Learning Algorithm with Self scaling Variable Metric for Neural Networks and Its Application for Predicting the Conditions of Polyethylene Terephthalate

In a complex chemical industry process, predicting the conditions of the process is one of the most promising fields for neural networks application. This paper is concerned with improvements of neural networks learning algorithm and its application for predicting the production conditions of polyethylene terephthalate (PET). On the basis of the analysis of the optimization methods, a new algorithm based on Quasi newton method with self scaling variable metric is proposed. Simulation results show the effectiveness and the good convergence of the new algorithm.