Modeling and optimization of turbidity removal from produced water using response surface methodology and artificial neural network

• Published in: South African journal of chemical engineering 2021-01, Vol.35, p.78-88
• Main Authors: Ezemagu, I.G., Ejimofor, M.I., Menkiti, M.C., Nwobi-Okoye, C.C.
• Format: Article
• Language: English
• Subjects: ANN; Box–Behnken design; Coagulation; Modeling and optimization; RSM; Turbidity removal
• ISSN: 1026-9185
• DOI: 10.1016/j.sajce.2020.11.007

•Design and optimization of turbidity removal using response surface (RSM) methodology via box-behnken design (BBD) and artificial neural network (ANN)-genetic algorithm (GA) was studied.•The RSM and ANN regression coefficients obtained showed good agreement with the experimental data.•Optimization of turbidity removal using BBD and GA increased overall process efficiency. In this study, results of parametric effects and optimization of turbidity removal from produced water using response surface methodology (RSM) and artificial neural network (ANN) based on a statistically designed experimentation via the Box–Behnken design (BBD) are reported. A three-level, three-factor BBD was employed using dosage (x1), time (x2) and temperature (x3) as process variables. A quadratic polynomial model was obtained to predict turbidity removal efficiency. The RSM model predicted an optimal turbidity removal efficiency of 83% at conditions of x1 (1 g/L), x2 (16.5 min) and x3 (45 °C) and validated experimentally as 82.73% with low model lack of fit F value of 0.6 and CV value of 8.22%. The ANN model predicted optimal turbidity removal of 83.01% at conditions of x1 (1 g/L), x2 (16.5 min) and x3 (45 °C) and validated as 82.98%. Both models showed to be effective in describing the parametric effect of the considered operating variables on the turbidity removal from produced water. However, the ANN described the parametric effect more accurately when compared with the RSM model, with a smaller PRE (percentage relative error) and AAD (absolute average deviation) of ±0.0241% and ±0.0139%, respectively.