This is my thinking, not based on Sense input.
- smart plugs supply ground truth on a 2sec (or 6 sec on my part because I have too many smart plugs) update, far different than the Sense monitor. So smart plug data is somewhat useful for telling Sense when devices are on and off, plus seeing the power usage patterns, but not useful for pulling out most of the parameters/features Sense uses to actually do detection or classification. That informs what Sense can and can’t do using the data - so no direct characterization of devices using smartplugs.
- I would suspect that Sense uses the envelope supplied by smartplugs for specific device types to study the characteristics of the mains during those periods to develop new detection algorithms.
- Sense could do machine learning by “training” on the standard features Sense extracts against the major transitions ground truth detected by the smart plug. That would enable them to build cluster maps for the specific devices. Those maps could help refine the clustering algorithms Sense uses as well as helping Sense assemble multi-component detections (washer, dryers and the devices that have multiple steps in an operation cycle).
- There might also be useful applications for progressive detection. TBD.
BTW - one of the things that confuses people about detection, is that they see the time domain waveform as the device “signature”, without really comprehending that the real signature is really a many dimensional cluster of points.
Just my thoughts.