A rule reduction methodology for fuzzy logic systems using modified orthogonal techniques

This paper presents a method for reducing the number of rules for large rule based fuzzy systems. While retaining the linguistic information of the original system, the reduction is achieved by attempting to select the most significant rules that have important contribution in the error reduction. Using a numerically efficient modified Gram-Schmidt orthogonalization routine, a measure called rule contribution in the error reduction ratio (RCERR), which allows the user to reorder rules according to their significance, is obtained. This algorithm provides substantial savings in running and implementation costs. In addition, it can find potential applications, especially in situations where knowledge of the order of rules is beneficial such as control, decision-making and socio-economic fields. The effectiveness of the proposed methodology is illustrated by considering some benchmark problems and comparing the results with existing works.