Study On Model Predictive Control Method Including Steady-state Calculation

As its good performance dealing with multivariable, constraint, coupling and purelag, model predictive control (MPC) gets more attention and has been widely used inprocess industry especially in refinery and petrochemical processes.MPC could pushthe productive process to the boundary of the quality constraints by improving thedynamic performance of process control and reducing the amplitude of processvariable. It makes the bounder control come true. So that it achieves the aim ofensuring the stability and security of device running, ensuring the quality of products,increasing the plant capacity and reducing the cost of running. It brings the remarkableeconomic benefits for enterprises.The enterprise-scale of modern process industry is huge so that the hierarchicalstructure is usually used to detail the commands of optimization and control.Complicated system would meet with the problem the position of optimal steady statebeing changed by the reason such as the system’s constraints, the operation of operatorand the impacts of disturbance. In this thesis, we studied the optimization and controlbased on two-layer predictive control from three aspects which are as follows.①Adopt Two-layer predictive control which is widely used in process control.To solve the problem of optimal steady-state’s position changing, the steady-stateobjective is recalculated at each sampling time considering the effects of measureddisturbance. Ensure that the steady-state objective is compatible with the velocityconstraints in dynamic MPC optimization. The difference between the modelprediction and the measured output at the current time is feedback to correct thesteady-state model.②Based on the two-layer predictive control, introduce the goal programmingmethod to study the soft constraints adjustment and target relaxation of system. In theprocess of steady-state objective calculation, classify the soft constraints and describeit with target priority factor. Transform the desire objective into objective constraintswhich should be classified too. Then, the soft constraints adjustment and targetrelaxation s coordinated. When the optimization problem is unfeasible or the feasibleregion is apart from the desired working point due to system dynamic, the abovemethod could be adopted. ③Model the TRMS with Newton-type method, recalculate the steady-stateobjective at each sampling time and use the terminal equality constraints to keepstability, we can get good control effect.