RADIAL BASIS FUNCTION NETWORK CONFIGURATION USING GENETIC ALGORITHMS

Most training algorithms for radial basis function (RBF) neural networ ks start with a predetermined network structure which is chosen either by using a priori knowledge or based on previous experience. The resu lting network is often insufficient or unnecessarily complicated and a n appropriate network structure can only be obtained by trial and erro r. Training algorithms which incorporate structure selection mechanism s are usually based on local search methods and often suffer from a hi gh probability of being trapped at a structural local minima. In the p resent study, genetic algorithms are proposed to automatically configu re RBF networks. The network configuration is formed as a subset selec tion problem. The task is then to find an optimal subset of n(c) terms from the N-t training data samples. Each network is coded as a variab le length string with distinct integers and genetic operators are prop osed to evolve a population of individuals. Criteria including single objective and multiobjective functions me proposed to evaluate the fit ness of individual networks. Training based on a practical data set is used to demonstrate the performance of the new algorithms.