Sense not effectively identifying devices


My sense has done a poor job identifying devices in my home, so far. For example, it identified 5 fridges (but I don’t have 5 fridges). My kitchen fridge doesn’t appear to be any of them. One of those fridges used to consume 90w now it’s consuming over 1000w which seems unlikely. Lots of unidentified devices. 80% of usage is still “unknown”.

Should it be doing better than that after 3 weeks, or do I need to wait?

Devices discovered, then lost and rediscovered as new devices

I’m 5 weeks in and it hasn’t identified any individual devices either – just things like “Unnamed Heat #1” which is triggered by half a dozen actual devices – electric kettle, toaster, dishwasher, EVSE…

Tech Support says it takes " a while". I suggested they define this a little better in their docs – is “a while” 6 weeks, 6 months? If I’d known to wait a couple months, I would not bother Tech Support with questions if something is wrong.


I’m not sure what the best way to handle it would be, but I try to be patient. I’ve had a few devices identified and then forgotten, and then rediscovered (or just partially rediscovered), etc. Since I’m not as concerned with device detection for now, I wonder if maybe infrequent emailings with lists of observations may be better, or if there’s some better way to amalgamate our individual problems.

For example, I have a water heater that is hit or miss. It’s on a timer, and Sense recognizes it’s on pretty reliably when it’s on, but it will occasionally think it’s shut off (grouping the power under unknown) when it’s still on. Or, sometimes it’ll call it half on when it’s really an all or nothing thing and the other half is grouped under unknown. The same also applies to my heat pump - most of the time it IDs it reliably, but sometimes it’s obvious it hasn’t really learned it. I also have a duplicated coffee pot…well, at least I think that’s what it is. It may be the toaster now that I think about it, but in any case, there’s possibly some overlap. None of this is really bothering me, so I don’t email about it - I anticipate things will get better. I was amazed how quickly and reliably it picked up my microwave and garbage disposal, so I’ll patiently wait for the learning system to catch up to my other appliances.

But you bring up a good point - would we all be better served by having a more efficient way to let them know the discovery problems we’re seeing than bombarding tech support with emails with every observed error?


Best way to handle what, exactly? It seems clear that there is absolutely nothing a user can do regarding detection. It reports what its algorithms think, and that’s all there is to it. One can ponder, offer musings and guesses about what it’s doing, share and compare observations with others, but you can’t affect it in any way. So there’s no “best way” to handle it, just one way – sit back and watch it.


What I meant was the best way to handle questions on errors and omissions. I’m sure tech support is watching for bugs, but are all of these bugs or are some simply learning? Are we helping by alerting them to things like I mentioned and you mentioned, or is that just part of the process? Like I said, [quote=“NJHaley, post:3, topic:321”]
would we all be better served by having a more efficient way to let them know the discovery problems we’re seeing than bombarding tech support with emails with every observed error?


Oh, I see what you mean. My impression is that the limits of successful device detection are just the limits of the algorithms at this stage, not specific bugs. And I also get the impression that they are not really interested in hearing much about specific errors or omissions. Presumably knowing that a given customer’s individual “device” is actually an amalgamation of devices doesn’t directly translate into actionable engineering insights and tasks – they are working at a higher level of abstraction. The resulting algorithms of course have to ultimately be verified at the specific device level, but that doesn’t mean hearing about every specific omission helps them much at this stage.

The very absence of a defined mechanism to report errors and omissions strongly suggests that it’s not useful to them – since obviously it would be simple to implement.

Of course, if they said “we believe Sense now perfectly identifies GE Toaster Model X”, and a customer saw an exception, they’d be interested in hearing that thru tech support, since that looks like a bug. But until that day comes, it seems like most questions about detection are met with a variation of “wait a while…” – either it hasn’t yet learned the device, or the algorithms are not yet capable of learning the device.

That said, I’ve only had it since December and may not have the full picture.


I think you’re absolutely right. That’s primarily why I’m not so concerned about whether my toaster is identified, while I am more concerned that it gets the basics right - like accurately tracking our ins and outs. I feel like without getting the basics correct it loses a significant percent of utility. Essentially, you have to make it functional at even the most basic level first so there’s at least some useful function, some useful data, while it’s learning appliance profiles. That’s been my only real source of frustration - not only do I have something that hasn’t yet matured to the stage of quickly identifying appliances (which I wasn’t expecting - heck, they’re leagues ahead of anyone else in this category even if my water heater is lost and found almost daily), it’s not even giving me an accurate reading of my meters (which I was expecting). They’re working on it, though, which I’m thankful for :slight_smile:


If Sense isn’t even reading meter correctly, that’s got to be a defect you should report to tech support.

For total ins and outs, there’s no “learning” or anything. It’s trivially simple.


They’re on it, hopefully close to a solution :slight_smile: I think one of the difficulties they’re running into is that for those of us with solar, the various set ups are mucking things up.


Given the complexities of identifying a device among the dozens (or hundreds) running in my house, I’m completely overjoyed when a new device is detected. Wow, how did they do that!?!

I purchased Sense for the fun of it. So far after 1 month, I’m having a blast.


I think that sounds pretty good for three weeks. Here’s the Jan 2017 status of my 4 months old Sense installation for you to review as a contrast.

closed #12