Cost-effective assessment of effluent COD and total nitrogen using routine online monitoring data: a hybrid deep learning strategy.

Reliable assessment of effluent chemical oxygen demand (COD) and total nitrogen (TN) is essential for regulatory compliance and operational optimization in wastewater treatment plants (WWTPs). However, conventional laboratory-based assays are costly, labor-intensive, and often delayed, limiting their utility for real-time management. To overcome these economic and temporal constraints, this study proposes a cost-effective assessment strategy that leverages routine online monitoring data-specifically flow rate, ammonia nitrogen, total phosphorus, and pH-as accessible surrogates to estimate complex water quality parameters. A novel hybrid deep learning framework, integrating Long Short-Term Memory (LSTM) networks with Extreme Gradient Boosting (XGBoost), was developed using long-term time-series data from a full-scale WWTP. Data quality was ensured through wavelet denoising and outlier treatment. The proposed strategy demonstrated superior performance on the test set, achieving coefficients of determination (R2) of 0.913 for COD and 0.923 for TN, significantly outperforming standalone Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) baselines. These results validate that the hybrid strategy effectively captures the nonlinear dynamics of the treatment process using only standard online sensors. Consequently, this approach serves as a reliable soft-sensing tool, reducing reliance on frequent laboratory testing and enabling high-frequency, cost-efficient effluent quality assessment.