An Approach Towards the Design of Interval Type-3 T-S Fuzzy System

This article providesa systematic approach for the design of an interval type-3 (IT3) Takagi-Sugeno (T-S) fuzzy logic system (FLS) using alpha- plane representation. An IT3 FLS is designed with the baseline of the general type-2 (GT2) FLS in a similar manner as an IT2 FLS was designed from the baseline of type-1 FLS. Hence, IT3 FLS evolved as a successor of GT2 FLS, where secondary membership function is an interval type-2 fuzzy set (FS), and values of tertiary membership are unity over the footprint of uncertainty of secondary membership. This extra degree of freedom in IT3 FLS provides better modeling capability as compared to GT2 FLS in the presence of a high degree of uncertainty and vagueness. The proposed system will be more appealing while dealing with uncertain information or data, which is supposed to be generated from uncertain sources; i.e., there exist uncertainties even in the presence of uncertainty. The computations needed for the design of IT3 FLS are derived using IT2 FS and GT2 FS mathematics. The design algorithms adopted for the baseline IT2 T-S fuzzy system are as per the modified interval type-2 fuzzy c-regression model clustering algorithm and hyper-plane-shaped membership function. The proposed methodology is applied to several benchmark examples and obtained results are compared with recently developed fuzzy modeling methods having a comparable number of rule bases. The proposed IT3 T-S FLS shows good performance in terms of accuracy when data is corrupted by noise and uncertainties related to missing or unvarying data exist. The computational cost is linear with design parameters and by optimum choice of alpha- planes, it is still bearable considering advantages and nature of applications.

Automated summary

This article presents a systematic approach for designing IT3 T-S fuzzy logic systems using alpha-plane representation. The proposed system has better modeling capability than GT2 FLS and is more appealing when dealing with uncertain information or data. The computational cost is bearable and the method performs well in benchmark examples.