An identification approach to nonlinear state space model for industrial multivariable model predictive control

Extending application of model predictive control technology has encountered new challenges from the chemical and polymer industries where the processes show strong nonlinear dynamic behavior and necessitate nonlinear dynamic models for MPC. This paper presents an approach to identifying nonlinear state space models from plant data. This approach uses a direct identification scheme and integrates several technologies including a hybrid linear-neural network model, PCA and PLS modeling algorithms and on-line adaptation to address the robustness of the identification and the resultant model. Two examples are presented to demonstrate the features of the approach.