Application of Machine Learning for FOS/TAC Soft Sensing in Bio-Electrochemical Anaerobic Digestion

Area of Science:

• Environmental Science
• Biotechnology
• Machine Learning

Background:

• Microbial electrolysis cell anaerobic digestion (MEC-AD) systems are crucial for wastewater treatment and energy recovery.
• Monitoring system stability, particularly the FOS/TAC ratio, is vital for efficient operation.
• Real-time prediction of key performance indicators can optimize MEC-AD processes.

Purpose of the Study:

• To evaluate the efficacy of various machine learning (ML) models for the real-time prediction of the FOS/TAC ratio in MEC-AD systems.
• To identify the most effective ML model for soft sensing of system stability.
• To assess the potential of ML in improving the operational efficiency and stability of bio-electrochemical systems (BES).

Main Methods:

• A 160-day trial was conducted using brewery wastewater in an MEC-AD system.
• Data collected over the trial period was used to train and test multiple ML models, including decision trees, XGBoost, support vector regression (SVR), support vector machine (SVM), and artificial neural networks (ANNs).
• An out-of-fold ensemble approach was employed to validate the performance of the selected model across the entire dataset.

Main Results:

• Artificial neural networks (ANNs) exhibited superior performance compared to other models, achieving an explained variance of 0.77.
• The ANNs effectively performed soft sensing of the FOS/TAC ratio, indicating system stability.
• The out-of-fold ensemble evaluation confirmed the robustness of the ANN model.

Conclusions:

• Machine learning, particularly ANNs, is highly effective for real-time FOS/TAC ratio prediction in MEC-AD systems.
• ML enhances the operational efficiency and stability of bio-electrochemical systems (BES), contributing to cost-effective environmental management.
• Implementing ML aids in maintaining microbial community health for biogas production and mitigating risks of system instability.