It's possible to profile and identify electrical loads by characteristics such as their current over time, their inrush current, the phase angle, noise spikes, duration, frequency of reoccurance, functional earth leakage currents present and power factor to name but a few. Basically by looking at several characteristics of an electrical load you can use that info to develope its signature. Compare that signature to a database of know load signatures and you can identify the load. The more characteristics you monitor, along with knowing if it's a domestic, commercial, industrial type load makes the accuracy surprisingly good. You can also utilise info about the weather, the time of day, number of people living in the house, their employment status etc to increase the accuracy ie if it's 2am and there's one full time employed male in the house it's unlikely to be a vacuum cleaner running at that time.
For example a linear load with no inrush current and steady current value is likely to be a heater element. A load that has a pf in the region of 0.7-0.8, a <50mS inrush current of around 3-4x the load current, a steadily slightly increasing phase angle and marginally decreasing run current over time, a run duration of around 10-25 mins and a reoccurance frequency of 10-40 times a day is likely to be a fridge compressor. Switch mode supplies also have very distinctive signatures with inrush, pf, functional leakage etc being very specific to them.
I have some unofficial experience with this field but only from about 10 years ago, the technology was already pretty sophisticated and accurate back then and I was involved from the opposite side developing an obfuscation system. Things will have progressed way beyond what they had then and with AI thrown into the mix I could imagine the accuracy now being almost 100%.
Yes. Big brother is watching you.
