0
$\begingroup$

I'm working on a machine learning project aimed at classifying electrical loads detected in a domestic electrical installation by a current transformer (CT) during daily activities. The challenge lies in the fact that various electrical devices contribute to the overall demand on the electrical supply grid, creating a complex and overlapping observation stream.

For instance, there's a nearly constant base load from the ISP installation or a TV on standby, lights and fans are intermittently turned on and off, and larger consumption spikes occur when devices like ovens and kettles are in use.

I'm seeking guidance on how to effectively model the demand of individual devices as they come online and turn off within this noisy data stream. Specifically, I'd like to know:

What machine learning techniques or algorithms are well-suited for this task of disaggregating the total demand into contributions from different devices?

How might I preprocess the data effectively to handle the overlapping sources of demand and ensure accurate classification? For the present I would like to note that 'this many devices of such and such a consumption pattern are present'. Identifying such as 'the refrigerator' is a longer term goal.

Are there any open-source libraries or tools that could streamline this process, or should I consider building a custom solution?

What are some potential challenges and best practices when it comes to labeling the data for supervised learning in this context?

Any insights, advice, or references to relevant research papers or projects would be greatly appreciated. Thank you!

So as far as I have got, the first differences on the data stream indicate the turning on/off of consumers. So the arrival of a e.g. 75Wh increment in the data stream may hypothesised to be tied to a later 75Wh decrement even if other loads have joined or left the mix. Massive caveat about the noise background...

Next up, is the sampling frequency. Where 2 or more consumers join between sample intervals their individual 'footprints' will be confused -- this is where the learning bit comes important as the system becomes aware of the patterns of usage

$\endgroup$

0

Your Answer

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