基于Kriging代理模型的迭代更新高效反演方法

为提高混凝土坝等大体积结构参数反演效率和精度,减少由于应用有限元进行大量正分析而产生的计算机时,建立了一种结合Kriging代理模型和粒子群优化(PSO)算法的迭代更新反演方法。通过拉丁超立方抽样(LHS)方法确定初始样本点的空间分布,并使用有限元正分析获取对应的响应值,构建粗糙的初始代理模型,结合具有全局寻优能力的PSO算法,反演大体积结构的分区弹性模量,随之再代入有限元模型中,计算获取新的位移响应,并将其作为新样本加入到样本集中,通过迭代更新获得局部更高精度的代理模型。工程实际算例表明,该方法对混凝土坝等大体积结构参数反演精度较高和适用性好,且能大幅减少传统有限元模型反演方法所需消耗的正分析机时,提高反演效率。

Efficient iterative updating inversion method based on Kriging surrogate model

To improve the efficiency and accuracy of inverse analysis of massive structures like a concrete dam,as well as to reduce the computational cost of finite element analysis,a new efficient iterative updating inversion method was constructed based on the recently developed Kriging surrogate model and particle swarm optimization (PSO) algorithm.The spatial distribution of the initial sample points was determined by the Latin hypercube sampling (LHS) method,and the corresponding responses were obtained by the finite element method to produce the samples with which a roughly initial Kriging model can be constructed.The regional elastic modulus of the massive structures is determined by using a combination of the surrogate model and PSO algorithm with excellent global optimization performance.The parameters obtained from the inverse analysis are applied in the subsequent finite element analysis to get the corresponding displacements and combining them as a new sample adding to the sample set to update the surrogate model for higher accuracy.Numerical examples show that the proposed hybrid method is suitable for inverse analysis of massive structures such as a concrete dam with higher accuracy and efficiency while remarkably reducing the computational cost of normal analysis in traditional model-based inverse analysis.