Measuring electricity demand flexing

For some time I’ve wanted to create a dynamic report that would allow me to see how well I am doing in flexing my electricity demand. This post documents my first attempt, and describes some of the conceptual and practical challenges I faced.

Background

I’m an Octopus Energy customer, with a smart meter. This means I can easily access my historical half-hourly consumption, via their API. I am on their Agile tariff, which means the price is different every half hour, and I make an effort to use electricity at the cheapest times of the day. On some occasions the price even goes negative! My biggest source of electricity consumption is a 150L Mixergy hot water tank, which gives me a lot of potential for flexing my demand to save money. (By flexing, I mean either shifting the time of use, or increasing/decreasing total demand. It is related to demand flexibility: flexibility measures how much you could flex.)

Remarkably, flexing my consumption has proven easier than measuring how much I’m flexing. However, I finally decided to do my best to come up with a reasonable measure of my demand flexing. This is very much a first attempt, so I’m keen to get feedback on better ways to measure flexing.

Some big questions

Should we encourage demand flexing?

The first big question I have to address is whether we should be encouraging people to flex their demand consumption. It probably won’t surprise people to learn I concluded ‘yes’. As we increase the proportion of electricity generated from intermittent sources like wind and solar, and increase use of electric vehicles and heating, we are going to need to harness as much demand flexibility as we can, in order to minimise the use of fossil fuels, the cost of additional generation and storage, and network infrastructure. So helping people flex their demand effectively makes sense.

However, when it comes to how we reward people for flexing, the question is more challenging. Ability to flex depends a lot on your home situation (do you own or rent, a house or a flat? are you on your own or sharing? do you have children? do you work from home? do you have an electric vehicle? what other technology do you have? can you code?). These are important questions to consider (see for example Powell and Fell, 2019; and Johnson, 2020). I take a Rawlsian view, in which it is important to ensure that the worst-off are better off as a result of any incentives to flex. I believe that society is better off for me flexing, even after I am rewarded, but I’m open to evidence to the contrary.

Given this difference in people’s ability to flex demand, I’d discourage people to use measurements of flexing to form moral comparisons of people (eg person A is doing better than person B because they are flexing more). This is much more about considering your own flexing. If seeing how other people flex encourages you, please do, but don’t feel bad if you’re not able to flex as much as someone with a Tesla Model X.

What should be the objective in flexing?

There are a number of different objectives customers might have in mind when flexing their electricity consumption.

Firstly, we can optimise in order to reduce emissions. National Grid forecast the emissions for each half hour, based on the generation technologies expected to be used. They report this on a regional level, and also a national average. My view is that deciding when to use electricity according to this makes rough sense, especially for any customers who aren’t on a time-of-use tariff. (It certainly makes a lot more sense than thinking because you’re on a green tariff, your consumption doesn’t contribute to any emissions!) However, it is important to remember that these numbers are average emissions for a given half hour — it doesn’t tell you how total emissions might change if you increased or decreased demand. For example, if you shift your washing machine to run when the grid is supplied 20% by gas rather than 50%, I’m not convinced it reduces overall emissions. There may also be times when there are grid constraints despite lots of wind and solar, and so you increasing demand may not help. But you’re certainly unlikely to making things worse by trying to reducing your overall emissions.

However, if you are on a time of use tariff like Octopus Agile, I’m recommend looking at cost rather than emissions. The price of electricity does tend to be lower when emissions intensity is lower. And on the occasions where they differ, I’m generally not convinced that you’d reduce overall emissions by paying more to consume at the time with lower emissions intensity (I wrote a blog post considering this question in more detail). You could always put some of the money you save towards improving your energy efficiency or other environmental investments.

It might therefore seem to make sense to minimise total cost, but this ignores the potential benefits of electricity. A more rational approach would be to maximise the value of the electricity less the cost of it. However, interestingly, my behaviour doesn’t always match this. Firstly, I sometimes don’t use electricity even where the benefit exceeds the cost, for example not making a cup of tea when the price of electricity is high. Secondly, when the price is negative, I tend not to maximise consumption if it means wasting electricity, even though that would lower my overall cost. My guess is that I probably should be slightly more willing to use electricity, but I am probably right not to waste electricity as it would incur actual costs.

Interestingly, a lot of the people I talk to about Agile tariffs don’t boast about how much they’ve reduced their cost as much as about how much they’ve reduced their average price (in p/kwh). I wondered why price would be a sensible thing to optimise. After giving it some thought, I think I’m able to make sense of it. If we assume that we will buy electricity when and only when the value outweighs the cost, then the lower the average price, the greater the net value. In other words, anything I can do to lower my average price, as long as it doesn’t involve me buying where with a net cost, or stopping buying when there would be net benefit, will increase my overall net benefit.

What should we compare consumption to?

The final big question is what we should compare our consumption to, that is, the appropriate benchmark or counterfactual. I considered a number of approaches:

1. Looking at a person’s living situation, and assess what demand profile we might reasonably expect them to have. This would inevitably be subjective, and I didn’t trust my ability to do this fairly.
2. Comparing with a customer’s historical demand profile. One downside of this is that it can be distorted by household changes. Another is that historical demand will depend on previous attempts to flex, perhaps leading to an incentive to reduce flexing in order to show improvement.
3. Letting the benchmark be zero. This would be objective and easiest to implement. It also combines changes in overall consumption with changes in consumption timing, which you sometimes want, but sometimes makes it harder to see the impact of flexing. I also worried that consumers would struggle to meaningfully interpret their total consumption/shape without a meaningful benchmark.
4. Using a benchmark of actual consumption, and applying it to a ‘default’ profile. In the UK, Elexon publish such a default profile, which is used for all domestic customers that don’t have smart meters. One weakness of this is that it doesn’t capture any flexing that increases or decreases overall demand. Another weakness is that it treats natural differences between a customer’s demand shape and the default profile as “flexing”, for example if you tend to use more electricity overnight because you have hot water. One could use a modified economy7 profile shape, but this starts getting towards picking needing to choose the appropriate benchmark for each customer.
5. A simplification of the previous approach would be to apply the overall consumption to a flat profile. However, I feel this would make the benchmark less meaningful to users.

In the end, I’ve taken an approach which combines 3 and 4. I will report the impact of the customer’s overall energy use, and the difference between that use and how it would have been had their demand been shaped according to the default profile. Neither of these approaches really capture how much an individual customer ‘should’ be able to flex, so I’d repeat my earlier warning to not use these metrics to ‘judge’ a customer’s flexing.

Measuring my flexing

Having considered those big questions, my next step was to perform some calculations. I built a webpage to do the calculations on a single month, so I could repeat them on multiple months, and so others to do the calculations on their consumption. Any Octopus smart meter customers can run it, my going to https://www.guylipman.com/octopus/consumption_analysis.html?month=2020-04&key=XXX&mpan=XXX&meter=XXX, replacing XXX with their security key, mpan and meter numbers. All the html/javascript code is in the html file, and if there’s interest I might write a follow-up blog post talking about how I built this.

The image below shows the results for each day in April 2020. The left section summarises my consumption, emissions and cost for each day. The next two sections show my consumption and the Agile prices for each 3 hour bucket. The buckets where my consumption was above average are highlighted yellow.

From this, I can see that I bought 113kwh of electricity for a cost of £3.76, or 3.3p/kwh. If this volume had been consumed according to the Elexon default demand profile, the electricity would have cost me an additional £4.76 or 4.2p/kwh. I am going to attribute this savings to flexing, however I do need to offer a few caveats:

• Even if I had made no attempt to flex, my consumption would likely been at cheaper times of the day than the default profile, due to my electric hot water tank. I don’t think it is unreasonable to include this, as my reason to have a hot water tank is to allow flexing.
• If I hadn’t been taking advantage of the periods with negative prices, my consumption would likely have less than 113kwh. I can see that of the three days with over 10kwh consumed, all had negative prices. On each of these days I heated my hot water tank to 100%. However, it is necessary to fully heat a hot water tank every 1–2 weeks to keep it clean, so likely without flexing I might have needed to still use this, just at a higher price. Also, I obviously do get some benefit from extra hot water. It would be good to try and adjust for these challenges, but I don’t think it is possible.

On a similar basis, I am attributing to flexing the fact that my emissions were 143g/kwh, 26g/kwh less than had I consumed according to the default profile.

Had I been consumed electricity according to the Elexon default profile on Agile, my consumption would have cost 7.5p/kwh, a further saving of 7.4p/kwh on Octopus’s standard variable tariff of 14.9p/kwh. I don’t consider this a saving from flexing. At a first approximation, 14.9p/kwh is what Octopus would have expected the prices to be over a whole year, and they would have likely hedged on that basis. They would have expected that customers would have been overpaying in April (which tends to have low prices) and underpaying in January (with higher prices). Secondly, they obviously didn’t expect a Covid lockdown when they set the price, which (along with lots of wind) has significantly reduced April 2020 prices even below what they expected. However, it could have just as easily gone the other way, which would make some of this a ‘lucky’ saving rather than due to my flexing or great wisdom.

There is some reason to think that perhaps Agile prices might be biased lower than standard variable tariffs. This is possible, perhaps reflecting lower cost of hedging, or the fact that Agile customers are quicker to switch to other providers. However, from the data available, any Agile ‘discount’ is likely to be small compared with the impact of April prices being so much lower than what was expected for the year when the fixed tariff was set, and it is impossible to separate these components.

The final piece of analysis I would like to consider is whether I might have flexed further, in order to save more money or further reduced emissions. The table below again shows the consumption and price profile for each bucket, and adds the emissions profile:

Looking at flexing to reduce cost, it is hard to tell from a table like this where I should have consumed differently, as it depends on my actual flexibility. While I set my hot water tank to heat at the cheapest time of each day, my washing machine can’t run overnight or it would disturb the neighbours. Similarly, while I’m generally prepared to wait until 7pm to cook my dinner, I’m not prepared to wait until 11pm just to save a few pence. I’d love to systematise this kind of logic, to allow me to identify what I might have done differently, but I’m not sure how possible it is.

Looking at emissions, while the highlighted buckets in which most electricity consumed tend to have lower emissions intensity than surrounding periods, this isn’t always the case, for example in the first bucket on the 13th or the fifth bucket on the 5th of April. However, as discussed earlier, I am not convinced that shifting consumption in these occasions would have lowered overall emissions.

Finally, I produced a table showing how my savings have varied over the full 16 months. My Agile cost has fallen much faster rapidly over the period, mostly due to increasing wholesale savings (the difference between the fixed price and the Agile price if your demand profile matched the default), but also due to a consistent flex saving, and a slight reduction in the fixed cost in May 2019.

Conclusion

This is very much a first attempt, a way to think about different approaches to measuring the impact of flexing my electricity consumption. It raised a lot of questions and challenges, several of which deserve further thought, but I’d be happy to hear anyone else’s thoughts on this.