Automatic model learning: how Optimaatti learns your home’s thermal behaviour

Optimaatti automatically learns how your home heats up and cools down. You don’t need to do anything — the system handles everything in the background. You can enable heating control right away, and the model improves on its own over days and weeks.

How does automatic learning work?

Optimaatti uses smart background algorithms that monitor temperature, outdoor weather and heating behaviour. When enough data has accumulated, the system identifies the thermal model automatically.

Fully automatic: the system periodically evaluates whether enough data is available and runs identification automatically when the time is right.
Smart scheduling: if normal usage data is not enough, the system plans a short identification test (ON/OFF cycle) automatically during a quiet period, such as at night.
MPC works from day one: the model does not need to be perfect from the start. Control always keeps the temperature within your set limits — only electricity price optimization is not yet at its best initially.
Continuous improvement: the model is re-identified periodically and keeps improving over hundreds of days.

Can I run an identification test manually?

In most cases you don’t need to — automatic learning handles everything.
You can start a manual identification test if you want to speed up learning for a new device.
Manual testing is useful after changing sensor placement, heating power, or zone structure.
If RolloutRMSE is much larger than RMSE, a manual test can improve things faster.
For floor heating and high thermal mass installations, shorter manual cycles can help 2R2C model identification.

What happens before the model is ready?

This is a common concern, but there is no need to worry. Optimaatti is designed to work safely even without a perfect model.

Control works immediately: the system uses safe default parameters that always keep the temperature within your min/max limits.
The only trade-off: electricity price optimization is not at its best during the first few days. In practice the difference is small.
Regular re-optimization: MPC optimization runs every few hours. Each run uses the latest weather forecast, electricity price and model state.
No risk: even if the model is initially inaccurate, temperature limits guarantee your home never gets too cold or too hot.
Model improves on its own: typically 3–7 days of normal usage data is enough for the first automatic identification.

Manual identification test (step by step)

If you want to speed up learning manually, the identification test is in MPC settings under the Model Quality tab. During the test the device cycles ON/OFF to produce a clear temperature response.

Open MPC settings Select the device and go to the Model Quality tab.
Start identification test Click "Run identification test". The system starts controlled ON/OFF cycling.
Monitor Make sure the temperature evolves sensibly and does not exceed your limits.
Stop if needed You can stop the test at any time.

What happens after the identification test?

Everything happens automatically. Once the test completes, the system runs model identification, checks quality, and enables MPC if the model is good enough.

Automatic identification The system identifies the thermal model automatically from the test data — you don’t need to do anything.
Quality evaluation If the model is not good enough, the system automatically plans a new test with improved parameters.
MPC activation When model quality exceeds the threshold, MPC optimization is activated automatically.

Safety and design principles

Automatic identification tests are designed to be safe: they always respect your comfort schedule temperature limits.
The system automatically chooses a quiet time (e.g. night) when small temperature swings are least noticeable.
When MPC is already active, the system uses "soft" identification: it steers temperature across a slightly wider range without a separate ON/OFF test.
Best data comes with minimal disturbances (doors, sun, sauna, fireplace, large internal loads).
You can check identification status on the Model Quality tab at any time.

Frequently asked questions

Do I need to do anything before MPC starts working?

No. Just set your temperature limits (min/max) and enable control. The system learns the rest automatically. Control works safely from day one.

Why doesn’t electricity price affect control right away?

Because price optimization requires accurate knowledge of how fast your home heats up and cools down. During the first few days, control focuses on keeping the temperature within your comfort zone. The model typically matures in 3–7 days.

How do I know when the model is good enough?

The Model Quality tab shows the confidence score and prediction accuracy (RMSE). The system also notifies you when the model is ready for full optimization.

What about old houses with high thermal mass?

The system automatically detects slow thermal dynamics. For high thermal mass buildings, learning may take a bit longer (1–2 weeks), but control works safely throughout.

Enable MPC control and let the system handle the rest. Return to the MPC guide to see all settings.

Back to MPC help