The Research On Improving Performance For Generalized Predictive Control

Generalized predictive control (GPC) is a model predictive control algorithm with strongrobustness. It can be applied in non minimum phase systems with delay and time varyingparameters. But in one part, the parameter model is hard to get for some systems. In another one,traditional generalized predictive control algorithm is more complex and the compute work isvery large. Third, some real industrial control is more sensitive to the overshoot, drasticovershoot will add difficult to the control or even lead to production accident. So, the researchfor an improved generalized predictive control algorithm which can restrain the overshoot,without the parameter model and have little compute work will have important realisticsignificance.Firstly, for the problem of restraining the overshoot, this thesis proposed an improvedgeneralized predictive control algorithm based on single step predictive output defference. Forthe SISO system, by introducing the single step predictive output defference into the costfunction, the overshoot is restrained effectively. Further more the single step predictive outputdefference is introduced into the cost function of multivariable systems. Simulation researchshow that the improved algorithm is both effective for the SISO system and the MIMO system.Secondly, for the problem of traditional generalized predictive control algorithm needparameter model and solving diophantine equation, this thesis proposed an improved adaptivepredictive control algorithm. Multistep predictive model is divided into two part. One part is thesoftened real system output, this will be the origin for the predicting. The other one is the effectestimating for the future input can contribute to the system, the estimate matrix can be identifiedby the recursive least square. This method for building multistep predictive model need not theparameter model, don’t care about the model order, needn’t to solve the diophantine equation,and have less parameter needed to be identified, so it is easier to be applied in the real controlprocess.Thirdly, for the problem of multivariable system control, this thesis proposed an improved multivariable adaptive predictive control algorithm. The multistep predictive model ofmultivariable system can also be divided into two part. One part is the softened real systemoutput, this will be the origin for the predicting. The other one is the effect estimating for thefuture input can contribute to the system, the estimate matrix can be identified by the recursiveleast square. Simulation research show that the improved adaptive predictive control algorithm isnot only effective for the SISO systems but also effective for the MIMO systems. Further more,the improved adaptive predictive algorithm is not only effective for the linear systems but alsoeffective for some nonlinear systems.At last, on the basis of the above research, combined the existing predictive PID controltheory, this thesis proposed an improved adaptive predictive PID control algorithm. Theimproved algorithm need not the parameter model and the system input have a traditional PIDtype. The PID parameter can be tuned in the control process. The overshoot in the controlprocess is also been restrained. Simulation research show that the algorithm proposed in thispaper have good effect when applied in fermentation temperature control.