Effluent parameters prediction of a biological nutrient removal (BNR) process using different machine learning...

Area of Science:

• Environmental Engineering
• Water Treatment Technologies
• Machine Learning Applications

Background:

• Wastewater treatment plants (WWTPs) require precise operational control for effective nutrient removal.
• Predicting effluent parameters is crucial for monitoring environmental performance and regulatory compliance.
• Machine learning (ML) offers potential for enhancing the predictive capabilities in WWTP operations.

Purpose of the Study:

• To propose and evaluate a targeted blend of ML algorithms for controlling WWTP operations.
• To predict key effluent parameters in a biological nutrient removal (BNR) process.
• To compare the performance of six ML algorithms for effluent parameter prediction.

Main Methods:

• Collected two years of operational data from the Plajyolu WWTP in Kocaeli, Türkiye.
• Applied six ML algorithms: Support Vector Regression Machine (SVRM), Random Forest (RF), EXtreme Gradient Boosting (XGBoost), Light GBM, and Hybrid Regression.
• Evaluated algorithm performance using metrics including Mean Absolute Percentage Error (MAPE).

Main Results:

• SVRM with a linear kernel showed high accuracy for Chemical Oxygen Demand (COD) and BOD5 (MAPE ~9% and 0.9%).
• RF and XGBoost were optimal for Total Nitrogen (TN) and Total Phosphorus (TP) prediction (MAPE ~34% and 27%).
• RF, SVRM (linear and RBF kernels), and Hybrid Regression generally outperformed other algorithms across all parameters.

Conclusions:

• Selective application of ML algorithms can effectively predict different WWTP effluent parameters.
• This approach can enhance the efficiency of WWTP environmental performance monitoring.
• Wider implementation of ML-based prediction can reduce resource demands for active monitoring.