Device Detection Home specific?


Many devices can be universally detected by an algorithm based upon the turn on signature. However, it seems that many devices can not.

So there are 2 ways I suspect that other devices can be detected.

  1. After the fact, the turn off signature (possibly in combination of other information) could indicate that a device was on. While not useful to know what’s currently on or off, it would be useful for the log associated with the device.

  2. If home-specific algorithms were applied, then devices could be recognized by the turn on signature, combined with the change in wattage usage etc. E.g. most TV’s today have an inrush resistor that limits the current at turn on time. This is useful from the TV, but may limit the detectibility for Sense. However, if applied to a specific home, may easily be recognizable from other devices.

While home-specific algorithms might be more difficult from a processing at a Sense server, I could see many generic house-specific algorithms matched to each home, which effectively would allow looking for a much smaller set of devices from the complete set of algorithms.

Also, for example, say there are 15 devices that match a turn-on algorithm, the Sense app could present all the devices, and ask which one applies to your home (or none match, then you can tell Sense what it is), and then use it as a house-specific algorithm.

Separate Unknown Detections

I did a Facebook post on some research that talks about two approaches to identifying household electrical loads.

“During the last two decades, several load monitoring techniques have been proposed. These techniques can be viewed into two classes. One class contains techniques which operate on highly accurate data (low sampling rate and high resolution). In such data, signatures (harmonics or turn-on transients) of the appliances are very prominent therefore simple soft computing methods are used to extract these signatures. Such techniques are proposed in [9, 23]. Another class consists of techniques which operate on low accuracy data (high sampling rate and low resolution). In such data, the signatures of appliances are not very prominent therefore efficient soft computing methods are utilized to detect the appliances. Such techniques are proposed in [10, 19, 20, 21, 22, 24].”

IIRC, Sense has intentionally taken the latter approach.

closed #3