Prediction of permeate flux and ionic compounds rejection of sugar beet press water nanofiltration using artificial neural networks

Artificial neural network (ANN) models were used to predict the permeate flux and rejection of ionic compounds (Na+, K+, Ca2+, Mg2+, SO4 2-, Cl-) of sugar beet press water through polyamide nanofiltration membrane. Experimental data was obtained at different transmembrane pressures (10, 15 and 20 bar), temperatures (25, 40 and 55°C) and feed concentrations (1-3 °Bx). The effect of the number of training points, the number of hidden neurons (H), type of transfer function and learning rule on the accuracy of prediction were studied. According to the results obtained for the best ANNs, 15% of the data was used to generate the model for the prediction of flux, and cross validation was performed with 40% of the total data. Independent flux predictions were also determined for the remaining 45% of the data. While for the prediction of the rejection of ionic compounds, 50%, 25% and 25% of the total data was used to learn the network, cross validation and testing ANN model, respectively. The modeling results showed that the overall agreement between ANN predictions and experimental data was excellent for both permeate flux and rejections (r = 0.998 and r = 0.974, respectively). Furthermore, sensitivity analysis indicated that temperature and Brix have the most effect on the prediction of flux and rejections (except for Ca rejection) by ANN, respectively.