摘要
该文提出了基于聚类分析的RBF(Radial Basis Function)网络建模方法:利用聚类分析确定RBF神经网络的隐层参数, 运用最小二乘法确定RBF神经网络的输出层参数。重点介绍了聚类分析的理论和算法。根据聚类分析和RBF网络结合后的优点以及中医证候大数据、大样本、多中心且无明确函数关系的特性,提出了用该方法建模应用于中医证候诊断,改进了 BP(Back Propagation)网络用于中医证候诊断建模的不足之处,并拓宽了RBF神经网络的应用。最后,用2-型糖尿病文献数据库验证了该方法的有效性和合理性。
The paper presents a modeling method of RBF Neural Network based on clustering analysis: the parameters of the hidden layer are defined by means of clustering analysis, the arguments of the linear output layer are determined by way of Least Squares Algorithm. Especially it introduces the theory and algorithm of clustering analysis. according to the advantage of RBF Neural Network based on clustering analysis and the characteristic of TCM syndrome with multi - data, multi - sample, multi - center, non - functiong relation. In this paper we put forward the modeling of Traditional Chinese Medicine(TCM) syndrome diagnosis based on the method, improve the modeling of TCM syndrome diagnosis by using the BP Neural Network, also broaden the application of RBF Neural Network. Finally, the result tested by the document Data Base of type - 2 diabetes mellites verifies the validity and reasonableness of the method.
出处
《计算机仿真》
CSCD
2006年第1期120-123,共4页
Computer Simulation
关键词
聚类分析
径向基神经网络
最小二乘法
中医证候
Clustering analysis
RBF neural network
Least squares algorithm
TCM syndrome