Research On Radial Basis Function Center Determination Based On Support Vector Machine

Radial Basis Function(RBF)neural network is a three-layer forward network with single hidden layer.It is widely used in pattern recognition,image recognition,natural language processing and other fields because of its advantages such as simple structure,fast training speed,and effective avoidance of local convergence.Determining the number,location,and width of each basis function center is the key to the design of a RBF neural network.Generally,it is necessary to determine the basis function center and their related parameters by a clustering algorithm.,where the most common application is the K-means clustering.But K-means needs to set a number of the initial condensation points in advance,which is difficult to achieve for many practical problems.The RBF neural network's structure constructed by the K-means clustering is particulary sensitive to the selection of initial condensation points and abnormal values.In view of this problem,some scholars have proposed a system clustering method to determine the center of the basis function,but system clustering and K-means clustering are essentially belonging to the greedy algorithms.The decision making in each step is the best for the current state while the final solution is probably not the global optimal solution,and the system clustering algorithm is also affected by the outliers and the clustering result is likely to be chains.In view of the above problems,the Support Vector Machine(SVM)clustering is proposed to determine the basis function center of RBF neural network in this paper,which makes up for the deficiency of traditional clustering algorithms such as K-means clustering and system clustering.This paper first introduces the structure,principle and learning algorithm of RBF neural network,and briefly introduces several methods to determine the center of the basis function,and introduces the basic principle of SVM.Secondly,the theoretical basis of the application of SVM in the classification problem is introduced,and the example of the classification effect of the iris data set with different kernel functions is given.Finally,the method of determining the center of basis function based on support vector machine clustering is applied to the construction of RBF neural network.And its performance is experimentally compared with the RBF neural network trained by system clustering,which performes well in this respect.The effectiveness of improved method has been verified.The main resarch results obtained in this paper are as follows:(1)This paper presents a new method to determining the radial basis function center of RBF neural network by using SVM clustering,and gives the calculation method and related model of this method.By analyzing the principle and process of the SVM clustering,and comparing with the RBF neural network trained by sysrem clustering,the conclusion is drawn that the network trained by SVM clustering is better in convergence speed,fitting accuracy,classification accuracy and prediction accuracy.(2)The feasibility of the algorithm is proved by programming,and the RBF neural network based on SVM clustering is designed and implemented by using MATLAB and PYTHON language.(3)Three examples are given to demonstrate the effectiveness of the RBF neural constructed by SVM clustering in soving agricultural engineering practical problems.The RBF neural network based on SVM clustering is applied to function fitting problem,classification problem and the nonlinear time series prediction problem,and a good result is obtained.The improved RBF neural network is compared with the RBF neural network constructed by the system clustering,which proves the feasibility and effectiveness of the new method.