I use Genetic algorithms to optimize the fuzzy logic system. Each individual in GA represents a fuzzy system that is a solution we need to find. In this representation, I use Gaussian membership functions for each rule. The boundary of two parameters of Gaussian membership functions are determined by [min,max] of training data. The obtained results are not good. I change the boundary that is subset of the [min, max] as the results that are sometimes obtained are good.

The question is this: Is there any methods to find out the good boundary of the parameters in this situation?