Study Of The Continuous-Time Generalized Predictive Control And Its Application Methods

Predictive control is a model based advanced control method which is originated directly from the industry process control.It combines the requirement of the practical industry processes and has good control performance.Generally the control techniques have both continuous-and discrete-time versions,and the control engineer,however,is free to choose the design approach according to the practical condition-either continuous or discrete.And predictive control is no exception.Since the continuous-time generalized predictive control(CGPC)is proposed it has attracted many attentions in control fields.It has properties similar to those of the discrete-time generalized predictive control.In addition the choice of sample interval is flexible and the control weighting is not essential for the control of non-minimum phase systems.However,there are still many open problems to be solved.Just from this view,the continuous-time generalized predictive control and its implementation methods are studied. First,the CGPC law with integration is derived to eliminate the offset.The closed-loop set-point tracking performance and disturbance-rejection performance are analyzed.The internal model structure of CGPC with integration is derived,and compared with the observer structure of CGPC the important role of observer polynomial is analyzed.A recursive computation method of polynomial parameters which can avoid the complex division computation of polynomial is proposed.According to the effects of tuning parameters to the system performance,some tuning rules are given based on the design model with integration,and they can simplify the design and implementation of CGPC.Second,for the coupling performance of the traditional cascade control structure,a special controller with decoupling robustness is proposed.Based on the CGPC algorithm,the original output predictor is replaced by a special predictor with cascade function by using the parameters identities.Thus only one predictive controller can perform the cascade control task.The effects of observer polynomial to the system performance are analyzed.The results show that the robust performance of the inner loop and the outer loop can be tuned separately by the corresponding observer polynomial while the set-point tracking performance is not affected.The compared simulations with the traditional cascade structure verify the performance of the proposed controller.Third,a CGPC method explicitly considering time delay (F-SDCGPC)is proposed.By using modified predictive output signal the delay compensator is incorporated in the control law with observer structure.The effects of observer polynomial to the disturbance-rejection performance and robustness are analyzed.A filter is added to enhance the robustness when considering large model error especially time delay error. The tuning of filter can enhance the initiative tuning ability of robustness. The design of filter does not affect the nominal set-point response,and it is more flexible than the design of observer polynomial.This property allows the design of the controller in two steps:firstly the controller parameters are chosen to attempt the nominal set-point performance specifications and then the filter is tuned to increase the robustness.The analysis and simulation results show that the F-SDCGPC has better robustness than the observer structure without filter when large time-delay error is considered.Forth,a new delay predictive solution is proposed for the multivariable continuous-time generalized predictive control(MDCGPC). For most of the physical realizable processes,the model parameter matrices can always be diagonally constructed,so that we can transform the multiple inputs and multiple outputs system into a set of multiple inputs single output model.By using the computation of the identities,the delay compensator is introduced in the control law.A filter is added to adapt delay error.Moreover the decoupling performance can be obtained by tuning the predictive horizon and the reference trajectory.The recursive computation method of polynomial parameters reduces the computation load and complexity.Simulations of Wood-berry and Shell models verify the performance.Fifth,the continuous-time model is identified by the NLJ method, and then based on this model the continuous-time predictive controller is designed.The CGPC simulation software is developed using VC which is the base to practical application.The CGPC law is considered as combination with the outputs of multiple transfer function blocks,and then the output of the block can be computed separately.This method provides some conveniences to add new blocks.