基于SVR方法的有限元模型确认中不确定性建模研究

SVR based uncertainty modeling method for structures finite element model validation

针对模型确认中结构有限元模型不确定性建模问题,采用D最优实验设计方法生成无明显分布规律的样本空间,提出用支持向量机回归方法建立结构响应和不确定参数之间的映射关系,给出了基于SVR方法进行不确定建模的一般步骤,引入了三级检验对支持向量机响应面模型的精度和泛化能力进行全面评价,以GARTEUR模型为仿真算例,比较了向量机响应面模型和幂多项式响应面模型,结果证明了该方法对模型的不确定性建模问题具有较高的精度,具有较好的推广性。

Aiming at the uncertainty modeling of FE model for structure,necessary sample space was provided by use of D-optimal experiment design method.Supported vector regression(SVR) method was presented to construct the implicit mapping between structure response and uncertainty parameters.The computational process of forward transfer of uncertainty parameters has been clarified.Three levels testing criteria were used to evaluate extensively the precision and generalization ability of RSM based SVR.Comparing with the polynomial RSM,the efficiency and accuracy of the proposed method were proved by the case study of GARTEUR.