Sequential growing and pruning of radial basis function network

We propose an algorithm for sequential structure adaptation of radial basis function (RBF) network. Our main goal is to obtain the model of the unknown time varying nonlinear mapping. Both parameter and structure adaptations are incorporated into the framework of an extended Kalman filter. Two approaches, the construction (growing) and pruning, are combined during adaptation of the RBF network structure. Examples of nonstationary nonlinear dynamical system modeling are presented to illustrate the proposed algorithm.