An optimal design approach for fuzzy inference system from data

The design process of a fuzzy inference system from data can be divided into two stages: the structure identification and the structure optimization. In this paper, the fuzzy system performance is optimized by partition refinement. A three-step method is employed to build a fuzzy inference system from data: Step 1 initiates the membership functions and system rules with a simple topology. Step 2 fine-tunes the input membership function parameters with data. Step 3 adds a new fuzzy set on the input region, which is responsible for the greatest part of the error. It is an iterative process from step 1 to step 3. Finally, this algorithm will generate different fuzzy system structures, which are with the different accuracy of the approximation and the different complexity of the rule set. It can selects from the different structures to obtain a fuzzy system that providing the best compromise between the accuracy and the complexity. The simulation results are compared with the equally partitioned fuzzy inference system.