Pattern-based identification for process control applications

In this paper, a pattern-based approach to process identification is presented. The process identification problem is formulated using a nonlinear regression model. An algorithm is proposed based on the modified Gauss-Newton search for a least squares estimate and the condition for the identification is derived. The algorithm is extended via the instrumental variable method to cater for possible correlation of residual error with a Jacobian function. Simulation results are presented to support the theoretical development for a typical range of industrial processes. The proposed method is also compared favorably with methods existing in the literature.