Forward-Propagation in RBF—Networks
P. Lang, D. Prätzel-Wolters, Frank Bärmann · 1996
One of the main tasks in chemical industry is the determination of physical and chemical properties of new complex compounds. The standard method to determine these properties is analysis through experiments. However those experiments are mostly expensive and time consuming. Alternatively, prediction formulas have been developed which lead to approximation of material properties based on physical laws and known data from previous analysed components of those complex compounds. These approximations are not always consistent with the true material properties and there is a strong demand for better prediction algorithms. In the last years Neural Network based algorithms have shown very satisfactory prediction properties. In this context the learning algorithms for RBF-Networks, developed in our paper, are applied to the following concrete problems arising in chemical industry: