Advanced treatment of industrial estate effluent using ozone-membrane processes based on optimized random forest model

Abstract

This study aimed to survey ozone-membrane processes for advanced treatment of secondary effluent in the industrial estate wastewater treatment plant. The performance of a pilot-scale system combining chemical treatment (coagulation, flocculation, sedimentation), sand filtration, ultrafiltration, activated carbon, and reverse osmosis were evaluated without and with ozonation at three different points of the set up. Ozone at the reaction time of 30 min and concentration of 24 ppm was injected before chemical treatment, before UF and before RO. UV254, COD, TOC, TSS, TDS, pH, turbidity and ozone efficiency were evaluated. For estimating the value of UV254, a random forest (RF) model was presented. The model uses supervised learning to estimate UV254 from an input vector consisting of step, location pH, temperature, electrical conductivity, and turbidity. The residue of organic matters, as UV254, that enter the membrane systems (UF and RO) were 0.125, 0.080, and 0.198 cm(-1), respectively, which is desirable compared to the ozone-free state which is equal to 0.251 cm(-1). Ozonation before UF, showed the lowest value (0.080 cm(-1)). TOC concentrations in the step of ozonation before UF were 12.1, 11.2, 10.8, 7.7, and 0.75 mg/l after SF, ozonation, UF, AC, and RO, respectively. Estimation of UV254 using RFBO method represents MAE, MSE, and RMSE indexes equal to 0.01632, 0.00044191, and 0.021022, respectively. The results showed better efficiency of ozone injection before UF, compared to the other two cases. In the UV254 estimation section, the model shows enhanced accuracy in terms of error indexes in comparison with SVM-BO and GPR-BO methods.