Simulation of anaerobic digestion process under variable feeding sludge using a hybrid machine learning model

Area of Science:

• Environmental Engineering
• Biotechnology
• Artificial Intelligence

Background:

• Anaerobic digestion (AD) is crucial for stabilizing wastewater sludge.
• Accurate simulation of AD performance is essential for optimizing treatment processes.
• Advancements in artificial intelligence offer new possibilities for modeling complex biological systems.

Purpose of the Study:

• To investigate a hybrid approach combining machine learning (ML) and Bayesian model averaging for simulating anaerobic digestion (AD) performance.
• To evaluate the effectiveness of five ML techniques (DT, RF, GB, SVR, MLP) in predicting biogas production, volatile solids removal, and total solids output.
• To enhance prediction accuracy for AD systems using laboratory-scale data.

Main Methods:

• Employed five machine learning techniques: decision tree (DT), random forest (RF), gradient boosting (GB), support vector regression (SVR), and multilayer perceptron (MLP).
• Used laboratory-scale data on input sludge conditions to predict biogas production, volatile solids removal, and total solids output.
• Applied Bayesian model averaging to combine ML model results for improved accuracy.

Main Results:

• Individual ML models showed varying performance, with the best model for biogas output achieving R² of 0.90.
• The lowest performance was observed for total solids output prediction (R² of 0.53).
• The hybrid approach significantly improved prediction accuracy, increasing R² to 0.98 for total solids output and reducing errors.

Conclusions:

• Machine learning models can effectively simulate key performance indicators of anaerobic digestion systems.
• A hybrid approach integrating Bayesian model averaging with ML models substantially enhances prediction accuracy for AD processes.
• This study demonstrates the potential of AI-driven hybrid models for optimizing wastewater sludge stabilization.