The research concludes that domain adaptation in NILM is feasible but warns against potential overfitting scenarios. The findings could guide future NILM studies, balancing between increasing training data and avoiding overfitting, thereby reducing expensive data collection.