Support vector recurrent neurofuzzy networks in modeling nonlinear systems with correlated noise

Abstract

Good generalization results are obtained from neurofuzzy networks if its structure is suitably chosen. To select the structure of neurofuzzy networks, the authors proposed a construction algorithm that is derived from the Support Vector Regression. However, the modeling errors are assumed to be uncorrelated. In this paper, systems with correlated modeling errors are considered. The correlated noise is modeled separately by a recurrent network. The overall network is referred to as the support vector recurrent neurofuzzy networks. The prediction error method is used to train the networks, where the derivatives are computed by a sensitivity model. The performance of proposed networks is illustrated by an example involving a nonlinear dynamic system corrupted by correlated noise.