Posted by: @jamespa

↑

Posted by: @jamespa

↑

Personally I don't think any interpretation of experimental results can be trusted unless it can be rationalised theoretically, and equally no theory can be trusted unless tested by experiment! 

 

 

By and large I agree, but I don't give the two (experiment and theory) equal weight. Given accurate enough experimental data, it always trumps theory, when experiment and theory disagree. In practice you must bend the theory to fit the data, not the data to fit the theory. The same principle is at work in Sherlock Holmes's "when you have eliminated the impossible, whatever remains, however improbable, must be the truth".

Please note I said any interpretation of experimental results.  That word is crucial.  The observations are the observations, without a doubt.  However it's the inferences one draws that need careful testing.  Robust experimental design is required if inferences are to be reliable, particularly when dealing with complex multi-variate and noisy systems like heating.

Following this discussion I'm now pretty sceptical that it's possible to answer the question posed in the title to this thread experimentally without quite a lot (weeks or months) of data from a _well designed_ experiment.  In the absence of this or a lab, theory seems currently our most reliable source.  If it's based firmly on thermodynamics and well peer reviewed for error (difficult I admit given the complexity), then it's very likely to be good, but of course can't be absolutely certain unless backed up by good experiment as there may be material engineering effects which aren't included in the theory.

However that brings us back to the well designed experiment involving quite a lot of data.   Until someone actually does that we will, I fear, need to rely on theory which can only partially be tested.  Personally, if forced to choose (which currently I am), I would prefer to rely on good theory over inferences drawn from insufficiently tied-down experiment.

 I'm absolutely not criticising the excellent data you have produced, much more than anyone else.  If nothing else it has shown us all whats possible, and the discussion it triggers has exposed the challenges in collecting data from which conclusions can reliably be drawn.

I'm more motivated now to see if I can find ways to make a model more transparent!  Hopefully you are motivated to post more good data that might help us all gain confidence in some conclusions.

You required a modeling tool to answer the questions.

IAT           10           Avg. 15              20

PI          10644         74262            55484

COP         5.64            2.84               3.24

PO         60030        211217           179968

LWT       21.28        21.14 / 55         42.9

IAT          10            10 / 19.37        20

@derek-m Is this a true representation of the average 15deg house. Surely the calculation should be based on 12 hrs @ 10 degC, 24 hrs @ 20 degC then finally 12 hrs @ 10 degC ? ie a full cycle rather than the half cycle.

@jamespa - I am reminded of another old saw of mine, better a good enough answer to the right question than an exact answer to the wrong question. I think that comes from my background, and what I do, medicine and navigation, which are both applied sciences, where one often has the awkward pairing of uncertainty with an absolute requirement to make a decision (both the sea and the grim reaper are pressing creditors). Sometimes 'wait and see' is a valid, indeed good, decision/response, as it reduces harmful meddling, sometimes it is not. As has been said, better a live problem than a dead certainty.

Another common problem in medicine is simply that there is no way you can do the perfectly controlled experiment. Even when you can, the results often  fail to apply to the real world, because unlike the lab, which has perfectly controlled subjects, the real world has people, who will keep doing the most extraordinary things in environments that constantly change. This is akin to where we are with heat pumps. No two people run their heats pumps exactly the same way, even the same person can and will change what they do, all against backgrounds of changing environments - different houses, different weather, different whatever, including those pesky unknown unknowns. One of the ways we deal with this in medicine is by having a hierarchy of evidence, with the randomised controlled trial at the top (even if they can and not infrequently do give an exact answer to the wrong question, the covid vaccine trials being cases in point) through observation studies (often excellent at getting good enough answers to the right question eg the key study that determined smoking causes cancer was an observational study) to, at the lower end of the hierarchy, case series (n=20) to anecdotes (n=1).

All of which is to say we might, in the absence of a robust theory (formula) that can be shown to apply to the real world, have to decide whether observational studies are good enough. The formula, if it can be assembled, is in the form energy in over time t = OAT IAT and all the rest we don't really know, and that is the problem, we really don't know, and that leaves us with our observations, and how to interpret them. 

For now, I will continue to collect data. I am confident the one we are most interested in, the energy in, is good enough, because I can check it against a manual reading of external energy meter, which I started doing on a daily basis just under two weeks ago, which now suggests the calculated energy in does underestimate actual energy by a fixed amount, calculated is ~80% of actual. Here's a scatter plot of the 12 days worth of data:

I will shortly make the correction to the calculated value permanent, once I am confident the relationship (under-reads by 20%) is stable. Why it under-reads I do not know, but my best guess is that Midea measures amps and volts as used in the heat pump box, and misses out things like the secondary circulating pump, which although powered directly from the heat pump, is not part of the heat pump. If that is correct, it is also a measure of the penalty of having a plate heat exchanger, and, of necessity thereby, a secondary pump. It also means, if the secondary pump is the reason for the under-read, that the correction factor will vary with heat pump output, at higher outputs the presumably constant secondary pump use will remain constant, while the total output increases, meaning the secondary pump will use a smaller proportion of the total.

I am also fairly confident my IAT measurements are good enough, as I have other ways of measuring it beyond the MD02 modbus sensor, and they are usually all with half a degree of each other, probably just measurement error, or possibly being in different locations.

The OAT is less reliable, as it is measured by (inside) the heat pump, and visibly is affected by the heat pumps running. I have looked into getting an outside modbus sensor, but there are problems. The sensor that could do it, the MD02, isn't weatherproof, and ones that are weatherproof either don't have modbus, or if they do, they require a higher voltage than that available (5V, because it actually comes from the USB port).

The rest of the data (LWT, RWT, flow rate etc) are all Midea's readings, and can be partially verified by taking readings, but I wouldn't swear as to the precision...  

Once we have enough data, then we can decide how to interpret it.    

Posted by: @iaack

↑

You required a modeling tool to answer the questions.

IAT           10           Avg. 15              20

PI          10644         74262            55484

COP         5.64            2.84               3.24

PO         60030        211217           179968

LWT       21.28        21.14 / 55         42.9

IAT          10            10 / 19.37        20

@derek-m Is this a true representation of the average 15deg house. Surely the calculation should be based on 12 hrs @ 10 degC, 24 hrs @ 20 degC then finally 12 hrs @ 10 degC ? ie a full cycle rather than the half cycle.

I performed the test that was requested, which was 12 hours @ 10C, then 12 hours @ 20C. I'm not certain what was the purpose of the test or its usefulness, since the IAT was never going to increase from 10C to 20C immediately, and in fact did not reach 20C after the full 12 hours of operation, even with the heat pump running at maximum for the whole 12 hour period, hence the lower efficiency and higher electrical energy consumption.

I am not certain if carrying out the test you propose would produce any meaningful results, since going from 20C to 10C would undoubtedly cause the heat pump to stop operating, and the fall in IAT would be dependent upon the thermal capacity of the building, which was not specified.

If the test was to investigate how the use of averaging can affect the results obtained, a much better test would be to run the heat pump at a constant IAT of 20C, and at a constant IAT of 10C, for a 24 hour period.

Then run the heat pump at an IAT of 20C and OAT of 5C for 12 hours, followed by 12 hours at an IAT of 20C and OAT of 15C, giving an average OAT of 10C over the 24 hour period.

How do you think that the results would compare?

@derek-m - I did not request any test, I directed a question at @jamespa, and with your razor sharp intellect you must have known it was a hypothetical question. It even started with the word 'imagine' and described what was clearly an imaginary situation. I'll reword and slightly rework it for you.

Proposition 1: given a steady state over a long enough interval (24 hours in this case), then the energy in will equal the energy out.    

Proposition 2: all other things, including OAT, being equal, a building at a higher average IAT will lose more energy than one with a lower average IAT.

Imagine (<= that tell's you this is a hypothetical question) four identical buildings, identical apart from:

A has a steady average IAT of 20

B has a steady average IAT of 15  

C has an average IAT of 15, based on 12 hours at 20 degrees (daytime comfort) and 12 hours at 10 degrees (nigh time saving)

D has a steady average IAT of 10.

It doesn't matter how C actually gets the average, all that matters is that the setback plus heating on periods average to 15 degrees, and during the hours of use, the house is at 20 degrees. Thus, in this hypothetical example, A and C have the same comfort level.

A will use the most energy, B and C will use the same amount and be in the middle, and D will use the least amount of energy.

True or false?

It is a general question, approaching the thread's core question from another angle: if you keep something hotter on average, then surely it must use more energy?

Posted by: @cathoderay

↑

@derek-m - I did not request any test, I directed a question at @jamespa, and with your razor sharp intellect you must have known it was a hypothetical question. It even started with the word 'imagine' and described what was clearly an imaginary situation. I'll reword and slightly rework it for you.

Proposition 1: given a steady state over a long enough interval (24 hours in this case), then the energy in will equal the energy out.    

Proposition 2: all other things, including OAT, being equal, a building at a higher average IAT will lose more energy than one with a lower average IAT.

Imagine (<= that tell's you this is a hypothetical question) four identical buildings, identical apart from:

A has a steady average IAT of 20

B has a steady average IAT of 15  

C has an average IAT of 15, based on 12 hours at 20 degrees (daytime comfort) and 12 hours at 10 degrees (nigh time saving)

D has a steady average IAT of 10.

It doesn't matter how C actually gets the average, all that matters is that the setback plus heating on periods average to 15 degrees, and during the hours of use, the house is at 20 degrees. Thus, in this hypothetical example, A and C have the same comfort level.

A will use the most energy, B and C will use the same amount and be in the middle, and D will use the least amount of energy.

True or false?

It is a general question, approaching the thread's core question from another angle: if you keep something hotter on average, then surely it must use more energy?

 

Don't panic. It was a hypothetical answer that I supplied, since the required conditions are not possible to achieve even using a modeling tool.

I also suspect that you are getting your 'use energy' and 'lose energy' confused.

B and C would 'lose' the same quantity of thermal energy over a 24 hour period.

Do you really spend 12 hours in bed, hypothetically?

Posted by: @derek-m

↑

Do you really spend 12 hours in bed, hypothetically?

Yes, to recover from your pointless bickering. I'm rather surprised you didn't have a go at the nigh typo. Lately I have need more like 18 hours, and if you carry on as you are, I will need 24 hours, 100% under the duvet. At least I can console myself that I will have saved some loose energy use.

Posted by: @derek-m

↑

@jamespa

I can now provide some figures that may be useful in understanding what may be occurring within a hypothetical home during an overnight setback.

The data is based upon a 14kW Ecodan, operating at a home with a calculated heat loss of 12kW, maximum heat emitter capacity of 24kW, and an assumed medium to high thermal mass of 250kW at an IAT of 20C. The system is operating in WC mode with curve settings of 53.57C @ -5C and 25.7C @ 20C.

With the system balanced at an OAT of 10C, the following values will be probable.

IAT = 20C, COP = 4.46, LWT = 37.46C, Energy In = 1120W, Energy Out = 4999W with a Heat Loss of 5000W.

Attached are the results from a 6 hour setback test, which shows the probable effects within the system. Table 1 being with no recovery boost and Table 2 with sufficient boost to return the IAT to 20C by hour 24.

Please feel free to make constructive criticism.

-- Attachment is not available --

 

I think this modelling merits a bit more discussion as the results are very plausible.

Unless someone can convince me otherwise the first of the two scenarios is not particularly interesting as something which one might practically do, since the house is severely depleted of stored energy, ie it doesn't actually recover.

The second however is interesting. 

Crudely one might first estimate saving in energy in = saving in energy out / cop.  But of course we know that is not the case because the hp has to work harder during recovery so, all other things being equal, cop must fall. 

According to your simulation the rather modest saving in energy in is only a little over half the crude figure, even though cop has fallen only fairly marginally.  Just eyeballing the figures tells us why, when cop is low, energy supply is high, so the reduced cop during recovery has a disproportionate effect.  All quite intuitive and plausible.

Now two further things immediately leap out intuitively:

Firstly your simulation is at constant oat.  It wouldn't take much if a variation in oat to change the cop (one way or the other) by a similar amount or even more.  In particular if recovery were confined to the daytime period the diurnal variation in temperature may work significantly in our favour.  Conversely if recovery starts too early (eg in time for breakfast), the result could be even less favourable.

Secondly, your result is at modest oat where variations in cop don't effect the energy in as much, because it's already low.  At lower OATs the variation in cop during recovery may well have a greater negative effect, reducing the relative saving still further.  I think your earlier results also suggested this as did my entirely independent and much cruder model.

I think some playing with these parameters will be interesting to understand whether the (modelled) behaviours do in fact follow these informed speculations or whether something else is going on.

Finally if we are to measure experimentally changes in energy in of 4% which your model predicts (a figure which is highly plausible), I really can't see how it can be done without lots of data from carefully designed measurements and well considered data processing.  The day to day noise has got to be greater than 10%, let's say around 20% so we are trying to dig a signal of 4% out of a noise of perhaps 20%.  Generally SNR goes as the square root of the number of samples, so very roughly we are going to need order 25, perhaps 50 samples to be sure.  We should perhaps use the model to find conditions when the saving is greater thus easier to dig out of the noise, but also expect a long, carefully considered, haul on the experimental verification of the theoretical predictions. 

The more I think about this the more I favour the synchronous detection technique outlined upthread, if anyone were prepared to run it.  There are some tweaks that are worth thinking about eg how to minimise the effect of the gradual shift in temperatures through the season.  This might influence experimental design or just data processing, I don't currently know (and will give it some thought).  Controlling it with something other than a 7day programmer would also be good, because 7 is a prime number thus not divisible neatly into regular periods greater than a day but less than a week (for example 2days), which I think would be optimum for collecting the max amount of good data in the shortest period of time.

@jamespa - I still think you are over-complicating this. How about trying this approach, which I can do with the data I have:

(1) running under weather compensation, there should be a clear relationship between average OAT over an hour and energy in over that hour

(2) if that is the case, I can test that with my data, then for any given hourly average OAT, I know what the energy in will be for that hour

(3) I can then look at a 24 hour period, and determine the expected energy in with no setback/recovery boost over that 24 hour period would be

(4) I can then compare that 24 hour expected energy in with the observed 24 hour energy in when I do run with a setback and recovery boost

(5) comparing the observed with the expected should answer the question at the core of this thread

When I have some time, I will try this, and see what results I get.   

@jamespa

I had arrived at similar conclusions so time ago.

Please find attached a table that I also produced some time ago to investigate the effect of thermal mass, variations in OAT, and different setback periods. Don't take the data as gospel since I may have made improvements to the modeling tool since the table was produced, but at least it provides some idea of the likely effect of changes to the above mentioned parameters.

The data in the table was merely copied from the modeling tool so don't bother searching for any formula. Also there is no 'recovery boost' utilised, hence the lengthy recovery periods.

If (1) is the case I potentially agree. Its not the case for my fossil fuel boiler (there is a correlation but it's very noisy indeed) so I don't see it's likely to be the case for a heat pump, but if course I'm willing to be proven wrong.

Perhaps you could plot, from your data, hourly oat Vs hourly energy out as a scattergram, we can then all take a look and assess options to simplify the experiment.  This is a question of experimental technique having regard to the nature of the data.

Posted by: @jamespa

↑

If (1) is the case I potentially agree. Its not the case for my fossil fuel boiler (there is a correlation but it's very noisy indeed) so I don't see it's likely to be the case for a heat pump, but if course I'm willing to be proven wrong.

Unless you have hourly or shorter time intervals for your data, it would be very difficult to analyse fossil fuel use in the way I suggest. In the case of my old oil boiler, it would be totally impossible , I had no idea what hourly oil consumption was. Nor do most fossil fuels systems use weather compensation. Instead, I am making use of particularities of heat pump systems.

I'm starting with the premise that in a standard heat pump setup with WC, the OAT sets the LWT via the WCC, with the WCC usually being a straight or possibly stepped because the numbers used appear to be integers linear relationship between OAT and LWT, and that the energy in is proportional to the LWT, so there should be a relationship. All usual caveats about other factors staying constant and can't extrapolate to other settings etc apply. That's a prediction from a hypothesis, and doing a scatter plot was indeed exactly how I was going to start testing the hypothesis. In practice it will be a bit tedious, because the WCC got changed from time to time, but I have records for when it was changed and by how much, and all I need to do is isolate the various periods and do the plots. 

I think you may have misread my comment, its about the relationship between OAT and energy IN, not energy OUT. It is easier to measure, can be verified by the external kWh meter, and at the end of the day or rather billing period it is what determines the amount we pay.