Study Of Effluent Quality Prediction By ANN Model In WWTP

Improving the operation and management of WWTPs is critical for protecting water environment.The continuous improvement of effluent standards has put forward higher requirements for the on-line detection and optimization control of wastewater treatment plants,but currently difficult to all operating variables with hardware sensor on-line detection,there are costly to run or may lead to data leakage problem.Thus,the use of the existing easily measured variable,can make real-time estimation of prediction model is established and forecast the water quality parameter such as total nitrogen,and other important variables and to guide the sewage treatment process control and fault prediction,can optimize the management of sewage treatment process.Recent advancements in artificial intelligence modeling suggested a modeling method of wastewater treatment process.Therefore,this study carries out the modeling the activate sludge process using artificial neural network（ANN）to contribute the maintainment and management for a WWTP.The main purpose of this paper is to study the modeling method of activae sludge process by using ANN.The main contributions of this paper are that dynamic ANN model is adopted to predict the effluent quality in a WWTP.For forecasting the effluent chemical oxygen demand(COD_cr)and total nitrogen（TN）,this paper adopted the multilayer perceptron network（MLP）and nonlinear autoregressive network with exogenous input（NARX）to build a single static model and a dynamic model,respectively.At the same time,a hybrid model was developed by using principal component analysis（PCA）and the performances between the static and dynamic model,the single and hybrid model were compared.Moreover,this study compared and analized the influence of various model parameters such as train algorithms on the performance of prediction.The simulation results show that a static model and a dynamic one had RMSE of 7.2 mg/L and 2.9 mg/L,respectively,when forecasting the effluent COD_cr,and 1.8 mg/L and 0.8 mg/L,respectively,when forecasting the effluent TN.Results revealed that a hybrid dynamic model has higher accuracy of prediction than static model,and a hybrid model is superior to a single model in all cases.Furthermore,PCA remains efficiency in dimension reduction and model simplification and plays a good role in improving the model performance.Overally,we found that a dynamic hybrid model can enough model the complicated activated sludge process belonging the extreme nonlinearity and hysterysis and it is significant in the study for successful operation and management such as full data-drive and diagnosis and prediction in WWTP.