Neuro-netwoks approach to the generation of the connected fuzzy knowledge data bases on rules and relationships

Abstract

The adaptive approach to generating composite IF-THEN rules based on the genetic and neural algorithm of solving fuzzy relational equations is proposed. It allows us to avoid rules selection and eliminate overlaps between classes. The essence of the approach is in constructing and training the specific min-max neuro-fuzzy network isomorphic to linguistic solutions of fuzzy relational equations, which allows adaptation of the rules set structure while the output classes’ bounds are changing. Resolution of fuzzy relational equations guarantees the optimal number of fuzzy rules for each output fuzzy term and the optimal geometry of input fuzzy terms for each linguistic solution.