Deriving status of devices (on/off) from total power consumption of house

The question is a bit off-topic, still hope this is the right place.

From the total power consumption of the house, it should be possible to derive which machine is switched on and consuming power under the condition that I have measurements from the power consumption of single devices (fridge, TV, pumps, washing machine) etc. Note that the consumption of the single devices is not currently measured, its data from the past. But still, the timeline should tell me which devices are used.

Having a look at picture 1 (Total Power Consumption), you can see regular curves from the fridge.
At 5:30, the heating also turns on (see picture 2). Now you have a much higher energy consumption, but you can still see the profiles from heating and fridge from the pictures.

So the aim is that I only measure the total heat consumption and can derive which devices are on from this single data.
Why? Because I do not want to measure every single device and because I have consumer which are mobile (iron) and the wife in the house does not want to take the power meter always with her. But I want to make sure the iron is switched off when we all leave the house.

What is the best way to achieve this?

  • Machine learning?
  • Hard coded algorithm to guess based on duration and amplitude of the curve?

This is the sort of thing that Fourier Transforms are meant to solve. However, your sample rate is likely way too low for them to work. This is one of those engineering problems that seem like it should be easy but in truth ends up being really hard, especially with the unknowns.