Study On The Application Of BP Artificial Neural Network For Ultrafiltration Membrane Flux Prediction

Membrane process as a new technology of separation, purification and concentration has some advantages such as low energy consumption, simple operation and little pollution to the environment. It has been became an advanced technology to deal with the problems of energy resource and environment protection. The membrane fouling is the serious issue to limit the application of membrane. The membrane flux prediction when the membrane is fouled becomes an urgent and important problem in membrane field.The membrane filtration of multiple substance from beginning to the end is a complex and unsteady process in the practical water treatment. The theoretical models hardly to describe this process. The statistical models based on analyses data appear as promising tools for developing simulation models. Among such tools, the model chosen in this work is the artificial neural networks (ANNs). This study tries to apply BP Artificial Neural Network to predict Ultrafiltration(UF) membrane flux. UF membranes (hydrophilic polyvinylidene fluoride(PVDF), Millipore, USA) with 100,000Da Molecular Weight Cutoff is used. In order to improve the precision, these data are recorded by a connected PC at a predetermined time interval. Three BP network models for flux prediction of ultrafiltration with time are set up. Among the three models, the BP network model based on Levenberg-Marquard algorithm is the optimum to predict nonlinear time series of flux variation.In addition, the mathematic software of MATHCAD is applied to calculate the parameters of the four theoretical models. The foul mechanism is analyzed. By comparing theoretical model with statistical model(ANN), the results show: mean relative error of ANN is 0.71%, less than that of theoretical model 0.81%...