Unsupervised Adaptive Event Detection for Building-Level Energy Disaggregation

The need for energy disaggregation increases with the need for a more detailed understanding and more accurate estimates about the energy usage. One of the main approaches for energy disaggregation is Non-Intrusive Load Monitoring (NILM). NILM refers to the analysis of the aggregate power consumption of electric loads in order to recognize the existence and the consumption profile of each individual appliance. While there exist non event-based NILM systems, many NILM systems follow the event-based approach in the sense that they rely mainly on the detection and classification of events in the aggregate electrical signal. In this paper, we describe our work on developing an event detector suitable for unsupervised NILM systems. The proposed event detector is capable of accurately defining the times limits of each transition interval in the power signal. This feature is very important specially for NILM systems that depend on transient features. The detector is tested on the publicly available BLUED dataset and shows event detection results more than 98%. Test results of a complete unsupervised NILM system using the proposed detector are also provided and show possible disaggregation up to 92% of the energy. Moreover, the system has been utilized in an energy-disaggregation competition held by Belkin and achieved a score within the top ten results with disaggregation value of 93.41% of the total time.