Hierarchical intelligent prediction system using RBF based AFS

In this paper we propose a hierarchical intelligent prediction system using radial basis function based-adaptive fuzzy systems (RBF based AFS). The proposed system employs a hierarchical structure that consists of low level modules, evaluation networks, and upper level judge modules. The RBF based AFS as the low level modules are presented according to different consequence types, such as constant, first order linear function, and general fuzzy variable. These provide versatility and generality to handle arbitrary fuzzy inference schemes for representing knowledge. An on-the-job classifier is used to evaluate the system's prediction performance (good or bad). The upper level judge modules use several blending techniques for multiple low level outputs such as mean, median, fuzzy, neural networks and neuro-fuzzy approaches. In simulation we present examples of chaotic time series predictions to illustrate how to solve these problems and to demonstrate its validity, robustness and effectiveness.