Demonstration abstract: Submetering by synthesizing side-channel sensor streams

Abstract

Detailed breakdowns of household energy consumption allow occupants to better understand their energy usage patterns and identify opportunities for energy savings. Current solutions are costly, invasive, and difficult to maintain. Sub-metering approaches rely on - and are hindered by - complex hardware. To address these problems, we demonstrate a sub-metering system that can estimate the power draw of individual loads by augmenting aggregate measurements with very simple sensors. These sensors wake up at a frequency proportional to the power draw of a neighboring load, and report these wakeups to a central server. We model the relationship between each sensor's wakeup frequency and the load's power draw as a monotonically increasing polynomial. We calibrate each sensor's function by constructing a linear least squares problem that allows us to discover the set of polynomial coefficients that minimize the difference between the estimated power draw and the power draw as derived from the aggregate measurements. After calibration, we can convert sensor wakeup frequencies to power draw in real time. This systems approach to sub-metering results in deployments that are easy to install and maintain, allowing users to gain a broad yet detailed view of their energy consumption and costs.