Model Predictive Control Of Conventional And Reactive Dividing Wall Columns

As a typical example of process intensification,the dividing wall column integrates two columns into a single shell and brings substantial savings in the energy consumption and capital investments as well as the reduction in the installation space.However,the highly coupled structure of the dividing wall column leads to the increase in the number of variables,among which complicated interactions exist,making it difficult for the dividing wall column to be controlled.A great deal of research has been done on the control of the dividing wall column and different control strategies have been proposed for different dividing wall column configurations.In this paper,steady state simulation of the conventional and reactive dividing wall column are carried out with Aspen Plus software.The PID control structures for these two kinds of dividing wall columns are constructed and tested on the software platform of Aspen Dynamics.The model predictive control of the dividing wall columns are implemented by a joint simulation of Aspen Dynamics and Simulink in Matlab.The multiobjective genetic algorithm function“gamultiobj”in Matlab is used for the weight tuning of MPC.The aim of this work is to explore the feasibility of the application of model predictive control to the dividing wall columns,with a view to significantly improve the control quality and ensure stable and efficient operation of the unit.Firstly,composition control and temperature inferential control of the conventional dividing wall column for separating ternary mixture of ethanol,n-propanol,and n-butanol are considered.Conventional PI scheme and model predictive control strategy are established for the process.The comparison between the control performances of these two strategies shows that for both composition control and temperature inferential control,the model predictive control delivers a better overall performances than the conventional PI strategy in resistingfeed disturbances and maintaining stable operation of the process.Then,the control of the reactive dividing wall column for the synthesis of n-butyl acetate and methanol through the transesterification of methyl acetate and n-butanol are discussed.Three control strategies,namely PI,PI with feed-forward control(PIFF),and model predictive control are proposed.The control performance of different control strategies are tested and compared.The simulation results show that for the control of the highly nonlinear and strongly interactive process like the reactive dividing wall column,all the three control strategies are able to handle feed disturbances and achieve the stable operation of the process,but the PIFF strategy tend to show a poor performance in terms of the overshoots and ISE of the controlled variables.Model predictive control generally delivers an improved overall performance in resisting feed flow rate disturbance when compared to the PI strategy,featured with reduced summation of ISE.While for the feed composition disturbance,the PI strategy still has its advantages.