支持向量机回归在矿区GPS高程转换中的应用

Application of Support Vector Regression in GPS Height Transformation in Mining Area

基于统计学习理论和支持向量机原理,提出了支持向量机回归应用于矿区GPS高程转换的方法用以精化矿区似大地水准面,研究了支持向量机回归、多项式、GA-BP神经网络3种模型在GPS高程转换中的应用,结果表明,支持向量机回归拟合数据的精度优于多项式和GA-BP神经网络,并且有效地解决了神经网络拓扑结构选择困难、过学习、无法避免局部极值等问题。

Based on the theory of statistics and the principle of Support Vector Machines（SVM）, it is put forward that Support Vector Regression（SVR） is applied into the GPS height transformation in mining area to refine the quasi-geoid in mining area. And investigation on the application of Support Vector Regression, polynomial and GA-BP neural network in GPS height transformation are carried out. The test results show that the precision of fitting data of SVR is superior to that of polynomial and GA-BP. Moreover, SVR can effectively overcome many shortcomings existing in neural network, such as difficult topology selection, over-learning, inevitable local extremum etc.