Predicting and interpreting nitrogen removal performance and functional microbial abundance of single-stage partial...

Area of Science:

• Environmental Microbiology
• Wastewater Treatment Engineering
• Computational Biology

Background:

• Single-stage partial nitrification and anammox (PNA) is an efficient nitrogen removal process.
• Controlling microbial communities and operational parameters is crucial for PNA stability and performance.
• Predictive modeling can optimize PNA systems by identifying key influencing factors.

Purpose of the Study:

• To develop machine learning models for predicting nitrogen removal rate (NRR) and functional microbial abundance in PNA systems.
• To identify key factors and their optimal ranges affecting PNA performance using SHAP and causal inference.
• To provide guidance for optimizing PNA technology applications.

Main Methods:

• Machine learning algorithms, including Artificial Neural Network (ANN) and Extreme Gradient Boosting (XGBoost), were employed.
• Shapley Additive Explanations (SHAP) and causal inference were utilized for factor analysis.
• Model performance was evaluated using R-squared values for NRR and microbial abundance prediction.

Main Results:

• ANN and XGBoost models demonstrated strong predictive capabilities for NRR (R² = 0.94) and microbial abundance (R² ≥ 0.57).
• Free ammonia (FA) and pH were identified as critical factors influencing NRR.
• Optimal conditions were proposed: FA > 5 mg/L and O₂ < 0.4 mg/L to inhibit nitrite oxidizing bacteria (NOB).

Conclusions:

• Specific microbial sludge types (Candidatus Brocadia or Candidatus Kuenenia) are recommended based on nitrogen loading, temperature, and pH stability.
• The developed models offer valuable insights for the practical application and optimization of PNA technology.
• This study highlights the potential of ML in advancing wastewater treatment processes.