Having used Sense now for 8 months now, a feature that seems to me would really be helpful is the ability to “lock a device signature in”
Problem I’ve witnessed:
Over time, an accurately identified device will morph into inaccuracy. The most common morph will be the device reporting more than total use. Almost taking a different power signature and blending it. My pool pumps for example turn on 15 minutes apart. I have 2 and they are on a sub panel from the main. Both were initially detected accurately but over time, the 1st one to come on has inherited some of the increased draw of the 2nd one starting. Even though the 2nd one was also an accurately detected device itself. I’ve swapped the timer settings making the 2nd come on 1st, now both device signatures exhibit similar “15 minutes in” increases. Obviously this causes total device usage reported when both are running to be greater than total house draw.
Proposed solution:
Rather than a device changing itself uncontrollably, allow us via the app, to lock a known good device into not changing itself. If later sense desires to change it, have the app altert us. Something like a message that tells us a “heads up” that sense believes this device signature has changed. This notification could give us some options like
proceed with the change
engage support to investigate
contact an electrician to check physical appliance
discard modification and consider new device creation
Options 2 and 3 would not immediately join the change to the existing device profile
Option 4 would create a new device leaving the existing locked one alone or discard the data as an anomaly
I would think this type of feature fits into the vision of sense reporting possible appliance issues before a complete failure of the appliance as well as greater sense accuracy.
@AZAaron,
You might want to make a distinction her between on/off identification signatures, which are currently very short in time, and the “on-power profile” which is the time dependent power usage of the device when it is on. Right now, as I understanding it, the on/off signatures that Sense looks for are less than a second in length. What I’m hearing from you is that you want to lock in the on-power profile associated with the device, not the on/off signature per se. Is that correct ?
Of course this could all change as Sense invests in classifying devices that have longer signature windows.
Here’s a screenshot of my described pool pump example. Initially this device in sense consumed just under 2k total usage for the duration of its runtime. The morphed that happened and shows in the graph carries through the remainder of the runtime and is totally incorrect.
This is one of multiple devices that have morphed over time in this way.
Are you sure the pump isn’t working harder now because it’s dirty, has a clogged filter, worn bearings, etc and therefor actually is using more energy?
Well the total wattage used by the house is less than the sum of the detected devices. And this isn’t an issue isolated to my pool pump. Since February when I started using sense, detected devices lose their accuracy.
Same sort of deal going on here, the first part of the pattern where it goes up and down is legit but as soon as my well pump turns on it flat-lines for the remainder of the cycle at 4489W.
Well I was just hoping to suggest a feature giving us more control of once accurate device signatures morphing but I guess we can discuss pump issues…
For the record, over the 8 months I’ve been using sense, I’ve seen similar behavior with my water heater and AC systems. Reporting ok for a couple months, then morphing into a behavior of claiming to use much more power. On one of my AC units, I could walk around the house turning other non detected, high draw items, on and off and watch the power draw of the AC increase and decrease with it. Obviously a broken model thay wasn’t broken until sense modified it.
I’d love a feature to stop that later modification and just say, “hey sense! This is good! Leave it alone!”
This is definitely something that we’ve thought about, but like most things related to ML, it’s a bit more complicated than it seems on the surface and ‘locking’ a signature in place could have unintended implications for further detection work. We’ll keep looking into it as it’s good idea for sure.
3 of my devices that were reporting too high, I deleted this week for hopefully better re-detection. 3 others that have been reporting too high, I’ve just this morning opened support issues on. 2 of wich were accurate for a month or so then morphed inaccurate.
Maybe they’re repairable without deleting. I really don’t like losing the historical data. But inaccurate historical data isn’t useful either.
