How about a 240V summary channel?

As a briefly mentioned a few weeks ago, Sense initially correctly found a “heatpump” since the day it was installed, but detection has been spotty ever since and the numbers are wrong (e.g. only one day of brief activity in the last week, even though it was running every day). This is a multizone minisplit system and the outdoor condenser is the only 240V consumer in my household (except for the sense device :smiley: ) and all five indoor units get their power (also 240V) from the condenser, each via an insulated harness of wires and pipes.

Condenser and all indoor units are variable speed inverter driven, making it hard to define an exact signature, of course. Still, just looking at the trace, my eyes can immediately tell when it is on!

I am not sure if this is really true, but I would expect a 100% correlation on the consumption of the two phases, which should be easy to sort out by the sense device/software. Couldn’t it add a second generic trace for “240V consumption”. In my case that would correspond to the minisplit system and would be all I need. I wonder how many false positives would occur if the math is implemented correctly?

I think it would be helpful for users to be able to see Power Meter breakouts for both legs, so you could observe 240V behaviors. I think there is a already a Wishlist item for that.

But I’m a bit stymied trying to think through the function that would match the 240V components on each leg given:

  • Random 120V “noise” on each leg
  • Possible multiple 240V device on and off in overlapping fashion
  • Unbalanced 240V consumption - some things like dryers and ovens have 120V functions like the controller, fans, motors, etc. tied to one leg so the numbers between legs can vary widely.

What kind of math function were you thinking of @altenbach ? Some kind of partial autocorrelation function between the legs ?

ps: I think the current linking algorithm relies on tying together on/off transitions that fit the native detection model, and are tightly correlated in time, not overall consumption.

1 Like

It is well possible that I am oversimplifying the problem in my mind and of course I am not sure how these units are actually wired internally (e.g. if some minor controller components or fans only use one leg), but I assume that the “one phase consumers” in there are minor compared to the 240V consumers and thus not really of interest.
I was very specific about your “multiple 240V device” concerns. I don’t really need to detect individual devices, just the generic global 240V consumption, i.e. the major consumer components (compressor, etc.). Starting with a power failure (or some other baseline), disregard anything that only changes in one phase, but count anything that comes on at the exact same time (and with very similar current) on both phases. Anything that is 120V (your random 120V noise, minor 120V components in the 240V device) will just count as 120V and ignored for that trace. Note that the global consumption trace should remain “as is” and include 120 and 240V. Maybe if just we had a special “sum of all 240V devices” in the device list?

My current issue is that my system is only partially and occasionally detected and there are no other obvious unknown devices that I could merge. I’ll keep watching. Maybe the detection algorithms will figure it out in a few more months. :smiley:

1 Like

I think Sense could apply some convolution / cross-correlation / autocorrelation techniques to get some kind of estimate of 240V usage from the 1/2 second power data. The challenge would be 120V “background noise” - the problems are that

  • you can’t just “subtract off” the 120V noise because the you don’t know what the 120V component looks like until you estimate the correlated 240V.
  • it’s likely that the same noise issues that affect Sense native detection would also introduce uncertainty into the 240V result.

Fun to think about.

I think the opening post is asking for transitions that occur on both legs simultaneously to be thus flagged. As kevin1 points out, the usage from each leg could vary widely and would not be easy to correlate. But I can see where it would be simple for Sense to spot transitions that happen on both phases simultaneously, as it is already looking for such items.

By transitions, I mean the little dips and jumps that are labeled only if you are watching the power meter in live time. @altenbach, do you see transitions when your variable speed HVAC unit cycles?

1 Like

I don’t see “transitions” on my Sense for several different variable speed devices, and Sense hasn’t been able to detect those either…probably for the same reason. So it seems doubtful that transitions on both 240v legs would be sufficient to allow Sense to make such a correlation.

I decided to take a look at bunch of my 240V devices that have been natively detected by Sense to see what the Power Meter makes of each on/off. Does it see it as a transition or not per the tags ? From the looks of it, there were changes made in the Sense firmware in Oct. that show up in the Power Meter tagging (more on that in a bit).

First off, Sense did tag a bunch of my 240V device going on and off. The first four significant bumps on the left are me turning on each floor heating loop. The next two bigger bumps are charring ramps for my Model S, then the one for my Model 3.

Looking a little closer at the heating loops, one thing jumps out at me. I know the biggest heating loops are around 5.6kW so the transition tags (on and off of 2.9kW) are too small by 2x, so the tag really seems to represent the 120V transition power. The good news is that it does show up as 5.6kW for the actual detection !

The EV charging tags are even more interesting. It used to be that Sense might pick up one very wrong part of the charging transition or nothing at all. Now Sense seems to capture longer (more than a second or two) stretches of the ramp, plus captures multiple steps in the ramp. Just looking at the the numbers again, it looks like Sense is using the 120V transition value for the tag, though again, it is getting power number for the device detection correct.

So the new behaviors I’m seeing from the tags:

  • The Sense monitor is inserting tags for events it didn’t used to flag as consistently (EV charging ramps)
  • The Sense monitor / mothership no longer seem to be substituting device names into the tags when it detects a known device, but if that’s a side-effect of better, more consistent native detection, I’m all for that.
  • The tags are also more persistent in the Power Meter. It used to be that if you move to any other view, previously inserted tags would disappear, and you would only see newly created tags as they were generated once you jumped back to the power meter. From what I can see now, the new tags start to be written once you enter the Power Meter, an remain in the Power Meter view for all on as you have the app running (there might also we a number of tags limit or time limit).

If this new behavior is long-lived and consistent, it should help people better know whether Sense is picking up some interpretation of the on and off transitions from their harder to detect devices.


Thanks for all your detailed discussion and comments. I don’t think there is an API to pull data for each leg separately, else we could play and see what comes up. Or is there? Haven’t really investigated…

(I do have extensive programming (examples) and signal processing experience but I can’t really do anything.)

It’s been a while since I have tinkered with the API here.

The API does have a realtime call that returns the voltages on both legs, but I think all the power / energy calls return a single value. I’m guessing that individual leg usage is available via realtime.get if you can figure out the right key for the request.

1 Like