Now more than ever I see energy monitoring devices like this one that uses "AI" to detect the loads connected to your home electrical network, like the washer or a hairdryer.

These devices use electrical signature measurements and compare them with a large database in the cloud to predict the type of load. Electrical signature measurements is not a new tech, we engineers had been using it to detect anomalies for example in motors and other loads as a nonintrusive means to explore the load.

You create a signature by collecting information about the frequencies with a fast Fourier transform, the phase, and power. And that's not rocket science. What really puzzles me is how can you identify a signature when there are several loads connected? I mean, I can create an individual electrical signature for my dishwasher, my 2 TVs, my Xbox and my fridge, but if several of those are running together the signature properties add so I will need to try every possible combination of signatures to predict what devices are powered on? Is it that brute forced the way these devices work? or there is an elegant solution to the problem?

  • 1
    \$\begingroup\$ There will be times when only one appliance will be active. During this time a signature could be acquired and recognized so, if the device can accumulate a list of appliances then there is only a relatively small number of finite possibilities to examine. I'm not making an answer, just an observation. \$\endgroup\$
    – Andy aka
    Commented Jul 30, 2018 at 12:19
  • \$\begingroup\$ @Andyaka I thought about it, but HOW can you be certain that there just one device connected? I mean maybe the electrical signature generated by my TV and hairdryer powered on match the signature of the GE Fridge model 123 according to the database... \$\endgroup\$
    – DomingoSL
    Commented Jul 30, 2018 at 12:22
  • \$\begingroup\$ @DomingoSL Key word in your comment is "maybe". Chances of it being actually true is low as the power profile of the TV and hairdryer are very different from the fridge. \$\endgroup\$
    – awjlogan
    Commented Jul 30, 2018 at 12:27
  • \$\begingroup\$ Devices get turned on and off. Differences in the total signature could theoretically be characterized on transitions. With enough data, individual devices could be recognized (although the mode a device is in may change its signature, making it indistinguishable from more than one device). \$\endgroup\$ Commented Jul 30, 2018 at 12:27
  • \$\begingroup\$ @CristobolPolychronopolis the transition is something I did not think of, in fact when something gets powered on or off you can measure the components that appear or disappear and consider that as a single device \$\endgroup\$
    – DomingoSL
    Commented Jul 30, 2018 at 12:31

1 Answer 1


It looks like they have 3 ADC's so they are doing more than current monitoring, and a CPLD, which probably means they have real-time constraints and need even sampling.

enter image description here Source: http://whatnicklife.blogspot.com/2017/12/sense-energy-monitor-teardown-sampling.html

Could there be a difference in this circuit? Souldn't be do hard from only looking at the graph of the voltage and the current.


simulate this circuit – Schematic created using CircuitLab

Even refrigerator doors can be detected opening: enter image description here
Source: The Determination of Load Profiles and POwer Consumptions of Home Appliances

On top of that each SMPS has a frequency signature from noise and harmonics.

How they are actually formulating the AI and neural network? Ask them.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.