Unsupervised Non-Intrusive Load Monitoring of Residential Appliances

Non-Intrusive Load Monitoring (NILM) is a technique in which electrical appliances are monitored through the disaggregation of the aggregate electrical signal gathered from a single metering point. NILM has recently attracted wide interest in research for application in both the commercial and residential sectors. In this paper, a completely unsupervised NILM system is proposed for application in the residential sector. The proposed system consists of four stages. First, events are detected in realtime using a bucketing technique and EM clustering. Features are then extracted from the transient behavior of appliances. Afterwards, mean-shift clustering is utilized in the detection of recurrences. Finally, an initial transition matching approach is proposed. The system is then tested on the publicly available BLUED dataset and showed high event detection and clustering results and reported a possibility of complete disaggregation up to 92% of the total consumed energy.