Modeling batch and column phosphate removal by hydrated ferric oxide-based nanocomposite using response surface methodology and artificial neural network

Batch and column phosphate removal was conducted by a commercially available nano-hydrated ferric oxide composite HFO-201 under varying conditions, and the performance was modeled and predicted with the aid of artificial neural network (ANN) model and response surface methodology (RSM). Initial pH, sulfate concentration, operating temperature, and adsorbent dosage were chosen as four variables for the batch study, while the removal efficiency was considered as the output. A central composite design (CCD) was referred to design 33 sets of batch experiments, and a RSM model was developed to compare with the ANN model. The three-layer feed-forward back-propagation network was established in MATLAB to estimate the phosphate removal efficiency. The positive behavior of both models was verified by Pearson and Spearman coefficient and mean squared error (MSE). Analysis of variance (ANOVA) tests and sensitivity analysis were performed on the models to find relative influence of four variables. Temperature was deemed as the least influential whereas the other three variables were considered significant to the output. Genetic Algorithm (GA) was employed to find optimum dosages for a desired removal efficiency under given conditions. ANN modeling was further attempted to estimate the breakthrough curves of fixed-bed adsorption, where pH, sulfate, temperature, flow rate (BV/h) and bed volume was considered as variables. Predictions made by the developed models were in reasonably good agreement with the test runs. This study suggested that ANN and RSM be considered as effective tools to model and predict trace pollutants removal by nanocomposite adsorbents. (C) 2014 Elsevier B.V. All rights reserved.