Event detection methods for nonintrusive load monitoring

Abstract

This paper proposes two novel algorithms to detect appliance switch-ON/OFF events, which is the first key step of a Nonintrusive Load Monitoring (NILM) software. The Window with Margins method uses the left and right margins of a window running over the house power consumption curve to locate power steps, while the Shifted Sample method relies on the derivative of the power consumption curve to identify power steps. The former may achieve higher event detection rates, while the latter presents very low complexity and requires a reduced computational effort. Extensive sensitivity studies were performed with the parameters of each method, and obtained results have shown the importance of an adequate parameter setting, as successful event identification rates may be increased up to 95%. The applicability and effectiveness of such algorithms have been verified using field measurement data.