Effluents quality prediction by using nonlinear dynamic block-oriented models : A system identification approach

The dynamic and complex municipal wastewater treatment plant (MWWTP) process should be handled efficiently to safeguard the excellent quality of effluents characteristics. Most of the available mathematical models do not efficiently capture the MWWTP process, in such cases, the data-driven models are reliable and indispensable for effective modeling of effluents characteristics. In the present research, two nonlinear system identification (NSI) models namely; Hammerstein-Wiener model (HW) and nonlinear autoregressive with exogenous (NARX) neural network model, and a classical autoregressive (AR) model were proposed to predict the characteristics of the effluent of total suspended solids (TSSeff) and pHeff from Nicosia MWWTP in Cyprus. In order to attain the optimal models, two different combinations of input variables were cast through auto-correla-tion function and partial auto-correlation analysis. The prediction accuracy was evaluated using three statistical indicators the determination coefficient (DC), root mean square error (RMSE) and correlation coefficient (CC). The results of the appraisal indicated that the HW model outperformed NARX and AR models in predicting the pHeff, while the NARX model performed better than the HW and AR models for TSSeff prediction. It was evident that the accuracy of the HW increased averagely up to 18% with regards to the NARX model for pHeff . Likewise, the TSSeff performance increased averagely up to 25% with regards to the HW model. Also, in the validation phase, the HW model yielded DC, RMSE, and CC of 0.7355, 0.1071, and 0.8578 for pHeff, while the NARX model yielded 0.9804, 0.0049 and 0.9902 for TSSeff, respectively. For comparison with the traditional AR, the results showed that both HW and NARX models outperformed in (TSSeff) and pHeff prediction at the study location. Hence, the outcomes determined that the NSI model (i.e., HW and NARX) are reliable and resilient modeling tools that could be adopted for pHeff and TSSeff prediction.Validerad;2021;Nivå 2;2021-06-03 (alebob)