Skip to content

PyADM1ODE Estimate

Python License: MIT Tests

State estimation for agricultural biogas plants based on the ADM1da mechanistic model (Schlattmann 2011, agricultural extension of ADM1), implemented in PyADM1ODE.

What this repository does

In live operation, the state of a real plant is estimated from incoming sensor data and a calibrated simulation model. The estimated states feed controllers that issue feeding recommendations to the plant operator.

Data flow:

Historical measurements ──► PyADM1ODE_calibration ──► calibrated model
                                                            │
Live sensors ──────────────────────────────────────► PyADM1ODE_estimate
                                                            │
                                                            ▼
                                                  State estimate x_hat
                                                            │
                                                            ▼
                                                     Controller / UI

Key features

  • 🎯 Mechanistic and data-driven — UKF on the full 41-state ADM1da plus
    planned deep-learning and fusion layers.
  • 🔌 Sensor-tolerant — channel gating for sparse lab measurements,
    OU drift for augmented input estimation.
  • 📊 Diagnostic — per-step NIS monitoring, per-channel innovation,
    Joseph-form covariance update for long runs.

Three estimation approaches

Approach Status Work package
UKF — mechanistic production AP 4.2
Deep Learning Ensemble skeleton AP 4.3
Fusion (Covariance Intersection) placeholder AP 4.4

Table of contents

  • Getting started — quick onboarding.
  • Installation — setup for different environments.
  • Usage — configure the UKF, load the calibration artifact.
  • Observability — which states can be estimated from which sensors?
  • Theory — ADM1da model, PINN concept.
  • Examples — end-to-end examples (twin, real plant).
  • API reference — auto-generated via mkdocstrings.

Quickstart

from pyadm1ode_estimation.estimation import (
    ADM1ProcessModel,
    InputSpec,
    ObservationChannel,
    ObservationModel,
    adm1da_full_spec,
)
from pyadm1ode_estimation.estimation.filters import UnscentedKalmanFilter
from pyadm1ode_estimation.estimation.observation_model import (
    extract_q_gas_total, extract_q_ch4_total,
)
from pyadm1ode_estimation.example_plants import build_multi_stage_plant

# 1. Build the plant (example or your own PyADM1ODE builder)
plant = build_multi_stage_plant()

# 2. Full 41-state spec with substrate augmentation
spec = adm1da_full_spec(
    digester_id="primary",
    substrate_inputs=[
        InputSpec("maize_silage",   substrate_index=0, initial_flow=4.74),
        InputSpec("solid_manure",   substrate_index=1, initial_flow=13.70),
        InputSpec("chicken_litter", substrate_index=2, initial_flow=1.09),
        InputSpec("slurry",         substrate_index=3, initial_flow=3.68),
        InputSpec("cereal_grain",   substrate_index=4, initial_flow=0.20),
    ],
)

# 3. Process + observation models
process = ADM1ProcessModel(plant, spec)
obs = ObservationModel(channels=[
    ObservationChannel("Q_gas", extract_q_gas_total, noise_std=10.0),
    ObservationChannel("Q_ch4", extract_q_ch4_total, noise_std=5.0),
])

# 4. Set up the SR-UKF and run it
ukf = UnscentedKalmanFilter(process, obs, spec)
for t, y in measurements:
    ukf.predict(dt=1/24)
    step = ukf.update(y, t=t)

More details: UKF in practice and twin experiments.

  • PyADM1ODE — base package with the ADM1da model.
  • PyADM1ODE_calibration —
    parameter calibration against historical data, writes the calibration artifact.