Study On Fuzzy Neural Network Based On Immune Genetic Algorithm

This article is accomplishmented by combining to "Reserch on genetic artificial intelligence control theory and apllicaton".Fuzzy logic,neural network,genetic algorithm are three important domains that have the best development future in the present artificial intelligence, moreover, artificial immune algorithm which has been developed and received more attention gradually in recent years also becomes an important one among them . They play more and more important roles in the domains of system modelling,pattern recognition,combination optimize,adaptive control and so on.But each algorithm itself still has such-and-such insufficiency, consummating the algorithms and exerting their better potential have the universal and profound significance. Among them, syncretizing one intelligent method to another or to other domain algorithms will make up for one's deficiency by learning from others' strong points mutually or solute certain difficult problems in algorithm design ,which has become the main research direction from now on.This article summarizes the basic principle,algorithm constitution, characteristics of fuzzy control,neural network,genetic algorithm and artificial immune algorithm, the syncretized way among them . In this foundation, studys the following four syncretized ways emphatically: first is the introduction of immune algorithm principle and fuzzy logic into genetic algorithm for solving the problem of precocious convergence cased by rapid drop of its population deversity ; second is the introduction of improved genetic algorithm into the specific algorithm - fuzzy c-mean(FCM) clustering in specific domain - fuzzy clustering, for improving its shortcoming that it is sensitive to initialization, easy to fall into local convergence; third is the introduction of the improved genetic algorithm and FCM clustering into fuzzy neural network which is the syncretized production between fuzzy logic and neural network,for determining rule numbers automatically and optimizing parameters in system identification; fourth is the introduction of improved genetic algorithm into one kind of commonly used fuzzy neural network controller which is based on Mamdani inference, for optimizing its structure and parameters simultaneously, and maintaining the integrity between membership functions and control rules, solving the problem "dimension disaster" effectively.In addition, this article makes new analysis,improvement as well as simulation confirmation in each syncretized way. The first is completed based on the foundation of thorough analysis and reserch on genetic algorithm and the insufficiency of similar algorithms in correlation literature . In the second way,designs a new syncretized way -double groups parallel, the algorithm not only increases the cluster precision, but also speed up greatly. In the third way, the rule numbers can be detemined automatically as result of joining the improvement FCM cluster algorithm, overcomes the shortcoming of trying again and again in experiment before, and designs the way of off-line training and online identification to overcome the contradiction between static state and dynamic state existed in fuzzy neural network when modeling. In fourth way, in order to optimize the structure and parameters of fuzzy neural network simultaneously, it designs a new genetic algorithm code method - real number and the mark mix code as well as new way of crossover and mutation, designs the process of online parameter identification based on analyzing the influnce that the quantification and proportionality factors on static and dynamic state of controller.Moreover, performs simulation experiment to each kind of new design and the improvement, and carries on comparison with other correlation method, confirms the superiority of algorithm in this article.