Sense already shows that it has the ability to detect changes in load real time, and often highlights them on the timeline. The request is to make these events interactive as follows:
Users are able to select an event and are presented with an screen
Users can enter data about the event to include:
a) Noting that a device turned on and adding a descriptor that may help identify the device.
b) Noting that a device turned off and adding a descriptor that may help identify the device.
c) Add just a personal note about the event or adding in details.
When a user saves the data from the interaction, then in the case of 2-a and 2-b, the data is added to the machine learning algorithm as “hints”. Further, the device with the identifier is added to the user’s database as a “possible” or “unconfirmed” device. The system still attempts to verify the device against known devices.
In this way, the end user can interact with the machine learning algorithm and possibly speed up detection of devices. If option c is used by the end user, it can help them in their tracking and end use of the data itself or allow them to flag unusual or “one-off” events.
@dunham.de,
All good ideas if the the human-visible events that you describe were the ones that Sense was actually using for identification today. But read this blog:
Sense is currently using sub-AC cycle measurements (one cycle is about 17 milliseconds) for their detection of on and off events. Right now the power meter display you see only displays on a half second interval. Until Sense offers a new UI that let’s us probe and flag events with millisecond accuracy, there’s probably not much value in the events we flag. The best you could do today is to flag a human-visible event that occurs hundreds of milliseconds away from the point of interest for the Sense probe.
But there is value in annotating events that Sense detects.
I read the post on why you can’t train Sense, and that all seemed reasonable.
I have an idea that would help me name devices that Sense finds. I’d like to tag certain events, such as a device coming on, with a label, and then have all my labels show up in the power meter history of a new device that Sense has detected. Even tagging points in time would be sufficient; It doesn’t need to be an “event” per se.
For example, I just turned on my electric teakettle. The overall power meter jumped by 1500 watts and stayed on for about a minute, then turned off. I would like to tag the start and end points of that episode with the label “kettle”. Then in the future when Sense says “I found something - name it”, I can see which of my tags line up with the thing Sense found.
Sense wouldn’t try to use these labels at all. It would simply allow me to create them, and then show them to me again. Seems feasible.
This addresses the pain point that I experience in which Sense has found something and I just can’t figure out what it might be from the waveform in isolation.
I just turned up my thermostat and my boiler came on. It would be nice to able to label that fact. My daughter turned on the toaster. I doubt I’ll recognize that later when Sense detects the toaster waveform. Just started the dishwasher. Just used the espresso machine. Just plugged in my car. Sure would be nice to tag these time points and see them later.
As a workaround, I guess I could take screenshots and then search through them later when Sense finds something.
I just tried the screenshot workaround but nope, I can’t judge the height of the event from the screenshot, So that’s not going to work very well. Plus, the searching process is going to be super tedious when Sense detects something.