Identification of Multiple Linear Models for Nonlinear Processes

Abstract

This work presents a nonlinear dynamic model based on several local linear models under different operating conditions. Response of the global nonlinear dynamic model is also de-rived by weighting the sum of all local linear model outputs. In addition, the fuzzy set theory is applied to account for the weighting factors for the local models. Also presented herein are two novel means of estimating the multiple linear models’ output: the parameter interpolation method and the output difference interpolation method. According to our results, these two methods are identical in terms of interpolating the difference of state vector, outputs, and in-outs. Some major identification methods, e.g., linearization of the first-principle model, identification of linear local models, and least squares algorithm, are proposed. Several typical nonlinear processes are used to demonstrate the effectiveness of the multiple linear model identification.