Fuzzy Granular Neural Network for incremental modeling of nonlinear chaotic systems

Abstract

Evolving intelligent systems are useful for processing online data streams. This paper presents an evolving granular neuro-fuzzy modeling framework and an application example on the modeling of the Rossler chaos. The evolving Granular Neural Network (eGNN) is able to deal with new events of nonstationary environments using fuzzy information granules and different types of aggregation neurons. An incremental learning algorithm builds the network topology from spatio-temporal features of a data stream. The goal is to obtain more abstract representations of large amounts of data, and thereafter provide accurate one-step predictions and insights about the phenomenon that generates the data. Results suggest that eGNN learns successfully from a data stream generated by the Rossler nonlinear equations. Additionally, eGNN has shown to be competitive with state-of-the-art data-driven modeling approaches.