Multiple neural networks modeling techniques in process control: a review

Abstract

AbstractThis paper reviews new techniques to improve neural network model robustness for nonlinear process modeling and control. The focus is on multiple neural networks. Single neural networks have been dominating the neural network ‘world’. Despite many advantages that have been mentioned in the literature, some problems that can deteriorate neural network performance such as lack of generalization have been bothering researchers. Driven by this, neural network ‘world’ evolves and converges toward better representations of the modeled functions that can lead to better generalization and manages to sweep away all the glitches that have shadowed neural network applications. This evolution has lead to a new approach in applying neural networks that is called as multiple neural networks. Just recently, multiple neural networks have been broadly used in numerous applications since their performance is literally better than that of those using single neural networks in representing nonlinear systems. Copyright © 2009 Curtin University of Technology and John Wiley & Sons, Ltd.