I am fully aware that you don't have a regression theory, but since you like to use the word 'theory' for my ideas and predictions, I thought that you would like me to use it.
Let me see if I can explain my statement in simpler terms.
During setback there is no energy being put into the home by the heat pump, but heat loss still takes place, so eventually IAT will start to fall.
At the end of the setback period, the heat pump restarts, but it now not only has to supply the present heat loss, but must also replace the lost energy if it is to raise the IAT back to the desired value.
So over the 24 hour period the quantity of thermal energy put into the home will be sufficient to achieve status quo.
The problem that I see with your prediction method, is that it still requires the same quantity of thermal energy during the 18 hour period after the setback, whether the setback takes place or not.
So the new total energy input into the home is the quantity during the 18 hour period after the setback would have taken place, plus the now predicted energy supplied during what would have been the setback period. Hence the new total is greater than that observed during the actual setback case.
This is where I find that your method starts to fall apart, since if energy is being put into the home during what would have been the setback period, then the IAT would not have fallen, so the heat pump would not have to work so hard to recover from the IAT reduction that would have occurred had the setback taken place. The energy input during the 18 hour period after the setback, would therefore not be the same as that observed when setback did take place, and would therefore be lower than your prediction.
Rather than using your usual 'rambling' method to try to confuse the issue, just state which, if any, of the above observations that you don't find to be correct.
The spreadsheet predictions only become noticeably low during the setback 'on' periods, suggesting there is something the spreadsheet model fails to take into account during those periods. The reason this matters is those too low values will in turn lower the 24 hour predicted totals, which in turn will have the effect of lowering the predicted savings.
Please explain how you know that the spreadsheet predictions are low, and that it is not that your predictions are high?
Obviously it is more convenient from the setback case if the spreadsheet is low rather than your predictions being high.
I am not sure what I am supposed to have not corrected. Most of it is raw data, just numbers made available by the Midea wired controller, inherent weaknesses and all, and the one calculated value which clearly does need a correction (energy in, because it doesn't match the external kWh meter) already has a long established correction. Maybe I will get a modbus enabled external kWh meter, maybe I won't, if I do, then I hope we can agree its data is as close to gold standard as is possible. The OAT I have left as it is, because it represents what actually goes into the heat pump, rather than some other OAT from air that never goes anywhere near the heat pump.
There are a number of weaknesses within the raw data.
The first is that the OAT sensor reading is affected by the operation of the heat pump by up to it would appear 5C. This causes problems within the spreadsheet and I would assume with your regression calculation, since the predicted heat loss from the home would be much higher than the true value.
The second problem is the frequent cycling that occurs, which again makes obtaining reasonably accurate hourly values more difficult to achieve.
From calculations I have performed it would appear that the V x I calculation, without the 1.18 correction, is quite a close match to the power input values shown within the manufacturers data tables. This therefore posses the question, what additional load could the external power meter be measuring in addition to the heat pump? The V x I input power will be multiplied by the COP value to produce thermal energy output, but the 0.18 of the 1.18 may only providing a COP of 1 if inside the thermal envelope, or worse still may be lost to the outside air.
@derek-m - it seems to me that you might be moving close to torrents of abuse territory. I can take it, but it is not pleasant, and I am sure others don't appreciate it. Let's try not to shout at each other.
So please explain why in the past have I received so much verbal abuse from you when reviewing your data and ideas.
For the umpteenth time, I advised you of the inherent problems with your system and the raw data produced from it. I think that it may have been over one year ago when I identified how the PHE would have a detrimental effect upon overall efficiency, and more recently the 'cold well' effect, again affecting efficiency, and also affecting the OAT sensor reading upon which your predictions are totally based.
You are obviously struggling to understand some of my explanations, so in future I will try to keep them as simple as possible.
I am not sure about 'so much verbal abuse'. Have I ever called you an idiot? Told you you were thick? Or an a**ehole? I don't recall ever having done that sort of thing. I did once suggest you needed to 'keep taking the meds' but that was in jest - I clearly don't think you are psychotic (though I might now have to revise that - another jest). Yet you feel free to opine that I am 'obviously struggling to understand your explanations' and so you will 'keep them as simple as possible' in future. This is about as close to saying you think I am thick as you can get without actually using the words.
The fact is I absolutely understand all these things, but I don't always agree with the importance or otherwise you may attach to them, which often boils down to a matter of opinion. For example, I have known from day one, long before it was ever discussed on this forum, on the day i saw it installed, that my PHE compromises performance. It's obvious. What is less obvious the quantification of that loss, and it is entirely a matter of personal opinion what I do about the loss in my system, ie keep or remove the PHE. As I have said before it does have some pros as well as the cons, and I remains undecided which wins out.
Likewise, I totally get the OAT thing, though I think we differ in the amount of the effect, again, as much a matter of opinion as fact,as we don't actually have all the necessary facts. I also totally fully 100% appreciate that my regression are done on what the OAT sensor records, but that's exactly what I was trying to do, get the relationship between what the OAT records, whatever it is, and the energy in. It doesn't matter that it is not the true OAT, whatever that is, all that matter is when the OAT says this, then the energy will, on average, be this.
I am fully aware that you don't have a regression theory, but since you like to use the word 'theory' for my ideas and predictions, I thought that you would like me to use it.
You've missed the point here, I view your approach as more theory based (as opposed to mine being observation based) and so call your approach theoretical, in contrast to my observation based approach. I do appreciate you start with the manufacturer's data, but what you don't know is how much of that is evidence based, and how much comes from models - maybe some of the table entries are modelled, who knows. In that case we have possibly have models within models, quite possibly all the way down. Who knows? Furthermore, although you start with the tabled data, the rest of you model is theory based, making use of various theories of how the heat/energy moves within and without the system.
The problem that I see with your prediction method, is that it still requires the same quantity of thermal energy during the 18 hour period after the setback, whether the setback takes place or not.
This still suggests to me you have not grasped what I am trying to do with this chart. I am only trying to test the predictive ability of our respective approaches. The only question is how close is the predicted value to the actual value for each given point in time. What I see is the regression based method gets closer, apart from during the recovery period, which is to be expected, because the regression method deliberately excludes the setback periods, because I want predicted values for when a setback isn't being used, as needed for the entirely separate longer time frame comparisons of 24 hour observed vs expected energy in analysis. In the current chart, I am not interested in longer time frames, just what happens at each time point: how close is the predicted value to the actual value?
It is in the other entirely separate observed vs expected analysis that I sum the observed values and the predicted values over each 24 hour period, and then compare them, to determine what difference there is, if any.
We also diverge on the next two paragraphs, because you are still discussing longer periods, whereas I in the predicted vs actual plots am only considering each corresponding point in time, with no need to sum over periods in time. That is not what these charts are for.
Now, having said all that, let's perhaps somewhat confusingly, actually use the chart (and its underlying data) to do the other analysis: what happens over 24 hours, comparing observed values with expected values. LibreOffice failed to get the dates on the x axis, and I didn't bother to correct that, because the dates/hours didn't matter on that chart, all that matters is the data point line up, which they do. Let us consider the first 24 period with a setback, and lets make it noon to noon. This is what it looks like (I have kept it as a line chart as the more correct bar chart is harder to assimilate, LibreOffice still won't label the x axis correctly), with the 24 hour values underneath:
This suggests a 6.06 kWh saving over the 24 hour period, or 22.91% saving over the expected (predicted) value. Clearly this is just one isolated 24 hour period, result seems optimistic etc etc etc, I post it just to show the method, not as long times series covering many days. In practice, for longer times series, the chart isn't needed, the 24 hour sum data only needs a pivot table to collect it.
Please explain how you know that the spreadsheet predictions are low, and that it is not that your predictions are high?
Obviously it is more convenient from the setback case if the spreadsheet is low rather than your predictions being high.
Hopefully the above explanation makes clear why this question doesn't actually apply. The test analysis (individual point in time value comparison) shows the regression predictions are closest to the actual values, and that is all we are concerned with: which prediction gets closest to the actual values, and then is it close enough? I think it is, and back that up with the R squared value and the residuals analysis etc as to why it is reasonable to use it.
The first is that the OAT sensor reading is affected by the operation of the heat pump by up to it would appear 5C. This causes problems within the spreadsheet and I would assume with your regression calculation, since the predicted heat loss from the home would be much higher than the true value.
The second problem is the frequent cycling that occurs, which again makes obtaining reasonably accurate hourly values more difficult to achieve.
From calculations I have performed it would appear that the V x I calculation, without the 1.18 correction, is quite a close match to the power input values shown within the manufacturers data tables. This therefore posses the question, what additional load could the external power meter be measuring in addition to the heat pump? The V x I input power will be multiplied by the COP value to produce thermal energy output, but the 0.18 of the 1.18 may only providing a COP of 1 if inside the thermal envelope, or worse still may be lost to the outside air.
The point about the OAT being out is that it doesn't matter for my purposes: the regression is for the energy in as predicted by the OAT as measured. The 'real' OAT isn't known, and isn't used: the heat pump, and the regression only know about the OAT as measured. As it happens, I disagree (a matter of opinion not fact) about what the real OAT is, I think it is usually only a degree or two out, but that is neither here nor there for this analysis, which only uses the OAT as measured. Before you start telling me I am thick and need an even simpler explanation, let me tell you I know most of the house responds to the real OAT, heat loss is determined by the real OAT etc etc, but it doesn't matter, because the regression and the heat pump only know about the OAT as measured by the heat pump.
The cycling isn't frequent, typically when not in defrost territory is is about once an hour, or just under. That does not count as frequent cycling.
Maybe the Midea wired controller makes up the V and I values, based on its hidden internal versions of the manufacturers data tables. Who knows? Diesel-gate happened, why shouldn't compressor-gate happen? All I know is that if I take those V and I values, and multiply them by 1.18, then I get the best approximation to the values shown on the external kWh meter, which shows the actual energy in, the actual energy consumed, and what I pay my bills on.
I have emailed various Midea support email addresses, including FHP, asking specifically about how and where the V and I values are measured, and what they actually measure. I will report back if I get any useful replies.
Midea 14kW (for now...) ASHP heating both building and DHW
If your heat pump uses more electrical energy in your predicted none setback scenario, why does it not produce more thermal energy output and cause the IAT to increase?
Since you are obviously enjoying our discussion, perhaps you would care to answer my final question.
One final question.
If your heat pump uses more electrical energy in your predicted none setback scenario, why does it not produce more thermal energy output and cause the IAT to increase?
Interestingly, if you have fancoils, then it is not necessary to raise the FT to increase the output from the emitters, so (1) can be circumvented. Makes me think that a couple of fancoils downstairs (which is where you want morning heat I would suggest) and maybe in the bathroom, pretty much guarantees a win from setback, unless its timed for a night shift.
Just some thoughts I might try to quantify some of this over the next week or two.
@jamespa Could you explain a bit more about fancoils as I haven’t been able to find out much about them? Do they work and do they fit any size of radiator? If they do, they might be one answer to the cycling at low output problem.
@derek-m - it seems to me that you might be moving close to torrents of abuse territory. I can take it, but it is not pleasant, and I am sure others don't appreciate it. Let's try not to shout at each other.
So please explain why in the past have I received so much verbal abuse from you when reviewing your data and ideas.
For the umpteenth time, I advised you of the inherent problems with your system and the raw data produced from it. I think that it may have been over one year ago when I identified how the PHE would have a detrimental effect upon overall efficiency, and more recently the 'cold well' effect, again affecting efficiency, and also affecting the OAT sensor reading upon which your predictions are totally based.
You are obviously struggling to understand some of my explanations, so in future I will try to keep them as simple as possible.
I am not sure about 'so much verbal abuse'. Have I ever called you an idiot? Told you you were thick? Or an a**ehole? I don't recall ever having done that sort of thing. I did once suggest you needed to 'keep taking the meds' but that was in jest - I clearly don't think you are psychotic (though I might now have to revise that - another jest). Yet you feel free to opine that I am 'obviously struggling to understand your explanations' and so you will 'keep them as simple as possible' in future. This is about as close to saying you think I am thick as you can get without actually using the words.
The fact is I absolutely understand all these things, but I don't always agree with the importance or otherwise you may attach to them, which often boils down to a matter of opinion. For example, I have known from day one, long before it was ever discussed on this forum, on the day i saw it installed, that my PHE compromises performance. It's obvious. What is less obvious the quantification of that loss, and it is entirely a matter of personal opinion what I do about the loss in my system, ie keep or remove the PHE. As I have said before it does have some pros as well as the cons, and I remains undecided which wins out.
Likewise, I totally get the OAT thing, though I think we differ in the amount of the effect, again, as much a matter of opinion as fact,as we don't actually have all the necessary facts. I also totally fully 100% appreciate that my regression are done on what the OAT sensor records, but that's exactly what I was trying to do, get the relationship between what the OAT records, whatever it is, and the energy in. It doesn't matter that it is not the true OAT, whatever that is, all that matter is when the OAT says this, then the energy will, on average, be this.
I am fully aware that you don't have a regression theory, but since you like to use the word 'theory' for my ideas and predictions, I thought that you would like me to use it.
You've missed the point here, I view your approach as more theory based (as opposed to mine being observation based) and so call your approach theoretical, in contrast to my observation based approach. I do appreciate you start with the manufacturer's data, but what you don't know is how much of that is evidence based, and how much comes from models - maybe some of the table entries are modelled, who knows. In that case we have possibly have models within models, quite possibly all the way down. Who knows? Furthermore, although you start with the tabled data, the rest of you model is theory based, making use of various theories of how the heat/energy moves within and without the system.
The problem that I see with your prediction method, is that it still requires the same quantity of thermal energy during the 18 hour period after the setback, whether the setback takes place or not.
This still suggests to me you have not grasped what I am trying to do with this chart. I am only trying to test the predictive ability of our respective approaches. The only question is how close is the predicted value to the actual value for each given point in time. What I see is the regression based method gets closer, apart from during the recovery period, which is to be expected, because the regression method deliberately excludes the setback periods, because I want predicted values for when a setback isn't being used, as needed for the entirely separate longer time frame comparisons of 24 hour observed vs expected energy in analysis. In the current chart, I am not interested in longer time frames, just what happens at each time point: how close is the predicted value to the actual value?
It is in the other entirely separate observed vs expected analysis that I sum the observed values and the predicted values over each 24 hour period, and then compare them, to determine what difference there is, if any.
We also diverge on the next two paragraphs, because you are still discussing longer periods, whereas I in the predicted vs actual plots am only considering each corresponding point in time, with no need to sum over periods in time. That is not what these charts are for.
Now, having said all that, let's perhaps somewhat confusingly, actually use the chart (and its underlying data) to do the other analysis: what happens over 24 hours, comparing observed values with expected values. LibreOffice failed to get the dates on the x axis, and I didn't bother to correct that, because the dates/hours didn't matter on that chart, all that matters is the data point line up, which they do. Let us consider the first 24 period with a setback, and lets make it noon to noon. This is what it looks like (I have kept it as a line chart as the more correct bar chart is harder to assimilate, LibreOffice still won't label the x axis correctly), with the 24 hour values underneath:
This suggests a 6.06 kWh saving over the 24 hour period, or 22.91% saving over the expected (predicted) value. Clearly this is just one isolated 24 hour period, result seems optimistic etc etc etc, I post it just to show the method, not as long times series covering many days. In practice, for longer times series, the chart isn't needed, the 24 hour sum data only needs a pivot table to collect it.
Please explain how you know that the spreadsheet predictions are low, and that it is not that your predictions are high?
Obviously it is more convenient from the setback case if the spreadsheet is low rather than your predictions being high.
Hopefully the above explanation makes clear why this question doesn't actually apply. The test analysis (individual point in time value comparison) shows the regression predictions are closest to the actual values, and that is all we are concerned with: which prediction gets closest to the actual values, and then is it close enough? I think it is, and back that up with the R squared value and the residuals analysis etc as to why it is reasonable to use it.
The first is that the OAT sensor reading is affected by the operation of the heat pump by up to it would appear 5C. This causes problems within the spreadsheet and I would assume with your regression calculation, since the predicted heat loss from the home would be much higher than the true value.
The second problem is the frequent cycling that occurs, which again makes obtaining reasonably accurate hourly values more difficult to achieve.
From calculations I have performed it would appear that the V x I calculation, without the 1.18 correction, is quite a close match to the power input values shown within the manufacturers data tables. This therefore posses the question, what additional load could the external power meter be measuring in addition to the heat pump? The V x I input power will be multiplied by the COP value to produce thermal energy output, but the 0.18 of the 1.18 may only providing a COP of 1 if inside the thermal envelope, or worse still may be lost to the outside air.
The point about the OAT being out is that it doesn't matter for my purposes: the regression is for the energy in as predicted by the OAT as measured. The 'real' OAT isn't known, and isn't used: the heat pump, and the regression only know about the OAT as measured. As it happens, I disagree (a matter of opinion not fact) about what the real OAT is, I think it is usually only a degree or two out, but that is neither here nor there for this analysis, which only uses the OAT as measured. Before you start telling me I am thick and need an even simpler explanation, let me tell you I know most of the house responds to the real OAT, heat loss is determined by the real OAT etc etc, but it doesn't matter, because the regression and the heat pump only know about the OAT as measured by the heat pump.
The cycling isn't frequent, typically when not in defrost territory is is about once an hour, or just under. That does not count as frequent cycling.
Maybe the Midea wired controller makes up the V and I values, based on its hidden internal versions of the manufacturers data tables. Who knows? Diesel-gate happened, why shouldn't compressor-gate happen? All I know is that if I take those V and I values, and multiply them by 1.18, then I get the best approximation to the values shown on the external kWh meter, which shows the actual energy in, the actual energy consumed, and what I pay my bills on.
I have emailed various Midea support email addresses, including FHP, asking specifically about how and where the V and I values are measured, and what they actually measure. I will report back if I get any useful replies.
Could you please provide a copy of the data that was used to produce the chart above.
If your heat pump uses more electrical energy in your predicted none setback scenario, why does it not produce more thermal energy output and cause the IAT to increase?
Again, this is the wrong question. The short answer is that it doesn't cause the IAT to rise because the 'predicted none setback scenario' never actually happens. If something never happens, it can't have any effect.
The slightly longer answer/explanation is this:
Running in setback, the heat pump does use less energy, and the does IAT fall (and then recovers using a boost).
We believe the recovery boost uses more energy than would be used in steady state running.
To estimate the net saving in energy use per 24 hour period, we need:
(a) actual observed energy used (easy, because it is the actual use, we have been recording it and just need to sum it for the 24 hours)
(b) what the energy use would have been without the setback. Since this didn't actually happen (we had the setback), we have to predict what the expected energy use would be without the setback
I do this using the established relationship (established by regression, drawing a chart and then fitting a line to the data and deriving the equation for that line) between the energy used and the OAT as measured by the heat pump. It does not matter whether this is the 'true' OAT or not, all that matters is whether the OAT as measured by the heat pump is a good predictor of the energy used. It turns out it is, R squared is 93%. R squared is a measure of how well the the predictor variable (OAT as measured by the heap pump) predicts the dependent variable (energy used). I then calculate the hourly expected energy use for the given OAT as measured by the heat pump were there no setback for that hour, and then sum those values over 24 hours to get what I would have used had there been no setback.
There is then the technical stuff about normal distributions, and whether the predicted values (which are means for that OAT as measured) are useful and valid values. I won't go into the detail here, as I have covered it more than once before, suffice it to say that, when summed over periods of time (days, weeks), I believe it is 'good enough' ie not perfect but close enough to be useful. Whether something is 'good enough' is a judgement call. There is no absolute permanent fixed line between what it and isn't 'good enough'.
I then simply calculate the saving as the difference between the expected energy use over 24 hours without a setback and the observed energy use over the same 24 hours with the setback in place. In the recent example I gave, the expected use was 26.46 kWh, the observed use was 20.40 kWh, giving a saving of 6.06 kWh.
Now, the key thing in relation to you question is this: in the setback scenario, which happens, the actual IAT does fall. The predicted no setback scenario never happens, but if it did happen (in a parallel universe?), then the IAT wouldn't rise, instead it would just stay the same, just as it normally does when the heat pump runs continuously.
All that said, all the usual caveats apply (one period in one house with its particular heating system, data collected processed by a mad scientist who doesn't know what he is talking about, forever and ever, world without end).
Midea 14kW (for now...) ASHP heating both building and DHW
Could you please provide a copy of the data that was used to produce the chart above.
I've already done that, it's in the December data I posted yesterday. You will need to do the regression, using only non-setback days, but that's no bad thing, it will verify (or not) my calculations.
Midea 14kW (for now...) ASHP heating both building and DHW
@jamespa Could you explain a bit more about fancoils as I haven’t been able to find out much about them? Do they work and do they fit any size of radiator? If they do, they might be one answer to the cycling at low output problem.
Fancoils are simply fan assisted radiators. The actual radiator bit is usually constructed a bit differently, you can add a bunch of computer fans to an existing radiator to get a very crude implementation of the same principle and at least one manufacturer does it exactly this way. Radiators operate largely by natural convection (not radiation!) and this limits the rate at which heat can be transferred from radiator to air. By forcing air over them more heat is transferred (just like the fan in your car, computer etc).
They wont help with cycling so far as I can work out, that occurs because the minimum output of the heat pump exceeds the total load from the house. What they do help with is the situation where a radiator of the required size at the design flow temperature is too large physically to fit the space available. They also have an interesting property (referred to above) that the output is not fixed at any given flow temperature, because the speed of the fan can vary. For heat pump applications this is of particular interest (to be honest I'm not sure we fully understand all of the features that could be extracted from this property because...
...Since they make a noise they aren't particularly popular in the UK, but some other countries use them a lot. They are quite expensive, but small for any given output.
I wonder if its worth turning the question round and asking instead, what are the circumstances where setback will use more energy?
This is absolutely the right question to ask, and then to follow up - so what should your should your strategy be for raising and lowering the set temp in your home?
The most likely ones, again purely from a physics point of view are (I think)
recovery is too aggressive (reducing COP because its necessary to raise FT higher to get enough out of the emitters)
recover happens at a time when its colder than during the setback period (ditto)
recovery doesn't happen fully, so the human (or machine) responds by whacking up the temperature or WC curve
the house has a very high thermal mass so acts as an integrator. Effectively setback doesn't occur (in the sense that the house doesnt cool) so the same amount of energy must be delivered in a shorter time, only possible by increasing FT (ie #3)
others that I cant yet think of
We can learn a few things about the right strategy here, for example, not to recover too aggressively. From 1: Rather than having a fixed setback of say 18 degC, and then bumping it straight back up to 21 degC for daytime, why not raise it in increments over a longer period? From 2: Why not have a negative setback when it's colder, and a positive setback when it's warmer? i.e. during the afternoon, at the highest outdoor temp, run your house a bit warmer. From 3: use PID control as well.
I think this is the natural future of heat pump controls: instead of maximising COP, they should minimise running kWh. And then you can get into minimising costs and/or CO2 beyond that, by looking at when electricity is cheapest/cleanest.
Thinking about installing a heat pump but unsure where to start? Already have one but it’s not performing as expected? Or are you locked in a frustrating dispute with an installer or manufacturer? We’re here to help.
