4. Validate your model

It is now time to validate your model. For performing such a task, Dymoval uses ValidationSession class objects.

A ValidationSession class object is created by using a Dataset instance as basis. Once instantiated, you can append as many simulated data as you want to the same ValidationSession class object. The validation metrics are automatically computed for each simulated dataset against the common stored Dataset.

The model validation process with Dymoval.

Dymoval validates your models in terms of

• R-square fit

• Residuals auto-correlation norm

• Input-Residuals cross-correlation norm

You can visually inspect both the simulations results with the plot_simulations() method and the residuals with the plot_residuals() method.

The coverage region can be shown through the plot_coverage() of the stored Dataset.

4.1. How to interpret the results?

As a rule of thumbs, your model is as good as the r-squared index is high and the residuals correlation norms are small (possibly less than 1).

Futhermore

• High values of the residuals auto-correlation -> your disturbance-to-output model needs improvement,

• High values of the input-residuals cross-correlation -> your input-to-output model needs improvement.

For more information on how to interpret r-squared fit and residuals, feel free to search the web or to read some good System Identification textbook.

At this point, you may consider to use Dymoval for performing unit-tests on your models and create building pipelines if you are using a CI/CD environment for developing models.