This all day. I have 8 smart outlets that report wattage like DCdyer, along with a DSB09104 that uses amp clamps like the sense that you can put on a single circuit (in my case a hot tub) that tells me exactly how many watts it is pulling all the time. This would be a great way for people with 240v applicances and cars to be able to help sense learn what is happening and what power is being used.
+1 on this thread. There is a TON of information in a smrtthings network that would make sense MUCH smarter. The smart plugs give direct energy listings (like TP-link) and the switch status provides energy insights. Iām sure the community would have an integration out in a matter of days if there was an open API. I have to believe smart things has a much greater bang for the buck vs. the existing integrations with TP-link or wemo.
Two issues with your basic assumptionsā¦
- Energy sampling is far more valuable than switch status (on/off) when it comes to āground truthā for machine learning.
- Fast, regular energy sampling is critical. The current HS110 and Wemo Insight smartplugs are sampled at somewhere between 1 and 2Hz (1-2 times /sec). I have seen numerous postings in SmartThings-land like the one below, that say Z-wave and zigbee meshes arenāt up to that taskā¦ Maybe Iām missing something, but are there SmartThings power sampling devices that communicate via WiFi ? Thatās probably the real gating factor.
On the flip side, there seems to be a big overlap between SmartThings users and Sense users, so there could be good leverage if the right parameters are met.
3 Likes
I would assume the customer base trumpās what is technically simple to implement. Smart things has a massive community whereas TP-link is a niche product.
Smart things requires a different approach, you wouldnāt want to sample at a few hz unless customer complaints about crashing the network were a project goal. One minute polling is probably a realistic goal. For many devices that would be plenty accurate to garner better ML info. This switch was on a minute ago and is now off, power dropped by X is a pretty strong signal, etc. In a smart things home, you can literally know the state of every light switch, window and appliance ā not to mention the integrations with other apps, like nest that tell you the heater is on.
Love the Cornell avatar 
1 Like
Hey thanks, on the avatar ā¦ Love the newish tough bearā¦
Agree that once per minute sampling would be sufficient for simple ādigitalā on/off devices with a single consistent power level like lighting, if the deviceās off and on periods are many multiples of 1 minute. But outside of lighting and small fans, I canāt think of anything else that behaves like that.
I have a few reasons for believing that the 1-2Hz sample rate was carefully chosen for maximum benefit.
- Many of the power waveform features, like spikes and dips I see in both the Power Meter and in detail from my smarplugs, are 5-15 seconds in size. 1 second samples are sufficient to catch them.
- Most recent research datasets for power disaggregation, like REDD or ECO use 1 second samples for ground truth.
- Going too small, also brings complications. Going much below 10x per second, you start worrying about sampling alignment with respect to the 60Hz power cycle and about accurate data interval alignment.
ps: I can see integrations with smart thermostats as being quite useful, since a significant amount of electricity flows to both heating and cooling.
Yeah, the bear in my day was pretty sad 
The power spike of a motor kicking in is definitely a 1hz phenomenon. I think you can see major appliances by the steady state in a longer time scale. An electric dryer is a pretty unique load as is a dishwasher. One option is to try to finger print the dynamic load, another is to look at power and dwell time. Nothing hovers at 4kw for 30-45min in my house except the dryer (followed by 10min cycles).
In any event, wouldnāt it be awesome if the sense SWE team solved this so we could end the debate?
Sense has solved the issue for me. My dryer is detected everytime.
Having a few smartplugs installed was eye-opening to me. I thought my washer might have a few simple modes/patterns, but a basic cycle looks like this:
Sure, there are places in the waveform where power is constant, but the multiple patterns are defined by the spikes and changes. A one minute sample would average most of the changes in this part of the patterns away.
Sense has detected some of our big loads like my dryer heating element as well as my furnace fans, but misses other components (dryer motor, furnace igniter and ignition blower). With the furnace (below), some of those auxiliary component features would only be resolved by a seconds magnitude sample (or smaller).
Even though it runs contrary to your home automation strategy, I would suggest you blow 30 bucks on a pair of HS110s and see what the unruly plug-in power consumers in you house look like.
Go Big Red !
1 Like
Its going to vary ALOT by washer. You can see mine kicking in here and its very easy to detect on a low duty cycle. The refrigerator compressor has the telltale spike and ~1hr duty cycle, whereas the dishwasher looks like a resistor ā this would be pretty easy to detect on 1/60Hz.
I have the TP-Link as well as smartthings. You can do a lot more (as its a much higher power standard than zwave). I agree its better, but Iām not going to rewire my home just to get a better sense signal ā so the question is can they build (or enable the community to build) an integration with smartthings that extracts what is knowable.
2 Likes
Yes there are. For example my Sengled Light Bulbs are connected over WiFi to my SmartThings hub and do report power usage through SmartThings. I do not know what their sampling rate is though.
That said, ZigBee/Z-Wave should be fine for energy sampling as Philips Hue does it over ZigBee and is supported by Sense, so why couldnāt others do this?
Zwave energy meters are far more configurable. One device could be set to send power every x seconds, another set to send power on x wattage change. Sense would have no idea how the device is configured by the user. This plus cloud delay of smartthings is probably going to be quite the challengefor something sampling as frequently as sense.
You are right Frank,
ZigBee/Z-wave is sufficient for the event-driven Hue-type power reporting model, where the controller, rather than a power measurement device, updates the power usage in response to changes in the switch setting (physical or virtual switches). That model relies on the devices to have a time-invariant power usage that is a simple function of the line voltage and their intensity setting (0% to 100%). Zigbee can handle this model because the bandwidth requirements are much lower, and latency can be much greater. So yes, Smarthings could report power usage via that model, to Sense.
But that model doesnāt work for most devices. Lighting, and specifically LED lighting, is very well behaved over time (though all lights have a warm-up cycle, where the load changes a little). Some single speed, constant load motors and perhaps other devices may also be consistent enough to use this model. But eventually you get back to the other challenge, learning. Even though Sense obtains the estimated power data from the controller, the data is less desirable for learning. Why ?
- Features - may lack critical waveform features, that sampling would have provided.
- Latency - reporting delays may skew the data. I have seen Hues and Smarthings that have a half second or more in switch to light latency.
- Redundancy - why try to learn the device if the data is already provided, but is less amenable to machine learning.
Yes totally agree with those points, but that said Sense is using Hue w/ these āknownā problems, so why not use SmartThings as well, at least for āthe time beingā until the ML gets to a point where these kinds of integrations are no longer āneededā as Sense can then detect them through ML!? I think the ML models are still far from accurate enough to even start to recognize lights, etcl.
Yes there may be certain SmartThing devices that have a variable power model, although I donāt know of any from the top of my head, nor do any of my 50+ SmartThings devices (that matter, ie that are connected to the power grid, and not run on a battery) have a variable power model as theyāre all light bulbs and a āsmart switchā.
1 Like
@frankwin.hooglander, that approach would work, and if it is easy to do, would be a nice addition. But most of my response was directed at the proposal to have Sense read Smarthing smartplug data ahead of TP-Link and Wemo WiFi-based devices. I still contend that that Zigbee/Zwave-supplied sampled data is less valuable to learning, than data flowing from WiFi smartplugs, due to bandwidth limitations and latency.
I agree with Staze. Iād like to be able to hook Sense into SmartThings.
In fairness, Iāve switched to Hubitat. So Iād much prefer integrating with thatā¦ =)
Alternatively, exposing individual devices to Alexa or Google Home via their integrations would work too. ifttt is just too slow for a lot of itās integrations. =/
Native ST or Hubitat integration would be great though. =)
1 Like
++
Also if any Sense engineer looking into this happy to work (for free) w.r.t. device integration. Have already done a few for SmartThings and donāt see a problem.
What I have seen going over all the reviews The main complaint is the long learning curve of the Sense to discover devices. Many have taken them back for that reason.
If Sense was to work with Smartthings and started pulling the information about the devices. The speed of learning should increase a great deal. We have over 100 devices that our Smartthings hub talks too! granted over half are battery devices which would not count. But the faster Sense can learn and the more deviceās it knows about the better not to mention increased customer satisfaction.
1 Like
I recommend reading through the rest of this thread. Itās unfortunately just not so simple. The benefit of the data SmartThings could provide is less helpful than you might think.
3 Likes
If Sense looked at the IP addresses of the devices on the network it would find 90% of my devices and what they are within seconds. Smartthings has all that info already.