基于POD-RBF代理模型的迭代更新反演方法

Iterative updating inversion method based on POD-RBF surrogate model

针对常规的水工大坝等大型工程结构参数反演需要耗费大量有限元正分析机时的问题,建立了具有较好反演精度和泛化性能的POD-RBF代理模型和快速迭代更新反演算法。基于有限元分析获得足量数据样本,利用POD提取本征向量,并使用RBF方法进行插值得到有限元模型的代理模型;同时结合粒子群算法的全局寻优能力和高斯-牛顿法的快速局部收敛优势,建立了一种新的高效率迭代反演方法,并应用于混凝土大坝分区弹性模量反演。结果表明,该方法适用于大坝等大体积混凝土结构的力学参数反演。同时,相较于传统的单一反演方法,该方法在反演效率和反演精度两方面均显示出优势。

To address the high computation workload in finite element analysis in the general inverse analysis of large engineering structures like hydraulic dams, a precise and robust POD-RBF surrogate model and an iterative updating inversion algorithm were constructed.Based on the data samples obtained from the finite element method, we can interpolate eigenvectors abstracted by POD method using RBF to get a surrogate model of the finite element model.Combining the global optimization ability of particle swarm optimization algorithm and the fast local convergence advantage of Gauss-Newton method, a new efficient iterative inversion method was established and applied to the zonal elastic modulus inversion of a concrete dam.The results shown the applicability of this method for mechanical parameter inversion of large concrete structures such as dams.Meanwhile, compared with traditional methods, this inverse framework consisting of the POD-RBF surrogate model and the iterative updating method can get good performance both on the inversion accuracy and the efficiency of computation.