Selection of the structure of radial basis functions networks
| Authors: | Chan Che Wai, The University of Hong Kong, Hong Kong
Choy K. Y., The University of Hong Kong, Hong Kong |
|---|
| Topic: | 3.2 Cognition and Control ( AI, Fuzzy, Neuro, Evolut.Comp.) |
|---|
| Session: | Intelligent Modelling and Identification II |
|---|
| Keywords: | Support vector regression, orthogonal least squares, radial basis function networks, nonlinear systems |
|---|
Abstract
The performance of radial basis function networks (RBFN) depends on the choice of its structure. The orthogonal least squares algorithm (OLS) and the support vector regression (SVR) are two popular approaches to choose the structure of the RBFN. The former considers only the modelling errors, whilst the latter also the model complexity. In this paper, a compari-son of the generalization results of networks selected from the OLS and the SVR is presented using a simulated nonlinear system, and river discharges and rainfall data of Fuji River. It is shown that the network based on the SVR performs better than that based on OLS, illustrating the importance of taking model complexity into account in the structure selection of the RBFN.
