Enhanced nitrogen prediction and mechanistic process analysis in high-salinity wastewater treatment using...

Area of Science:

• Environmental Engineering
• Wastewater Treatment Technologies
• Machine Learning Applications

Background:

• High-salinity wastewater poses challenges for conventional biological treatment processes.
• Membrane bioreactors (MBRs) are effective but require optimization for saline conditions.
• Predictive modeling is crucial for understanding and controlling nitrogen removal in MBRs.

Purpose of the Study:

• To develop an interpretable machine learning framework for predicting nitrogen removal in MBRs treating high-salinity wastewater.
• To link predictive accuracy with operational decision-making for saline MBR systems.
• To elucidate the impact of salinity and operational parameters on nitrogen removal mechanisms.

Main Methods:

• Integration of Categorical Boosting (CatBoost) for predictive modeling.
• Application of Shapley Additive Explanations (SHAP) for model interpretability.
• Analysis of key parameters including salinity, dissolved oxygen, pH, chemical oxygen demand, temperature, and carbon-to-nitrogen ratio.

Main Results:

• CatBoost model achieved high accuracy: R² of 0.88 for effluent ammonia nitrogen (NH₄⁺-Nout) and R² of 0.91 for effluent total nitrogen (TNout).
• SHAP analysis revealed salinity inhibits nitrifying enzymes and disrupts carbon metabolism.
• Dissolved oxygen, pH, and COD removal efficiency were identified as key regulators; temperature and C:N ratio modulated nitrogen dynamics.

Conclusions:

• The SHAP-CatBoost framework provides accurate predictions and interpretable insights for high-salinity MBRs.
• Understanding salinity's impact is critical for optimizing nitrogen removal processes.
• The model facilitates predictive modeling and mechanical process control in challenging wastewater treatment scenarios.