It has been about two weeks since I installed my sense and so far it’s detected about a dozen devices. Surprisingly it has identified all three refrigerators I have in my house (mini fridge, my old fridge which is now in the basement and my kitchen fridge). I noticed the other day that when I have my 3D printer on and printing it seems to confuse Sense. Sense will properly detect when my kitchen fridge comes on but if the printer is going then it doesn’t seem to know when the fridge turns off. My guess is that the 3D printer is somehow masking the “off” signature for the fridge. I can go into the “Report a Problem” and tell Sense that the Fridge is not on, otherwise Sense doesn’t sort out on it’s own that the fridge is off until shortly after the print is done. My question, is there a better way to train Sense to better recognize the difference between the 3D printer and the fridge besides continually going in and manually telling it that the device is not on?
Two weeks is a very short time in Sense-time; your best bet is to submit “Device is not on” problems every now and then. You won’t get immediate feedback since the machine learning takes some time to process the data and learn the differences between device signatures.
It should get better over time. You don’t have to catch every “Device is not on” problem for it to eventually figure it out, so don’t feel like you have to catch it in the act every time.
My opinion is that you should not submit the “device is not on” feedback if the 3D printer is running. That printer is adding random usage “noise” to your overall power usage pattern. This is going to make it inherently more difficult for Sense to identify on or off patterns, as you suspect. By submitting feedback (training Sense, in a way) when noise is present, it will degrade the accuracy of its model for your fridge.
Unfortunately there’s not really a good way to deal with that additive noise problem. You could put a smartplug on the printer to clean up your bubble display (the printer usage will be directly measured and accounted for in a dedicated device bubble) but that won’t subtract the usage noise from the measurements Sense is making to your main power lines. The time resolution on smartplug data is way too low for that purpose.
I think it’s safe to submit these reports if the device is truly not on. Home power is noisy, Sense keeps telling us it’s noisy and complex, and these noisy times are the when Sense can use the most help.