Posted by: @jamespa

↑

Posted by: @jamespa

↑

Posted by: @derek-m

↑

Don't forget, the DT values supplied are integers, but I doubt that they are in the real World. Since heat pumps like to operate at a nominal DT of 5C, a 1C difference due to one value being rounded up and the other being rounded down, immediately introduces a 20% error.

Good point.  Depending on how good the control loop is the actuals may be a bit noisy, which on average might improve the resolution, but that's clutching at straws.

Actually, on second thought, it's the actual temperatures which are measured and reported as integers.  Given that those do vary, and given that flow rate is roughly constant and thus dt varies, the average dt should have better resolution.

So I don't think that its likely to explain the discrepancy between 8kW measured and 11kW specified.

 

There are possibly two likely answers, or even a combination of both.

I think it has already been established that the OAT sensor is reading low by several degrees, and is less accurate the harder the heat pump is working.  The true heat loss is therefore less than that indicated by the OAT sensor reading.

The PHE is probably reducing the water temperature at the heat emitters by 5C, so this dictates how much thermal energy is going into the home. From the Detailed Data tables the required LWT at an OAT of -2C would be 52.5C, which would indicate a heat loss of 11.3kWh.

If the actual water temperature at the heat emitters is 47.5C, this would indicate the quantity of thermal energy being supplied to the home is more in the order of 8.6kWh.

Posted by: @cathoderay

↑

This chart shows the heat pump not quite meeting demand on the left hand side where the OAT is around zero, leaving the IAT below design IAT (19 degrees), nadir 17.3 degrees, and then recovering as the OAT rises, with the IAT rising to meet the desired IAT (had got to 18.6 degrees by noon today): 

-- Attachment is not available --

 

@cathoderay Having looked again at the graph I'm not sure I agree with your analysis.  The lwt achieved appears to be equal to the lwt set, so the heat pump appears to be doing what you are asking it to do, rather than failing to do so due eg to lack of capacity. 

Are you sure that the lwt set is high enough for your emitters to deliver the required output?  Graph suggests not, and heat pump can only supply what the system demands.

See also @derek-m above reminding us that the ft seen by the emitters may well be 5C lower.

Posted by: @jamespa

↑

Actually, on second thought, it's the actual temperatures which are measured and reported as integers.  Given that those do vary, and given that flow rate is roughly constant and thus dt varies, the average dt should have better resolution.

So I don't think that its likely to explain the discrepancy between 8kW measured and 11kW specified.

At the moment, it seems to me there is not enough second thoughts going on, and instead there is a lot of noise repeating stuff we already know - in some cases have known for years - and the signal is getting lost. I for one don't need to be told yet again that PHEs interfere with heat transfer, or that Midea claim my heat pump can deliver outputs around 11.5kW at high LWTs and low OATs, or that accurate measurements are difficult and expensive to do, but conversely if you want delta t rather than absolute values then the errors may cancel out leaving you with not too bad an estimate of the delta t, and generally Midea's habit of using integers is a pain in the arse. All of this and more has been known for ages. Merely repeating it is just creates more noise without a signal.

What we are interested in here is the practical question of whether setbacks can save energy without compromising comfort. That has led to some secondary questions like what is the real OAT, and what is the real heat pump output. On the face of it there are some pretty serious discrepancies on this latter question: 8kW max output from a 14kW unit might well be enough of a discrepancy to put a sparkle in the eye of a local trading standards officer.

Posted by: @jamespa

↑

Having looked again at the graph I'm not sure I agree with your analysis. The lwt achieved appears to be equal to the lwt set, so the heat pump appears to be doing what you are asking it to do, rather than failing to do so due eg to lack of capacity. 

Are you sure that the lwt set is high enough for your emitters to deliver the required output?  Graph suggests not, and heat pump can only supply what the system demands.

Well, I am not sure I agree with your analysis. When the actual LWT is varying all the time, it is not the peaks that reflect the actual LWT over time, but the average. Here's the chart again:

On the left, during the defrost cycles, the actual LWT only briefly touches the set LWT some of the time. The average actual LWT over time is less, considerably less, and can be calculated from the raw data (though I haven't done that, the Mk 2 Eyeball Integrator is good enough for me here). On the right, where there is 'normal' cycling, the average actual LWT is close to the set LWT. I think it is possible Midea know their heat pumps almost always cycle in normal use, and so have the control logic set up to achieve an average actual LWT that matches the set LWT. Even if they don't, that is what appears to happen in practice during 'normal' cycling. When defrost cycles appear, all that goes out of the window. I suspect the average actual LWT on the left during the defrost cycles may in fact be a bit less that on the right hand side, during the 'normal' cycles. The heat pump is not doing what is being asked of it, instead, it seems heat pumps, at least my heat pump, have/has a serious flaw built in, it in effect fails when it is needed most. You can see that happening in the chart. The average LWT (and so ultimately the energy transfer, because of the steady flow rate) on the left is the same or a bit less than that on the right, despite the fact that on the left the reported OAT is a whole 8 degrees less. 

Assuming the Midea reported modbus collected raw data is tolerably accurate - and it probably is, last time I checked, the calculated energy out was close to that based on the Midea total lifetime energy out: for the last 24h, calculated energy out is 133.71, that based in the lifetime energy out is 133 - then the burning - or perhaps that should be failing to burn - question is why does my heat pump only deliver ~8kW max power when on paper it is supposed to be capable of ~11.5kW power? What appears to be happening is that it delivers at the 'norm' capacity level (CL - see table below, zero OAT, Set LWT 50 = 6.6 - 11.7kW out). Of course, the actual energy delivered to the interior of the house is even worse. For me, the question is why is it apparently delivering at the 'norm' capacity level when it needs to deliver at nearer the 'max' capacity level? Raising the Set LWT isn't the answer, it is already all but maxed out, and in fact raising it to 60 degrees actually reduces the power output (COP goes through the floor when LWT goes above 55, 'norm' capacity level at zero OAT falls from 8.8kW at LWT 50 to 7.2kW at LWT 60).  

The chart (the graph, not the table) above also suggests that the 'real' OAT is not 'several degrees' above the reported OAT. Since the real OAT is unknown, strictly speaking there is no way of knowing what the difference is. However, when we look at the right hand side of the chart, the recorded OAT only rises by one degree during the periods when the compressor is off. That is not 'several degrees'. The picture on the left hand side is complicated by the defrost cycles, during which heat from the house is actively and directly delivered to the heat pump. Of course the apparent OAT goes up, because the OAT sensor is inside the heat pump which is being heated to melt the ice. Those spikes do not tell us anything about the real OAT.        

@

Posted by: @derek-m

↑

There is normally notes attached to the data tables to indicate if defrosting cycles have been incorporated in the results.

The string 'defrost' does not appear to exist in the Midea Engineering Data manual. I have plotted the 'norm' power outputs over the range of OATs at a LWT of 50 and it does appear there is a dip over the defrost range of OATs, but it is not that marked. Make of this what you will: 

I’m a bit lost on this thread can someone enlighten me if the topic has changed? It was :

“Do setbacks save energy without compromising comfort?”

From what ive just read above it might be that an accurate modelling comparison with real life data is absolutely relevant but a clarification of the mission might help. What I’m not sure about is - is cathoderay’s setup a typical setup. Unless the mission is about identifying various compromised setups. 

If the latter is the objective of the thread then surely all the known compromised elements are essential to be reminded of.... .Or is my brain just frazzled now 🧠💥

Posted by: @cathoderay

↑

Posted by: @jamespa

↑

Actually, on second thought, it's the actual temperatures which are measured and reported as integers.  Given that those do vary, and given that flow rate is roughly constant and thus dt varies, the average dt should have better resolution.

So I don't think that its likely to explain the discrepancy between 8kW measured and 11kW specified.

At the moment, it seems to me there is not enough second thoughts going on, and instead there is a lot of noise repeating stuff we already know - in some cases have known for years - and the signal is getting lost. I for one don't need to be told yet again that PHEs interfere with heat transfer, or that Midea claim my heat pump can deliver outputs around 11.5kW at high LWTs and low OATs, or that accurate measurements are difficult and expensive to do, but conversely if you want delta t rather than absolute values then the errors may cancel out leaving you with not too bad an estimate of the delta t, and generally Midea's habit of using integers is a pain in the arse. All of this and more has been known for ages. Merely repeating it is just creates more noise without a signal.

What we are interested in here is the practical question of whether setbacks can save energy without compromising comfort. That has led to some secondary questions like what is the real OAT, and what is the real heat pump output. On the face of it there are some pretty serious discrepancies on this latter question: 8kW max output from a 14kW unit might well be enough of a discrepancy to put a sparkle in the eye of a local trading standards officer.

Posted by: @jamespa

↑

Having looked again at the graph I'm not sure I agree with your analysis. The lwt achieved appears to be equal to the lwt set, so the heat pump appears to be doing what you are asking it to do, rather than failing to do so due eg to lack of capacity. 

Are you sure that the lwt set is high enough for your emitters to deliver the required output?  Graph suggests not, and heat pump can only supply what the system demands.

Well, I am not sure I agree with your analysis. When the actual LWT is varying all the time, it is not the peaks that reflect the actual LWT over time, but the average. Here's the chart again:

-- Attachment is not available --

 

On the left, during the defrost cycles, the actual LWT only briefly touches the set LWT some of the time. The average actual LWT over time is less, considerably less, and can be calculated from the raw data (though I haven't done that, the Mk 2 Eyeball Integrator is good enough for me here). On the right, where there is 'normal' cycling, the average actual LWT is close to the set LWT. I think it is possible Midea know their heat pumps almost always cycle in normal use, and so have the control logic set up to achieve an average actual LWT that matches the set LWT. Even if they don't, that is what appears to happen in practice during 'normal' cycling. When defrost cycles appear, all that goes out of the window. I suspect the average actual LWT on the left during the defrost cycles may in fact be a bit less that on the right hand side, during the 'normal' cycles. The heat pump is not doing what is being asked of it, instead, it seems heat pumps, at least my heat pump, have/has a serious flaw built in, it in effect fails when it is needed most. You can see that happening in the chart. The average LWT (and so ultimately the energy transfer, because of the steady flow rate) on the left is the same or a bit less than that on the right, despite the fact that on the left the reported OAT is a whole 8 degrees less. 

Assuming the Midea reported modbus collected raw data is tolerably accurate - and it probably is, last time I checked, the calculated energy out was close to that based on the Midea total lifetime energy out: for the last 24h, calculated energy out is 133.71, that based in the lifetime energy out is 133 - then the burning - or perhaps that should be failing to burn - question is why does my heat pump only deliver ~8kW max power when on paper it is supposed to be capable of ~11.5kW power? What appears to be happening is that it delivers at the 'norm' capacity level (CL - see table below, zero OAT, Set LWT 50 = 6.6 - 11.7kW out). Of course, the actual energy delivered to the interior of the house is even worse. For me, the question is why is it apparently delivering at the 'norm' capacity level when it needs to deliver at nearer the 'max' capacity level? Raising the Set LWT isn't the answer, it is already all but maxed out, and in fact raising it to 60 degrees actually reduces the power output (COP goes through the floor when LWT goes above 55, 'norm' capacity level at zero OAT falls from 8.8kW at LWT 50 to 7.2kW at LWT 60).  

-- Attachment is not available --

 

The chart (the graph, not the table) above also suggests that the 'real' OAT is not 'several degrees' above the reported OAT. Since the real OAT is unknown, strictly speaking there is no way of knowing what the difference is. However, when we look at the right hand side of the chart, the recorded OAT only rises by one degree during the periods when the compressor is off. That is not 'several degrees'. The picture on the left hand side is complicated by the defrost cycles, during which heat from the house is actively and directly delivered to the heat pump. Of course the apparent OAT goes up, because the OAT sensor is inside the heat pump which is being heated to melt the ice. Those spikes do not tell us anything about the real OAT.        

  

Thank you very much for your data.  I am personally pleased that, with a bit of processing and incorporation of the fabric heat capacity into the calculations, it has been possible to establish what is, at least to me, a remarkably good correlation between energy supplied to the house and OAT/IAT, certainly much better than I have seen previously in any field measurement.

As to some of your comments, if you would like others to provide insight into what you are seeing, you may find it helps not to get angry with them.  I can understand your anger with the tentative conclusions, but others on the forum are not responsible for those, and are merely trying to offer some insight which they hope will be helpful; obviously you do not have to take any notice. 

For what it is worth I will add, as a final observation on your data, that it seems unlikely that the Midea heat pump targets by design average FT, its much more likely that it has a simple feedback loop which attempts to achieve the set FT.  There was some discussion of this, in the context of defrost, on another forum and the general conclusion seemed to be that possibly only Nibe currently modifies their WC curve to account for the 'lost output' during defrost.

I hope you solve your mystery.

Posted by: @sunandair

↑

I’m a bit lost on this thread can someone enlighten me if the topic has changed? It was :

“Do setbacks save energy without compromising comfort?”

I agree, and every now and then I as the thread starter try to remind folks what this thread is actually meant to be about (eg an hour or so ago I wrote: "What we are interested in here is the practical question of whether setbacks can save energy without compromising comfort"). This objective does get regularly compromised by certain individual(s) who shall remain anonymous in this particular post who like to gloat over point out the deficiencies in my heat pump system, despite the fact the said deficiencies have been established and then beaten to death countless times over the last couple of years. Apart from remaining as a background caveat (my system is not necessarily a typical system), they do not need to be constantly repeated.

There is also a direct clash between the empirical approach (me) and the modelled approach (@derek-m), with I think it is fair to say @jamespa sitting somewhere in the middle, but leaning perhaps a bit more towards modelling. I suspect this is due to our backgrounds: I am totally familiar with using messy real world data to make pragmatic decisions (medicine/epidemiology), whereas the other two come from engineering/physics backgrounds where models and precision are greatly valued. Both approaches of course have value.   

I also take the view that even if my system isn't typical, it still remains the case that I (and @kev-m) are the only ones who have collected and published comprehensive raw minute by minute data as well as many charts derived from the data, meaning that this is the only data we (as a group) have to work with - currently it is my/@kev-m's raw data or none - and furthermore, while the specifics of my system may not be typical, it might nonetheless be able to answer the general and pragmatic real world question posed in the thread title. In other words, having a PHE doesn't mean I can't answer the question in the thread title. The specifics may be different, but the overall answer might be general.    

Posted by: @sunandair

↑

I’m a bit lost on this thread can someone enlighten me if the topic has changed? It was :

“Do setbacks save energy without compromising comfort?”

From what ive just read above it might be that an accurate modelling comparison with real life data is absolutely relevant but a clarification of the mission might help. What I’m not sure about is - is cathoderay’s setup a typical setup. Unless the mission is about identifying various compromised setups. 

If the latter is the objective of the thread then surely all the known compromised elements are essential to be reminded of.... .Or is my brain just frazzled now 🧠💥

I think there has been some scope creep but much of it (not all, I grant) is very relevant to the actual problem in the title. 

@derek-m has created a model which attempts to answer the question (answer - probably there is a small saving of a few percent (<5%) in some, perhaps many circumstances, but also the potential for setback to cost money in certain circumstances).  However others have expressed a very sensible desire to have experimental verification.  Unfortunately any experimental data so far produced (not just on this forum) appears very 'noisy' in relation to the expected saving.  So noisy in fact that trying to measure any saving due to setback is likely to require months of carefully considered data collection, and then some carefully considered data processing.

One way to help with this is to understand the source of the 'noise'.  If it can be modelled (ie its not 'actually' noise, its a manifestation of some other effect), then it can be accounted for in any comparison between actual energy used when set back is in operation, and actual energy used when it is not.

Well it turns out (see my post from early yesterday) that, at least with the data we have, it can largely be modelled by accounting for the energy stored in the fabric.  There is (experimentally) a very good correlation indeed between energy delivered to the house and OAT provided you take into account energy stored in the fabric.  I do this in the results posted above, and the correlation in the experimental data is materially better than if you don't take into account the energy stored in the fabric.  This matters if we want to interpret experimental data accurately in order that any conclusions are robust.

Hopefully that helps.  

For the avoidance of doubt I neither prefer models nor experimental data.  I'm a pragmatist so what I will use is whichever approach gives results or a combination of the two, ideally (in all cases) supported by some theoretical understanding as well. 

At present we simply do not have experimental data which tells us the answer to the question, but we do have a model which might.  So for now the model results are what we have to go on.  Unfortunately even the model results are fairly specific, which is where the theoretical understanding comes in as it helps us understand how to extrapolate the model results to slightly different situations.  I am pretty confident we will only ever get a small sample of experimental results (because few people will have the time or inclination to conduct a properly controlled experiment), so for general conclusions we will remain reliant on at least a degree of theory and/or modelling, hopefully confirmed and certainly informed by the experimental results we actually get.

Posted by: @jamespa

↑

much of it (not all, I grant) is very relevant to the actual problem in the title

I do of course by and large agree, much (but definitely not all, by a long shot) is relevant, because it turns out that to answer the original question (do setbacks save energy without compromising comfort?), we need to have some way of establishing what the energy use would have been without the setback. The problem is that, unavoidably, when we have a setback (and recovery boost, to satisfy the confort requirement), we don't have real world data on what the energy use would have been without that setback. In effect, the experiment destroys the control. The core clash is over how to set about getting an answer to that question: on the one hand, start with theory and derive a model, on the other, start with real world data, and try to determine relations that exist within that data (which I agree can be called a model). But the underlying argument is I suspect philosophical, about what we trust. I just happen (based on what I have seen over the years) not to trust models, unless and until they have been validated against the real world. I am instead 1000 times more comfortable starting with real world data, and trying to make sense of it. And I do agree that your approach of factoring in the thermal mass of the building in the way that you have has borne real fruit. As I see it, there is only one more step required, to get from the middle chart, which has sum (IAT - OAT) for the day as the predictor (x axis) variable, to one which has sum (IAT - OAT) for the hour as the predictor variable, which means we can then estimate the power use for any hour, given the sum (IAT - OAT) for that hour, which, in a setback and recovery period, should be what that hour would have used had there not been a setback.

The only obvious remaining question I have is this: what was the sample used? Looking at the spreadsheet, I think it uses all data, which means it also includes days when there was a setback and recovery boost, and so is strictly speaking not representative of the subset of days when there was no setback. Here is the original version of your chart: 

and here is a version in which I have removed the setback days (5th to 27th November):

It is now I dare say visually even tighter (and R squared has increased, from 0.951 to 0.969, whether that is significant I do not know), though inevitably the sample size is smaller. 

I do think this chart has within it 'good enough' information to mean that we can estimate what hourly energy use would be given normal 24 hour running. It just needs to be modified to cover hours rather than days, and I may be wrong, but I don't think that can be achieved by just dividing the days results by 24, because we will lose a lot of detail that is in the raw data; instead, the pivot table needs to be 'grouped by' hour rather than day, and I can't see how to do that (I'm using LibreOffice Calc, it won't load properly in my version of OpenOffice).

Alternatively, perhaps we can skip the hourly stage, and jump right to daily use given that days sum (IAT - OAT)? Or does that lose the hourly (and necessary) detail?

Despite our differences, we have come a long way, and we should be grateful for that.       

Posted by: @cathoderay

↑

The only obvious remaining question I have is this: what was the sample used? Looking at the spreadsheet, I think it uses all data, which means it also includes days when there was a setback and recovery boost, and so is strictly speaking not representative of the subset of days when there was no setback.

I used all the data from the start of the heating season as I did not know when setback was initiated and ceased and was not inclined to inspect all the records to deduce when that was.  

Posted by: @cathoderay

↑

I'm using LibreOffice Calc, it won't load properly in my version of OpenOffice

Im also using LibreOffice.  I have also exported it to excel but my version is 2003 which accepts only 64K rows, thus truncates the data, so I have not used the exported version.

Posted by: @cathoderay

↑

Alternatively, perhaps we can skip the hourly stage, and jump right to daily use given that days sum (IAT - OAT)? Or does that lose the hourly (and necessary) detail?

My feeling is this

• use the minute data to work out the energy delivered and degree minutes, sum to daily
• Work out the expected energy required on a daily basis, accounting for the fabric

I don't think hourly data helps us, it will almost certainly introduce more noise, whilst not giving us any information needed in the final result.

In principle we might be able to skip the second step and just compare the slopes and intercepts of two sets of data, one with setback and one without.  Id want to be certain that any difference found is significant, having regard to the remaining noise.

Here are the results of running the Libreoffice regression analysis on two subsets of the data, corresponding to when there was and when there was not setback.

Currently the 95% confidence limits on the slopes overlap (albeit only just), so we cannot be confident that they are in fact different.  Also note the very high intercept in the setback data.  This must mean something, I don't know what, and if the value is maintained when we have more data it definitely matters to the answer to the question posed in the thread!

Although this is not yet an answer, it certainly has the makings of a way to answer the question with more data, perhaps a relatively modest amount.  Of course the answer, if we do get one, applies only to your system.  This is where the theory and modelling comes in to help us understand how it might apply to other systems, where the owner wont actually do the experiment.

@jamespa

Well done James, I could not have said it better myself.