Modelling of nonlinear dynamic systems using support vector neurofuzzy networks

摘要

Though neurofuzzy networks can approximate a nonlinear system with arbitrary accuracy, good generalization results are obtained only if the structure of the network is suitable chosen. It is therefore important that there are reliable techniques to select the 'best' structure of the network. The Support Vector Neurofuzzy Network (SVNFN) derived here from the Support Vector Regression (SVR) has the special feature that its structure is selected for the error bound specified by the user, similar to that of the SVR. As the output of the network is linear in its weights, the weights can be estimated by the linear least squares method. Important properties of the SVNFNs, such as bounded modelling error and unbiased estimate of the dynamics of the system are also derived. To illustrate the performance of the SVNFN, it is applied to model a nonlinear system.