Multiple model predictive control of a styrene polymerization process

Abstract

In this paper, multiple piecewise linearizations of a nonlinear process in different operating points are used within a Kalman filter bank which computes the conditional probabilities of various hypotheses that are modeled by the filters. State estimates provided by the Kalman Filters and local model parameters are weighted using conditional probabilities and then used within the predictive control framework. The proposed strategy is tested on the complex model of styrene polymerization process.