Document Type : Research Article
Department of Water Resources Engineering, Faculty of Civil Engineering, University of Tabriz, Tabriz, Iran
In this study, black box artificial intelligence models (AI) including feed-forward neural network (FFNN), support vector regression (SVR), and adaptive neuro-fuzzy inference system (ANFIS) were used to predict effluent biological oxygen demand (BODeff) and chemical oxygen demand (CODeff) of Tabriz wastewater treatment plant (WWTP) using the daily data collected from 2016 to 2018. In addition, the autoregressive integrated moving average (ARIMA) linear model was used to predict BODeff and CODeff parameters to compare the linear and non-linear models' abilities in complex processes prediction. To improve the prediction of BODeff and CODeff parameters, the data post-processing ensemble method was also used. The input data set included daily influent BOD, COD, total suspended solids (TSS), pH at the current time (t), and BODeff and CODeff at the previous time (t-1) and the output data included BODeff and CODeff at t. The results of the single models indicated that the SVR model provides better results than the other single models. In ensemble modeling, simple and weighted linear averaging, and neural network ensemble methods were applied to enhance the performance of the single AI models. The results indicated that using ensemble models could increase the prediction accuracy up to 15% at the verification phase.