Research On Distributed Model Predictive Control Of Process Networks

With the development of the intelligent manufacturing in process industry,there are more attention to online large-scale complex control system implementation issues.The process network modeling method provides a new way to describe and analyze the typical processes,and it provides the necessary theoretical basis for distributed optimization and control.Distributed model predictive control(DMPC)method can decompose complex dynamic optimal control system into several subsystems,it is widely used in process industry with the combination of chemical process network modeling,which can decrease the computational complexity of control system,and the implementation of the online calculation.It can solve the large-scale complex control system implemented online,and provide an advanced control method to realize the intelligent chemical factory.At present,the industrial application of distributed model predictive control has become a hot topic in the field of control.In this paper,the process network structure is divided into series system and parallel system.There are many series systems in the process network,such as the atmospheric and vacuum distillation tower in the petrochemical production process,the nuclear industry cracking tower and so on.According to the process of industrial environment,structure characteristic of cascade control system,the typical tower series production process as the research object in this paper,a distributed model predictive control algorithm is designed based on Nash optimality,and the optimal convergence conditions are given.According to the model of parallel system in process networks,such as ammonia fertilizer production process and gas boiler heating system,a distributed model predictive control strategy is proposed based on a typical gas boiler heating system with a definition of the competitive relationship in this paper.Considering the competitive coupling characteristics of the parallel system process network,and the optimization of performance based on competitive constraints for parallel system is improved,at each sampling time optimization problem of each subsystem is not only a local optimum,but also in the global performance,and centralized control consistent global performance prediction.The algorithm can reduce the computation time and save the cost,and ensure the stability and global optimization of the whole process.Considering the complexity of the actual situation,the problem of uncertainty is added to the design of distributed model predictive controller.By using the idea of variable transformation and LMI methods,the infinite time domain “min-max” optimization problems are converted into linear programming problems.A piecewise continuous output feedback control law is obtained in each subsystem and the sufficient conditions for the existence of this control law are given.It is proved that the robust stability of the closed-loop singular systems is guaranteed by the initial feasible solutions of the optimization problems,and the regular and the impulse-free of singular systems are also held.A simulation example shows the effectiveness of this method.