Stochastic modeling applications for the prediction of COD removal efficiency of UASB reactors treating diluted real cotton textile wastewater

Abstract

A three-layer Artificial Neural Network (ANN) model (9:12:1) for the prediction of Chemical Oxygen Demand Removal Efficiency (CODRE) of Upflow Anaerobic Sludge Blanket (UASB) reactors treating real cotton textile wastewater diluted with domestic wastewater was presented. To validate the proposed method, an experimental study was carried out in three lab-scale UASB reactors to investigate the treatment efficiency on total COD reduction. The reactors were operated for 80 days at mesophilic conditions (36-37.5C) in a temperature-controlled water bath with two hydraulic retention times (HRT) of 4.5 and 9.0 days and with organic loading rates (OLR) between 0.072 and 0.602 kg COD/m3p/day. Five different dilution ratios of 15, 30, 40, 45 and 60% with domestic wastewater were employed to represent seasonal fluctuations, respectively. The study was undertaken in a pH range of 6.20-8.06 and an alkalinity range of 1,350-1,855 mg/l CaCO3. The concentrations of volatile fatty acids (VFA) and total suspended solids (TSS) were observed between 420 and 720 mg/l CH3COOH and 68-338 mg/l, respectively. In the study, a wide range of influent COD concentrations (CODi) between 651 and 4,044 mg/l in feeding was carried out. CODRE of UASB reactors being output parameter of the conducted anaerobic treatment was estimated by nine input parameters such as HRT, pH, CODi concentration, operating temperature, alkalinity, VFA concentration, dilution ratio (DR), OLR, and TSS concentration. After backpropagation (BP) training combined with principal component analysis (PCA), the ANN model predicted CODRE values based on experimental data and all the predictions were proven to be satisfactory with a correlation coefficient of about 0.8245. In the ANN study, the Levenberg-Marquardt Algorithm (LMA) was found as the best of 11 BP algorithms. In addition to determination of the optimal ANN structure, a linear-nonlinear study was also employed to investigate the effects of input variables on CODRE values in this study. Both ANN outputs and linear-nonlinear study results were compared and advantages and further developments were evaluated.