22nd March 2020

Domestic energy usage patterns during social distancing

David Sykes, Head of Data Science

corona_banner-02.png

Analysis and findings by Jaye Cribb, Igor Gotlibovych and David Sykes.

Key findings

  • As of Wednesday 18th March, UK domestic energy consumption changed markedly from previous patterns as people stayed home throughout the day.
  • Up to 30% of households have changed consumption pattern markedly (note non-changers will include retirees, homes where one member is usually home all day, key workers who are continuing to work out of home)
  • Among households with changed patterns, average electricity bills are increasing between £1.34 and £2.85 per week and gas bills between £1.65 and £1.93 (assuming they pay the same level as the government price cap). Note that gas consumption will fall rapidly as the weather improves.

Executive Summary

The week commencing 16th March saw a rapid escalation in measures designed to curb the spread of Covid 19. On Monday 16th, the government issued advice for everyone to avoid social contact as far as possible, to work from home and to not go to pubs, clubs, restaurants and theatres. By comparing anonymised smart meter usage data from 115,000 customers from this week and last week, we’ve identified an increase in daytime energy usage which we attribute to more customers being home during the day. This effect appears to start on Tuesday after the announcement on Monday 16th and becomes more pronounced through to Thursday.

Covid 19 – electricity consumption by day of week

Energy usage on weekdays increased by between 2-4% across the whole group of customers as a result of the measures. We’ve used two approaches to try and isolate the impact on those customers who have stayed at home with the below results.

Electricity

Method Percentage of customers Average incremental usage (kWh per day) Average incremental usage (%) Cost of energy (p/kWh) Average incremental cost per working week
All customers average 100% 0.5 4% 17.80 £0.45
Mixture method 30% 1.5 13% 17.80 £1.34
Consistent increasers method 17% 3.2 32% 17.80 £2.85

Gas

Method Percentage of customers Average incremental usage (kWh per day) Average incremental usage (%) Cost of energy (p/kWh) Average incremental cost per working week
All customers average 100% 2.4 4% 3.50 £0.42
Mixture method 25% 11 20% 3.50 £1.93
Consistent increasers method 8% 9.4 20% 3.50 £1.65

Data

For this report we’ve used anonymised half hourly smart meter readings gathered from 115,000 customers from the week commencing Monday 16th March. We compare this data to readings from the same cohort of customers from the week commencing Monday 9th March.

To ensure the 115,000 customers were representative of UK domestic users, we compared them to Elexon’s standard load profiles which are used to estimate usage for customers without smart meters. The Octopus Smart meter customers are noted to have slightly higher than average consumption but follow a very similar shape throughout the day to Elexon’s deemed profiles. The higher nighttime usage is accounted for by the presence of economy 7 customers in the Octopus Smart Meter group.

Average half-hourly consumption comparison

Average effect across all customers

To understand the impact on UK domestic energy consumption due to more of the population working from home we compared the consumption of our smart meter customers during the week of 16th March with the week prior. The below plots show the average energy usage profiles of our smart meter customers along with the total usage per customer each day. In both electricity and gas we see similar patterns:

  • Monday showed no sign of increased daytime consumption
  • Tuesday and Wednesday and Thursday show increasing daytime consumption presumably as more customers work from home
  • Friday shows slightly reduced daytime consumption increases perhaps as Friday is a more common day to work from home in a typical week
  • On the most pronounced day, Wednesday, the customers used an average of 4.3% more electricity and 4.6% more gas.

Average consumption be half-hour (electricity)

Average consumption be half-hour (gas)

Gas consumption was corrected for weather and seasonality by multiplying the consumption on week 16th by the ratio of Xoserve’s LPA coefficients for each weekday.

Identifying the effect of those who did work from home

The ~4% increases reported above are for the whole group of smart meter customers, many of whom may not yet have changed their consumption behaviour. The signal we see above is therefore a diluted effect of those customers who are spending more time in their homes and those who haven’t changed behaviour. To be able to effectively plan for any increased usage and spending on domestic energy, we need to disaggregate the two groups so we can estimate the increased consumption of those customers who are now working from home.

In the absence of any labelled data, this poses a challenge. The issue being that a single customer’s consumption profile is stochastic. There is a natural variation in a customer’s usage profile week on week regardless of any underlying behavioural change. This is illustrated by the plot of 10 randomly selected customers below. Without labelled data, selecting who of these we think has changed behaviour and started working from home is highly subjective. Below we take two approaches to try and isolate the working from home cohort and estimate their increased energy consumption.

Consumption profiles

Mixture model

For each customer we define two variables:

  • X = their energy consumption between 9am and 5pm on a given weekday in week commencing 9th March
  • Y = their energy consumption between 9am and 5pm on the same weekday in week commencing 16th March

The plot below shows the distributions of X and Y for all customers:

Binned daily consumption

To estimate the percentage of customers staying at home and how much extra they used, we simulated a proportion of customers p staying at home with an increased consumption ΔQ. We chose the combination of p and ΔQ that minimised the Kullback-Leibler divergence between our actual distribution and our simulated distribution. The below plot shows the KL divergence plotted against p and ΔQ for electricity. From this we infer a best estimate of 30% of customers staying home using an average of 1.5 kWh extra during 9am - 5pm.

Simulated KL divergence

The best fit simulation is plotted below:

Best fit simulation

The same approach was taken for seasonally corrected gas consumption. A summary of results is shown below:

Mixture method results Electricity Gas
Estimated proportion of customers staying at home 30% 25%
Estimated increase in consumption 9am-5pm 1.5 kWh 11.0 kWh

Consistent increasers model

To attempt to isolate those customers who stayed at home we used a crude rule of selecting customers who showed increased 9am-5pm consumption for all four days Tuesday to Friday.

For electricity:

  • 15% of customers (17,000) fulfilled these criteria. Below shows their comparison of the week of 16th March vs the previous week
  • These customers show a ~30% increase in daily consumption

Customers staying at home

For gas:

  • 8% of customers (7,800) fulfilled these criteria. Below shows their comparison of the week of 16th March vs the previous week
  • These customers show a 20% increase in daily consumption
Customers staying at home (gas)

Summary of results

From the whole group average, we see a noticeable trend in higher domestic energy usage in the middle of the day. There is no perfect method to identify which customers are staying at home and how much extra energy they are using. We’ve applied two methods to try and isolate the signal from those customers staying home and get the below findings.

Electricity

Method Percentage of customers Average incremental usage (kWh per day) Average incremental usage (%) Cost of energy (p/kWh) Average incremental cost per working week
All customers average 100% 0.5 4% 17.80 £0.45
Mixture method 30% 1.5 13% 17.80 £1.34
Consistent increasers method 17% 3.2 32% 17.80 £2.85

Gas

Method Percentage of customers Average incremental usage (kWh per day) Average incremental usage (%) Cost of energy (p/kWh) Average incremental cost per working week
All customers average 100% 2.4 4% 3.50 £0.42
Mixture method 25% 11 20% 3.50 £1.93
Consistent increasers method 8% 9.4 20% 3.50 £1.65

The mixture method is likely to be an underestimation of how much extra energy some households are using since it models all stay at home households with the same incremental consumption. The consistent increasers method is likely to overestimate how much extra energy some households are using since it analyses the cohort with the strongest signal.

If you are applying these results to forward projections of energy consumption or to customer segmentation for bill relief it is important to be cognisant that:

  • The results are derived from one week of observations and whilst we don’t expect the pattern to change, the magnitudes of change will depend heavily on the temperature
  • The signal appears to be developing with time as society rapidly changes its behavioural patterns in response to the crisis
  • Although we have reduced the incremental usage to a single number, in reality it is a distribution with some customers using far more than the average stated here and others using less
  • The nature and magnitude of change in energy usage is unlikely to be easily attributed to clearly defined demographic segments. It is pervasive across society and the magnitude of change will be a function of the household behaviour and heating and devices in the the home
image of David Sykes

David Sykes

Head of Data Science

Hey I'm Constantine, welcome to Octopus Energy!

×