KP115 and HS300

If I connect a device to one of these smart plugs, and sense eventually detects and identifies the device, does the device need to remain plugged into the smart plug or can I move the smart plug to another device?

Essentially, what I am asking, or what I am trying to do is force sense to detect devices in my house rather than have them stuffed into always on or other, or split between the two (or more).

You can move it as at that point sense has natively discovered it. There’s a few posts where people have done so. But the plug doesn’t help sense discover them natively.

The answer is that putting a device on a smart plug, may or may not eventually help with a native detection, but you shouldn’t count on that. Here’s why - most of the devices that you are likely to put on smart plugs are DC electronics. Those almost always don’t conform to the main Sense native detection model, that looks for quick and clear on and off transitions. And even putting a device that does use the type of transitions that Sense looks for, it’s not a direct learning process - Sense collects the data over the long haul, and eventually enhances their general models based on the data. They don’t learn the specific device in your house.

Given that, my approach to smart plugs is to use them on most of my electronics.

1 Like

Two sort-of conflicting responses. I think Todd understood me better. It doesnt make sense - pun intended - to put all your devices on smart plugs if it doesn’t help sense discover them, which is precisely what I was asking.

I can’t even get my Tesla detected, or my window AC units, my refrigerator, my sump pump, my televisions, or my stove (100%). I have many appliances that just don’t get “detected” and I want to fix that. I’ll never understand why, if a device has a nameplate, and it’s current draw is consistently the same, I can’t just manually add it. But whatever, I guess the device is good for a general overview.

Maybe I can give a little more clarity. There are different technical types of device identifications in Sense-land.

  • Native detections - for device that have clean on and off transitions in the timeframe Sense is looking for (1/2 second transitions). Sense learns these.
  • Special Native Detections - for devices that use lots of power, but don’t have on and off transitions that fit Sense’s traditional detection time windows, mostly DC-based HVAC and EV charging. Sense has developed special models for these, but they are very dependent on picking up the complex power ramps and waveforms that specific vehicle / AC /Heat pump models kick out. My Tesla detections have been found, then lost numerous times because Tesla software often changes charge ramps.
  • Smart devices assists with detection - Sense uses readings from a smart device to assist with learning. The actual device still needs to have clear on and off transitions, but Sense gets clues from data coming from the device. Two flavors of this that I am aware of:
    • NDI (network device identification) - Sense monitor uses network traffic to spot ons and off of some very specific smart TV models.
    • Ecobee Historic - Uses Ecobee 5 minute updates to better refine HVAC detection. The data from the Ecobee comes too infrequently to be monitored on Sense directly (bubble would be off even though AC has kicked in), but is very useful in training Sense and improving Sense HVAC models.
  • Smart device measures “detection” - where the smart device regularly updates Sense with the actual usage based on a measurement or calculation from the device. This data can be used as “ground truth” for training Sense, but isn’t directly used to “learn” the specific waveforms of the connected devices. Measurement comes in a few flavors today:
    • Smart plugs like HS300 and KP115 - report back to Sense monitor every 2 seconds or so
    • Hue lights - the Hue hub sends calculated usage info to the Sense monitor.
    • DCM (dedicated circuit monitoring) - additional Sense CTs that report power usage on 1-2 specific 120V/240V circuits in your house.
  • Sense is working on an improvement called Progressive Device Detection - more on it here:
    [Video] 2021 Data Science Device Detection Updates

Because they are doing actual measurements, the second to last category (with smart plugs and DCM) include the Always On for a device, the rest likely don’t. Example - I have Ecobee Historic and Sense now does a darn good job of identifying my AC units, but it still misses the 6W or so that are always flowing to the electronic control board and thermostat.



Here’s a perspective from someone who has the same issue. I myself don’t have my stove detected, my cooktop, my TV’s, my EV charger (although it picks up 1 second of charging every night at the very start). I have a total of 2 natively discovered devices, my AC and my Well Pump.

I sometimes wish that I could instantly tell Sense what is what, but I CAN do that with my Phyn. And from experience let me tell you, at least Sense gets it right consistently. With Phyn, I tell it what something is and a week later it thinks it’s something else. So I personally think the Sense people are on the right track, it’s just a slow track and patience is not easy sometimes. I understand that.

I also feel bad for Justin, Kevin and the rest of the folks at Sense who get the same discovery time questions day in and day out. Being that I work in customer driven development, it’s not fun fielding the same request for speedier discovery and I think if they could do it without compromising accuracy they would.


I mean, thats why I brought my question here. God forbid if I have to ask questions directly to the people working at Sense. lol


I’m not criticizing you for asking a question, be realistic. My point, which obviously must not be clear is that if they could make device discovery something that would happen on day 1, I’m sure they would because that way they wouldn’t have to field the same questions.

Thanks for the very detailed response.

From what you said, wouldn’t you agree that after maybe 6m to 1y, my sense monitor should have picked up the things I’ve listed? Things like a sump pump, with a solid on\off cycle when there is water, and what I would expect to be a relatively consistent current consumption, shouldn’t be too hard to pick up, right? My A/C units, on, fan, compressor, or off. Refrigerator basically the same. TVs are just on or off and on the network(more of a standby?). Etc.

I totally understand the complexity and breadth of learning that needs to be done. I just feel like the learning could be helped if we could specify non-complex items. I think the most reliable item that appears on the bubble chart is my dehumidifier in the basement, which is almost as non-complex as it gets :smiley:

1 Like

No doubt. I just wish there were things we could do to help it along instead of feeling completely helpless. Make it feel like a kickstarter product.

Like I said I share your frustrations. I didn’t join this community as a pleasant person. I joined because of necessity because I had a defective CT and I was quite the unhappy camper.

Over time I’ve understood a bit more about the team and that they actually really care about making it “right”. So I’ve stuck around and offer some advice when I can here and there based on what I’ve learned. I’ve also screwed over some people probably with my incorrect assumptions, but let’s imagine that never happened.

I worked with the IBM Watson team many years ago and would frequently sit in a sky scraper in SF going over how AI works. I’m not assuming you don’t know yourself, but the basics of it is this. You have to have really large sample sets in order for it to work. You also have to have ground truths, in other words you have to correlate something to a fact. This is basically what you are asking to do, correlate an electrical signature to a fact. The bad part about doing that is that you can screw up the whole thing for everyone else if you’re not 100% sure. I’m not assuming you would mess it up, but I can guarantee one of their customers would. It would probably be me because I like jumping the gun.

So let’s say 100 customers have sump pumps and 75 have a Grundfos brand, Sense can probably figure out after looking at a lot of energy signatures that match that it’s something that’s recognizable so it asks in the app, hey what is this. Customers then say “oh that’s my pump” and then they have some ground truth. But if you have that one single odd ball pump, there’s nothing to compare to, so it just holds that signature in it’s database waiting for more and more and more.

As Sense grows as a company the data will grow and the recognition of devices will be better. I hope that in a few years it will be much better and I’m confident that it will be.

1 Like

I agree in spirit, but I know that we, collaboratively, would screw up the recognition of devices.

I agree that mid-sized to big AC motors should be picked up fairly quickly. Sump pumps might have the problem of not turning on enough - a few hundred cycles might be need. And one has to be very careful in separating out AC and DC motors. Many washing machines, fridges, etc have sizable motors but they are DC and sometimes variable. Same for battery back garage door openers. The electronics that control these DC motors can effectively “hide” the on/off characteristics.

I think we all wish we could teach Sense simple stuff, but machine learning hasn’t progressed to the point where it can on and off infer behaviors under a wide range of conditions. The actual detection it learns is a multi dimensional grouping/cluster of maybe 20 on / off transition parameters, not directly the time domain shape of the waveform.

that never happened
what? :wink:

So, lets use our Tesla vehicles as an example. Im not sure if you’re a part of the public beta, or have the current autopilot. Lets say the car makes an error while driving. You can report it, some guys like yourself in the dungeon of the tesla buildings review the error and tweak on the ML IO to fix the problem (or in many cases in my experience, not fix anything). I am more than willing to take the extra steps to help it along, if we’re going to help the development of “AI” or machine learning and in the end provide us with a better product (assuming there is some end in sight, knowing theres really no end to driving scenarios or consumer product current wave forms).

Yeah I totally get that, but regardless of how often it turns on or off, it uses a very limited range of current, so you can very easily infer how much power it is consuming.

Sometimes you make me laugh, so that’s a good thing.

I am part of the Beta and my EV is not recognized by Sense. But honestly I don’t care if it is because Chargepoint gives me the data I need on it anyway.

Maybe you can be trusted by helping, but let me share an example from today. Someone in another thread claimed that their Dyson stick vacuum is pulling 100w of power. They would have classified that as the vacuum, they said as much. A couple of us chimed in and said “what …that’s a little high”. You have to realize that not everyone is gonna get it right, and a lot of people can screw up a lot of stuff in no time flat.

It absolutely does make “Sense” to use Kasa smart plugs to monitor all those devices (like my complete computer center) that Sense never will find anyway. That lets me monitor usage and lowers the mysterious “always on” collection. I’ve never hoped that Kasa monitoring would “teach sense” in any way.

1 Like

This is the best way I’ve heard this summarized… for those of you that have been aware of energy monitors for a while, several are no longer around that switched to a “manual” (or semi) detection methodology. There are reasons that this approach isn’t being pursued at scale.

We’re going to have some more informative ways to present you with data that can help you early on - the foundation is centered on a reworking of real-time detections our current real-time detection methodology that @kevin1 mentioned further up in this thread. We’ve also done some significant work to make future integrations easier for our team to implement, and believe this is one of the primary ways we can impact device detection moving forward.


@addohm ,
Here’s one way to “see” what Sense is seeing, at least as far as traditional detection is concerned - let your Power Meter go for a while in iOS/Android app (specifically NOT the web app). The app Power Meter “tags” what I believe to the be the “transitions of interest”, the ones that meet Sense’s criteria for traditional detection, with either transition power numbers or with the name of the already-detected device. Here’s a view of some of the tagged transitions, both undetected and detected, in my house.

If you have your AC or sump pump turn on and Sense doesn’t tag the on transition, or breaks it up into several tags, then the transition doesn’t fit the model for traditional detection. Also watch the off, for a similar negative tag, because Sense can’t detect an on without a matching off, and vice versa.

If your AC or sump has both on and off transitions being tagged, then Sense either doesn’t have enough repeats, or the on/off data doesn’t pass the uniqueness criteria - that would mean that it’s on and off signatures smear together with what seem to be other devices, or are so broad in variation that it might not be one device.

1 Like

@addohm ,

And here’s a partial view of what happens when those “transitions of interest” reach Sense’s “brain” back at the mothership. Each transition has around 20 measured parameters or features associated with it, and Sense drops each transition point into a corresponding 20 dimensional space - here we’re seeing only 2 of those dimensions, some measure of power on the y-axis, and some kind of phase measurement on the x-axis.

Over time, the one transition points create clusters. If the cluster is confined enough and dense enough, and if it has a matching off cluster, it becomes a detection. I think colors here reflect different clusters, though you have to remember that we’re only seeing 2 dimensions of a 20 dimensional cluster.

Picture excerpted from video, here:

1 Like