Radial basis function(RBF)neural network is a kind of local approximation of three layer forward neural networks,compared to the other forward neural network has a simple structure,fast convergence rate and won't get into local minimum point,by the great attention and has been widely applied in many fields.In the process of the construction of the RBF neural network,the use of traditional k-means clustering method to determine the center of the basis function learning algorithms need to give the initial clustering center in advance,when given the initial clustering center is not at the same time,get the basis functions of center may be different,lead to unstable network training results,and the number of neurons in hidden layer network need is given in advance,but often the network structure is not predetermined.In order to solve this problem,this paper puts forward the center of the base function determined by system clustering method,so as to effectively solve the RBF neural network is sensitive to the initial clustering center of the problem.This article first introduces the basic principle of RBF neural network,RBF neural network for different structure and performance are analyzed,and points out the characteristics of various network and the problems needing attention.Some common used RBF neural network learning algorithm is studied,analyzes several methods to determine the basis function center work processes and their respective advantages and disadvantages.Analysis the basic principle and operation steps of system clustering,this paper introduces the determining basis function center calculation in the process of sample spacing and spacing of a variety of methods,according to the changes of the class in the process of clustering distance clustering stop condition is given,and describes its basic ideas and operation method.Will use the system clustering to determine the center of the basis function method is applied to neural network building,introduces the detailed steps of network training and improvement process.Based on the theory to improve the method of program design,with examples to validate the effectiveness of the improved method,finally obtained the main results are as follows:(1)To study and put forward the system clustering is used to determine whether a new method of radial basis function center,and gives the detailed calculation model and steps of this method.Will this method compared with other methods,analysis the superiority of this method are given.Through the analysis of the principle and process of system clustering,proved that the new methodis compared with the traditional method does not need to advance the center of the basis function are given initial point,effectively avoid the network structure is sensitive to initial value selection basis function center.(2)The study presents a new method to determine the number of radial basis function.In the study of various clustering system and class spacing calculation method on the basis of sample spacing,presents a system using clustering to determine the center of the radial basis function iteration class interval,the change rule of put forward with the relationship between the class interval variation as a condition of iteration is stop judgment,proved that the new method is no longer need the number of neurons in hidden layer is given in advance,can the construction of the network structure of the organization.(3)Through the programming to realize the algorithm,prove the feasibility of the algorithm.Through to the traditional based on k means clustering method and improved method of the MATLAB implementation of neural network and MATLAB toolbox provides the function of comparison,proved the feasibility of the improved method.(4)By using three examples to verify the improved method proposed in this paper is effective in solving practical problems.At the center of the basis function were determined by system clustering method to build the RBF neural network was applied to function approximation,classification and time series prediction problems,obtained the good effect. |