Dynamic Simulation And Research On Advanced Control Of Acetic Acid Dehydration Azeotropic Distillation Column

Distillation column is one of separating devices, which are applied widely in the oil refining and chemical industry, and it directly decides the quality of products and productive power of enterprises, commonly. And at the same time, distillation column is a high energy consumption device. However, Distillation process is a typical nonlinear, coupled, time-varying and multivariable system, and it's automation control has been always a research hotspot and difficult point in engineering field and control field. Because of the complexity of distillation system and gradual strict request for control, conventional control method can not already satisfy these conditions. Therefore, how is advanced control technology applied successfully to distillation process for the objective of stabilizing the product quality, increasing the throughput and energy conservation, has important realistic significance.The main research contents of this thesis are outlined as follows:(1) Steady state simulation of acetic acid dehydration azeotropic distillation process was implemented by using process simulation software called ASPEN PLUS, and a way to obtain the optimal location of sensitive plate in stripping section was presented basing on the simulation, further, dynamic simulation was achieved by using ASPEN DYNAMICS.(2) With the acetic acid dehydration azeotropic distillation process as background, advanced control strategies such as internal reflux control, neural network based predictive control and decoupling control were applied to the distillation column. Three advanced control strategies were designed and compared with each other, according to the different requests of distillation control and the strong coupling of multivariable system:â‘ Internal reflux control on the top of column and neural network based single-variable predictive control on the bottom. The temperature of the sensitive plate in stripping section could be controlled on the setpoint precisely, but control of product quality on the top of column was not considered in this method;â‘¡Neural network based multivariable predictive control. It realized a multivariable control for the product quality of the two ends of the column, by using two independent groups of RBF neural networks as the multi-steps predictive models of the temperature on the top of column and the temperature of the sensitive plate in stripping section respectively, and with the optimization strategy of finding a optimizing direction via gradient information of RBF neural network and obtaining a appropriate step size via golden section method. Because the coupling of multivariable system could not be removed completely, the steady state error existed and precise control couldn't be achieved in this system, though the temperature on the top of column and the temperature of the sensitive plate in stripping section were both controlled in a certain range;â‘¢Neural network based predictive decoupling control. The multivariable system was decoupled to SISO subsystems by a feedforward compensator, and two neural network based single-variable predictive control systems were designed respectively for the two decoupled subsystems. This strategy could control the two temperatures precisely and rapidly. Simulation research on the advanced control strategies above was implemented by using a combined platform with ASPEN DYNAMICS and MATLAB SIMULINK on the basis of dynamic simulation. The results indicated that, neural network based predictive decoupling control had a simple control structure, setting flexibility of controller parameters, strong decoupling ability and satisfactory control precision.