Modeling and Control of Wiener Systems Using Multiple Models and Neural Networks: Application to a Simulated pH Process

Abstract

This paper describes an efficient multiple models control scheme based on a bank of neural Wiener models and its application. Each Wiener model consists of a linear dynamic part in connection with a nonlinear static part which is approximated by a radial basis function (RBF) neural network. The control algorithm is developed from a Clarke performance index and can copy with unstable zero-dynamics problems. An extended system that is associated with the criterion function is defined for system parameters identification and control law design. After establishing multiple neural Wiener models for several distinct operation regions, a robust switching mechanism is introduced for the selection of the best controller. The overall scheme is applied to a simulated pH neutralization process. The set-point tracking, disturbance rejection, and robustness performance of this scheme are compared to several alternatives.