Examples — overview

Methodological end-to-end examples for using the UKF on an agricultural biogas plant. The examples show the methodology without plant-specific details — a concrete plant is documented separately and privately.

In the repo (scripts)

Planned

• Validation report — a summary of how the methods (UKF, later
  Deep Learning Ensemble, later Fusion) were tested against reference data. Contents: accuracy, robustness, consistency diagnostics (NIS), convergence behaviour.

Procedure for your own plant

1. Plant topology built via the pyadm1.configurator API, or imported
   from a private plant-builder repo.
2. Calibrated artifact loaded with load_artifact(...) and applied with
   apply_to_plant(...).
3. State-vector spec tailored — not all 41 ADM1 states are estimable
   with typical SCADA sensors; see Observability → Literature review.
4. Observation model wired to the available sensor channels.
5. UKF instantiated; in the online loop, alternate predict() and
   update() calls, and monitor NIS.