Research On Point-to-Point Tracking Algorithm In Batch Processes Based On Iterative Learning Predictive Control

The point-to-point tracking is a general problem in the tracking control of the batch process,and the tracking effect of several key points determines the product quality.Iterative Learning Model Predictive Control(ILMPC)algorithm has received extensive attention in point-to-point tracking of batch processes.However,the ILMPC algorithm does not fully utilize the degrees of freedom of non-key points,and the convergence speed is slow.Secondly,difficult modeling caused by collinearity and high dimensionality among variables,and large computation load,have brought challenges to the promotion and application of ILMPC.Based on the strategy of parameter optimization,and the requirement of in the multivariable system and the fast batch process,this dissertation proposes the corresponding point-to-point tracking control method.The main research contents of the dissertation include:(1)Aiming at the slow convergence speed in the point-to-point iterative learning model predictive control(PTP-ILMPC)algorithm,a point-to-point iterative learning model predictive control algorithm is proposed within which the learning gain is optimized alternatively.The reference trajectory is determined according to the predefined the learning gain,and the iterative learning model predictive controller is designed to track the current batch reference trajectory to obtain a control law.Then,the performance index function is constructed according to the tracking error to determine the optimal learning gain,and the next batch of reference trajectories is updated by the optimal learning gain.The track of the reference trajectory may be achieved at a faster speed by iterative optimization of the learning gain and the control law.The simulation results on a three-axis gantry robot model show that the proposed method can track the reference trajectory quickly and efficiently.(2)To solve the problems of coupling and high dimensionality among variables in multivariable systems,a latent variable PTP-ILMPC(LV-PTP-ILMPC)algorithm is proposed based on parameter optimization.The algorithm captures the main information of each batch through a latent variable model constructed by the Dynamic Partial Least Squares(Dy PLS)algorithm,and designs the PTP-ILMPC controller in the latent variable space.At the same time,a reference trajectory update rule based on parameter optimization is designed in the latent variable space,and the reference trajectory is continuously updated according to the tracking error information.The algorithm is suitable for multiple-input multiple-output(MIMO)systems with strong coupling and severe collinearity.The simulation results on the numerical model and electromechanical platform verify that the proposed algorithm can not only decouple the variables,but also reduce the computational cost.(3)To meet the requirement of real-time control for a class of fast batch processes,the predictive function control(PFC)is introduced into PTP-ILMPC,and a point-to-point iterative learning predictive function control(PTP-ILPFC)algorithm based on parameter optimization is proposed.The method transforms the control input into a linear weighted sum of predetermined basis functions,thereby transforming the high-dimensional control input solution problem into a low-dimensional linear weighted coefficient solution problem.To obtain the control variable,the control input can be calculated only by obtaining the linear weighting coefficient at each sampling interval.This method not only maintains the advantages of the original PTP-ILMPC,but also effectively reduces the calculation time of optimizing the control input and improves the real-time performance of the control.The simulation on the three-axis gantry robot model proves that the algorithm effectively shortens the online computing time of the original algorithm by reconstructing the control input.