Aprendizado genético de funções de pertinência na modelagem nebulosa.

Abstract

The success of Fuzzy Rule Based Systems is due to their ability to represent vague and uncertain knowledge and to the facility to express the system behavior in a language easily interpretable by human beings. The knowledge acquisition of a Fuzzy Rule Based System is done from a human expert, which is known as the direct approach to knowledge acquisition, or through methods that automatically extract this knowledge from numerical information that represent samples or examples of the problem. Genetic Algorithms have demonstrated to be a powerful tool to rule bases construction, rule bases optimization, membership functions generation and membership functions optimization. The main research focus of this work is the investigation of fuzzy systems automatic generation approaches applied to pattern classification problems, using genetic algorithms to the definition and tuning of fuzzy sets that belong to the fuzzy partitions of the domains involved, considering interpretability maintenance of the linguistic values. The genetic learning is employed only in the fuzzy system data base, that is, in the membership functions, either through the traditional way of tuning previously defined functions after the rule base has been defined or through the more recent approach of constructing membership functions before the rules definition. This kind of investigation requires the use of different rule generation methods, the rules being generated before or after the use of genetic algorithm. The combination Fuzzy Systems + Genetic Algorithms, known as Genetic Fuzzy Systems, has great acceptance in the scientific community, since this systems are robust and capable of finding good solutions in complex and irregular spaces.