Data-Driven Structure Identification of Takagi Sugeno Fuzzy Models Using a Bounded Error Approach

Takagi Sugeno fuzzy models are a widely used model class for data driven identification of nonlinear systems. Besides good approximation quality, the local model approach allows for good interpretability and the utilization of linear control strategies in dynamic models. However a major challenge is the identification of the model structure, i.e. determining the number of local models and the partitioning parameters. A new method for identification of Takagi Sugeno fuzzy models inspired by bounded error methods is presented. The method finds partitions of the data that are defined by bounding linear inequalities and therefore identifies locally linear system behavior. The error bound as a tuning parameter is used to adjust the compromise between model accuracy and sparsity. The presentation of the method is complemented by two case studies.