带多个核函数的多尺度径向基函数网络  

Multi-scale radial basis function networks with multiple kernels

作　　者：付丽华[1] 李宏伟[1] 张猛[2] 黄娟[1] 

机构地区：[1]中国地质大学数学与物理学院,湖北武汉430074 [2]华中师范大学计算机科学系,湖北武汉430079

出　　处：《华中科技大学学报（自然科学版）》2010年第1期39-42,共4页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(60672049);华中师范大学自然科学基金资助项目(2008A13)

摘　　要：构造了一种带多个核函数的多尺度径向基函数网络.这种网络结构同时使用多个核函数,能根据不同的数据分布特点,自适应地选取核函数的尺度与其他参数.利用k均值聚类方法得到核函数中心参数备选项,用经验确定核函数的尺度,用高斯函数与小波函数构造原子库,最后利用正交最小二乘算法训练网络.实验结果表明,与传统的单核径向基函数网络相比,新网络模型结构更稀疏、泛化能力更强.A multi-scaleradial radial basis function （RBF） network with multiple kernels was proposed. The new network structure adapts several kernel functions in a model. According to the different distributions of different datasets, the model size and the centre of each term were determined by a k-mean clustering method. And each kernel scale was adaptively selected. Orthogonal least squares algorithm was also used to train the network. The experimental results show that the new network is much more compact than the conventional RBF network with a single kernel, and the abilities of learning and generalization is enhanced greatly.

关 键 词：径向基函数网络 正交最小二乘 多核 小波 回归模型 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]