Genetic Algorithm Research And Application Based On AFS Fuzzy Logic

AFS (Axiomatic Fuzzy Set) theory was firstly proposed by professer Xiaodong Liu in 1995. In AFS theory, AFS algebras and AFS structures, the fuzzy theory based on AFS algebras and AFS structures has been established. In essence, the AFS fuzzy system provides an effective tool to convert the information in the training examples into the membership functions and their fuzzy logic operations, and the membership functions and their logic operations are directly determined based on the distribution of original data. For ages, there exist the arguments for the basic issues exist in fuzzy set theroy about how to establish the membership function of the fuzzy concept with a rigor and uniform method and the accurate representions of the fuzzy logic operations. In order to deal with the above issues, AFS theory analyze and study these issues further. Now, AFS theory has been developed further and applied to fuzzy decision tree,credit rating analysis , pattern recognition and hitch diagnoses,etc.These theory studies and their applications show that AFS fuzzy logic theroy is more close to the thought of humanity than others.Genetic algorithms (GA's) are search algorithms that use operations found in natural genetic to guide the journey through a search space,it shows powerful capabilities for automatically designing fuzzy systems from data. Genentic algorithm has the significant theoretical meaning and practical value in the research fields of artificial intelligence, such as machine learning, data mining and intelligent control and so on.This paper proposed two kinds of genetic algorithm combining with AFS fuzzy theory, in order to solve the classification problem with two phrases.First, we built the AFS structure (M,τ, X) and define the Zadeh fuzzy membership funciton of fuzzy concepts by applying the AFS theory; second, based on the idea of adjusting the rule length,a new genetic algorithm is introduced to select some elite rule sets demanded by the second phrase for classification,and avoiding the the premature phenomena; finally, we modify the known multi-objective algorithm that used for classification to search the fuzzy rule sets which with the high accuary and good interpretability. Experimental results indicate that it is a feasible algorithm which is easy and adaptable to the evaluation of the performance,also has gain a good classification effect.