Structure Selective Updating for Nonlinear Models and Radial Basis Function Neural Networks

Selective model structure and parameter updating algorithms are introduced for both the on-line estimation of NARMAX models and training of radial basis function neural networks. Techniques for on-line model modification which depend on the vector-shift properties of regression variables which are widely used for linear models cannot be applied when the model is nonlinear. The first is based on selectively updating the nonlinear model structure and this leads to a reduction in computational cost. The second is based on selecting specified regression variables to efficiently obtain nonlinear models for applications such as one-step ahead controller design. Efficiently on both simulated and real data sets are used to demonstrate the performance of the new algorithms.