Application of Weighted Latent Variable Model Predictive Control in Batch Process Temperature Control

Abstract

This paper presents a Weighted version of the Latent Variable Model Predictive Control (WLV-MPC) to address the control solution instability of the original LV-MPC algorithm that is related to the loading matrix decomposition. The suggested idea is firstly applied in a system identification framework where a modified version of an iterative Least Squares (LS) technique supported with the Upper Diagonal (UD) factorization algorithm is implemented in model parameter optimization. The second part illustrates the derivation of the WLV-MPC through penalizing the loading matrices that form the basis of the designed cost function. The use of the D matrix to penalize the formulated Hessian matrix in Quadratic Programming (QP) has significantly improved the solution stability. The performance of the proposed approach has been verified through a numerical example and in the temperature control of a batch process benchmark.