基于混合试验设计方法的支持向量回归机参数优化研究  

PARAMETERS OPTIMIZATION FOR SUPPORT VECTOR REGRESSION WITH HYBRID EXPERIMENTAL DESIGN METHOD

作　　者：杨希祥[1] 杨慧欣[1] 张为华 

机构地区：[1]国防科技大学航天与材料工程学院,长沙410073

出　　处：《机械强度》2012年第5期706-711,共6页Journal of Mechanical Strength

基　　金：中国博士后科学基金(200801491);国防科技大学科研计划项目(JC12-01-05)资助~~

摘　　要：研究高斯径向基核支持向量回归机参数优化问题。推导线性和非线性支持向量回归机公式,分析影响支持向量回归机精度的主要控制参数,将拉丁超立方设计方法与Powell法相结合,提出一种快速有效的支持向量回归机参数优化方法。将支持向量回归机用于近似建模,提供仿真算例,并与Kriging函数和径向基函数近似性能进行比较。结果表明,设计的支持向量回归机能实现近似精度和近似效率的良好折中,参数估计简单,易于编程实现,是有效的近似建模方法,可为飞行器多学科设计优化用近似建模方法研究提供理论参考。Parameters optimization of support vector regression with Gauss kernel function was researched. The formula of linear and nonlinear support vector regression was deduced, main control parameters affecting the accuracy of support vector regression were analyzed, and an efficient optimization algorithm for parameters selection of support vector regression was proposed by combining Latin hypercube design method with Powell method. Support vector regression was lead to the field of approximate modeling, simulation cases were provided, and comparison results with Kriging and radial basis function indicate that, the support vector regression designed can make a good balance between approximate accuracy and approximate efficiency, parameters evaluation is simple, and programming is easy, so it is an effective approximate method. The research can provide theoretical reference for the application of approximate method in multidisciplinary design optimization of flight vehicles.

关 键 词：支持向量回归机 参数优化 拉丁超立方设计 Powell法 近似建模 

分 类 号：TP302[自动化与计算机技术—计算机系统结构] TP18[自动化与计算机技术—计算机科学与技术]