摘要
针对径向基函数(RBF)神经网络构造时其结构和参数难以确定的问题,结合可拓理论对输入样本和基函数的中心向量建立物元模型,并借鉴第2类型可拓神经网络(ENN2)的聚类思想,根据样本分布,采用可拓分析及可拓变换动态调整隐节点数目和基函数中心,从而提出基于可拓理论的RBF(ERBF)神经网络.同时,通过UCI标准数据集进行了测试,并通过应用实例进行了验证,结果表明,ERBF结构和参数的确定方法简单、收敛速度快,且泛化精度、鲁棒性和稳定性均显著提高.
During the construction process of radical basis function（RBF） neural network,the structure and parameters are hard to be determined.Therefore,combining with the extension theory,an extension theory-based RBF（ERBF） neural network is proposed,in which the matter-element models including input samples and center vectors of the basis function are established,the clustering method of extension neural network type 2（ENN2） is introduced,and the hidden layer nodes number and center vectors of the basis function are dynamically adjusted by using extension analysis and extension transformation according to the sample distribution.Meanwhile,UCI standard data sets are tested,and application object is validated.Through the verification and comparison,the proposed ERBF algorithm has the advantages of simple calculation and fast convergence,which significantly enhances the generalization accuracy,robustness and stability.
出处
《控制与决策》
EI
CSCD
北大核心
2011年第11期1721-1725,共5页
Control and Decision
基金
国家自然科学基金项目(60774079
61074153)
国家863计划项目(2006AA04Z184)
关键词
径向基函数神经网络
可拓理论
回归
建模
radial basis function neural network
extension theory
regression
modeling
