BT not heating to target temp

[SebboGit]

So I just set up BT yesterday evening, I selected Offset Based and AI Time Based and I am well aware the adaption might take some days to fully work. However, I am wondering if it really is this "bad" at the beginning.

7am: external temp_sensor is at 20.2 degree
BT was set to 19 degree during the night, now set to 23 degree, radiator is getting warm

8am: an hour later the external temp_sensor shows 20.6 degree, so it took a full hour to increase the temp by 0.4 degree
BT still set to 23 degree heating

8:30am: the external temp_sensor now has been dropping since 8am and reached 19.9 degree

It is now roughly 10am and the current temp from the external sensor is 20.1 degree

So after 3 hours of "heating" I've lost a total of 0.1 degree and I am still 3 degrees off my target temp.
Radiator is only slightly warm when I would expect the valve to be a lot more open considering we are still way off target temp after 3 hours of heating. This is my 5m² bathroom by the way which did heat up in less than 20 minutes on schedule before I switched to BT.

[SebboGit]

Different room, so different values but:

Bedroom set to 20 degree heating, current temp from external sensor is 18 degree.

Local temperature in Zigbee2MQTT shows 25.1 this is clearly the internal temperature measured by the device, so given these values the valve stays at 0% and the room does not heat up. There is also nothing set as remote temperature here:

But the whole point of switching to BT for me is that I want the TRV to be based on the external temp sensor not the internal one and based on the first picture the current temp is correctly set to my external temp sensor. I don't see this reflected in the Zigbee2MQTT settings for the TRV though.

[cl445]

TL;DR: Yes, the first 1–2 days really do look bad with AI Time Based — it needs full heating cycles to learn before the math becomes useful. For now switch to Normal mode and observe; you can re-enable AI Time Based later once it has data.

---

Why (for the curious):

"AI Time Based" maps internally to CalibrationMode.HEATING_POWER_CALIBRATION. It computes a target valve position with:

valve_pos = 0.019 * (temp_diff / heating_power) ^ 0.946

heating_power (°C/min when heating is active) starts at the default 0.01 and is updated only after complete heating cycles (HEATING start → HEATING stop with peak-temperature tracking, see utils/thermal_learning.py::HeatingPowerTracker). Until BT has observed a few full cycles the value stays at the default and the formula misjudges the room.

There is also a special branch at calibration.py:723-734: when hvac_action != HEATING, the mode forces valve_percent=0 for TRVs that support direct valve control. If your TRV reports idle (its internal sensor reads target reached while the external sensor disagrees), BT actively pins the valve closed.

For your specific setup (Offset Based + AI Time Based on TP-Link KE100):

• KE100 has local_temperature_calibration but no direct valve control, so the fallback path at calibration.py:760-775 applies:
  new_calibration = current_calibration - (local_cal_min + cur_trv_temp) * valve_pos
• With external 20.2 °C and your target ~21–22 °C the temp_diff is small and the still-default heating_power makes valve_pos modest → the offset push is gentle and the TRV is convinced it's close enough.

Pragmatic options:

1. Switch to Normal (CalibrationMode.DEFAULT) for the first days — predictable, no learning required.
2. If you want the adaptive behavior, enable debug logging (custom_components.better_thermostat: debug), let BT run 1–2 days through several full heat cycles, then switch back to AI Time Based once heating_power has converged (visible in the BT entity attributes as heating_power_norm).
3. Hybrid mode combines offset and direct valve control where supported (not on KE100).