Prediction of effluent quality in a wastewater treatment plant by dynamic neural network modeling

Improving the operation, management, and consequent performance of wastewater treatment plants (WWTPs) for conserving the water environment is crucial. Recent advancements in artificial intelligence (AI) modeling have shown the potential to solve the non-linear simulation of processes in WWTPs and facilitate real-time operational adjustments. In this study, a dynamic nonlinear autoregressive network with an exogenous input (NARX) model was established for predicting effluent quality. The performance was optimized with different time-delay parameters and training algorithms. Then, a PCA-NARX hybrid model was established for high performance and comparison with two static artificial neural network (ANN) models. The BR algorithm exhibited the highest performance among the four training algorithms for the NARX model. The dynamic PCA-NARX model was significantly superior to static models in modeling effluent quality. The PCA-NARX model predicted the effluent chemical oxygen demand (CODcr) and total nitrogen (TN) with high accuracy (RMSECOD = 2.9 mg/L, RMSETN = 0.8 mg/L). Therefore, we propose a stable and sensitive dynamic neural network model for predicting effluent quality and potential real-time adjustment of wastewater treatment operations.