Non-linear ensemble modeling for multi-step ahead prediction of treated cod in wastewater treatment plant

Abstract

The paper proposes the application of data-driven models, including wavelet neural network (WNN) and multilayer perceptron (MLP), for multi-step ahead modeling of treated chemical oxygen demand (CODTreated ) using neuro-sensitivity input variables selection approach. Afterward, two non-linear ensemble techniques were applied to increase the prediction performance of the single models. Daily measure data obtained from new Nicosia wastewater treatment are used in this study, the performance efficiency of the models was determined in terms of Nash–Sutcliffe efficiency (NSE) and root mean squared error (RMSE). The obtained results of single models showed that WNN increased the performance accuracy up to 7% and 8% over MLP in both calibration and verification. The results also revealed the reliability of non-linear ensemble models in multi-step ahead prediction of CODTreated, hence, ensemble modeling could efficiently improve the performance of WNN and MLP models.