Knowledge Extraction in Fuzzy Relational Systems Based on Genetic and Neural Approach

Abstract

In this paper, a problem of MIMO object identification expressed mathematically in terms of fuzzy relational equations is considered. We use the multivariable relational structure based on the modular fuzzy relational equations with the multilevel composition law. The identification problem consists of extraction of an unknown relational matrix and also of parameters of membership functions included in the fuzzy knowledge base, which can be translated as a set of fuzzy IF-THEN rules. In fuzzy relational calculus this type of the problem relates to inverse problem and requires resolution for the composite fuzzy relational equations. The search for solution amounts to solving an optimization problem using the hybrid genetic and neural approach. The genetic algorithm uses all the available experimental information for the optimization, i.e., operates off-line. The essence of the approach is in constructing and training a special neuro-fuzzy network, which allows on-line correction of the extracted relations if the new experimental data is obtained. The resulting solution is linguistically interpreted as a set of possible rules bases. The approach proposed is illustrated by the computer experiment and the example from medical diagnosis.