Application of feedforward and recurrent neural networks to chemical plant predictive modeling

Abstract
The authors explore the use of backpropagation and recurrent backpropagation neural networks for the indentification of the dynamics of a chemical process. The authors build predictive models for plant variables and compare the performance of the feedforward and recurrent neural networks on this prediction problem. The authors also consider the training efficiency of two recurrent backpropagation learning laws-namely, R.J. Williams, and D. Zipser's teacher forced learning law (1989) and F.S. Tsung's law (1991). In this study, the Tsung law performed significantly better (faster learning speed and lower ultimate error level) than the teacher forced learning law.