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 fc-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.
展开▼
