temperature smoothing: ema implementation vs ha-average sensor

[maia]

I've compared the current EMA implementation of versatile_thermostat with the implementation being used in https://github.com/Limych/ha-average (installed via HACS) which creates sensors for average values. This implementation is better than the HA statistics sensor, as this will cause issues when the sensor does not report any new data for longer than the defined average time span (instead of assuming the temperature simply didn't change).

The average sensor is more exact and I wonder if this could help in improving versatile_thermostat too:

Here's the configuration I've been using:

sensor:
  - platform: average
    name: "bedroom smoothed temperature"
    unique_id: "bedroom_smoothed_temperature"
    entities:
      - sensor.bedroom_govee_temperature
    duration:
      minutes: 2
    precision: 2

[jmcollin78]

Hello @maia,

This is very interesting. Thank for this. I wonder how will the auto window open detection works with this average sensor in a "noisy" environment like you have. It is not so easy to integrate and replace the Vtherm Ema with an external sensor but if you think it could be interesting, I can help to implement this.

Clearly the VTherm ema is more smooth in your graph. Have you change the parameters or is it the default parameters ?

[maia]

I didn't yet consider replacing the ema with an external sensor, rather I thought it could be helpful to analyze the code used in ha-average, as it could be also used for vtherm.

The ema parameters are the default values.