IMPROVING THE PREDICTION CAPABILITY OF RADIAL BASIS FUNCTION NETWORKS

Radial basis function networks (RBFNs) have been well established as a class of supervised neural networks. Because they do not require a pr iori process information and they possess localized interpolation prop erties, they are attractive for the empirical modeling of complex nonl inear multivariable processes such as those associated with the chemic al industry. In this paper, classical regularization theory is used to develop a technique for improving the prediction capabilities of RBFN s by incorporating process knowledge obtained from physicochemical rel ationships through modification of the objective function. An analysis of this new objective function is provided and compared to the conven tional least-squares objective function. Several chemical process exam ples are provided to demonstrate the improved predictive capability of this modified RBFN.