Application Of Artificial Neural Network In Simulation Of Extractive Distillation And Reactive Distillation

Extractive distillation and reactive distillation have been widely used for their low cost, low energy consumption and high separating efficiency. However, as the special distillation processes, the simulation of extractive distillation and reactive distillation is comparatively complicated and is of a lot of shortages. As a result, a simulation method which is simple and more efficient is significant for reducing cost and enhancing efficiency.In this dissertation, artificial neural network (ANN) model which is trained by momentum back propagation with variable learning rate are presented to simulate extractive distillation and reactive distillation without the consideration of the specific process and mechanism in the columns. The results indicate that excellent effects can be achieved by ANN models with high simplicity. The feasibility of CHEMCAD for simulating extractive distillation and reactive distillation is validated by comparison of the simulated data and experimental data. So CHEMCAD can be used to provide sufficient data for ANN to make up the shortage of experimental data when setting up ANN models. Then by comparing the different predicted results of the training set and testing set to establish the optimal structure of ANN models, which the different predictive results are produced by ANN with different transfer function and nodes in hidden layer.In this thesis, the extractive distillation for separating methanol-acetone mixture and reactive distillation for synthesizing of methyl acetate are taken as examples to demonstrate the simulation method, and the results show the ANN models trained by momentum back propagation with variable learning rate has higher precision and less training time than the ANN which be trained by common BP algorithm for simulating this two processes.