Online data-driven fuzzy modeling for nonlinear dynamic systems

In this paper, a new method for online learning of Takagi-Sugeno (T-S) model from input-output data is presented. It is based on a novel learning algorithm that recursively updates T-S model structure and parameters by combining supervised and unsupervised learning. The rule-base and parameters of the T-S model continually evolve by adding new rules with more summarization power and by modifying existing rules and parameters. To reduce the complexity of fuzzy models while keeping good model accuracy, some approximate similarity measures and simplification methods are presented. Using these methods, the redundant fuzzy rules are removed or merged. The simplified rule base is computationally efficient and linguistically interpretable. The consequent parameters of the T-S model are identified and optimized by Kalman filter. The approach has been successfully applied to T-S models of non-linear function approximation and dynamical system modeling.