Research On Latent Variable Model Predictive Control Algorithm For Batch Processes

Batch processes,characterized by flexible production,are widely used in cosmetics,food,pharmaceutical,and other fields.Owe to the advantages of iterative learning control(ILC)and model predictive control(MPC),the iterative learning model predictive control(ILMPC)method has been widely used to track key process variables in batch processes.However,in the original variable space,it is difficult and time-consuming to develop the predictive model for ILMPC,due to the collinearity and high dimension of variables.Moreover,the online implementation of ILMPC is suffered from too much calculation.Meanwhile,the existing ILMPC controller is not directly used for the end-point quality of batch processes.Therefore,the dissertation is based on latent variable techniques,such as principal component analysis(PCA)and partial least squares(PLS)to improve the ILMPC control method.The main research contents are as follows:(1)Considering that ILMPC is difficult to model and optimize in the original space,a latent variable iterative learning model predictive control(LV-ILMPC)algorithm based on the dynamic PLS modeling method is proposed.The controller is designed in a low dimension latent variable space by using the dynamic PLS model.A MIMO system can be automatically decomposed into multiple SISO subsystems.For each subsystem,the state-space model is implemented and the LV-ILMPC controller is designed independently.Because of the decoupling of the subsystem model,the parallel operation of each controller not only reduces the dimension of the controller but also reduces the calculation amount.(2)To reduce the online calculation of LV-ILMPC,a latent variable iterative learning predictive function control(LV-ILPFC)algorithm is proposed by employing predictive function control to replace MPC.The control input of latent variable is constructed as a linear combination of some predetermined basis functions.Each latent variable space can be designed with different base functions.The control input can be calculated by calculating the linear weighting coefficient at each sampling time.This method not only has the advantages of LV-ILMPC,but also further shortens the amount of online calculation.(3)Considering that the existing ILMPC controller is not designed directly for the end-point quality of batch processes,an LV-ILMPC method is proposed which can improve the end-point quality of batch processes.First,the dissertation uses the key process variable data to define the qualified operating space.Secondly,a predictive model of manipulated variables and key process variables is established.Finally,based on this predictive model,a LV-ILMPC statistical controller that can improve the consistency of the end-point quality of the batch process is designed,and the SPE control limit is used for monitoring.