Three quick answers, but once again I’m answering as a user with a little knowledge of power systems, DSP, and machine learning. So my comments are based on learning how Sense does things from public information like blogs, educational videos, etc, plus me putting the pieces together. I presume that most of the in-depth details are proprietary, though you can certainly glean more from their filed patents.
-
Not sure how anyone would “query the majority of internet enabled smart devices on the network”. There are literally 50+ different company-specific protocols for querying power / energy information from smart devices that provide power/energy usage data (you see many of these protocols in the various Home Assistant integrations). Matter might help, but 90% of the devices in use today don’t support Matter today. And some of the data supplied today isn’t an appropriate fit for real-time monitoring. And some devices, like the Wiser/Square D plugs, mix in a little garbage with good data, so one can’t just read the data as golden.
-
I can’t tell you all 20 parameters or “features” used for detection, but this screenshot from one of their internal tools indicates clustering based on at least 17 features.
There’s a video associated with this screenshot in my perspective on how Sense detection currently works. Worth reading if you really want to know more.
- My “perspective” also highlights some of the challenges of this identification methodology - variability & clustering, home noise, and devices that don’t fit the simple on/off model, like a dishwasher, which is really many devices in one (heating element, DC motor, controller).