Predicting single-cell protein production from food-processing wastewater in sequencing batch reactors using ensemble...

Area of Science:

• Biotechnology and biochemical engineering
• Environmental science and engineering
• Machine learning applications in environmental systems

Background:

• Single-cell protein (SCP) production from food-processing wastewater offers sustainable resource recovery and animal feed.
• Accurate prediction of SCP system performance under variable conditions is crucial for operational efficiency but challenging due to complex bioreactions.

Purpose of the Study:

• To develop and evaluate ensemble learning algorithms for predicting the performance of a continuous-inflow, sequencing-batch-reactor-based SCP system.
• To identify key features influencing system performance for model optimization and operational decision-making.

Main Methods:

• Trained and tested ensemble learning algorithms: ensemble Support Vector Regression, ensemble Gaussian Process Regression (GPR), Random Forest, and Extreme Gradient Boosting.
• Utilized industrial soybean-processing wastewater as influent in a sequencing-batch reactor (SBR) system.
• Performed interpretable analysis to validate feature significance for model optimization.

Main Results:

• Ensemble learning models, particularly GPR-based ensembles, significantly outperformed linear regression in predicting effluent and biomass variables.
• GPR-based ensembles with influential features achieved a coefficient of determination (R²)=0.72 for biomass production prediction, demonstrating superior performance and reduced overfitting compared to linear regression (R²)=0.4.

Conclusions:

• Ensemble learning, especially GPR, provides a robust and accurate method for predicting SCP system performance under dynamic conditions.
• The developed models offer valuable insights for operational decision-making, enhancing the efficiency of resource recovery and wastewater treatment processes.