Combined genetic algorithm optimization and regularized orthogonal least squares learning for radial basis function networks

The paper presents a two-level learning method for radial basis function (R BF) networks. A regularized orthogonal least squares (ROLS) algorithm is em ployed at the lower level to construct RBF networks while the two key learn ing parameters, the regularization parameter and the RBF width, are optimiz ed using a genetic algorithm (GA) at the upper level. Nonlinear time series modeling and prediction is used as an example to demonstrate the effective ness of this hierarchical learning approach.