Study Of Distributed Model Predictive Control In Polymerization Reaction

Polymerization reactions are nonlinear systems which have the characteristics of strong coupling, uncertainty and complicated dynamic mechanism. They are always consisted of many subsystems which have complicated association. Model predictive control strategies can deal with various kinds of constraints effectively, they are used in polymerization reactions widely. With the rapid development of computer network, model predictive control strategies are not limited to the centralized control, they are substituted by distributed control more and more, all this bring new challenge to conventional control problems. Based on large scale polymerization reactions, this thesis has studied dynamic characteristics of polymerization and the design of distributed model predictive control, the main achievements as following:●The dynamic mechanism of polymerization reaction is analyzed taking styrene polymerization for example. The important controlled variables are discussed. The model of thermal bulk polymerization of styrene is given and reactions in CSTR are analyzed. The analysis of the steady-state multiplicity behavior is presented by simulation. The region of feasible temperature operation, the relationship between reactor temperature and number average molecular weight (NAMW) are discussed. The results show the complexity of polymerization reaction, multiple steady-state and infeasible region may occur with the change of parameters. The temperature of reactor determines NAMW and its distribution of polymer, it is the important operating factor in polymerization.●The typical model predictive control methods of large scale system are studied. A distributed model predictive control strategy is proposed for grade transition in polymerization reaction. Setting up the specific performance index based on target following, the index including effect of other subsystems, the characteristic of subsystems are described by combined model and the detailed derivations of control law are given. The analytical form of constraint quadratic programming is given. Finally, the performance comparison with many other predictive control strategies is given to illustrate the effectiveness of the proposed approach by the application on styrene polymerization reaction. The approach makes important contribution to the application of distributed model predictive control on polymer grade transition.●The on-line multiple models identification strategy is presented based on data-driven and local-modeling for a class of unknown-structure nonlinear systems just on the basis of numerous historical database. A database searching strategy based on KHM clustering is presented which insensitive to initial value. The approach is easy and feasible, it can shorten searching time and improve searching efficiency. A new approach to determine data information (determine the neighborhoodΩk (x)) is proposed, it can improve precision of the data which used to get optimal local model. The goodness of fit criterion and rapid selection method to determine bandwidth (h) are given. Finally, characteristic analysis and simulation studies were done, the simulation test illustrates the validity of this approach.●Combining the multiple models thus developed with predictive control, the multi-model distributed cooperative predictive control strategy is proposed, thus solving the control problem of a class of unknown-structure polymerization reaction. At every moment, after getting the parameter model, using a model structure transition and structure recombination method, to get the internal information of subsystems and the interacted information with other subsystems. Then expand the dimension of state vector to get combined model which includes information of other subsystems. The all design process of multi-model cooperative model predictive control strategy is given. This approach can update model parameter on line easily and feasibly, it not only solve the distributed predictive control problem of a class of unknown-structure polymerization reaction, but also make up for the defective of sole overall linear model which can't express nonlinear characteristic well in existing distributed model predictive control. By the application on polymerization, this approach has illustrated obvious improved dynamic performance.●When process disturbance and measurement noise exist for constraint nonlinear system, moving horizon estimation method is studied, which can deal with constraint effectively. Based on the principle of separation, combining moving horizon estimation with distributed model predictive control, the MHE distributed model predictive controller is designed. The main parts of controller are given. The precision of estimated state is important to controller performance. With the application on polymerization reaction, the simulation test illustrates the validity of this approach.