Improving ADM1 predictions via Bayesian analysis for continuous anaerobic digestion

Area of Science:

• Environmental Engineering
• Biochemical Engineering
• Computational Modeling

Background:

• Anaerobic Digestion Model No.1 (ADM1) calibration is challenging with limited long-term data.
• Accurate model calibration is crucial for optimizing anaerobic digestion processes.

Purpose of the Study:

• To develop a Bayesian inference framework for ADM1 calibration using only initial digester performance data.
• To enable reliable model predictions and risk-informed design in data-scarce settings.

Main Methods:

• Developed a custom Python implementation integrating global sensitivity analysis, Bayesian calibration, and parameter identifiability.
• Utilized Random Balance Designs-Fourier Amplitude Sensitivity Test (RBD-FAST) for parameter refinement.
• Employed informative priors derived from existing ADM1 studies.

Main Results:

• Calibrated ADM1 with less than two hydraulic retention times of data.
• Achieved accurate predictions for pH (1.10% error) and total chemical oxygen demand (5.38% error).
• Captured biogas production trends within the 95% credible interval for 63.14% of observations.

Conclusions:

• The Bayesian framework provides reliable ADM1 calibration and prediction using limited early-stage data.
• Informative priors significantly improved predictive accuracy compared to uniform priors.
• The approach supports improved safety, sustainability, and optimization in anaerobic digestion operations.