MIMO model predictive control with local linear models

Abstract

The performance of a Model Predictive Control (MPC) algorithm depends on the quality of the derived model. Using a divide-and-conquer strategy process operations were partitioned into several operating regions and within each region, a local linear model was developed to model the process. This set of locally linearized models was simply and effectively combined into a global description of a multivariable nonlinear plant. To save on computational load, a linear model was obtained by interpolating these linear models at each sample point and then this linearized model was used in a Generalized Predictive Control (GPC) framework to calculate the future behavior of the process. Thus, time-consuming nonlinear quadratic optimization calculations, which are normally necessary in nonlinear predictive control, can be avoided. Modeling and controller design procedure was demonstrated using a simulated pH neutralization process with two inputs and two outputs.