Clever algorithms + ST = magic (possibly)


(Gray) #1

This kickstarter sounds really interesting and clever:

My understanding is that it plugs directly into your circuit breaker and communicates with the cloud and (hopefully) ST through wifi. The cloud stuff is supposed to allow it to use algorithms to analyze what changes in your consumption levels mean. Did your laundry finish? Did you turn on your dishwasher, meaning you’re heading to bed? Their stated intention seems to be to allow you to trigger ST events based on those things.

I have no idea how well they can make this work, but I’m definitely curious. If they really are good at data mining and analysis, it could be incredibly powerful–and a simple way to pull a lot of information into ST. If nothing else, it would be a $125-$150 device to pull your total power consumption information, but it could be a lot more.


(Todd Wackford) #2

The campaign video is pretty vague. If someone has more details I’d love to here them. Otherwise I’m gonna hang back on this one.

@Gray, did you back them?


(Gray) #3

@wackware: I thought the video was incredibly vague too. There’s a lot more information further down the project page, and it looks like they’ve been pretty responsive to questions, if a bit vague.

I’m backing it at this point, though my working strategy for KS these days is to back and see how well they respond, and how much detail they provide when they do. But it seems like an interesting enough idea that it could be worth $125 to me.


(Todd Wackford) #4

I’ll live the KS through you on this one. It just feels too much of a solution in search of a problem.


(Gray) #5

Couldn’t all of this Internet of Things stuff easily be classified as a solution in search of a problem, depending on your situation? :slight_smile:

I see it as a solution to the problem of putting a sensor on every device. Particularly for appliances like washing machines and dishwashers, there aren’t easy sensors I can install. And for things like TVs and stereos, I’d rather not have to buy a sensor for each one. So if a clever approach like this can deduce when those are on from the main line, I can get all of that information without having to put a sensor on everything.

Of course, whether they pull it off remains to be seen. Particularly since I threw $100 down the Instacube well (encouraged by the endorsement of, ahem, SmartThings), I don’t have super high expectations of any Kickstarter. Though Spark Core does appear to be on track to follow through on all of their promises, albeit a couple of months late…which is months ahead of schedule in Kickstarter time!


(Chrisb) #6

I’m not sure what advantages Neurio brings over SmartThings. About the only thing I can see for sure if the ability to detect which large appliances are turned on/off. Now, to be sure this would be nice to know if some cases, but it’s really not that majorly helpful and it’s also something that could be done with SmartThings.

Unless I’m majorly underestimating neurio, I can’t see how it would be able to detect if a light was turned on or off, and certainly it wouldn’t be able to tell me which light. It looks like it just monitors for the amount of energy that is being used. Maybe it can tell which circuit breaker it’s coming from as well, but I have multiple rooms on one breaker in some cases. How can it tell that the light is in my son’s room vs. my daughters room vs. the living room? Or if it’s a radio instead of a light?

Plus there are ways with SmartThings to tell if my washing machines is on or not… a vibration sensor would be able to do that. A vibration sensor or temp sensor in the right place on my dryer would tell me if it was running. Likewise the dishwasher or stove.

Other than actually getting a handle on how much energy I’m using and when, I’m not sure what I gain with this.


(Gray) #7

@chrisb:

Unless I’m majorly underestimating neurio, I can’t see how it would be able to detect if a light was turned on or off, and certainly it wouldn’t be able to tell me which light. It looks like it just monitors for the amount of energy that is being used.

The point of data mining is to be a little more sophisticated than that. It should be able to figure out that the type of current being drawn is powering, for example, your A/C unit, plus a TV and two lights. Of course if multiple lights are similar, it can’t tell them apart, but it should be able to tell whether two or twenty are on.

Plus there are ways with SmartThings to tell if my washing machines is on or not… a vibration sensor would be able to do that

You know, I’ve tried that. It hasn’t worked at all for me. Has that worked for you?

I don’t mean to defend this project that I have nothing to do with, but I’m intrigued by it. It’s not a ST substitute but a complement, and a solution to a particular problem: It’s hard to put a sensor on everything. It’s expensive to put metering plugs behind every appliance, and difficult/impossible to do with bigger appliances like washing machines and electric stoves.

Say I just want to know when these things are on/off:

  • washing machine
  • dryer
  • A/C
  • refrigerator
  • dishwasher
  • 2 TVs

Even if I stop the list there, I’m pretty sure there’s no way to put sensors on those things for $150.


(Chrisb) #8

@gray

You know, I’ve tried that. It hasn’t worked at all for me. Has that worked for you?

Nope. Haven’t had the need for something like that personally, but I’ve heard others have done it. Don’t know their results. I suppose it would take some calibration. How much vibration is enough to say it’s on vs. just got bumped? How long does vibration need to be off before it says it’s off vs. just between cycles?

I don’t mean to defend this project that I have nothing to do with, but I’m intrigued by it. It’s not a ST substitute but a complement, and a solution to a particular problem: It’s hard to put a sensor on everything. It’s expensive to put metering plugs behind every appliance, and difficult/impossible to do with bigger appliances like washing machines and electric stoves.

I definitely agree with everything you said here… I’d just be interesting to know how they can tell what’s on or not based on a given number. I suppose if you were able to calibrate it finely enough so that you had a specific number for each thing in your house, then you might be able to do it:

“The meter shows 736.82 Kwatts used. The only way to get exactly 736.82 is if you add the TV, this light, the dishwasher, and that radio.” *(Please don’t laugh if that value is just silly… not an electrician… just throwing a number out there for example. :slight_smile: )

But can it calibrate that accurately? And do devices draw that power statically enough so you can say with precipitation it’s this item that draws 15.25 vs. that item that draws 15.03? How can it tell if you’re running the dishwasher and a bunch of other electronics or just the dryer and a handful of electronics?


(Gray) #9

@chrisb:

I can’t even get a vibration sensor on top of my washer and dryer to say it’s detecting vibration except when I slam the door. So yeah, I guess they’re not sensitive enough. Or I need to balance my laundry less? :slight_smile:

I definitely agree with everything you said here… I’d just be interesting to know how they can tell what’s on or not based on a given number.

Two things:

  1. I don't mean to be glib here, but this is the power of big data! This is actually a case where the cloud can be incredibly useful. I see this operating like the Nest thermostat has. Basically, they have anonymous usage data for every Nest user, and have been able to mine that to construct algorithms to predict when people are leaving, when they're getting home, etc. Here they could do the same thing (though of course we have presence sensors and mobile presence for that), and they could also use it to identify what's going on. If they've seen enough TVs to know that this instantaneous change in power is a TV with 99.98% probability, well . . . I'm pretty sure they're more accurate than my ST sensor. :-)
  2. Also, I don't know enough about the mechanics of this because I'm no electrical engineer, but they have more to analyze than just watts drawn:
    Measures active power, reactive power, energy consumed, energy delivered, voltage, frequency, power factor, and current for each channel.

    They’re sampling every second, so unless you’re flipping things on/off in very quick succession, the sensor should be able to pick up with a high level of confidence when each things goes on/off sequentially. If multiple states change in the same second, that’s when the algorithms should be able to do their work. At least in theory.

I think you’re right that it would depend on the accuracy, as well as the quality of their data mining. But presumably the sensor will be accurate enough, and I would hope their data mining is pretty good…


#10

I followed the link this morning and you’re right, it is another interesting gadget to complement ST. I’m curious how much of a job it is to actually attach the CT to the electrical panel. This posting from one of the backer caught my attention… makes you wonder if each electrical load (even the small ones like a light bulb) does generate a unique signature enough to be identified by Neurio.

This is the post I was referring to.

“My guess is that the algorithm looks at the changes in power consumption and phase angle (current versus voltage), e.g. to profile appliances by amount of current used and how inductive/capacitive the load is. For example, when your refrigerator’s compressor and fan turn on, it usually has a very distinctive load (highly inductive, medium current load). I’m further guessing that you have to go around your house and “teach” your Neurio by turning devices on/off and identifying them.”


(Ali Kashani) #11

Hi guys,

I’m Ali from the Neurio team. I thought I should chime in.

@gray thank you so much for starting the conversation. And for taking the time to respond to questions.

First a bit about myself. I studies pattern analysis at university. Half of our team at Energy Aware are masters and PhD’s in that subject. And the other half takes care of our hardware and firmware.

I should point out that we are not competing with SmartThings at all. We’ve already had great conversations with the ST team and are very excited to work together. The way we see it, ST guys have done a great job making smart devices a reality. We don’t want to do something that’s already done well. We like to add value to ST users by bringing in the idea of “home intelligence.” Nest showed the value of leaning things, but why stop at a thermostat?

If you think about, wires in our homes are analogous to a nervous system that connect everything to a central place. Except, in that central place (the breaker panel) there’s no intelligence – no monitoring of what’s actually going on. That’s what Neurio likes to be. If Neurio can watch the home as a whole, it can put the pieces together and understand user behaviour and intent. It can then share that with the user through notifications (your washer is done but dryer hasn’t started, or your oven is still on), or share it with connected devices like ST (no one’s home, so let’s adjust the thermostat and turn off vampire loads).

Admittedly, this is a new concept in many ways, so I completely understand any skeptical views. Hence why I am here!

@chrisb, regarding the algorithms, you’re absolutely right that we cannot see every single device, or tell the lights apart. But for much of what we’ve discussed, you actually don’t need that level of granularity. Detecting if a home is at its baseload is not difficult, so the moment something is turned on (an unusual spike in the baseload power), we know that there’s presence (someone just got home). It didn’t matter which light or switch it was – we can still deduce that someone’s gotten home to turn that switch.

Now let’s add some historical modelling into the mix: looking at the last month, Neurio could detect the hours user leaves home and comes back, goes to bed, etc. (based on watching when the home is at its baseload pattern). So once Neurio notices an event, it can also validate it through its historical models to see if what it has detected “makes sense”.

When it comes to pattern analysis, it’s not just about permutations of appliances adding up to a certain power level. We monitor each phase, which cuts down overlaps by half. We then look at some obvious parameters, such as the jumps on the power data. As it happens, within the same home, there’s usually not that many major appliances that have a similar power spike. Your dryer is likely 4kW, your dishwasher more like 2kW and your washer more less than 1kW. So that already gives us a quick idea of what we should be looking for. We also consider external parameters such as time of day or the weather (a sudden 5am spike is less likely to be AC). We can then introduce things like power factor, or frequency of change in the signal, the duration of usage, the shape of the pattern, the energy use (not just power), etc.

We have a couple of videos that visualize the algorithms to the best we could, so feel free to check them out. We’ve been working on these algorithms for two years. And we’re going to keep working to make them better.

@twack, sorry if the video is vague. We tried to get into more details in the content page. I’d be more than happy to get even more technical here, if you there are any questions I can help with.

Thanks again for the conversation guys. Check out our campaign; I’ll keep an eye for more questions: Neurio: Home Intelligence

Cheers,
Ali Kashani


(Chrisb) #12

@kashani,

Thanks for stopping by and helping to explain a few things. While I’m skeptical, that doesn’t mean I’m right… just ask my wife, I’m wrong more than I’m right I’m sure. :slight_smile:

A few questions:

First, what’s involved in connected the device to your home wiring? Will this be something an electrician has to do?

Second, how “natively” do you envision nuerio to work with SmartThings? For instance, on your project page you have two examples:

Let Neurio control your connected thermostat. Select 3 temperatures: home, away, and asleep. Neurio will set the temperature dynamically based on what you’re actually doing, rather than a time-based prediction, so you waste less energy. This system uses Neurio Cloud, and a WiFi-connected thermostat.

Using an Arduino Uno with Ethernet, a relay shield, and a power bar, you can have Neurio disconnect wasteful appliances when you’re not home. When Neurio detects your return, everything gets powered back up.

I have a z-wave thermostat and many z-wave switches and outlets tied to SmartThings. Will neurio be able to talk to SmartThings to turn things on/off or adjust temp?


(Gray) #13

Thanks for stopping by, @kashani! I’m glad to see that I was at least on the right track in explaining this as Nest for the whole home.

@chrisb: I’m sure Ali can answer more comprehensively, but I believe one answer is that not only would neurio be able to talk to ST to turn things on/off, but it would have to. Neurio itself won’t be capable of controlling anything. You would need something else, like a WiFi-connected thermostat, Z-wave devices hooked up to ST, or a spark core controlling something with a relay.

That’s why I said that it wouldn’t replace ST, but would simply complement it. You can think of it as a set of sensors to add to ST. Smart sensors built on pattern recognition, lots of data, and eventually behavior prediction–but basically just sensors.

I’m also really curious about how this would be installed. I see that they plan to provide a video after it’s backed, but this is what they’ve said so far:

1) Flip the main breaker on your panel, to de-energize it. 2) Remove the cover on your breaker panel. 3) Mount the Neurio Sensor inside the panel using the included screws. 4) There's no cutting or disconnection of wires. Just snap the current transformers around the two phase lines coming in the top of your panel, and plug them into the Neurio Sensor. 5) Connect the voltage cable to breakers from each phase, and the neutral bus bar. 6) Close up the panel, and turn the main breaker back on. 7) Use the mobile app to connect the Neurio Sensor to your WiFi Network. 8) Use the Neurio web portal to make an account and connect to your Neurio Sensor.

I certainly hope there are going to be lots of pictures of how different circuit breakers would look so that non-experts (who at least know how to cut the power and open a breaker box) can install it confidently.


(Ali Kashani) #14

Sorry if it takes me a while to respond guys.

@eibyer there’s certainly a lot of unique signatures when you watch the data. if you get into too small of a load, say things like chargers, then that might not be detectable. Or if you have a bunch of 60W lights, we cannot say which one is which. But something to keep in mind is that 80% of our energy use comes from 8 or 9 main appliances, so there’s plenty of things that can be done without detecting the smaller things. And as I said previous, sometimes we don’t need to know the device to know the behaviour (e.g., someone coming home and turning something on).

@chrisb, cheers! I welcome skeptical eyes, it’s healthy :slight_smile:
re: installation, it’d be best to do let someone qualified do it, someone who’s comfortable with wiring. if you have ever installed a light switch or a plug, chances are you’ll be comfortable doing this. there’s no need to cut any wires; it’s a clip-on sensor that goes around the lines.
re: ST integration, sorry I missed responding to that earlier. yes what we’re working on is to have Neurio send messages to ST devices based on those events we detect. say, we detect no one is home, we can shut off some plugs or adjust the thermostat based on user preference. We can talk to ST without the need for z-wave or other radios, because we do our integration with them at the REST API level.

@gray, yes absolutely, we’re calling it the Learning Home, as opposed to just the Learning Thermostat.
and you’re also correct about the need of other Wifi or Zwave connected devices in the home. We don’t make control hardware, but ST does that already, hence the perfect team to work with.
regarding installation material, we’re definitely going to be as comprehensive about this as possible. The good news is that we’ll have a few hundred units of these installed already in a different pilot, so by the time we send these out, we’ll have a lot of experience on the types of panels and setups.

Thanks for the questions guys! I’ll keep monitoring this thread.


(Todd Wackford) #15

Welcome Ali,

I’ve read up and thought some more about your product. Sorry I was a bit pessimistic at first. I had a wireless power monitor called “energy aware” from my utility (SMUD) it didn’t much but display current and total consumption. Cool for a month or two, but that was it. The fact that HE and HA don’t talk to each other is a shame.

Anyway, I thought of a couple of use cases I would use your product with.

  1. Shut off power to everything except emergency lights when fire detected. I have a home sprinkler system and we all know what happens when electrons and water mix.

  2. When power Co’s start charging tiered pricing based on current power draw instead of total power draw, you’ll be able to throttle by shutting down circuits.

  3. Tell me that a circuit breaker popped when I’m away. I want the one. Will this be doable?

Thanks in advance,


(Gray) #16

@wackware: Keep in mind that this wouldn’t actually be able to shut off the power to anything. Also it doesn’t access information on individual circuits; it’s only attached at the mains.

But if there were something in particular that you were interested in knowing if it lost power (e.g. a server?), I think the idea is that you could set it up to send you a message when that happens.


(Ali Kashani) #17

@gray, precisely!

@wackware, we’ll be monitoring the entire home to detect patterns, both of the user, and of the individual devices. Other than major appliances, it’s even possible to detect a fridge compressor breaking, for example. That probably won’t be available on day once, hence why it wasn’t mentioned in the video; but it should be a natural addition after we hit the ground. So in that case, if you fridge is about to break, you could get a message before it makes a huge mess!


(Mike T) #18
Plus there are ways with SmartThings to tell if my washing machines is on or not… a vibration sensor would be able to do that
You know, I’ve tried that. It hasn’t worked at all for me. Has that worked for you?

Never worked for me either. This is one thing I really hoped would work since I have an old-fangled washer without even a buzzer. They talk about it on the website as if it really worked but I think it is just wishful thinking.


(Jeremy Whittaker) #19

If anyone is familiar with Bayes’ theorem or any other type of machine learning using statistics then you know this task to identify devices on the “power grid network” in your house is absolutely possible. Belkin even had a competition to do this - http://www.kaggle.com/c/belkin-energy-disaggregation-competition


(Lucas) #20

@gray,

I like the idea but in the KS I notice that they mention a more granular way to monitor per breaker ($300,000 Stretch Goal - Submetering). I am sure unique signatures are present but current is current and a 10 amp widget is going to look like another 10 amp widget. I really don’t see a benefit over this http://www.theenergydetective.com/resproducts unless its cost effective to monitor each breaker. I know for sure what is on each breaker.