@afward2000 , welcome to the Community ! Gonna try to answer your question, but first wanted to change your thinking about two important things about how Sense works:
- Sense doesn’t use matching per se. Because things in your house can affect waveforms (your house voltage can vary by 5%) and the on/offs of devices can change a fair amount from one off on/off to the next on/off, Sense is forced to do “fuzzy matching” - much harder than normal matching, partially because it has to figure out which patterns are distinct, unique, and constitute an on/off pair, before it can do a detection. Fuzzy matching looks far different from the matching process people are typically familiar with.
- Sense doesn’t fuzzy match time-based waveform patterns, even though that’s what we see. It tries to match a many-dimentional pattern of data extracted from the half-second on transitions and off transitions. If you want more info on what this pattern data looks like inside of Sense, you can look here.
Given that foundation, let me try to answer your questions:
Sense can use Sense-compatible Kasa, Wemo and Wiser smart plugs (and other) to do this today. Add one of those smart plugs, and Sense adds that plugged in device to its list of devices using the measured waveforms. But Sense can’t directly learn from that smart plug. The “patterns” that Sense gathers are from millions (or at least hundreds of thousands) of measurements Sense does in the half second of an on or an off. These compatible smart plugs are only sampled by Sense every 2 seconds so there’s no way they provide the info Sense needs to build the needed patterns. I do think that Sense can use data from these smart plugs to improve their models back in the “lab”, but no direct learning is possible.
Yes and no - Sense doesn’t keep a library of patterns that can be “looked up” for a couple important reasons. Device patterns look different between houses. And Sense can’t possibly characterize every make and model of every device type. But fuzzy matching does allow Sense to essentially measure how “far” a newly detected device (unique, distinct pattern with on/off pair) is from categories of devices that have been detected and identified by other users. That’s what you are seeing when this pops up:
That’s fuzzy matching in action !
I’ll also point out that Sense deals with EV charging differently because the on and off ramps are far slower than Sense normally watches for. For EVs, Sense does use the time pattern of the ramp, and each EV make, model and even software revision has a specific ramp pattern.