Volunteers needed for T&TEC Account Number Collection for AI project in TTLab

[Rajeevr84]

Hi guys,

We at TTLab (lab.tt) are requesting volunteers to provide your T&TEC Account Numbers and household size for a research project.

See submission form here -> https://forms.gle/aggmFPJSf2zYdjTZ6

We are seeking to use T&TEC smart meter data to predict household size using Artificial Intelligence/Machine Learning. TTLab has already liaised with T&TEC who (pending internal approval), will then provide TTLab with the last 3 years of hourly readings for the electrical usage data associated with your account.

Project Goal

To create an alternative method to obtain census data for population estimation. We posit that by creating a model that can predict household size based on electrical usage data, we can estimate population size, density, and observe changes in population movement/growth over time. This can be a great tool for governments across the world in obtaining more real-time population estimates as well as changes in spatiotemporal population shifts.

Please know that your information will NOT be made public.

Please feel free to share with your friends and family as the more T&TEC accounts we collect the better our results!

[Devourment]

submitted, interesting project pal.

But what about all the unmetered connections? Wouldn't that mess up your results in population estimations?

[ProtonPowder]

A very nice idea, a huge limitation will be this simple thing: air conditioning.

Most standard size houses without ac units run around 150-300 per billing cycle with 2-4 persons in the household

While houses with 1-3 units can be around 500-800 per billing cycle. Entirely airconditioned houses can be around 1300+ while those with pools can be around 1800-2000 per billing cycle, all with the same amount of persons in the house.

Problem is in Trinidad all those houses can be neighbours on the same street, regardless of neighbourhood.

Another giant variable is usage of smart phones and computers. Good luck though, if you can find a correlation between just those 2 numbers it would be remarkable.

[Rajeevr84]

Good questions!

@ProtonPowder, by using Deep Learning and some smart feature extraction, such as differences between peaks and troughs, during the day, sustained usage etc, we hope our models can learn features associated with a person count.

An issue we will have to deal with is the disparity in low and high-income areas. I imagine they'd have different patterns of usage, which our model will have to learn.

A final issue (which is why we're going back 3 years with our data), is that covid-19 changed habits a lot! Everyone is home now so patterns will be significantly different in 2018 and 2019.

@Devourment, it will present a problem for sure, but perhaps if we had a rough estimate of those with unmetered connections we could still get close to an approximate census population figure.

I know Prof Hosein has wanted to do this project for some time. Other than an older and not very good, Australian study, there's very little research in this area so I hope our method can quite novel!

[sMASH]

It seems like a proper challenge for the AI.
And a proper AI should be up to the task of working it out.

[Rajeevr84]

It seems like a proper challenge for the AI.
And a proper AI should be up to the task of working it out.

Once we get enough data the Deep Learning models (LSTMs) should be able to model this function. We'll likely aid it by doing some interesting feature engineering as well.

However, a model is as only as good as the data used to train it.

So please share your T&TEC Account numbers and household size! Thanks.

[sMASH]

Mind if I post this on fb? I think this is a very intriguing use case to attempt. U might get more data to work with.

[mazdatt]

How are you getting the data from the smart meters? Is T&TEC sharing this information with a third party without the consent of customer?

[sMASH]

How are you getting the data from the smart meters? Is T&TEC sharing this information with a third party without the consent of customer?

as i said, its INTRIGUING.

[Rajeevr84]

How are you getting the data from the smart meters? Is T&TEC sharing this information with a third party without the consent of customer?

In this study, no, we need YOUR consent for this, hence, why we're asking for volunteers to share their data

In the long run, if this prototype is successful, T&TEC along with the relevant government agencies can use your data for census prediction.

Your electrical usage data SHOULD not be sold or distributed to any third party without the customer's consent. However, census data might be considered public information.

Data privacy is an important and complicated issue, but as of now I can assure you it wasn't easy to get T&TEC to partake in this study, Prof Hosein has been trying for years.

[Rajeevr84]

Mind if I post this on fb? I think this is a very intriguing use case to attempt. U might get more data to work with.

Definitely, feel free to share amongst as many people as you can.