A novel stochastic wastewater quality modeling based on fuzzy techniques

Measurement and prediction of wastewater quality parameters are crucial for evaluating the risk to the receiving waters. This study presents new methods for the identification of outlier data and smoothing as an effective pre-processing technique prito to modelling. This new data processing method uses a combination of the autoregressive integrated moving average (ARIMA) model and -the adaptive neuro fuzzy inference system with fuzzy C-means clustering (FCM) (ANFIS-FCM). These new pre-processing methodsare compared to previously employed non-linear approaches for modelling of wastewater influent/effluent 5-day biochemical oxygen demand (BOD₅), chemical oxygen demand (COD) and total suspended solids (TSS). Linear modelling of each parameter, 242 linear models, were investigated, and a linear model for each parameter was selected. The results of the non-linear models led to an acceptable prediction for qualitative parameters so that the high coefficient of determination (R ) was observed for the influent and effluent BOD and TSS, respectively. The range of the R for all models was recorded as 0.8-0.87 and 0.83-0.89, respectively. By a combination of the linear and non-linear mothods a hybrid model was introduced. The proposed hybrid model for the influent BOD with the highest correlation between the observed and predicted values, and limited scattering was identified as the optimal model (R² = 0.95). The use of hybrid models to predict wastewater quality parameters improved the performance and efficiency of the models. In addition, a comparison of the hybrid model with the recently developed models in the literature indicates that the developed ARIMA-ANFIS-FCM outperformed other models.