LSSVM改进测地距离的核函数算法研究  

Research on the Improved Geodesic Distance Algorithm of LSSVM Kernel Function

作　　者：吴登国[1] 黄宴委[1] 李竣[1] 

机构地区：[1]福州大学电气工程与自动化学院,福建福州350108

出　　处：《自动化仪表》2011年第12期5-8,12,共5页Process Automation Instrumentation

基　　金：福建省自然科学基金资助项目(编号:2010J050132);福建省教育厅科技基金资助项目(编号:JA10034)

摘　　要：针对满足一定形状的非线性不均匀分布样本点的分类与拟合问题,提出一种改进的测地距离算法。该算法综合利用传统的k近邻法和ε半径近邻法来确定样本点的近邻关系点,提高了计算样本点测地距离的精确性。将算法应用于最小二乘支持向量机的核函数,通过数据分类仿真测试以及在结构健康检测中丢失数据重构的回归应用,提高了分类与拟合的精度。最后,经试验证明了所提算法的优越性。For the classification and fitting for specimen points in nonlinear and inhomogeneous distribution and meet certain sharps, the improved geodesic distance algorithm is proposed. The algorithm determines the neighbor points of the specimen points by combining the traditional k-nearest neighbor method with e-radius neighbor method, thus enhances the precise of calculation of geodesic distance for specimen point. The algorithm has been applied to the kernel function of least square support vector machine （LSSVM）, through data classification simulation test, and reconstruction of lost data in the structural health detection, and the precise of classification and fitting is enhanced. Finally, the experiments verify the superiority of the proposed algorithm.

关 键 词：K近邻法 ε半径近邻法 最小二乘支持向量机 数据重构 数据分类 核函数 

分 类 号：TP274.2[自动化与计算机技术—检测技术与自动化装置]