Wemo plugs as a training device

Since Sense is Wemo smart plug aware, would it be possible to have Sense learn from the usage changes when a wemo device (plug or switch) turns on and off?

1 Like

Would you mind rephrasing this?

Sounds like you’re asking for wemo to learn from wemo?

Only the Wemo Insight smart plug works with Sense - that’s the only model that does power monitoring. And Wemo switches don’t talk to Sense at all. Sense probably does use data from Wemo Insights to improve models for devices, but it’s not like a smart plug is used to learn the exact waveform of just your device on that plug.

1 Like

This should be clearer.

Never type suggestions when you are tired :wink:

I have to admit, I was a little confused initially as well :slight_smile:

Yeah, I agree. I too work in ML and AI and I find the inputs to this very limiting. I have several ML training machines and gaming PC which the Sense fails to see. Several of these use up to 900w and I have three computers listed and even tried to “train” the Sense to learn that 140w or 520w (if dual GPU is in use). I also have some pool equipment that the sense also fails to see which is a large portion of my Other or Always On. I’m not sure I can find value out of this as I already know my two AC units and primary pool pump eat most of my power but clearly not true when I have 2000w which Sense can see, even when it is directly told what it is with a compatible smart plug… very disappointed. I feel one should be able to define a max and min “measured” power value using one of these smart plugs and enter it beside the data one has in their home. Perhaps the model would be better trained if it actually “knew” my PC computers are rarely going to match most… there must be a way

Ok let’s try to break this down…

If you have a Wemo or Kasa plug on a device then Sense get’s that reading from the plug itself and reports it in the app. So could you clarify what you mean here?

Again, I’m a bit confused. If the plug reports your power usage is 900w then it’s 900w… Setting a limit on that would be nullifying the usage above the limit which I don’t understand why you would want to do. You can however set a limit to trigger when the device is considered to be “off”. This is only for reference and doesn’t affect the data reported by the plug or the data saved with Sense.

I have multiple deepstack and facebox machines running in my house and I work in the industry. I’m again confused as to what you are trying to say here. What “inputs” are you referring to. The data from the plugs is not used in any algorithms or training for Sense AI detection. The only input that Sense learns from is the data reported from the CT clamps.

Very correct in that if you have a signature that is abnormal and rarely/never seen then it will not be recognized quickly or even at all.

Sense however doesn’t allow for training and hopefully never will allow users to train. That would not end well.

900w Gaming PC, typically runs at 900w plugged into Kasa plug. I have defined a computer, three of them actually. I’d really like to unplug the Kasa and have the Sense learn on its own. Nobody is turning off computers with graphics these days, pays too much NOT too!

Does that make sense? :slight_smile:

I’ll hop in here.

A Wemo smart plug, not a current sensing one, is identifiable on the network and has its transitions visible to the network. In many cases a name is associated with the Wemo Packet. My router sees it, and a packet sniffer sees it. As that smart plug turns on and off, Sense sees changes in the load on a given phase in the house, in temporal proximity to the device command. It seems that if Sense looked at the network for smart plug activity, it could associate the consistent load changes and the on command or off command with a given smart plug, and in that way, pull the load from the “Other” category to a device, initially called “Wemo1, 2, 3, …”.

In essence, I am suggesting that a second category algorithm be added to the Sense palette, that rather than identifying a waveform, looks for network activity that associates with load changes.

I fully recognize that this is very easy to say, sitting at my desk, but is another thing to actually do.

If you are really keeping your PC/GPUs on all the time at a consistent 900W, you can add them to your Always On device list at 900W, using the + in the upper right. Then you can use your smart-plug on something else.

And here is where your problem lies. If the computer is just using power then it doesn’t have an on/off on/standby signature and it just represents always on. That data is the hardest for Sense to see because it’s just a constant flow of power to “somewhere”.

I would suggest doing what @kevin1 is saying to do. Since you know the constant power that it consumes just allocate that device to 900w in the always on.

@ofangonline Let me explain how this works. For starters your Wemo or Kasa plugs are known on the network because of a number of different things including the mac address associated with the OUI vendor and also the response to ARP and NetBIOS queries. Every device that connects to the network does this and the plugs are no different. Sense sends out a broadcast UDP to the router via a broadcast IP, which in turn broadcasts that query to all devices on the network. For example the Kasa plugs listen on port 9999, so Sense will broadcast using that port. The broadcast UDP requests a number of things from the plug, including on/off and power consumption. The plugs respond back to the Sense monitor using unicast TCP. So Sense can see the network plug activity and can control it as well. This is all currently doable with Sense. It is also doable to pull that load out from the “other” … basically it’s: Total Power Consumption - Known Devices (including plugs) = Other (grey bubble). Likewise this applies to the always on. When a plug is connected to Sense it analyzes the always on load and includes that in the provided data.

@kevin1 is the electrical engineer I think he would be better suited to explain the waveform comment. I believe there’s a technical different between a waveform, that you suggest, and the electrical signature that Sense recognizes.


Two thoughts:

  1. Sense does try to do something similar to what you are describing with NDI - it can associate the on and off power transitions of some smart TVs by looking at the network traffic. But from what I have seen, there are a number of challenges with this approach - different makes/models use different protocols, sometimes the transitions aren’t so clear.
    How does Network Device Identification make Sense smarter?

  2. Many devices that one could put on a smart-plug don’t have clear on/off behaviors, when a smart-plug is activated or turned off. For instance, a computer on a smart-plug might jump to a low-standby level when the plug is turned on, but that’s not indicative of the actual power usage in operation.

1 Like

I would posit that most devices controlled through a non-current-sensing smart plug would have relatively stable draws: lights, fans, and etc. Those that do not have stable draws would be likely to have more unique signatures: devices with compressors, computers, appliances,…

In the end precision is not what a person using Sense should reasonably expect because Sense is prone to misidentification precisely because it is looking for signatures. This is OK. The value I find in sense is it helps me identify power quality issues, exceptional device issues and it gives me an understanding of my use patterns. Where I need precision I have used probes (my car charger), and current sensing smart plugs.

Curtis K
(Removed personal information)


For context here, we have several points to monitor if desired with these types of matters. For one, these do get reboots and do so typically weekly for security patches, etc. So these is an ON/OFF happening a few times. Also, these are Corsair AX series power supplies and therefore I have full insight into their actual usage and draw overall as a system along with further telemetry from the UPS systems they are plugged into. Now I have the Kasa plugs on top of this.

Thanks Kevin. Will the Sense metrics still provide me the insight as to how much actual energy they consume after I remove the plug and still permit me to track it for notifications if it goes offline?

Complicated answer if I go back to your original question about PCs with GPUs that are Always On. You have two options in my mind.

  1. Keep the PCs plugged into your smart plugs long term if you want to see their exact usage patterns and consumption over time. I do this for my household PCs.
  2. If the smart plugs show them to be Always On at a nearly constant power levels, you can let them slip into the aggregate Always On (take them off smart plug) then just do as I suggested earlier - add them as Always On components.