Multimodel Decomposition of Nonlinear Dynamics Using Fuzzy Classification and Gap Metric Analysis

Abstract Identification of accurate nonlinear models is central to the success of the nonlinear model based schemes. Approximation of the nonlinear system dynamics in a multiple linear model framework has been well addressed in the literature. However, such multimodel decomposition can result in unstable local models. Additionally, the number of local models to be selected for the nonlinear identification is critically dependant on the partitioning approach. This paper proposes a novel gap metric based fuzzy decomposition of nonlinear dynamics using multiple, locally linear models. Such a decomposition is shown to result in a stable and parsimonious model set which can be deployed for online control. A simulation case study involving nonlinear polystyrene reactor, is presented to illustrate the proposed approach.