Yes, also the higher OAT end is possibly a smidge high too.

This fits with what I have just said, the WCC, which was the same for both 2025 and 2026, although the lower IATs in 2025 will have triggered more auto-adaption, needs to be lower at the lower OAT end with no setback running. But if the same WCC is used with setback running, then the low OAT 0700 IATs will almost certainly go lower. Another example of how tricky the 'natural' experiments can be!

Also that experiments often need repetition to incorporate discoveries.

I noticed them as well, they are clearly outliers, and I meant to look into them, suspecting they may be from a few days at the very end of April 2026 when the heating was off but I forgot to delete the relevant rows. I have now done that, and the outliers have disappeared, see below.

I also did a double take on the subplot titles, 0700 and 0900 on the same one above the other plot???? Then I realised you had very neatly copied and pasted together the two morning plots...

Thanks, that would explain them!

Indeed, I can well imagine that a heat-pump based system for our 1930ies semi with about 90m^2 should work well with some simple control, including weather compensation; most internal doors stay always open due to the cats wanting to walk around freely in the house without scratching doors. To arrive at this state, it would need some preparation in advance to get most bits right, like radiator sizes and power rating, then monitor and tweak after installation. This is were external software solutions can help, Havenwise or others such as Adia which comes with its own hardware.

Ideally I would like Adia's control software and controlled radiator valves (but not their unit) in tandem with a Viessmann VitoCal 151-A (or equivalent self-contained pre-plumbed 600mm x 600m unit). This would allow controlling temperature profiles for different rooms separately, because "well-balanced" radiators and a single control loop means there is only one common profile moving in unison, even if some rooms might be cooler than others; but their temperatures are correlated. Alternatively, I might get started with Home Assistant and compatible devices for sensors and actuators.

Regards, peter

Yes, as technology geek, I can't help wanting to understand the system under control, first by studying what is going on through instrumentation and measurements, then building models to help in control algorithm development. For instance, I am quite intrigued by Adia Thermal's way to control and dynamically balance radiators.

You are right to be intrigued by it IMHO.  SOFAIK it is the only system currently available that both controls the heat pump flow temperature and the flow through the radiators, in addition to measuring temperatures of individual rooms.  This of course is 'ideal' if you want to have proper control of the system as a whole.  By making various measurements it purports to be able to tell you which radiators would benefit from upsizing and what the effect would be as well as control it day to day.  This is believable, it has all the data it just needs a clever way to process it it.  The fact they are based in Cambridge and called Adia (as in adiabatic) suggests they may well know what they are doing!  I have spoken to them a couple of times and its clear that they are bright! 

They could accumulate a lot of data over time from systems they monitor which may well tell us more generally useful stuff!

That said, operating a heat pump on pure (well adjusted) weather compensation with reasonably well balanced radiators without TRVs works very well for many, including me!  The problem with this approach is that it takes some tweaking, which cant readily be done in the time between when installers finish the plumbing and when many want to go home.

My initial thought is do you actually need this level of detail (room and rad temps for every room etc)? It might be fun to collect it (and doing something just because it is fun (as long as it is also harmless) in the great British eccentric tradition is fine by me), but at the moment you are missing some key variables: short interval energy use, flow and return temps, central living area IAT (I use a modbus temp sensor for that) and possibly OAT (depends how the Viessman measures it). Is there any way you can hack your boiler to get some data from it without blowing it up?  

@cathoderay , indeed, I do not need that level of detail, and I think from my first analysis I have more or less the information I was after: a) required power rating of a heat pump using various approaches; b) some quantification of the consequences when not upgrading my radiators.

Yes, as technology geek, I can't help wanting to understand the system under control, first by studying what is going on through instrumentation and measurements, then building models to help in control algorithm development. For instance, I am quite intrigued by Adia Thermal's way to control and dynamically balance radiators.

Cheers, peter

As for my supplier, we are with British Gas, and I have asked for the data several times; usually I get the response that I can use the web page to "see" the daily pattern, day by day, but they do not have a public API, say combined with an access token, that I could simply write a python script and download the data. Perhaps I will switch to Octopus, and if it is only for that reason.

Kind regards,

peter

I agree, but in assessing the 'saving due to setback' its essential IMHO to take into account the 19:00 IAT deficit in order to compare the setback and no setback cases as close to like for like as is reasonably practical.

I too agree with what you say, and am grateful to you for looking into it.

Sorry that was less for you and more for others that have suggested that your findings provide a general proof of setbacks saving energy/money.

Fair enough! I sometimes do the same thing include something not aimed at the person I am replying to, but other potential readers.

We did only look at short periods but the linear WC "curves" used by many heat pump manufacturers, which provide acceptable levels of control/comfort when heat pumps are defrosting, strongly suggests that no special case is needed to account for defrosts.

I agree, but I would also like to see evidence.

Since the higher mean IAT without setback may well be, indeed almost certainly is, the mechanism by which no setback running uses more energy at moderate to lower OATs (it does, it is in the charts), this begs the question: have I once again not compared apples with apples?

 

Unfortunately I think so, as can be seen in the 07:00 chart below. The setback IAT is clearly decreasing with OAT, while the no setback IAT is rising with falling OAT.

Might this be explained by the lower OAT end of the WCC being a bit too high? With no setback, the heat pump more than keeps up, but with the setback, it has to do the recovery, and the lower the OAT. the more recovery it has to do, and so the lower the IAT at 0700.

The simplest change would be to lower the no setback WC curve to produce IATs that correspond to the recovered setback IATs, and not the elevated IATs in the current data.

This fits with what I have just said, the WCC, which was the same for both 2025 and 2026, although the lower IATs in 2025 will have triggered more auto-adaption, needs to be lower at the lower OAT end with no setback running. But if the same WCC is used with setback running, then the low OAT 0700 IATs will almost certainly go lower. Another example of how tricky the 'natural' experiments can be!

The hours when it matters will depend on the household, many (the majority?) will be up and awake between 6-8am five days a week having breakfast and getting ready for work/education. Such households don't have the "luxury" of wating until after 9am for the heating to be restored to a comfortable level after a setback. 

My normal getting up time is 0700, which is not that unusual, and it means I am in bed during the hour when the 0700 value applies. But I fully take the point that households vary greatly in their habits, but all want the same thing, comfortable warmth when they get up in the morning.

I do have a question though, what is up with the two no setback data points circled in red below?

I noticed them as well, they are clearly outliers, and I meant to look into them, suspecting they may be from a few days at the very end of April 2026 when the heating was off but I forgot to delete the relevant rows. I have now done that, and the outliers have disappeared, see below.

I also did a double take on the subplot titles, 0700 and 0900 on the same one above the other plot???? Then I realised you had very neatly copied and pasted together the two morning plots...

Sorry that was less for you and more for others that have suggested that your findings provide a general proof of setbacks saving energy/money.

I would go further and say essential. The whole point about collecting long time series data is to achieve a sort of regression to the mean, I have the noise cancel itself out. In the previous discussions, we only really looked at short periods. How do we know they were typical? We don't. A long data series will give us the answer.  

We did only look at short periods but the linear WC "curves" used by many heat pump manufacturers, which provide acceptable levels of control/comfort when heat pumps are defrosting, strongly suggests that no special case is needed to account for defrosts.

Part laziness (WCC adjustment fatigue), part whatever the actual IAT is, it is good enough for me (not too hot, not too cold), and more recently, a deliberate decision not to alter anything so that Spring 2026 could be more reasonably compared with Spring 2025, ie both had the same baseline WCC plus any auto-adapt adjustments.

Since the higher mean IAT without setback may well be, indeed almost certainly is, the mechanism by which no setback running uses more energy at moderate to lower OATs (it does, it is in the charts), this begs the question: have I once again not compared apples with apples?

Unfortunately I think so, as can be seen in the 07:00 chart below. The setback IAT is clearly decreasing with OAT, while the no setback IAT is rising with falling OAT.

In other words, should I not lower the WCC during no setback running to achieve the same mean IAT as happens with setbacks? Or alternatively raise the setback WCC to achieve parity of mean IATs? The problem of course is in the doing of that. I will have to wait until next spring, and then tweak the WCC up or down depending on whether I am using setbacks or not to match the mean IATs from a period when I was using the opposite state. 

The simplest change would be to lower the no setback WC curve to produce IATs that correspond to the recovered setback IATs, and not the elevated IATs in the current data.

More generally, I think within all that there may be a general conclusion. It is that all other things being equal (in particular, no change to the WCC), then setbacks do save energy, by reducing the mean IAT. I am mindful of the fact that at moderate to low OATs, the savings were around 10%, with around a one degree difference in mean IAT, which is broadly consistent with the lower your heating by one degree and you will save 10% off your bill rule of thumb. The key question is whether that lowering of mean daily IAT can be achieved while maintaining adequate mean hourly IATs during the hours when it matters.

Note that the values are the means for the preceding hour eg the 0700 plot values are the means for 0600 to 0700. I am not often up at that hour. 

The hours when it matters will depend on the household, many (the majority?) will be up and awake between 6-8am five days a week having breakfast and getting ready for work/education. Such households don't have the "luxury" of wating until after 9am for the heating to be restored to a comfortable level after a setback. 

If there is a general conclusion I think it is along the lines of: the saving from a setback reflects the reduction in mean IAT but if the recovery from a setback can be achieved by the time required by the occupants will depend on the household's daily routine and IAT requirements when the occupants are awake.

Unfortunately, this suggests something of an apples vs oranges comparison, the no setback data has an average IAT ~1.5C above your target IAT, while the setback IAT is still below your target IAT even 6 hours after the setback (and after most household's breakfast time) at lower OATs.

I agree, and have partly addressed this above. Because these are natural experiments / observational studies, we have what we have, and have to wait until the next natural iteration if we want to change what we can eg the WCC to achieve a different mean IAT. But I still think the general conclusion suggested in the paragraph above that starts 'More generally, I think...' remains valid, that all other things being equal (in particular, no change to the WCC), then setbacks, at least in my type of property, do save energy, by reducing the mean IAT. But at the same time I also suspect that if either the setback WCC was tweaked up to get the mean IAT to no setback levels, or vice versa, then the energy saving may well disappear. But here's the rub: in the former case, raising the setback WCC to get a mean IAT that matches the no setback mean IAT, may cause daytime hourly mean IATs to go above what is necessary, so as to get the mean daily IAT higher.

The no setback WC curve could simply be lowered, especially at lower OATs, to produce IATs similar to the comfortable setback IATs. That would result in a more apples vs apples comparison, and yes, would likely see the energy consumption difference disappear.

Edit 1145 16 May 2026: plot correction: Although the upper X axis in the plots above go below zero, such values got missed out. Here is the correct version, which does show more cold mornings when the IAT was at the lower limit of my comfort zone:

Thank you for doing the 07:00 and 19:00 charts! I do have a question though, what is up with the two no setback data points circled in red below? Was the heating switched off at some point during the no setback trial period?

The point I am making is that there may have been a measurable setback IAT penalty, but does that necessarily translate in to a human comfort penalty?

I agree, but in assessing the 'saving due to setback' its essential IMHO to take into account the 19:00 IAT deficit in order to compare the setback and no setback cases as close to like for like as is reasonably practical.

This may also be a rare occasion when it is worth not anchoring the Y axis to zero. Of course, everything gets greatly exaggerated visually, which is why marketing types like non-Y-zeroed plots, and is equally why generally I don't like them, but it does mean you can see individual data points better, which may have some virtue on this occasion. Here are the plots expanded to fill the available space. Note that both the x and the Y axes are not the same in range or scale, and so the two plots are not directly comparable. Instead, each should be considered on its own.

I can see eight data points in the upper plot where the mean IAT is between 17.5 and 18.0°C between 0600 and 0700. At the time, I would have been in bed. The setback comfort penalty almost certainly went unnoticed. By 0900, there was only one point below 18°C, and it was only just below. Again the axes differ, plots should be considered on their own:

The point I am making is that there may have been a measurable setback IAT penalty, but does that necessarily translate in to a human comfort penalty?

I will try to zoom in to work out the deltas between the two lines unless you have an easy way to do that (which would be helpful). 

I can get both the fitted values, which are row specific ie depend on the OAT (ambient) value for that row, so in effect they are random, or the regression line equations, which may be more useful to you. Here are the equations in all their multiple decimal places glory with some clutter removed for these from the 0700 / 1900 data set. Naming conventions: df = data frame, sb = setback, nosb = no setback, 07 = 0700, 19 = 1900, the regression is as you can see a simple (x ~ y) one, the first (sb_07) equation is IAT =  18.85257 x (0.09351 x ambient):

model_sb_07 <- lm(IAT ~ ambient, data=df_sb_07)

summary(model_sb_07)

Coefficients:
            Estimate
(Intercept) 18.85257
ambient      0.09351


model_sb_19 <- lm(IAT ~ ambient, data=df_sb_19)

summary(model_sb_19)

Coefficients:
            Estimate 
(Intercept) 18.81057
ambient      0.06697


model_nosb_07 <- lm(IAT ~ ambient, data=df_nosb_07)

summary(model_nosb_07)

Coefficients:
            Estimate 
(Intercept) 21.36956
ambient     -0.10908


model_nosb_19 <- lm(IAT ~ ambient, data=df_nosb_19)

summary(model_nosb_19)

Coefficients:
            Estimate 
(Intercept) 19.52138
ambient      0.06626

That's great thanks. 

I dont have the time today to 'process' this but will in the next couple of days (which also gives me time to think more clearly about how to use the data!). 

My gut feel is to correct for the 19:00 deficit (on the grounds that the same comfort level could also be achieved at 19:00 with the non setback case by reducing the temperature, which will reduce the energy required - and by making a correction the comparison is closer to apples for apples.  My gut feel is also toregard the 07:00 deficit as a comfort factor to consider, on the grounds that I think a deficit at 0700 is inevitable (and thus a 'cost' of setback) if you reduce the temperature meaningfully overnight.  The exception might be if you have a fan radiator in the 'breakfast room' (which, as it happens, I do), however thatys a whole load more complexity!