Paper Number



In this paper, we address the topic of energy disaggregation for real-world implementation. The idea behind energy disaggregation is to utilize the aggregated energy consumption data from smart meters to identify individual appliances in the load by decomposition. However, appliance level usage prediction for residential areas is currently not practical. Yet Smart Living environments might benefit from non-intrusive load monitoring (NILM) adoption. We therefore first lay a theoretical foundation by working out the core challenges for NILM adoption in real-world scenarios. We evaluate the semi-supervised model performance on a publicly available dataset REFIT. By reducing the sampling rate to 32 seconds as well as limiting the available labeled data for training, we give significant insight into the usability of semi-supervised NILM under real-world conditions. The availability of unlabeled observations seems to strengthen model prediction performance, allowing it to outperform a fully trained supervised model for certain appliances.



When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.