基于支持向量机的径向基网络结构优化

RBF Neural Network structure optimization method based on SVM

为了解决径向基网络(RBFNN)结构设计的随机性,进一步优化RBF网络性能,提出一种基于支持向量机(SVM)的径向基网络结构优化方法。通过训练得到的SVM确定径向基网络的隐层节点个数、隐层权值和阈值;同时利用SVM对输入向量进行特征变换,进一步对输入向量进行维数约简。通过齿轮箱的故障诊断实验表明,优化后的RBF网络具有更精简、稳定的网络结构,能得到更准确的诊断结果。

An approach of Radial Basis Function Neural Network（RBF NN） optimization based on support vector machine was proposed to solve the randomness of the network structure and the unstableness of the network＇s performance.The number of the hidden neurons,the value of the weight and bias in RBF neural network are determined by the trained SVM.The approach is adopted in modeling and diagnosing the fault of gear case.In the experiment,the input vectors are transformed with the SVM character extraction approach firstly and then trained with the optimized RBF.Experimental results indicate that the optimized RBF network can obtain stable performance and more accurate diagnosis.