The "perfect climate control" - ndrwrbgs's version

[ndrwrbgs]

ndrwrbgs's climate vision

Forked-from #16

excerpt

As another "nice to have", would allowing you to have smooth transitions between nodes be helpful? We have underfloor heating and radiators so it's very slow to apply requested changes so having stepped control is fine, it looks like you have more frequent transitions though so could be useful to add interpolation between set points?

The transitions you saw in my chart were not goals themselves, but approximations and work-arounds for other goals. They only addressed the true concerns indirectly, and there may be better solutions than supporting the work arounds.

Below are the goals I have in my climate automation. I imagine much of what I want is niche, and that I'll be dropping into python and making virtual devices.

Environment

• Humidity/Temperature sensors
• Central System (Gas Furnace + Electric A/C) - Smart Thermostat
• Electric oil radiator - Smart Plug [spot heating]

Desires & concerns

Central System (Furnace + A/C)

Concerns:

🪟 Rooms have != quality of insulation

Therefore sensor.room_temp ≈ f(sensor.outside_temp, climate.thermostat.temperature) and not just the thermostat.

To work around this, I use a generic thermostat and do not try to set the Physical to a target temperature, but instead, process the heat "on/off" signals and "over-set" the thermostat. E.g. when HA sees that it is 69 and the target is 70, I set the Physical to 74.

🛏️ I "care" about different rooms at different times of the day

For instance, I am only in the bedroom at night.

I've worked around this by adjusting what I set the thermostat to, but note that due to sensor.room_temp ≈ f(sensor.outside_temp, climate.thermostat.temperature), this is an approximation that breaks down when it's -20C like tonight 🥶☃️

🦠 I want to avoid mold/control for dew-point

Mostly what I care about here is { window/surface temperature, dew point }. I am approximating it with just { temperature, humidity }. But as a long-term goal, I'd like to handle it better. My approximation usually works well enough, but we just got a cold snap and I woke up to condensation on some windows.

I'd have preferred a "system" which heated the place more slowly (to permeate surfaces better), or to a lower temperature, over having the condensation. If that were not possible, I am able to lower the humidity since that's controlled(/influenced) via Home Assistant as well.

👉 I want to control how _fast_ temperature changes [This is what you're seeing approximated with smoothed transitions]

I read online™️that structural damage to an old building is exacerbated by fast temperature changes. Even if that were not an issue, I've noticed my human perception is better at noticing "changes" than absolute values - so a sudden drop in temperature feels very cold, whereas with a gradual one I don't feel it until much lower.

📅 Desire for a typical "day"

1. Sleep: it "can be" cold.
   E.g. in winter, no need to heat so much. However, in summary I want it to cool down. I call this out as "can be" cold, because the temperature I allow it to drop to during winter, is not the same as the max temperature I'd want it to cool to during summer.
2. Waking up: I want it to be warmer
   Warmth helps me "get moving"
3. During the day: whatever (within reason)
   I wear layers, so as long as my hands aren't frozen (which always happens at this darn computer!) I don't much mind what the temperature is. If I'm sedentary at the computer though, then I want it warmer than otherwise.
4. Sunset (4:30 pm right now!): warmer
   During summer, I use the sun as a bed signal, but during winter that's far too early to go to sleep. I've noticed (and previously started the research into) sunset makes it suddenly "feel" colder. I want to counteract this at sunset, when sunset is too early.
5. Bed time: cooling off
   Decreasing temperature throughout the night (slowly, so it's not a sudden "brr"), helps me signal to my body that it's time to shut off and get into bed -- avoiding a "second wind"

⚡Short "on" times are wasteful on the furnace

The furnace has a "start up" period, which delays start up, and uses electricity without providing any heat. It also has a "cool down" period running the fan, to cool its components. This means that the shorter the "on" time, the less efficient the heating.

I use the hysteresis & min cycle to approximate this in home assistant.

Electric-based oil Radiator

🐌 It heats up slowly

This is relevant/noticed in charts when it "turns on" but the temperature continues to decrease for some time. Ideally, it'd be closer to a PID system, and therefore be able to turn on ahead of time.

I let the generic_thermostat hysteresis work around this

♨️ It continues to heat after shutting off

After a radiator turns off, it continues to heat the room. How much is not static, and depends on how much heat it has stored internally (approximated by how long it's been on for? But a bit more complicated than that, since being on for 2s doesn't mean it's 'cold' if the pattern was on:10h, off:2s, on:2s).

I let the generic_thermostat hysteresis work around this

Thanks

Thank you for the provocation to finally formalize and write up my thoughts!

I tried to use formatting above to differentiate the "importance" of different parts of the content and have "details" be less distracting.

[ndrwrbgs]

Now that I have a proper write-up, I can organize the goals. Many of these are in direct conflict with one another.

• Security/safety
  • Avoid frozen pipes
  • Avoid mold
  • Avoid thermal shock
  • Minimize thermal stress
• Cost
  • Minimize overall use of thermal system
  • Maximize intentional heat/$
  • Don't wear the furnace out
• Comfort
  • Use heat to wake (morning & sunset)
  • Use heat for sedentary activities
  • Use cool to sleep
  • Maintain comfortable humidity
  • Avoid heat/cold stress on houseplants
    (e.g. my tropical plants want min 60 at night, but/so they're in a better insulated room than the bedroom)

[ndrwrbgs]

Looping back to the question...

I think optional interpolation could provide a nice experience particularly when I make a manual/override change.

If I overrode the schedule for guests and did not want the temperature to immediately drop to the night temperature afterwards, I could use an external component to limit such passive temperature changes.

But active decisions which are made with more than the current state - e.g. pre-heating for the morning, require at a minimum some signal from whatever the scheduler that there is an upcoming change. Interpolation could provide a way to approximate this, but if there's another way I'd probably prefer to use the information directly.

I could imagine many users being perfectly satisfied with interpolation - my question would just be whether those same users would be just as happy without it.

[kneave]

Lots to consider, nothing impossible but just needs thinking through to get right. I've been thinking in similar terms though and would love to have the whole thing able to create it's own schedules based on forecasts and user preferences.

Keep writing down any thoughts, definitely good to have like minded folk using it to help me think these things through.

[kneave]

One thing I'm just adding to at least give us a head start on some of this are derivative sensors for all the entities controlled by the scheduler. This means we'll be able to get the cooling/heating rates for everything we're controlling. I'll need to add in the logic to combine these results with if we're in a heating or cooling state but if the data is there at least we have options.

[ndrwrbgs]

I can't wait to see how it looks!

I've been trying to do derivatives for my heart rate data and it's been a trade off between responsiveness and accuracy wrt. short term noise and larger changes. One issue I had was that if I have a sine wave cycling every 3 min, then a 6 min lookback suggested no change at all. Particularly a problem with Heart Rate since the periods are not consistent in length.

Perhaps climate data is less nefarious.

If it provides any help - the best luck I've had so far has been using filtering on the original series (moving average, EWMA, lowpass) and then processing the result of that but I know that this is a hack. What's that saying - "Months of coding can save you days in the library"!

[kneave]

By the way, all the sensors have been in there for a while now but aren't exposed in the UI. If you add a history card and search for "climate_schedule_" when looking for entities to add you should find all of them. There's also a coldest/warmest sensor that shows the warmest and coldest entities it's monitoring. It has the temperature and the name in different attributes which is handy.

[ndrwrbgs]

I see threads from other people and assume they are of more actionable content, so closing this guy to get it out of "the list"

[kneave]

Leave it open, all good. I can dip and and out to get ideas this way and if it's closed I'll forget its here

[ndrwrbgs]

The release notes mentioned added support for IHP; do you think that gets some of the smart desires I was looking for, or have you played with it?

(Context: sending before bed, so if it’s not an L1 cache hit in your processor, I can google it later 🤣)

[kneave]

I've not played with it at all and I'm not sure I'm providing data in the right format to be honest. If you have a go with it let me know how you get on.