The Application And Study Of Fuzzy Neural Network

Fuzzy Neural Network combines the advantage of Fuzzy system and Neural Network system. It has so great superiority in processing problems of nonlinear. fuzzy and so on and so huge potentialities in the field of intellect information processing that more and more scholars and experts devote themselves to this field, and have made considerable headway. However most study of Fuzz). Neural Network are based on the Algorithms' establishment. improvement and perfection, few summarize documents make generality, which makes it difficult for those who first enter the field to understand the concept of Fuzzy' Neural Network and apply' it. The author collate summarize and study all kinds of relevant theory and knowledge basing on reading a lot of documents to make a systematic summary and some elementary researches on Fuzzy Neural Network.The paper is made up of tv~'o sections in fact. The first section makes a systematic summar\' to Fuzz' Neural Network. A new Algorithm is presented and implemented in the second section.FuZz\ Neural Network is a nev~ concept. The paper discussed its origin and growth from Fuzzy system and Neural Network's history, demonstrated the possibility and necessity of its generation, and presented the software and hardware Techniques on Neural Networks and Fuzzy logic in brief. It also defined the Fuzz~' Neural Network on the basis of Fuzzy Neural unit concept. discussed the Function approximation ability of Neural Network system and Fuzzy system, both of which can approximate continuing real functions with any degree of accuracy. A brief introduction to Algorithms and modeling which have mature theor'~' is made. Basin~ on the problems existed in the implement of Fuzz~' Neural Network, the author discuss networks' study ability', capacity and framework.A two-phase hybrid learning algorithm based Fuzzy Neural Network is proposed in this paper. In the first stage. a algorithm which combined Fuzzy Inference System and Genetic Algorithms is used to determine the subject functions' parameters needed according to the training data, the Genetic Algorithms is used to search the optimize solutions of parameters in the domain, the Sugeno Fuzzy' Inference System is selected as our inference modeling. In the second phase, the author determine the network' framework, according to the training data train network by the B? Algorithm tuning the network's bias and weights. For avoiding local minimal phenomenon and speeding up its convergence, the author select Gradient descent w/ntomentum & adaptive lr backpropagation as training algorithm.In order to generalize the application of our implementation of the Genetic Algorithm, the author constructs a C-H- general Genetic Algorithm class library' using some advanced technology of C-H- such as template and abstract class. It supports the optimization of one dimension or multi-dimension functions. While optimizing multi-dimension functions the lengths of genes of each dimension can be different or uniform. It supports solid or alterable aberrance rate and soliditeration times or ending iteration in certain condition.The author implements the algorithm using the Fuzzy and the Nnet toolboxes in MatLab. The simulation results show that the Algorithm has merit of high efficient, fast convergence and high modeling scheme.