Salty Water Desalination Modeling Using Electrodialysis Process and Neural Network Technique

In this study, a model is presented using the black box neural network technique. In this model, the effects of four basic parameters, including temperature, voltage consumption, flow rate, and feed concentration on salt separation percentage from salty water, have been investigated. With the help of trial and error, the training method, transfer function, and the optimal number of neurons on each layer are selected to provide the best network performance. Multi-layer neural network, backpropagation, and Levenberg-Marquardt algorithm are utilized as the optimum options. In this study, 135 data were used, allocating 60% of them (81 data) to network training, 20% (27 data) to training data evaluation, and the remaining 20% (27 data) to assess the network generalizability as test data. Finally, a comparison of the model results with independent laboratory data indicates that the optimal network arrangement is 4:5:8:1, and the model with an error of less than 1% can predict the process behavior.