Predicting wastewater treatment plant quality parameters using a novel hybrid linear-nonlinear methodology

Abstract

Biochemical oxygen demand (BOD), chemical oxygen demand (COD), total dissolved solids (TDS) and total suspended solids (TSS) are the most commonly regulated wastewater effluent parameters. The measurement and prediction of these parameters are essential for assessing the performance and upgrade of wastewater treatment facilities. In this study, a new methodology, combining a linear stochastic model (ARIMA) and nonlinear outlier robust extreme learning machine technique (ORELM) with various preprocesses, is presented to model the quality parameters of effluent wastewater (ARIMA-ORELM). For each of the studied parameters, 144 different (144 × 8 models) linear models (ARIMA) are presented, with the superior model of each parameter being selected based on statistical indices. Moreover, 48 nonlinear models (ORELM) and 48 hybrid models (ARIMA-ORELM) were considered. The use of linear and nonlinear approaches to model the linear and nonlinear terms (respectively) of each time series in the hybrid model increased the efficiency and accuracy of the predictions for all of the time series. The influent wastewater nonlinear TSS model and the effluent COD and BOD models attained the best performance with a high correlation coefficient of 0.95. The use of hybrid models improved the prediction capability of all quality parameters with the best performance being achieved for the effluent BOD model (R2 = 0.99).