Unsupervised Event Detection and On-Off Pairing Approach Applied to NILM

Non-intrusive load monitoring (NILM) is a technique of energy disaggregation which takes an aggregate signal from one point in the home and determines usage consumption of each appliance from that aggregate signal. Many NILM algorithms are event based as they provide improved disaggregation results when labeled training dataset with exact timestamps is available. The proposed unsupervised approaches of event detection and on-off pairing are suitable for event based NILM systems that eradicate the pre-training process and hence, show the potential of scalability. Event detector provides exact time stamps of the events in the aggregate signal and shows correct test results more than 99% using publicly available event based real dataset BLUED. On the other hand, on-off pairing approach is robust to detect up to 95% of the total activations by using a small value of the on-off pairing threshold. The total NILM system, in which the event detector and on-off pairing are used, shows successful disaggregation of more than 97% of the energy of phase A of the BLUED testing dataset. The proposed approach is a part of the semi-supervised NILM system which helps consumers to utilize the energy resources in an efficient way. Studies show that disaggregation feedback system can cause reduction in energy consumption by energy awareness.