Alternative "weather compensation" control ideas

[dpmiller]

@ProDave here's some blurb from Honeywell-

https://youlearn.honeywellhome.com/uploads/documents/FAQ_-_Optimisation.pdf

[TerryE]

1 hour ago, SteamyTea said:

Still be interesting to run a Chi Square ...

 

Luckily I still remember enough of my stats modules 48 years on from my undergrad days at Cambridge -- which more than can be said for my daughter who did maths with stats at Oxford 30 years later. ??  So what you propose as the null hypothesis?  A constant set-point?  This isn't our goal.  The reason for the ripple was a policy decision trading off the temperature tramline to save on overall heating costs: We cut our heating bills by about 30% by doing our heating using E7 off-peak vs single rate tariff.  Yes having 22.7 ± 0.5 °C wastes under a kWh compared to a tight 22.4 °C, say, but we don't really notice the ripple; I do notice the savings on my electricity bill.

 

This is fundamentally a control system problem.  I look at the temperature plots over the winter months and the system is remarkably stable: essentially the same temperature throughout the house 24×7.  We love it.

Edited February 20, 2022 by TerryE

[SteamyTea]

1 hour ago, TerryE said:

So what you propose as the null hypothesis?  A constant set-point?

Was thinking more along the lines of deviation from binned prediction ranges.

So something like this with made up numbers:

 

Δ Temp Range	Predicted Energy Input	Actual Energy Input
0	0	0.0
1	0	0.0
2	0	0.0
3	0	0.0
4	0	0.0
5	0.1	0.1
6	0.2	0.1
7	0.4	0.1
8	0.5	0.1
9	0.6	0.6
10	0.7	0.8
11	0.8	0.7
12	1.0	1.1
13	1.1	0.8
14	1.2	0.6
15	1.3	1.3
16	1.5	0.1
17	1.6	0.2
18	1.7	1.5
19	1.8	0.2
20	2.0	0.8

 

Giving a chart like this

 

Which give a Chi Square result of 0.985, when tested on values greater than 0.

As it is greater then 0.05, then the null hypothesis is not rejected.  Though that does not mean the hypothesis is correct of course.

[TerryE]

I did a least squares fit over the first years heating days to get my first order functions.  But remember this is a control algo with feed forward and feedback terms, so if it does put 5 kWh too little in for one day, say, then the the average house temp drops by 0.1°C and the 5 kWh is added in to compensate on the next day. Hence (visitors and offsite visits aside) the daily average temperature rarely drifts by about a  0.1°C 1-sigma.  The only reason that I haven't put in the visit compensation term is that it just isn't worth the hassle.

[SteamyTea]

32 minutes ago, TerryE said:

reason that I haven't put in the visit compensation term is that it just isn't worth the hassle.

Yes.

And it could be testing the MET Office prediction as much as anything else.

I am going to ponder the whole problem more as I think there is merit in a simple self learning algorithms that can put the correct amount of energy into a building.

This may become more useful with variable ToU tariffs as you can bank prior to a known expensive time, or take a bit of a hit until a cheap time come.

[Radian]

38 minutes ago, SteamyTea said:

I am going to ponder the whole problem more as I think there is merit in a simple self learning algorithms that can put the correct amount of energy into a building.

 

 

Which is why I'm going through this right now: https://www.futurelearn.com/courses/introduction-to-machine-learning

[SteamyTea]

10 minutes ago, Radian said:

 

Which is why I'm going through this right now: https://www.futurelearn.com/courses/introduction-to-machine-learning

I have just downloaded, for free, Python Machine Learning.

Personally I don't think it needs to be too difficult. Really just a simple feedback loop based on basic thermodynamics, which is similar to an RC circuit in electronics in this instance.

[Marvin]

I hope see what you started @ProDave!

[SteamyTea]

33 minutes ago, Marvin said:

I hope see what you started @ProDave!

I think is is a great topic.

Chance to meld mathematical modelling and process control.

[Dan F]

So i've been thinking about this too.  I'm very interested in @TerryE's ideas, but am currenlty running things with the ASHP controller's weather compenastion to see how it behaves:

 

This is what I've observed:

- The Vaillant heat curve of 0.2 tells ASHP controller to use a flow temperature of 25C with the current external temperatur.

- The power output by the UFH is a function of, not only the flow temperature, but also the pump speed though.  Given the ASHP controller doesn't know anything about the pump speed, the heat curve required needs to be re-adjusted if the pump speed is changed.

- Vaillant has various weather-compensation modes. The one I'm using considers internal temperature, but doesn't turn the heating off when a set-point is reached, but rather keeps running with a lower flow temperature.

- In practice what happens is that the UFH pump runs continously with a mixer valve being adjusted by controller to maintain 25C UFH flow. The ASHP itself powers up every hour or so for 10-15mins and raises buffer temperature to around 28C, and then powers down.

 

So whats the issues with this?  Well, if you look at the supposed efficency of the ASHP for A7W35 it's 4.8.  But, if I look at data from 6am to 6pm today, for example the actual COP is just 2.9.  So it looks like leaving the system on all day and allow it to adjust temperature isn't particulalry efficient and, given UFH is in a passive slab, it would be more efficient to allow ASHP to run at a steady state for longer periods of time and then turn off the heating circuit.  This requires either continuing to use weather compensation, but using the mode that turns off the heating when setpoint is reached, or moving to a differnt, probably externally controlled, heating strategy.

 

 

EDIT: The energy figures on the right are kWh/2min.  So ouput climbs to 4kW for 2-3min and then starts dropping off.

 

Edited February 20, 2022 by Dan F

[Radian]

3 hours ago, SteamyTea said:

I have just downloaded, for free, Python Machine Learning.

Personally I don't think it needs to be too difficult. Really just a simple feedback loop based on basic thermodynamics, which is similar to an RC circuit in electronics in this instance.

 As you were discussing with Terry earlier, I'm also trying to use regression to give me the optimum heat input for the required room temperature but I'm definitely feeling the strain on the brain trying to piece it all together with the training data I have at hand. Good fun though.

[dpmiller]

I'm kinda surprised that @Dan F's unit doesn't have a variable-speed pump, are the big-name ASHPs not generally equipped with this? Our Cool Energy does, and it's well used as part of the modulation strategy.

[Dan F]

1 hour ago, dpmiller said:

I'm kinda surprised that @Dan F's unit doesn't have a variable-speed pump, are the big-name ASHPs not generally equipped with this? Our Cool Energy does, and it's well used as part of the modulation strategy.

Are you referring to ASHP pump or UFH pump?  The ASHP pump is variable-speed although the primary modulation approach is compressor speed.  I was referring to UFH pump whose speed directly impact kW delivered.

[dpmiller]

@Dan F I was referring to the ASHP pump.