Using The Fuzzy Clustering To Improving The Membership Of The Rule Layer Of The Fuzzy Neural Network

The advance prediction of high dimensional data has become a hot research topic.Neural networks and fuzzy systems are two common methods for predicting nonlinear functions of uncertainty,but there are some defects in both.The fuzzy system lacks self-learning and adaptive ability;Neural network is difficult to express,and its ability to approximate data is limited by their network structure.Fuzzy neural network is a kind of fuzzy logic in the aspect of knowledge structure with neural network self-learning ability of combining the advantages of a local approximation network,deal with the neural network fuzzy data-processing and fuzzy logic in the study of related problems,can make it more perfect processing nonlinear problems related.Detailed introduces the related theory and model of fuzzy neural network,is used to show that fuzzy neural network is determined by the output of the rule layer learning mechanism,the approximation capability of the data depends on its network structure and the membership functions;When processing high-dimensional nonlinear data,the rules layer of fuzzy neural network can become complex and large,which seriously affects the operation speed of fuzzy neural network and reduces the computing power of network.This article will research on the cluster analysis of the rule layer,rule layer by clustering carry less information in some structured together,and will these nodes into zero,use any number and the characteristics of zero multiply all zeros,to reduce the complexity of the rules of the whole layer,forming a new easier calculation rules of the layer.In specific research compared the K-means and fuzzy clustering of two different clustering method,through comparing the clustering results found:K-means clustering results is relatively coarse,and the initial value of K is difficult to determine,therefore decided to adopt the fuzzy clustering analysis to the rule layer clustering neat;By using MATLAB to set up a forecast of the nanjing Yangtze river water level of the t-s fuzzy neural network model,the original rules of fuzzy neural network and improved after layers of fuzzy neural network are applied to the prediction of water level in the Yangtze river,through the comparison of experimental results show that this improved method can be in the case of little impact on prediction accuracy to speed up the computing speed of the fuzzy neural network.