Latent Variable Iterative Learning Model Predictive Control For Nonlinear Systems

Batch processes are widely used in manufacturing,pharmaceutical production,food and chemical industries due to their flexible production,short cycle times,and rapid transformation.Latent Variable Iterative Learning Model Predictive Control(LVILMPC)has good trajectory tracking and interference suppression capabilities,and can decouple and reduce the dimensionality of process variables.It has been widely used in multivariable batch processes.However,LVILMPC only targets linear batch processes.When this method is applied to time-varying nonlinear batch processes,the control performance is not ideal due to low model accuracy.In this paper,Dynamic Nonlinear PLS is used to improve LVILMPC algorithm and extend LVILMPC method to nonlinear batch processes.This paper studies Nonlinear-Latent Variable Iterative Learning Model Predictive Control(N-LVILMPC)based on nonlinear latent variable model.Specific research contents include:(1)Aiming at the time-varying nonlinear characteristics of practical batch processes,a nonlinear latent variable iterative learning model predictive control(N-LVILMPC)method based on linear variable parameter(LPV)models was proposed.The LPV model is used to fit the PLS inner model relationship to describe the nonlinearity and dynamics of batch process,in each low-dimensional latent variable space,the state space model is realized and the N-LVILMPC controller is independently designed to track the projection of the reference trajectory in the latent variable space,finally,the control low is back-projected into the original space to act on the controlled system.The model that this method relies on can accurately describe the actual dynamics of the nonlinear system,improve the tracking effect and convergence performance of the controller.(2)In order to reduce the online computational burden of N-LVILMPC based on LPV model,the T-S model is introduced into the dynamic nonlinear PLS to fit the inner model relationship,and a N-LVILMPC algorithm based on T-S model is proposed.In the low-dimensional latent variable space,a local LVILMPC controller is designed for each fuzzy rule of the T-S model,and then the global N-LVILMPC controller in the latent variable space is obtained by weighting it with membership.The parameters of the prediction model in this method are constant at each optimization moment,reducing the online computation and improving the real-time performance of the control.(3)Due to the actual operating environment and other reasons,it is difficult to ensure high repetition of different batches in batch processes,and models established solely based on single batch data information often cannot describe the dynamics of the ent ire batch process.The paper further proposes a nonlinear latent variable iterative learning model predictive control(N-LVILMPC)method based on multi batch fusion modeling.First,the data information of multiple batches is projected into the latent variable space to obtain the input and output latent variables of multiple batches,and then these latent variable are fused into a new two-dimensional data matrix,and T-S modeling method is used to identify the model parameters,and then N-LVILMPC controllers are designed.This method utilizes the multi-batch difference characteristics of batch process to improve model robustness and control accuracy of the controller.