The Optimization For The Distillation System Of Methanol-acetone Based On Response Surface Methodology And Neuralnetwork

This paper aims at the large energy consumption problem in themethanol-acetone-water distillation in the pharmaceutical industry. Taking the actualprocess of some factory as the research object, the better parameter combination is obtainedand the energy consumption is reduced using the response surface methodology andartificial neural network.The process model of extraction tower, acetone tower and methanol tower is establishedwith the actual operation data of industry. Determining the target function and theoperational parameters, and does the single factor analysis for each tower, and basing onthose establishes the center composite design.Using the response surface optimization method to optimize extraction tower andacetone tower. The mathematical model of extraction tower and acetone tower is obtainedby Basing on the center composite design to make regression for the target function and theoperation parameters, and gets the mathematical model of extraction tower and acetonetower. Applying the sequential quadratic programming method to optimize themathematical model, and gets the better operation parameter combination and targetfunction value. Then using the artificial neural network to optimize extraction tower andacetone tower. With experimental groups of the center composite design as the trainingsamples, and the parameters of extraction tower and acetone tower as the input variables,using tower kettle energy consumption and acetone content respectively as the targetfunctions, establishing two artificial neural networks, choosing the transfer function ofhidden layer and the training function by comparison to train the networks, finally gets twomathematical models which can describe the relationship between operational parametersand tower kettle energy consumption with acetone content. Also using the sequentialquadratic programming method to optimize the mathematical models, gets the betteroperation parameter combination and target function values.Comparing the decision coefficient in the modeling of response surface methodologyand artificial neural network with the predicted RMS error in this two methods, obtains theconclusion that the modeling of neural network is more accurate. Under the condition ofassuring the quality of products, extraction tower and acetone tower can economize energyconsumption0.1673×106kcal/h after optimizing by the neural network,22.89%of energy can be saved compared with un-optimized.Finally using the artificial neural network to optimize methanol tower, and methanoltower can economize energy consumption0.0889×106kcal/h after optimization,21.21%ofenergy can be saved compared with un-optimized.After optimizing, methanol-acetone system technology can save22.28%energyconsumption in total.