CONIX Publication

Appliance-level energy usage information is useful to encourage energy-saving behavior and to inform future purchasing decisions. To get the most accurate consumption data, individual sensors can be mounted to each device, but manually recording the location and label of each sensor in a large deployment is infeasible. Therefore, I am working toward the automatic location and labeling of such sensors based on the data they collect. I present a classification method that uses transfer learning to automatically identify the type of appliance being monitored. I also explore some tradeoffs in cloud v.s. edge computing based on the available resources and the desired accuracy, and I construct a succinct feature space to address the problem of transmitting high-frequency data over limited bandwidth. I end by considering the resources necessary for local classification and consider optimizations that could push such computation closer to the edge.