Prediction, Uncertainty Quantification, and ANN-Assisted Operation of Anaerobic Digestion Guided by Entropy Using...

Area of Science:

• Biotechnology and biochemical engineering
• Machine learning applications in environmental science
• Process optimization and control

Background:

• Anaerobic digestion (AD) is prone to instability due to feedstock variability and biological fluctuations.
• Existing methods struggle to manage the nonlinear and disturbance-sensitive nature of AD processes.
• Predictive modeling and uncertainty quantification are crucial for enhancing AD operational stability.

Purpose of the Study:

• To develop an entropy-guided machine learning framework for AD parameter prediction and uncertainty quantification.
• To evaluate the performance of Support Vector Machine (SVM), Random Forest (RF), and Artificial Neural Network (ANN) models for AD process prediction.
• To assess the impact of the developed framework on operational stability, economic viability, and environmental footprint.

Main Methods:

• Collected six months of industrial AD data (~10,000 samples) for model training and validation.
• Compared SVM, RF, and ANN models for predicting biogas yield, fermentation temperature, and volatile fatty acid (VFA) concentration.
• Utilized entropy-based uncertainty analysis and feature importance analysis to evaluate model performance and identify key influential variables.

Main Results:

• The ANN model demonstrated superior performance (accuracy = 96%, F1 = 0.95, RMSE = 1.2 m³/t) with the lowest prediction error entropy, indicating reduced uncertainty.
• Feature entropy and permutation analysis identified feed solids, organic matter, and feed rate as the most influential variables (>85% contribution).
• Real-time implementation of the ANN model reduced gas-yield fluctuation from ±18% to ±5%, improved operational stability by ~23%, and decreased process entropy.

Conclusions:

• Combining machine learning with entropy-based uncertainty analysis provides a robust approach for stable and low-carbon AD operation.
• The ANN model, guided by entropy, effectively enhances AD process stability, reduces uncertainty, and offers significant techno-economic and environmental benefits.
• The developed framework serves as a reliable decision-support tool for optimizing AD plant performance and sustainability.