Soft Sensor And Research On Advanced Control Of Acetic Acid Distillation Column

Distill column is the center equipment during the production of chemical products. The measurement and process are always restrained and energy wasted. Soft sensor tech and advanced control can be implemented in the control of a complicated process where traditional control methods perform less efficiently, so it receives more and more attention by researchers.In this research paper, the main focus is the process of acetate dehydration distillation. Through detail analyzing the mechanism of the acetate dehydration process, the control problem of the column is studied including stationary and dynamic modeling of the column, soft sensor methods, inferential control and predicting control.Following is the main research focus of the paper:(1) The process of the acetate hydration distillation is modeled by Aspen Plus. Based on the steady modeling of the column, the sensitive plate is analyzed which is an important parameter for implementing control strategies. Also the influence of different operation conditions on the products of the column is analyzed. Then the dynamic model of the column is established by transforming the state model into the environment of Aspen Dynamics.(2) The concentrations of products from the acetate distillation column are hard to be measured directly. To solve this problem, a soft sensor method based on additive support vector machine is proposed. The proposed additive support vector regression overcomes and reduces the bias made by traditional support which comes from the problem of data sparsity in the high dimensional input space (the curse of dimensionality). The proposed additive support vector regression is described by a quadratic programming (QP) formulation. Simulation results showed that the additive SVR model had better predication performance than traditional SVR and least square support vector regression (LS-SVR).(3) Aiming at the problem of unknown disturbances during the control of the distill column and also the problem of hard measuring of the concentrations of its products, and for better control accuracy, inferential control and prediction control based on support vector regression soft sensor modeling are proposed,and furthermore the combinations of the two methods are studied. The simulation results show some surplus will occur due to the characteristics of the multiple variables control. But the latter control method has the merit of double-ended control, which can have the products of both the top and the bottom of the iistill column been controlled more quickly.