摘要
本文将支持向量分类机(SVC)引入到结构可靠度计算分析中,采用拉丁超立方抽样法进行初始输入训练样本的实验设计。将支持向量分类机作为响应面函数,并利用遗传算法进行参数优化,最后结合蒙特卡罗模拟提出了基于支持向量分类机的改进响应面法。其主要思想为:定义"重要性"判定函数,在迭代过程中,按判定函数值从抽样样本中选取新的训练样本,使支持向量分类机的模拟功能函数在抽样点分布区域内能更进一步地接近真实功能函数,从而大大提高可靠度分析的精度以及效率。
The support vector classification machine-SVC was introduced into the study of structure reliability. The experiment design of initial input training sample was implemented by using LHS method, the SVC model was treated as the response surface function, and the parameters were optimized by using genetic algorithm. Combined with Monte Carlo simulation method, the improved response surface method which based on the SVC was proposed, in which the main approach is ： defining an ＂important＂ critical function, a few of test sample points were selected as the new additional training points according to the value of critical function, and the approximation of SVC would be more close to the real limit state function in the region which has a greater contribution to the failure probability, so that the accuracy and efficiency of reliability analysis will be improved rapidly.
出处
《科学技术与工程》
2011年第31期7714-7720,共7页
Science Technology and Engineering
关键词
支持向量分类机
遗传算法
拉丁超立方抽样
改进响应面法
蒙特卡罗
可靠度
support vector classification machine genetic algorithm LHS improved response surface method Monte Carlo reliability
