Comment by Charlotte Avery, Data Scientist at Energy Systems Catapult.

A quick and easy way to detect heat pumps and EVs in Smart Meter data.

Home heating and car transportation account for around 14% and 12% of total UK carbon emissions respectively. Low Carbon Technologies (LCTs), such as heat pumps and Electric Vehicles (EVs) will play a significant role in accelerating the UK towards a Net Zero future as these technologies rely on energy from the electricity grid, which is being supplied by an increasing proportion of renewable power.

In a bid to reach Net Zero emissions, researchers and industry are looking to understand what the barriers for the uptake of EVs and heat pumps are, and how consumers change the way they interact with energy in response to having LCTs in their homes. This is particularly important for energy suppliers as they design and offer consumer-centric, flexible tariffs and create demand-side flexibility services for homes with EV chargers and heat pumps, helping to ensure the grid can keep the lights on.

To aid these efforts, we need a way of linking the presence of LCTs with electricity consumption data. Of course, one way is to rely on consumer surveys; or alternatively, you can model the electricity consumption profiles from smart meter data.

Accurately identifying signatures of heat pumps and EVs in homes within smart meter data is non-trivial given the unpredictable way in which different consumers use energy throughout the day. However, using the half-hourly smart meter readings from a sample of 298 homes in our Living Lab, over half of which we know have a heat pump and/or an EV (based on consumer questionnaires), and the other half have neither a heat pump or EV, we were able to identify a straightforward way of quickly determining the presence of either an EV or heat pump in homes.

Simply put, we found that an effective way of identifying if a home has either an EV or a heat pump (or both) is to sum up the electricity usage overnight (between 11pm and 6am).

If the total night time usage is greater than 3.6 kWh, on average over the course of a week, this was a good identifier for our Living Lab homes having an EV and/or heat pump.

Using this rule-of-thumb, we were able to correctly identify LCTs with an accuracy of about 85%. Details of the methodology are provided at the end of this blog.

The caveat to this simplistic method is that it will only be applicable if the consumers were charging their EVs or using their heat pumps during the period over which the data was collected. For this reason, our training and testing was done using smart meter data during the months of January and February 2023, when consumers were more likely to have their heat pumps turned on.

Furthermore, letting us know whether you have a heat pump or EV is not a necessary requirement to joining our Living Lab, therefore we expect there to be some participants who are mislabelled as not having an LCT which are present in our sample. Or, equivalently, there may be some participants with EVs who do not own an EV charger at home, therefore they do not show the signatures of EV charging in their smart meter data. For these reasons, we expect the 85% accuracy could be improved upon knowing this extra information.

From the average consumption profiles over 24 hours, we can clearly see that Living Lab participants tend to charge their EVs overnight. Furthermore, heat pumps tend to use more electricity in the early hours of the morning. These are good indicators of these low carbon technologies in homes.

The methodology details …

The dataset: the data consisted of half-hourly electricity smart meter readings from 298 homes in the Living Lab. For each home, we took the daily consumption profiles averaged over each of the 8 weeks during the months of January and February 2023. This gave us, for each home, a daily electricity profile for each week making up the dataset.

Breakdown of the LCT in Living Lab homes chosen for our sample:

Up to 70% of households in the dataset set were randomly selected for training a decision tree model with maximum depth 1, leaving 30% of the households for testing the model. Random selection was done such that the fraction of homes with EVs only, heat pumps only, and both EVs and heat pumps were consistent in the training and test sets.

A variety of features were fed into the model, including: the time of day when consumption was maximum, the maximum electricity usage, the inter-quartile range of electricity readings, the difference between the usage at night and day, the total usage during the day and during the night, the variation of the usage (standard deviation squared) and the ratio of the variation during the night versus the day, the number of hours the consumption was above 1kWh.

What about distinguishing homes with EVs from homes with heat pumps?

Developing a model which classifies homes into ‘only has an EV charger’ versus ‘only has a heap pump’ versus ‘has both a heat pump and an EV’ is more challenging although it is possible with more sophisticated modelling techniques.

However, because the charging of EVs causes a large increase in usage overnight compared to the daytime usage (see figure above), a simple rule of thumb might be to assume homes with a lower night:day ratio of energy consumption do not have an EV charger, but instead have a heat pump.

What next?

Calling industry:

• Use this simple method to help consumers find cheaper and greener tariffs, or if developing your own method of identifying LCT in datasets, use our simple rule-of-thumb as a benchmark to improve upon.

Calling homeowners:

• Learn more about heat pumps for domestic use.
• If you have a smart meter but haven’t linked it up to our Living Lab, and you want to be a part of this sort of research, you can get involved here.