Posted by: @derek-m

↑

The raw data consistently shows that some time after the heat pump stops, the OAT reading increases by several degrees, and when the heat pump restarts the OAT reading falls by several degrees, both happening over a short time period so would be unlikely to be a real change in ambient temperature. How would this affect your statistical approach?

I have no idea how to account for this other than the obvious observation that it is likely to compromise hp performance. 

I'm not sure I'd describe the statistical approach as 'mine', it's data from @cathoderay !  Personally I prefer to account for as many of the material known variables as possible before drawing scattergrams (which I don't much like but sometimes the real world isn't as tidy as I would ideally want). 

However I also have to acknowledge the good correlation between oat and energy in, which, as I refer to upthread, tells us something about the mode in which the system is operating.  For me the question has, as a result, altered from 'why was so much energy apparently saved' to 'why didn't the house cool more? with a side of 'was that much energy _really_ saved?'.  Perhaps your modelling will explain or at least give us further clues where to look.

Posted by: @derek-m

↑

We could do with some actual ambient temperature data to compare with the OAT reading provided by the heat pump.

Solar gain, even on cloudy days can be sufficient, to affect the heat loss from the building. 

Indeed.  ' South of M4' is probably good enough to download some data and @cathoderay tells us he has some independent local data in addition.

Posted by: @derek-m

↑

The raw data consistently shows that some time after the heat pump stops, the OAT reading increases by several degrees, and when the heat pump restarts the OAT reading falls by several degrees, both happening over a short time period so would be unlikely to be a real change in ambient temperature. How would this affect your statistical approach?

We could do with some actual ambient temperature data to compare with the OAT reading provided by the heat pump.

The 'several degrees' is simply not true, unless your version of several includes a few, ie one or two possibly very occasionally three degrees. See the chart below covering the last two days (all visible changes in OAT are one or two degrees), and the countless charts I have posted showing the minute data with changes in the OAT visible as the heat pump stops and starts. The only exceptions are the defrost cycles, where there is a very transient but slightly greater rise in the OAT, presumably because of the heat released during the defrost.

More generally, for the regression analysis and its intended use, we don't need the 'true' or 'real' ambient. The regression tells us what the predicted (expected) heating energy in will be for any given ambient as measured by the heat pump. If I then predict the expected energy in from the ambient as measured by the heat pump at any given time, I do nothing wrong. The fact there is another real but unmeasured and undone regression based on energy in and 'real' OAT is neither here nor there, because I am not trying to model that particular set of circumstances; instead, I am using the regression I do have as a tool, to predict what they energy in will be, using measurements I do have, the OAT as measured by the heat pump. 

I put 'real' in quotes because I am not sure what the 'real' OAT is. Is it on the north or the south side of the building? In the open or in the shade? It will always an arbitrary judgement call as to what the 'real' ambient temperature is. What matters is not the particular ambient chosen, but rather that a consistently measured ambient is used, be it the one on the north side, the south side, or as measured by the heat pump.

I have just done the obvious next step of combining the three control periods (each of which is a sample of the population of all OAP vs energy in points, and each of which will be subject to both random and very likely systemic error ie sampling or other bias) into a larger sample (which will still have random and very likely systemic error ie sampling or other bias, but hopefully less) and repeated the regression analysis on this data set, and got this:

Nothing remarkable, pretty much as expected, more data means better picture etc. The extraordinarily high R squared values remains (what it means, for those less used to statistics, is that 93% of the variation in energy in is explained by the OAT, only 7% of the variation is caused by other 'unknowns'). The residual plots aren't too bad (meaning no obvious/gross bias), though the right hand tails are interesting: is this the transition from a variable state to a steady standing load?. There is also some evidence of flat lining at the right hand end of the main plot: is this possibly also evidence of the real but hitherto unknown standing load? Although the trend line, influenced as it is by all the points, sits at about 0.3kWh (300W for an hour) at the right hand end, the 'flat line' points are at around 200kWh (200W for an hour), which to me seems a plausible value for a standing load (circulating pumps x 2, other small unknown additional loads, roosting vampires etc).

@jamespa - I'm struggling to understand what is troublesome in this analysis. All three control sets, covering varying external conditions, fall pretty much on the same line. The R squared value is impressive, a full 93% of the variation in energy in is explained by the OAT, leaving only 7% unexplained, and the residual plots give no cause for alarm. It is not meant to be a perfect model of the operation of all heat pumps everywhere, but it is a tool I can use to predict what my expected energy in will be, given a particular OAT, and that is all I ask it to do. As time goes by, I will be able to add more data points, and I imagine, but can't yet know, that it might (or might not) become an even tighter plot, I will just have to wait and see. Is there any reason why this analysis - regardless of confusions stemming from other analyses which don't make sense, for reasons that don't apply to this analysis - doesn't pass my 'good enough answer to the right question rather than an exact answer to the wrong question' test?    

Posted by: @jamespa

↑

scattergrams (which I don't much like but sometimes the real world isn't as tidy as I would ideally want). 

Very interesting! Perhaps I because I normally deal with medical/epidemiological data, which is invariably a very long way from tidy, I do like scatter plots very much, as a way to look for patterns and trends, and sometimes nonsense - for example, the scatter plots for energy out vs variations on OATs suggest to me there isn't a lot to be gained from pursuing that data, it is just simply too scattered - something, somewhere is wrong. When there is a pattern, a scatter plot plus equation and R squared is far more informative for me than just the equation and the R squared. For example, a scatter plot tells you very quickly whether extrapolation will take you down the rabbit hole into the realms of the absurd, something you can't easily see in the bare equation. Visual presentation of information is also an extremely effective way of communicating information to those not used to, or sometimes even averse to (lies, damned lies and) statistics. I almost always do a scatter plot before anything else when looking at the relationship between two variables, and it then remains close at hand during any further analysis.

What I am getting at here in a roundabout way is that while we both are used to working in disciplines that like to imagine they fall within the realms of science (though sometimes i wonder for mine), the particularities of our different disciplines are indeed very different. I for example end up being OK with messy data from observational studies, because that's what we have most of the time (RCTs being the obvious exception, and even they are far from perfect) in epidemiology and medicine, whereas perhaps - I only make a tentative suggestion - you are more comfortable with tidy data, tight models and controlled tests and experiments, because the nature of the work you do allows, even encourages, such an approach. In medicine, such an approach is well nigh impossible most of the time. We still don't really know, to give some recent examples that have been flogged to death and still don't habe an answer, whether masks prevent respiratory infections, or lockdowns control pandemics, because the necessary controlled experiments to get a definitive answer just are not possible in the real world. At the end of the day each one of us has nothing more or less than an opinion, based on our assessment of the messy uncontrolled data that rolls around in our analyses. That's why I have my good enough answer to the right question than an exact answer to the wrong question principle, it allows me to make the best sense I can of what I do have (and note the caveat that it has to be a good enough answer, not any old answer, if it is not good enough you don't use it), rather than get lost in futile pursuit of an exact answer to the wrong question. I also suspect this is influenced by the fact that illness and death are pressing creditors, and doctors often have to make decisions NOW, rather than have the leisure of doing more experiments and analysis. We have to decide, even in the face of uncertainty and imperfect evidence.     

By the way, neither is right or wrong, they are just accidents of circumstances. But maybe they can lead to unnecessary clashes, when our firm but often unstated way of making sense of the world somehow feels threatened.       

@cathoderay

I don't follow your logic. You expect to use the OAT reading to predict the Energy In to your heat pump, but this OAT reading would appear to be changed by the operation of the said heat pump. Three degrees may not seem much, but if the maximum ambient operating range of the heat pump is probably 20C, then a 3 degree variance is 15%. I don't know the normal ambient temperature variation on a daily basis, but I don't think it will be more than 10C on a given day, which would mean the inconsequential 3C, is now a 30% change.

If you are happy with the way your heat pump is operating, then please forgive my impertinence for trying to help you improve it.

Just out of interest, what is the statistical probability of your heat pump keeping you warm come the colder weather?

@jamespa

Attached is a copy of the test results from the raw data for the period 5th November 2023 to 20th November 2023.

I don't see any point in continuing the exercise with the present data.

If you need clarification on any point please feel free to ask.

Posted by: @derek-m

↑

I don't follow your logic. You expect to use the OAT reading to predict the Energy In to your heat pump, but this OAT reading would appear to be changed by the operation of the said heat pump. Three degrees may not seem much, but if the maximum ambient operating range of the heat pump is probably 20C, then a 3 degree variance is 15%. I don't know the normal ambient temperature variation on a daily basis, but I don't think it will be more than 10C on a given day, which would mean the inconsequential 3C, is now a 30% change.

If you are happy with the way your heat pump is operating, then please forgive my impertinence for trying to help you improve it.

Just out of interest, what is the statistical probability of your heat pump keeping you warm come the colder weather?

It's not my airy-fairy logic, it's the regression which shows the OAT as measured by the heat pump can predict the energy in. It doesn't (within reason, of course) matter where you measure the OAT, as long as you measure it consistently in the same place for both the regression analysis and then predictions made from the analysis. Lets say we have, from measurements and analysis (OATp is OAT as measured by the heat pump):

OATp     Energy in

7          1.75 

9          1.4 

11        1.1 

etc

We then find we can do the same for the 'real' OAT (OATr) which is let's say 3 degrees higher than the OATp. This time we get:  

OATp     Energy in

10          1.75 

12          1.4 

14         1.1 

etc

As long as we stay consistent with the measurement used (OATp or OATr) we will get the same result from predictions:

When the OATp is 7, the energy in will be 1.75

When the OATr is 10, the energy in will be 1.75

etc.

This is OK, because I am not trying to define a perfect model, instead, I am just using a variable I can measure easily to make predictions about another variable. It is 'god enough' using the OATp because, in short, it works, I don't need to have the harder to measure OATr.

Rolling my statistical dice, I estimate the probability of the heat pump failing to keep me fully warm come the colder weather to be around 100%. 

@cathoderay

But what happens if the OAT reading varies from 0C to 3C deviation with heat pump loading?

Posted by: @derek-m

↑

But what happens if the OAT reading varies from 0C to 3C deviation with heat pump loading?

Counter-intuitive perhaps, but it still doesn't matter. I suggest you are trying to get an exact answer to the wrong question, instead, what matters is can the OAT as measured by the heat pump accurately predict the energy in, and the answer, visible in the scatter plots and the regression statistics, is that it can, at an accuracy that is 'good enough'. We have a good enough answer to the right question. It might be interesting to know how it does that, but we don't have to know, the fact it can do so makes it good enough for the intended purpose, predicting the expected energy in for a given OATp.      

Posted by: @cathoderay

↑

Posted by: @derek-m

↑

But what happens if the OAT reading varies from 0C to 3C deviation with heat pump loading?

Counter-intuitive perhaps, but it still doesn't matter. I suggest you are trying to get an exact answer to the wrong question, instead, what matters is can the OAT as measured by the heat pump accurately predict the energy in, and the answer, visible in the scatter plots and the regression statistics, is that it can, at an accuracy that is 'good enough'. We have a good enough answer to the right question. It might be interesting to know how it does that, but we don't have to know, the fact it can do so makes it good enough for the intended purpose, predicting the expected energy in for a given OATp.      

 

You are perfectly correct, it does not matter. Well not to me, anyway.

Posted by: @cathoderay

↑

@jamespa - I'm struggling to understand what is troublesome in this analysis. All three control sets, covering varying external conditions, fall pretty much on the same line. The R squared value is impressive, a full 93% of the variation in energy in is explained by the OAT, leaving only 7% unexplained, and the residual plots give no cause for alarm. It is not meant to be a perfect model of the operation of all heat pumps everywhere, but it is a tool I can use to predict what my expected energy in will be, given a particular OAT, and that is all I ask it to do. As time goes by, I will be able to add more data points, and I imagine, but can't yet know, that it might (or might not) become an even tighter plot, I will just have to wait and see. Is there any reason why this analysis - regardless of confusions stemming from other analyses which don't make sense, for reasons that don't apply to this analysis - doesn't pass my 'good enough answer to the right question rather than an exact answer to the wrong question' test? 

There is nothing troublesome about the results or analysis per se if the only interest is in your system in the particular circumstances of the experiment.  However the title of this thread is not 'Do setbacks save energy without compromising comfort on the system owned by @cathoderay', it is 'Do setbacks save energy without compromising comfort?'.  

Given your frequently expressed concern about extrapolation, I would have thought that's its entirely obvious to you that extrapolating the results to any other system (or any other circumstances even on your system) is out of the question whilst we are unable to reconcile them with the physics.  Put another way, until they can be reconciled with the laws of thermodynamics,  they give us no help at all in answering the question that forms the title to the thread.

Perhaps in medicine you don't attempt to do this because the underlying engineering of the human body is currently too complex to understand.  In simpler disciplines, which are closer to physics, its an essential step in interpreting any experimental results, and certainly in any attempt to apply those other than in the circumstance in which they were measured.

Posted by: @derek-m

↑

@jamespa

Attached is a copy of the test results from the raw data for the period 5th November 2023 to 20th November 2023.

I don't see any point in continuing the exercise with the present data.

If you need clarification on any point please feel free to ask.

 

-- Attachment is not available --

Thanks

I've had a scan and think I possibly understand what they are saying, but can you summarise what conclusions you draw as you know better your spreadsheet and its features and limitations.  Also there are some rows without labels and what is 'the raw data' in this case?

Many thanks

Posted by: @jamespa

↑

Posted by: @derek-m

↑

@jamespa

Attached is a copy of the test results from the raw data for the period 5th November 2023 to 20th November 2023.

I don't see any point in continuing the exercise with the present data.

If you need clarification on any point please feel free to ask.

 

-- Attachment is not available --

Thanks

 

I've had a scan and think I possibly understand what they are saying, but can you summarise what conclusions you draw as you know better your spreadsheet and its features and limitations.  Also there are some rows without labels and what is 'the raw data' in this case?

 

Many thanks

 

Hi James.

The raw data is the hourly records provided by CathodeRay recently, I think that you have been working with a copy.

I took the data on face value, assuming that it had been correctly averaged from the minute data.

The two rows of numbers above each table are the IAT values for that particular day extracted from the raw data, and the offset value required to produce the correct IAT value in the spreadsheet, to match that from the raw data. I included these values with the tables so that I could easily replicate the results by entering these values back into the relevant cells.

I made several initial attempts, but then realised that the results where being compromised by the PHE effect, in that the spreadsheet was producing the wrong Energy Supply value from the heat emitters, because it was using a LWT value that would be higher than that in the real World, so I improved the spreadsheet by adding a PHE DT value in the calculation of Energy Supply. I set the PHE DT value to 5C, which I think is a common value used by PHE manufacturers. I also added a Other Heating Source (OHS) value to the Energy Supply calculation, this is to take account of the other electrical devices, and humans, that generate thermal energy within the building. I set the OHS at 8kWh of thermal energy for each day, spread evenly across all 24 hours.

I have arranged for both of these two parameters to be changed if necessary, or removed from the calculation by setting their value to zero.

The settings used in the Initial Data sheet were 11kW Heat Loss, 23kW Heat Emitter Capacity and 246kWh of Thermal Mass/Capacity.

For each day the results are contained in three tables, the lower one is from the raw data values for a Setback with Boost, this is used as the reference. The middle table is with a Setback but no boost with the requirement that the specified IAT value is obtained by 9pm. The upper table of the three is with the Setback removed, so 24 hour operation, again with the specified IAT being achieved by 9pm.

I was too busy producing the results so did not look at them in close detail. 

As expected, the Setback without boost always produces an energy saving when compared to the reference Setback with Boost. The highest energy saving predicted being 2.64kWh on the night of 8th to 9th of November. Of course the extra energy saving being achieved by having lower IAT's.

Running without Setback was a bit of a mixed bag, with a prediction of an energy saving of 1.84kWh by running continuously on the night of 6th to 7th November, but then an energy saving of 2.1kWh by adopting a Setback on the night of the 7th to 8th November. So no clearly defined results.

As expected, in all cases continuous running put more thermal energy into the building, with as much as 12.5kWh on the night of 10th to 11th of November. Whilst not increasing the final IAT measured at 9pm, it is used to keep the IAT more even throughout the 24 hour period, sometimes with a slight increase in Energy In, but also sometimes with a slight reduction.

One thing that I did note is that the spreadsheet consistently predicted higher COP values than those in the raw data. There could be several possible reasons, but I would not care to speculate as to the reason why.