Modeling And Optimization Study On Solvent Recovery & Dehydration Distillation Column

In this thesis, the modeling and optimization of the solvent recovery & dehydration distillation column are studied based on statistical theory and neural network technique. The main contributions of the work can be summarized in the following:1) Using on-line collected data, a soft-sensor model of the column compositions is constructed via multiple linear regression (MLR) and principal component regression (PCR) technique.2) The soft-sensor model of the column composition is also founded using the neural network technique with BP algorithm, together with the validation using on-line data.3) PCA-BP, the approach combining statistic theory and neural network technique, is systematically analyzed. Ad hoc approaches for choosing some key parameters of the model are presented. Their application in the modeling of the solvent recovery & dehydration distillation column is demonstrated with satisfactory results.4) An off-line optimization for the minimum cost operation of the distillation column is presented.