Research On Probabilistic Constrained Dynamic Matrix Control

Model predictive control is a type of advanced computer optimization control algorithmsbased on model prediction. The typical algorithms of model predictive control are modelalgorithmic control (MAC); dynamic matrix control (DMC); generalized predictive control(GPC). They have same principle, such as model prediction; rolling optimization; feedbackemendation. The trait of model predictive control is: the mathematic model of controlled systemis easy to acquire, and it is not need to complicated system analysis and precise modeling; sincerolling optimization based on feedback emendation instead of conventional optimization control,then the controlled system can conquer the uncertainty influence, increasing the robustness andcalculating burden on-line is opposite simple.For thirty years, predictive control has been great developed in model predictive controlwith disturbances or constraints and their stability; robustness; feasibility have been researched.Non-linear model predictive control (NLMPC) and constrained model predictive control (CMPC)have been researched increasing. The research refers to field as shown in follow:(1) Modify the basic algorithms which have been existent.(2) Extend single variable to multi-variables; use the algorithms only being the same withsteady objects to non-self-balance systems; spread the predictive control applicationarea to non-linear and distributed parameter system.(3) Choose optimization aim function.(4) Select prediction model.(5) Introduce big system method.(6) Combine advanced control theory with the basic control algorithms. Furthermore,predictive control research also refers to feasibility, stability and robustness of system.Research on reducing the calculate burden on-line and making the control algorithmstransparent is also significative.The key of model predictive control is optimization rolling on-line. On a great deal of Resear-ch Literature published in magazine or conference, this optimization problem could be predigestquadratic performance index optimization without constraints, it is differ from practical industrialapplication. On industrial process, multi-variables and constrained system is inevitable, so thenresearching on constraints predictive control problem is significant.My work in the paper applies myself to research on constrained system optimization controlproblem. At first based the controlled system mathematic model, describe the states constraints atevery time as probabilistic form, according to Gauss equation, then use an approximatedistribution for the state estimate error to convert these probabilistic constraints into deterministicconstraints on the conditional mean of the state; Meanwhile, design a special predictive observerto compensate the time delay problem; then also using the feedback information, add tail errorvariety rate to observer to modify the predictive state. Finally computer simulation approves thetheory has feasibility and advantage.