The Flux Prediction Of Nanofiltration Membrane Based On Neural Network

With the rapid development of science and technology, it is possible to make direct drinking water with the city tap water. Because of low operating pressure, high water production and high-quality water, nanofiltration(NF) is getting the favor of people. In the actual water treatment process, many factors affect the NF separation performance and the complex non-linear relationship exists between them. A lot of theoretical models have been developed based on the specific solutions, which often have their own application conditions range or have a lot of structural parameters and the computational complexity.There is no good theoretical model of flux prediction for multi-component tap water, because of factors including water quality, the performance of membrane and operating conditions lead to the flux decline and interacted each other. It’s an important and urgent problem to accurately predict the water flux and provide a theoretical basis for the determination of membrane fouling in the NF field. Neural network with its powerful nonlinear computing power and direct analysis the experimental data gradually get the attention of scientist of membrane.The commercial NF membrane used in this study was supplied by Vontron Co., Ltd, Beijing, China. The spiral-wound NF membrane VNF-1812is made of polyamide. The inlet water comes from the Dalian City tap water. The NF experiment was carried out about720hours. This article first studied the influence of operating pressure, time, temperature and recovery on the separation performance of NF membrane. Then the neural network models were built with the parameters of inlet and output of time, temperature, pH, pressure difference, water conductivity, inlet flow, TOC and flux respectively. Meanwhile, this article also studied the parameters setting of models and determined the optimal network of flux prediction. Finally, the flux was estimated by the non-equilibrium thermodynamic model of NF and made the pros and cons compare with the neural network.The results show that the neural network model has good generalization ability of the flux prediction than the theoretical model after learning and trained neural network. The BP network based on LM algorithm and the RBF network have the advantages of fast computing speed, high accuracy and easy to operate; The time series neural network has good practical application and provide real-time predictions for changes in membrane performance.The error of water flux prediction based on theoretical model is larger than the neural network model. Therefore, the theoretical model needs to improve when it was applied in the flux prediction for the nanofiltration process of tap water.