Temperature slope jumps with stable temperature

[gregorko]

Hello,

something behaves wrong with my temperature slope and causes the window detection feature in auto mode to trigger frequently.
I noticed this in a room with what I would call fairly stable temperature. The temperature goes up/down 0.1°C frequently but stays in a small range mostly between 18.7°C and 18.8°C over hours. Still the temperature slope jumps wild between +5°C/hour and -7°/hour.
This triggers the window detection routine frequently.

Any ideas what I can do about it and what causes the slope to jump so much on stable temperature?

[caiusseverus]

This is because your temperature sensor only has 0.1ºC resolution. The actual temperature is somewhere between 18.7 and 18.8, so the thermometer is jumping between the two values. It's this rapid oscillation that is causing the slope to be unreliable. If you check the sensor entity options, you can try to increase the precision to 2 decimal places but this won't help if the underlying sensor doesn't report values with more than one decimal place.

You can mitigate this by using a filter helper to smooth the temperature reading out. Use the moving average (timed based) and start off with it set for a few minutes. If it's not smooth enough, then increase the time slightly.

[jmcollin78]

Hello @gregorko ,

I can confirm what @caiusseverus said. The oscillation of the room temperature is the root cause of your issue.
The temperature is smoothed with an ema (exponential mobile average) function but it seems this is not enough for you.
You can try the filter helper.

What is your temperature sensor brand ? If the filter don't work, you will have to change. There is good temperature sensor here: https://github.com/jmcollin78/versatile_thermostat/blob/main/documentation/en/presentation.md#thermometers

Sonoff SNZB are very good for this !

[gregorko]

Ok... although I personally think that 0.1ºC oscillation should not be an issue, specially when looking at the documentation saying:

This slope is smoothed and filtered to avoid aberrant thermometer values that could interfere with the measurement.

Anyway it's fine for me to apply the filter. With the filter there is still some oscillation but now it's less frequent and with higher precision, so only 0.01ºC range now. Still I'm seeing sudden temperature slope drop to -8ºC. Personally it seems to me the slope calculation is much too sensitive.
It might maybe be related to the temperature drop from 18.80ºC to 18.65ºC around 01:55, about 10 minutes earlier, but this is a drop of 0.15ºC in around 5 minutes, so it's still only around 2ºC in an hour. When considering the earlier drops (01:32, 01:42) it's even less.

[jmcollin78]

0.1 in 10 sec means 36°C/hour.
This is not the 0.1 which is important but the how much divided by how long.

At the very start of this feature (two years ago) I try to deal with such cases but If i filter more, then the dectection just don't work anymore. So it was difficult to tune for everyone.

I think you should display the ema temperature which is used to slope calculation and not the root temperature. See how it behaves.
You have it as an entity (at home: sensor.clim_salon_ema_temperature

[jmcollin78]

When looking at the code, it seems you can try something if you are able to change the VTherm's code.
In the custom_integration/versatile_thermostat/ema.py file, change the MIN_TIME_DECAY_SEC at the top of the file to something like: MIN_TIME_DECAY_SEC = 20. Then too near temperature values will be ignored.

But the drawback is that, you can miss the last temperature value and this could have impact on window detection.

You don't tell me what you do with this slope ? Do you do some open window detection with it ?

[jmcollin78]

@KipK because you use also slope for auto TPI, this discussion can be interesting.