New Semi-Supervised and Active Learning Combination Technique for Non-Intrusive Load Monitoring

Abstract

Non-intrusive load monitoring (NILM) is a procedure that is used to disaggregate the contributions of different appliances in a building. Various traditional techniques use supervised and unsupervised learning for disaggregation. Very few papers utilize the semi-supervised or active approaches. The two approaches are used in case of training data scarcity. The semi-supervised approach is prone to errors as training dataset is small, while the active leaning approach relies completely on the user to get correct labels. A technique that leverages the semi-supervised and active learning together is applied. This can reduce the number of samples required to the users, also can increase the accuracy of the semi-supervised algorithm while more reliable samples can be added. Many cases are studied and the results showed the robustness and reliability of the new combination. The proposed combined technique showed that the number of queries can be reduced to fifth without losing more than 4% of the accuracy. Also, by combining the semi-supervised and active techniques, the accuracy increased by 6% compared to the stand alone semi-supervised approach.