Disaggregating household loads via semi-supervised multi-label classification

The essence of Non-Intrusive Load Monitoring (NILM) is to extract electricity consumption details of individual appliance from an aggregated house-level electrical measurement at the main panel without sub-metering each appliance. In this paper, an Expectation Maximization (EM) based semi-supervised multi-label classification technique is applied in NILM. It requires a one-time registration of individual appliance to obtain few samples during the training stage. After that, the total electricity is utilized to detect the states of each appliance and analyze electricity consumption information of individual appliance for each instance via the help of semi-supervised learning method. Experiments on house 1 and house 3 dataset of Reference Energy Disaggregation Dataset (REDD) verify the effectiveness of application of Semi-supervised learning techniques in NILM.