House Thermal Inertia and Roomstat Setback (some cautionary notes)

[HeatStig (Daikin Althema 3 8kW)]

Dead right about solar gains👍

I’m pondering how to have a temperature sensor on the north wall (already have) and another one on the sunny side, and use some external logic to integrate / compare the readings and tell the HP what to do in a more applicable way.
The north side only is way too blunt an instrument.

[SarahH (Sarah Hanson)]

I’m no control/electrical engineer but comparing the model in your link to @langestefan’s post #52 the three segments may simply be 1 boundaries (walls, floors etc), 2 furnishings/contents, 3 air, each of which can be modelled differently.
As a pragmatist I still prefer to lump everything together for simplicity (as I did in the original post) but if there’s mileage in a more sophisticated approach, then go ahead .

[langestefan (Stefan de Lange)]

@djh

More fundamentally I don’t understand why the model has three segments - why not one or two, or four or five? Do the segments correspond to something physical in the real world or are they arbitrary features created for some reason? (and how is the number 3 arrived at?)

This is a free decision. To get something practical and robust you typically want between 1-5 masses (capacitors). One mass, like @SarahH showed does not do well in most cases. But it is a great pedagogical tool.

In reality your house consists of many components, the more you know about how they store and transfer heat the better your predictions will be. You can’t really know what the masses and resistances are, unless you know every detail about your property. So you lump a bunch of components together and pretend it represents a single physical aspect of your house, represented by one mass and one resistor. So you could have distinct masses to represent the mean air temperature, your furniture, the envelope and walls, the emission system and maybe even the sensor itself. You can imagine that if you have underfloor heating for example that the value of the emission systems starts to become very critical in forecasting temperature.

Whether you need all of these depends on what your data looks like. But the general goal is to find the simplest possible model which explains the data. This is referred to as the bias-variance tradeoff and whole books have been written about it. I think Bacher, Madsen give a good overview of all the different model options.

@djh

I’ve previously experimented with a model by Simon Rouchier and published at Chapter 11 A simple RC model (Python) | Building energy statistical modelling and related pages. But I’ve had a lot of trouble getting reliable predictions from it. Hence looking to understand things in more depth.

That looks very interesting, thanks. I will definitely be trying out PySIP on my own data.

[djh (Dave Howorth)]

@HeatStig

Dead right about solar gains👍

I’m pondering how to have a temperature sensor on the north wall (already have) and another one on the sunny side, and use some external logic to integrate / compare the readings and tell the HP what to do in a more applicable way.
The north side only is way too blunt an instrument.

Yes, I’ve found it’s quite difficult to get a sensor somewhere where it doesn’t get any direct sun (because the readings shoot up then) but isn’t too enclosed and can get useful readings for the temperature of walls that do see sun. FWIW M.Rouchier thought my PV generation records were probably a reasonable approximation for the insolation component in his model.

[djh (Dave Howorth)]

@langestefan

So you lump a bunch of components together and pretend it represents a single physical aspect of your house, represented by one mass and one resistor. So you could have distinct masses to represent the mean air temperature, your furniture, the envelope and walls, the emission system and maybe even the sensor itself.

I can see the idea of general blobs representing notional things. I don’t understand why they are all in series though. The air doesn’t have to go through the furniture to get to the walls for example? My floor slab and walls see very different environments on the outside? (Lots of EPS followed by ground versus external air and sky radiative background)

@langestefan

Whether you need all of these depends on what your data looks like. But the general goal is to find the simplest possible model which explains the data. This is referred to as the bias-variance tradeoff and whole books have been written about it. I think Bacher, Madsen give a good overview of all the different model options.

Thanks for the link. I’ll study that paper.

[langestefan (Stefan de Lange)]

@djh

The air doesn’t have to go through the furniture to get to the walls for example? My floor slab and walls see very different environments on the outside? (Lots of EPS followed by ground versus external air and sky radiative background)

The masses are parallel and heat conduction happens through the thermal resistances. So lets say you have 3 masses connected by 2 resistors (so C1 - R12 - C2 - R23 - C3). You add heat to C1, which increases its temperature T1. Once T1 > T2, heat flows from C1 to C2 through R12. And again same story for C2 and C3. If T2 > T3 then heat will flow to C3. Conceptually this is how your radiators work as well. Your heat pump heats the radiators C1, which then release heat to the interior C2. And finally the interior heats your walls and envelope which is C3.

And it’s a linear system, so the equations are super nice and simple. We can use all the mathematical tools we have to study and control this system.

[SarahH (Sarah Hanson)]

Thanks as ever, @langestefan.

@langestefan

a great pedagogical tool

1. Riddle for you. Is having to look up the meaning of “pedagogical” itself somewhat pedagogical ?
2. This discussion of sophisticated modelling tools begs the question “How do I evaluate all the separate masses and specific heats (i.e. capacitances) and heat transfer coefficients (resistances) to set up my model in the first place?” Do I have to weigh my sofa? And how do I weigh a wall?

[langestefan (Stefan de Lange)]

@SarahH

This discussion of sophisticated modelling tools begs the question “How do I evaluate all the separate masses and specific heats (i.e. capacitances) and heat transfer coefficients (resistances) to set up my model in the first place?” Do I have to weigh my sofa? And how do I weigh a wall?

I wish that were possible, and people do make calculations like that for research, but if you want to build a commercial product you’ll have to make do with the data you have available, estimating the model parameters based on data is called model identification

Take for example homely. When it is first installed it has a two week ‘learning’ mode. They won’t tell you this but that is when they control your heating to create a good dataset to do the model identification with.