Ten days of Sense

Here is my Sense story after ten days of use:

I installed my Sense unit myself. The install was not difficult and went without a hitch (to be fair, I am an engineer).

Here is what the system has detected so far after 10 days of use:

  1. The first thing it detected (within 24 hours of install) was “Heat 1”, a mysterious device that used ~800 watts of electricity for exactly one minute, spaced 6 minutes apart. It would do this 30-60 times, then stop for ~24 hours. After much searching, this turned out to be the sump pump in my crawlspace (not a basement). What’s interesting is that the pump hasn’t functioned for years, but apparently it’s still been turning on occasionally. I unplugged it since this represents a potential fire hazard. Thanks Sense!

  2. The second thing it discovered was my HVAC (no surprise there). It’s not 100% accurate though, sometimes my A/C is on and it’s still grouped in the “Other” bubble. Sometimes it thinks my A/C is my dryer.

  3. Next it discovered Heat 2, which is my Keurig Coffee Maker and my Toaster Over. Yes, if either of those things turns on, it’s listed under this device. This isn’t surprising as both are heating elements of similar wattage on the same breaker loop.

  4. It also discovered my clothes dryer, though sometimes it thinks the A/C is my clothes dryer.

  5. It discovered my microwave and so far is pretty accurate at flagging this device correctly when used.

  6. It discovered “Stove” and “Heat 3” simultaneously, which are various elements or combinations of elements on my stove top. I haven’t quite figured out how to trigger them–when I turn them on one at a time, it doesn’t flag either of these devices. I combined them both into one device I call Stove.

  7. Finally, it discovered the light in the master bathroom. I don’t know what’s special about this light in particular. It usually correct identifies when I turn it on, but it misses when I turn it off about 40% of the time. It also falsely triggers from other devices in my home.

In summary, it found 8 devices in the first week (but none in the past 3 days). Some of these devices it triggers off of accurately, others not so much. However, it did lead me to my faulty sump pump, which had been completely wasting money for years.

I’ve worked extensively with machine learning algorithms as part of my job, so I know what a difficult problem unsupervised learning is, especially when the input data is mixed. However, I hope both the detection process and the accuracy is improved over time. Either way, it’s a solid device for measuring overall electricity use. I do wish there was a way of downloading the raw data collected by the Sense so I could analyze it myself (is there?)


That’s actually not too bad for 10 days.

Glad we were able to get you sorted on that sump pump. Always nice to hear those stories.

I do think that Sense will eventually be able to correctly parse out that Keurig and Toaster Oven as well as the AC and clothes dryer. Same as with the other issues. In the vast majority of cases, Sense improves over time. I won’t go into the whole ML thing, since you get it, but the more data Sense acquires, the better it becomes.

And no way of downloading raw data at this time (and since we collect at 1MHz, that’d be a lot of hard drives :slight_smile: ), but we are working actively on a data export feature that would compress the data somewhat. I’m curious, what would you like to see out of a data export feature? What sort of data would you like to analyze on your end?


Interestingly, shortly after I posted this, Sense found two additional devices. One was a “Heat X” device, that I assume is another burner on my stove, and the other was my TV, that it correctly called “LG TV”. LG must have a particular signature for Sense not only to recognize that it’s a TV, but one made by LG. So far, the TV is 100% accurate in being flagged for on/off. On the depressing side, I can now see exactly how long the TV is on each day.

I’ll be interested to see if it gets my water heater, since mine is unusual in that it’s the ultra-efficient variety that uses a small heat pump to heat the water instead of a high-wattage coil.

I would need to know what data are available to decide which I’d want to export. I am sure there are features that Sense is deriving from the raw time series that would be nice to see, though they may be considered proprietary. At a bare minimum, I’d like to see the total electricity use per main updated no less than 1 Hz. Then I could compute my own trend metrics. In a perfect world, I’d like to see all the data–but that is probably infeasible, though I suspect it could be compressed substantially (tons of redundancy at 1 MHz, I’m sure).


Thanks for this. It’s very helpful.

As for the TV, I’m guess you have Network Identification turned on under Data Sources in Settings. That’s likely how it found your TV and was able to fill in the model information.

1 Like