Device detection - Option to not be a part of the data pool

I am new Sense but I understand there probably have been thousands of requests to name your own device when you turn the device on. Can there be a mode for a person to name all their own devices within their home? In doing so, Sense would ignore their data. One would be divorcing themselves from the Sense community. If they want to jump back to the Sense community, the mode would be switched off and their device settings would be wiped clean.
This would allow a person to quickly understand their home and correct any big issues/power drains. Once the problems are addressed, then one can jump back on board. Otherwise, I think Sense is missing out on a bunch of customers tired of waiting for device detection.

Sense has never detected my dehumidifier.

Hard for me to understand your request. Are you suggesting a mode where Sense does initial detection and identification solely based on your home data, vs. relying on all previous data and models from the whole user base ? I’m guessing that could be a very long wait and might not even be possible because you wouldn’t have any tagged data (pre-identified on and off-events) to bootstrap the learning process.

1 Like

I will do my best to explain. I am suggesting that in this “mode”, the owner would NOT utilize the database or contribute to it. A lot of users are technically capable of inputting make/models of their equipment. For example, I can cycle my sump pit pump and see the spike on the energy draw. I would like to label that spike immediately. I personally can do this in short order on a majority of my appliances. Question for the team; would I have to cycle the pump x amount of times to get the right signature? Does an appliance have roughly the same signature every time it cycles? If so, I can cycle each piece of equipment that I am concerned about to understand consumption. I can probably do this in short order versus waiting for Sense to do it for me.

You stated: “I’m guessing that could be a very long wait and might not even be possible because you wouldn’t have any tagged data (pre-identified on and off-events) to bootstrap the learning process.” Under my scenario, would this still be the case as I am forcing the on/off event and registering the device.

I guess I don’t see why it is so hard to create a manual learn mode as long as it does not muddy the database. The app could ask to toggle a piece of equipment on and off, denote the spike, and the homeowner could label it. Then repeat the process to your heart’s content.

Machine learning doesn’t work like you seem to assume. Here’s some worthwhile reading that explains why “manual learn mode” really isn’t possible, even though it seems like a cool thing.

Several companies that have tried to add “manual learn mode” (Neurio is one of them) have since given up on device learning.

Understood. Yeah, if it was simple, then it would have already been done. Chalk it up to a newbie in the Sense world. The device signatures must be a pain in the @ss.

Thank you for taking the time to explain. I am looking forward to seeing what is possible

1 Like

I think a lot of people have the same question / idea because Sense does such a good job of presenting power data to humans in a way that is easy for us to process. What you see in the Power Meter is nice half-second-by-half-second snapshot of a complex mathematical computation.

What Sense really sees is 1 microsecond snapshots of current (i) and voltage(v) alternating current (AC) waveforms that look like this with 16,667 snapshots per 60Hz cycle.

I don’t know the exact approach Sense takes, but the number you see is some form of the average power (Average Power) over some number of cycles, not the instantaneous power § at that exact microsecond in time.

And the computation can even get more complicated if you have a big motor in your house, where the coils on the motor force the voltage waveform to lead the current waveform by a small time difference or phase angle (phi).


But Sense looks at all of this data in great detail and uses the exact voltage (v), current (i) and phase angle (phi) data from every microsecond over a fairly short span of time, at most 1/2 second or so, to detect a device turning on or off.

So when you are able to recognize a pattern in the Power Meter, remember that that is a highly processed and tuned perspective on power/energy usage, that makes it easy for you to see what is going on in timeframes humans understand.

What’s interesting is that Sense is now working on looking at longer patterns to help with recognition, but that entails figuring out what data Sense needs to pay attention to, and what it can ignore from the gigabytes of data from 3 hour changing cycle for an electric car, or the 20 min wash cycle of a washing machine. Sense does 4 million samples per second of voltages and currents, using 2 bytes per sample, or nearly 29 GB per hour. Of course the Sense monitor in your house doesn’t send all of that data back to Sense central.

1 Like

After reading up on the problem, the baseline signature seems to be a moving target at any given moment. Hats off to the team trying to make sense, ugh, of it all.

1 Like

This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.