Leveraging smart meters for residential energy disaggregation

Smart meters promise to provide residents with the information required for using their appliances in the most energy efficient manner. We demonstrate how smart meter data can be used to infer individual appliance energy consumption by extracting low-frequency transient features which correspond to appliance turn-on and turn-off events. These features are a combination of a transient shape metric and the summary statistics; and are used to detect and classify appliance events. The detected events are fed into a disaggregation algorithm that is robust to missed detections. The combination of our transient features and disaggregation algorithm allows us to achieve a disaggregation accuracy of 94% on a test set of 4 major appliances without tuning to that home. This demonstrates our approaches ability to generalize to unseen homes. Our scheme uses low frequency transients, yet provides performance equivalent to more complex and costly high-frequency sampling approaches.