Related Experiment Video
Updated: Nov 24, 2025
Author Spotlight: Advancing Anaerobic Microbiota Research Using a Novel Respirometry Protocol
Published on: April 26, 2024
Performance prediction of ZVI-based anaerobic digestion reactor using machine learning algorithms
Weichao Xu1, Fei Long2, He Zhao3
1Department of Biological and Ecological Engineering, Oregon State University, Corvallis, OR 97333, United States; School of Chemical and Environmental Engineering, China University of Mining and Technology (Beijing), Beijing 100083, PR China; Beijing Engineering Research Center of Process Pollution Control, National Key Laboratory of Biochemical Engineering, Institute of Process Engineering, Innovation Academy for Green Manufacture, Chinese Academy of Sciences, Beijing 100190, PR China.
Machine learning models accurately predict methane production in zero-valent iron enhanced anaerobic digestion (AD) systems. XGBoost initially excelled, but deep learning showed superior performance with more data, identifying key factors like feedstock solids and ZVI dosage.
Area of Science:
- Environmental Engineering
- Biotechnology
- Computational Science
Background:
- Zero-valent iron (ZVI) enhances anaerobic digestion (AD) for improved methane production and stability.
- Accurate modeling of ZVI-based AD reactors is crucial for optimization and industrial design.
- Current modeling approaches can be time-consuming and laborious.
Purpose of the Study:
- To evaluate the feasibility of machine learning (ML) algorithms for predicting the performance of ZVI-based AD reactors.
- To identify dominant operating parameters influencing methane production in these systems.
- To compare the predictive accuracy of Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Deep Learning (DL) models.
Main Methods:
- Collected operating parameters from 9 published articles on ZVI-based AD reactors.
- Trained and evaluated RF, XGBoost, and DL algorithms to predict total methane production.
- Utilized feature importance analysis from XGBoost to identify key influencing parameters.
- Incorporated digestion time and expanded the dataset to predict cumulative methane production.
Main Results:
- XGBoost achieved the highest initial accuracy in predicting total methane production (RMSE: 21.09).
- Key parameters identified by XGBoost include feedstock total solids (TSf), soluble chemical oxygen demand (sCOD), ZVI dosage, and particle size.
- With expanded data, Deep Learning outperformed RF and XGBoost, showing the lowest RMSEs for both control (11.83) and ZVI-added (5.82) reactors.
Conclusions:
- ML algorithms show significant potential for modeling ZVI-based AD reactors.
- XGBoost effectively identifies critical parameters affecting methane yield.
- Deep Learning demonstrates superior predictive capability with larger datasets, offering a robust tool for optimizing AD processes.
