Process modeling with neural networks using small experimental datasets

This paper reports some work done to improve the modeling of complex proces ses when only small experimental data sets are available. Various solution strategies based on feed-forward and radial basis function (RBF) neural net works have been tested for three problems including two wood pulp applicati ons. Experimental data sets obtained from D-optimal design and from a rando m selection throughout the experimental space were compared for their abili ty to lead to the proper model. In addition, the influence of activation fu nctions, the number of levels in stacked neural networks and the compositio n of the training data sets have been studied. The study shows that designe d training data sets are more desirable than random experimental sets, due to their higher orthogonality. The use of neural network is a powerful tool for modeling complex processes even when only a small set;of data is avail able for training. However, special care must be exercised to insure that g ood predictive models are obtained. (C) 1999 Elsevier Science Ltd. All righ ts reserved.