Calibration Tutorial

The short version of notebooks 01 to 03.

1. Load and clean the data

from pyadm1ode_calibration import MeasurementData

measurements = MeasurementData.from_csv("data.csv")
measurements.remove_outliers(method="zscore", threshold=3.0)
measurements.fill_gaps(method="interpolate", limit=3)

The index must be a timestamp, and the columns carry the channel names the objectives refer to (Q_gas, Q_ch4, pH, ...).

2. Pick the parameters

Fit few of them. Start with the hydrolysis rate of the dominant substrate fraction, since it is what most directly moves gas production. Configuration lists the usual candidates with their bounds.

3. Fit

from pyadm1ode_calibration import InitialCalibrator

calibrator = InitialCalibrator(plant)
result = calibrator.calibrate(
    measurements=measurements,
    parameters=["k_hyd_ch"],
    objectives=["Q_gas"],
    validation_split=0.2,
)

4. Check the result

result.success only says the optimizer terminated. What matters is whether the error on the held-out part dropped as well: compare result.validation_metrics against the training metrics. A large gap between the two means the fit followed noise, not the plant.