علي ملاح زاده

Modeling a distillation column can be challenging due to the inherently nonlinear nature of the process, multivariate interactions, unstable behavior, and time delays. Over the years, this challenge has been solved in various fields of application with the help of artificial neural networks. Furthermore, optimizing the sequence of distillation columns is a more complicated process, which has a longer computing time and more possible structures. In this thesis, firstly, the nonlinear behavior of the distillation column and the development of a feed-forward neural network model with input, hidden, and output layers are discussed to predict certain variables for three-component feeds in two case studies, C4-6 and BTX. The necessary training data sets are obtained from multiple simulations with the help of Aspen Plus software. The eva‎luation and selection of variables are made so that the maximum possible process data is obtained in the post-processing stage. After checking the performance of the developed neural network, it will be used to check the accuracy of predicting variables in any simple distillation column in a direct or indirect sequence. The development of such a model will free the user from training other networks in a distillation column sequencing. Then the use of the final model for the process of optimizing the sequence of distillation columns will be investigated, and the results obtained with the help of a neural network will be compared with the results obtained from rigorous simulation. Based on the comparison of the obtained results, it can be acknowledged that the neural network as a simulator shows a good performance in terms of the accuracy of predicting the variables. The amount of MSE error obtained for each system varies between 1e-7 to 1e-5 orders, indicating the developed networks' high precision. The study and interpretation of the results of the application of the final neural network in optimization problems show the superiority of the neural network over the link mode between Aspen Plus and MATLAB software in terms of computing time. According to the obtained results, the neural network can reduce the optimization time from about 200 minutes to 2 minutes by performing more optimization calculations and, according to its algorithm, which is a significant time reduction in optimization calculations. Moreover, the accuracy of the neural network in performing calculations is satisfactory according to its nature; however, the development of a more accurate model is still needed in these issues.

فرآيند تقطير , چيدمان برج‌هاي تقطير , شبكه‌هاي عصبي مصنوعي , مدل‌سازي و شبيه‌سازي , بهينه‌سازي

Distillation process , Distillation Columns Sequence , Artificial Neural Networks , Modeling and simulation , optimization

Ali Mallahzade

Dr. Kasiri