MVLSC HomeIssue Contents

Granular RBF Neural Network Implementation of Fuzzy Systems: Application to Time Series Modeling
Milan Marcek and Dusan Marcek

In this study, we are concerned with fuzzy systems for mapping input fuzzy sets to output fuzzy sets. TheRBF(Radial Basic Function) network or other neural networks are endowed with some properties that make them more flexible and logically appealing. At first, we discuss the basic structure of the fuzzy system. RBF neural network architectures are proposed as techniques for performing fuzzy logic inference in fuzzy systems. Then,
we show a new approach of function estimation for time series model by means of a granular RBF neural network based on Gaussian activation function modeled by cloud concept. The learning aspects of RBF networks
are presented in accordance to supervised learning in which the rule weights are adjusted following the gradient of a certain objective function. An application is included to illustrate the approximation performance of these approaches.

Keywords: Probabilistic time-series models, fuzzy system, classic and soft RBF network, cloud models, granular RBF network, supervised learning.

Full Text (IP)