New Sense user here. Going on about 3 weeks now. I have a question on device detection. I’ve seen a few posts where users will compare their Always On numbers and discuss turning things off to lower it. My question is about how Sense itself treats Always On versus Other from a detection perspective.
Am I assuming correctly that Sense is attempting to find matches to devices within the Other category while ignoring the Always On because there is no Start/Stop signature to identify?
If the above assumption is correct, wouldn’t device detection be most successful if we just left our Always On devices alone until our Other category was zero? And only then start playing with putting Always on equipment on timers or somehow powering them down to help sense “see” them start and stop?
Or is there no difference, at least to Sense, between the two categories and how it’s looking for waveform matches?
In my efforts to “promote” device detection, because I hear loud and clear NOT to “train” , I’ve been trying to leave my Always On simply on, while minimizing the cycling of all the unknowns at overlapping timeframes when possible.
My quick thought. Sense is really only likely to identify devices after they go through many on and off cycles under many different background conditions. So tweaking some of the background conditions daily (i.e. trying to turn off some long-running home devices) to lower Always On, will have minimal impact on detection. Most of the devices, or should I say device power components in the true Always On category, are ones that Sense will never detect because they never have distinguishable on-events or off-events - things like TVs, stereos, computers in sleep mode, garden timers that turn on the lights/sprinklers, etc.
I had a bunch of clusters of devices in my house that had 1-2 devices that needed to run continuously (router, cable modem, server, switches, TiVo, Hue controller, AppleTV running HomeKit), but most of the devices in the cluster could be powered down, so I rewired the ancillary devices (not always on - TVs, AV stereo, other AppleTVs, etc. ) to smartplugs that turn off at night. The smart plugs have an added benefit of tracking power usage of the remaining devices in the cluster. One more for on/off consideration - your hot water circulation pump.
@kevin1, so are you saying that by putting those things on timers it has allowed Sense to detect them for you?
What I would really like is for the “Always On” to be a bubble that can be “clicked” in to… Meaning, I would really like my PC that I leave on 24/7 to be detected… I understand since it never really turns off or on, that it probably never will… But… If I put it on a timer or just manually rebooted it from time to time, it could get detected, right?
But, then it’s “Always On” too… So would it just always show up as a bubble? I really want to track it’s power usage but it will never get found since it’s on 24/7…
@kmccb,
No, not exactly. I’ll give you an example of what I’m saying. I have a AV cabinet that includes TV/stereo/AppleTV/TivoMini/etc. plugged into one Elgato Energy plug. Thanks to the Elgato, I can see that all these devices use something like 25W, even when all of them are turned off (sleep mode). I can also see that in typical operation (watching TV), they use something like 117W. By using the timer/smartplug, I can do several things:
I can drop the 25W to 0 watts from midnight to 6AM using the timer - that lowers the Sense Always On by 25W
I also know that that 25W is really being transferred from Always On to Other, at least from 6AM to Midnight. Sense would never find that baseline “sleep power” anyhow.
It remains to be seen whether Sense will detect/identify the TV, stereo, or AppleTV going from sleep mode to active mode, but at least I have a rough read of how much power is going to the TV/stereo/AppleTV which is part of the Other bubble until if and when Sense does detect them.
@kevin1
Your example is sort of along the lines of my question. I have a few things on timers that also report power usage for the equipment plugged into those timers. As well, I’ve scheduled my NAS’s to power down during quiet evening and early morning hours just to save some electricity. My Always On runs around 400W if I just let everything run. When I enable all of my scheduling, that drops to high 200s, low 300s but shifts an extra 100W to the Other category.
What I’m really curious about with regard to Sense device detection is does powering down Always On equipment at quiet times, thus making it show up in the Other category, create “noise” for Sense to deal with while attempting to discover the devices like AC, Pool Pumps, Water heaters, etc? Or does it make no diference bacause the categories of Always On and Other are more for us humans than Sense’s actual machine learning?
Since installing Sense, I’ve disabled my timers to keep more categorized as Always On and leave less in the Other grouping hoping to “help” sense with discoveries but if it makes no difference to how Sense works, I’ll just re-enable my timers and go back to saving a bit of electricity.
One interesting possibility here (Sense folks are you listening?) would be for Sense to offer WiFi smart plugs - I imagine they would cost 25-50 bucks or so - that report back to the Sense monitor and create a bubble and nameable “device” for each plug. That way we could get a reading for, say, the entire entertainment center separate from “always on” and “other”.
@AZAaron,
I don’t believe that two short transitions a day per timer are going to create enough “noise” in the power waveform data to stop or significantly slow Sense learning. In fact, seeing the on-off cycle of other devices with different steady state power baselines might actually help Sense build a more robust model for each device detection. If it helps, I have several timers going on and off and I haven’t seen a negative impact over the period just after I first purchased the Sense and hadn’t put any timers into use.
@luterra,
The more I see, the more I agree. I’ve been investigating what researchers are doing worldwide to train and evaluate their disaggregation-based non-invasive load monitoring (NILMs) systems. In almost all cases, they use a wall plug or breaker box based data acquisition system to capture the “ground truth” for each device in the household. What’s interesting is that most research groups don’t have primary hardware or plug hardware that samples anywhere near the sample speed of the Sense (12KHz for example in the case of CMU’s BLUED dataset vs. Sense’s MHz data gathering), so most researchers aren’t likely using the same algorithms as Sense.
For those who want to delve more, here are a couple of very interesting resources.
A NILM Wiki that points to:
NILM datasets - data used to train and evaluate detection algorithms. Give a better idea of what training data actually looks like.
NILM data on different appliance types
NILM products, including Sense - See appears to be one of the most advanced. Lots of do it yourself stuff as well as industrial stuff.
NILM resources - including technical forums, blogs, plus pointers to conferences on the topic. Next big conference is at UT Austin next month. http://wiki.nilm.eu
Welcome to the community @AZAaron - great to have you here! And thanks to @kevin1, @kmccb, and @luterra for weighing in on this one.
It sounds like the system you’ve put together with timers to reduce your “Always On” load is pretty darn good, so I might recommend you just leave that active to save money and energy. That being said, it’s likely that those devices won’t get listed as “Always On” by Sense, and end up over in the “Other” category as you describe. While this could add some amount of noise when those devices are on, constant noise sources like these “Always On” devices are easier to compensate for than variable noise from devices who’s consumption patterns are quite sporadic.
@BradAtSense
That is helpful. Thank you for some insight as well as everyone else. I’ll just let things fly shutting things down at night and see how it goes. After about 9 discovered devices and some multi-Fridge arguing with Sense, discoveries seem have stalled for almost 2 weeks now so I’m hoping the dry spell turns around soon. I still have quite a few large draw devices that are used frequently. I just wanted to make sure I was doing what I could to help it along.
, I’ve been trying to leave my Always On simply on, while minimizing the cycling of all the unknowns at overlapping timeframes when possible.