ePrints

Developing robust neural network models by using both dynamic and static process operating data

Lookup NU author(s): Dr Jie ZhangORCiD

Downloads

Full text for this publication is not currently held within this repository. Alternative links are provided below where available.

Abstract

Neural network models trained by dynamic process data alone can lack static representation capability. In process control applications, it is desirable that process models be able to capture both the dynamic and the static relationships between process input and output variables. This paper presents a technique for developing neural-network-based process models using both dynamic and static process operating data. The network training objective is to simultaneously minimize both the dynamic prediction errors and the static prediction errors of the neural network model. Studies in this paper show that model representation capability can be significantly enhanced by using only a very limited amount of static process operation data as additional training data. The developed technique is demonstrated by applications to a simulated water tank and a simulated neutralization process.

Publication metadata

Author(s): Zhang J

Publication type: Article

Publication status: Published

Journal: Industrial and Engineering Chemistry Research

Year: 2001

Volume: 40

Issue: 1

Pages: 234-241

Print publication date: 02/12/2000

ISSN (print): 0888-5885

ISSN (electronic): 1520-5045

Publisher: American Chemical Society

URL: http://dx.doi.org/10.1021/ie000286g

DOI: 10.1021/ie000286g

Altmetrics

Share