Watch one of our data scientists, Etienne, trace the electrical signals from an example house and show you how he identifies them as devices, even when signals from multiple devices overlap.
Etienne on why identifying devices is hard:
“Well I guess there are three things that make it challenging. Probably the thing that was most obvious when I show those electrical traces, and that’s that these devices overlap with one another all the time. You know I used to work with speech recognition previously, and if you’re doing speech recognition and two people start talking simultaneously, we just say ‘well, this is too hard, we’re not even gonna try.’ But with electrical devices, you know your refrigerator is going to turn on, while your microwave oven is on, while the dishwasher is going, and so you have to deal with this overlay of appliances all the time. So that’s one thing that makes it hard.
“The second thing is that your time scales can be so different. So I was showing you like, the way the dryer’s motor turns on, which is a second or less, and then for an hour the oven will run, so you have to get these things right on these very different time scales, which is quite a challenge.
“And then the third one is kind of related, that also your wattage scales can be very different. Something can be ten watts in size, and be crucial to understanding what this eight thousand watt device is. So all of that makes quite a challenging problem.”
Etienne on why identifying devices is fun:
“So, two things make it real fun. One thing is that, you deal with this device, it looks really complicated, and you start writing algorithms and you start extracting parameters, and then it gets better and better and you finally get to the point where you release it on a brand new house, and it does a good job of recognizing the microwave oven or the electric dryer. And that’s really quite a satisfying thing.
“But the other thing is that you also- sort of- with my colleagues I sit and look at these traces, and then you see this really puzzling device that makes no sense, and then little by little, people start figuring it out. And so, my colleagues Shakwan and Caleb are really good at that, they’ll say, ‘well you know here’s a motor and it’s running at more or less this frequency and it’s this size, well, wouldn’t a bread maker have something like that?’ Then you start putting two and two together: yes, there’s a heating element, and it was doing a little bit of this and a little bit of that, and then suddenly it all falls into place, and then we as humans at least know, ‘yeah, this is certainly a bread maker.’ Then, developing the algorithms and finding ways to identify that automatically is of course a further challenge, but there is some satisfaction in at least recognizing what these traces are.”