A Design of Direct Granular Model based on Takagi-Sugeno Fuzzy Model

Abstract

In this paper, a direct granular TS fuzzy model (DGFM) trained from numerical data is proposed. The DGFM extracts a set of weighted data from the rules of numerical TS model and generates information granules from weighted data as the output of the granular model. The DGFM can evidently reduce number of parameters in model training process and can reduce computational complexity during model construction and derivation. In addition, an improved performance index for the granular model is designed, which can reflect the deviation between the output information granule and the target value by improving the coverage criterion and making the evaluation of the granular model more comprehensive. Experimental results are reported for both synthetic data and real-world datasets to show the performance of the DGFM.