Modeling of the pervaporation process for isobutanol purification from aqueous solution using intelligent systems

Abstract

ABSTRACTIn this study, intelligent systems (ANN-GA and GMDH) was employed to develope a model based on experimental data to predict the performance of the pervaporation process. The ANN system was coupled with the genetic algorithm (GA) to choose initial connection weights and biases of the multi-layer feed forward neural network (FFNN). The input parameters were the feed concentration, membrane thickness, and Reynolds number, while the outputs were total flux and permeate concentration. The RMSE of the estimated total flux for the ANN-GA and GMDH were 0.09170 and 0.0903, respectively. Also, the RMSE of estimated permeate concentration for the ANN-GA and GMDH were 0.0994 and 0.0975, respectively. The results indicated that the models had sufficient accuracy, but that GMDH could provide a better outcome. Finally, the relative importance of input variables on the network outputs was determined. Sensitive analysis showed that the membrane thickness and feed concentration are the most effect on the total flux and permeate concentration, respectively. Other variables also have important effect on the PV process and cannot be ignored.