结构可靠性分析的LCVT-SVR方法

Structural reliability analysis with LCVT-SVR method

为了克服结构可靠性分析中隐式极限状态函数和高计算量问题,提出LCVT-SVR结构可靠性分析方法.该方法利用拉丁质心Voronoi网格化(LCVT)抽样方法生成空间均匀性较好的训练样本,采用均匀映射得到一定数量的失效样本.基于训练样本,建立极限状态函数的支持向量机回归(SVR)代理模型,用于结构可靠性分析.在同一组SVR训练参数条件下,对多种抽样方法进行对比研究.结果表明,基于LCVT样本构建的SVR代理模型具有精度高和鲁棒性好的特点.利用2个工程算例,验证了所提方法的性能及实用性.

A LCVT-SVR method was proposed for structural reliability analysis in order to solve the problems of implicit limit state function and high computational effort.The method adopts the Latinized centroidal Voronoi tessellation(LCVT)sampling method in order to generate the training samples with better uniform coverage.The uniform mapping was employed to obtain a certain number of failure samples for a proper training process.The investigated limit state function(LSF)was approximated based on these training samples by a support vector regression(SVR)surrogate model which was applied to structural reliability analysis.Various sampling methods were compared under the same setting of SVR training parameters.The computational results show that the SVR surrogate model based on LCVT samples has high accuracy and robustness.The performance and applicability of the proposed method were validated by two engineering examples.