基于支持向量回归机的结构非概率可靠性分析

Structural non-probabilistic reliability analysis using support vector regression

将支持向量机回归技术引入隐式极限状态结构的非概率可靠性分析,基于未确知信息的分段均布描述模型,设计了训练样本抽取策略.为了统一样本尺度,根据分段均布模型与标准化区间均布模型的双射关系,将基本变量区域中的样本数据转化成标准区间变量域中的样本数据,保证了支持向量机训练的稳定性.给出了SVR预测模型算法,并实现了在标准化区间变量域中直接抽取训练、测试及预测样本,使得样本抽取和蒙特卡罗模拟计算更便于实现.通过算例对方法的精度和可行性进行了验证,结果表明:该方法可解决隐式极限状态结构的非概率可靠性分析问题,且应用简便.

The regression technology of support vector machine（SVM） was introduced to analyze the non-probabilistic reliability of structures with implicit limit state function.Based on the fragment description model of unascertained information,the training data sampling method was proposed.The sample data in basic variable range was transformed to that in norm interval variable scale,and the dimensions of training samples were unified.So the stability of SVM could be assured.The algorithm was offered,and the training,test and prediction sample could be directly drawn in norm interval scales.So the sample drawing and Monte Carlo simulation became easier to realize.The accuracy and feasibility of this methodology were proved through two given examples.The problem of structural non-probabilistic reliability with implicit limit state function can be solved using this technique,and which is easy to use.