基于谱元边界积分法的二维跨分层高效电磁全波反演研究

    Efficient Electromagnetic FWI of 2D Objects Straddling Multiple Planar Layers Based on The SEBI Method

    • 摘要: 本文开发了一种基于谱元-边界积分(Spectral Element-Boundary Integral, SEBI)和列文贝格-马夸尔特算法(Levenberg-Marquardt, LM)的高效混合方法来解决跨越多个水平分层的二维各向同性和各向异性散射体的电磁(Electromagnetic, EM)正演与反演问题。在正演计算中,采用一维光滑边界包围跨越多个水平分层的散射体,在区域内部使用谱元法(Spectral Element Method, SEM)求解二维电磁场分布;同时在边界上使用边界积分方程(Boundary Integral Equation, BIE)作为精确的辐射边界截断SEM求解域。针对横电(Transverse Electricity, TE)和横磁(Transverse Magnetic, TM)两种模式,分别用BIE求解等效电流\磁流,最后结合分层二维并矢格林函数计算接收器上的散射场。在全波反演(Full Wave Inversion, FWI)中,为了分别重建TE模式中各向同性散射体模型参数和TM模式中各向异性散射体模型参数,分别计算质量矩阵相对于各向同性模型参数的一阶导数和刚度矩阵相对于各向异性模型参数的一阶导数,并构建敏感度矩阵。随后,采用LM优化算法迭代调用SEBI正演求解器完成反演过程。最后通过多组数值实验验证了所提出的SEBI-LM算法在正演计算和FWI中的准确性和高效性。

       

      Abstract: This paper developed an efficient hybrid algorithm based on the spectral element-boundary integral (SEBI) method and the Levenberg-Marquardt (LM) method for solving both the electromagnetic (EM) forward and inverse scattering from isotropic and anisotropic objects straddling multiple planar layers. In the forward scattering computation, a one-dimensional smooth boundary is used to enclose the scatterers straddling multiple layers. The spectral element method (SEM) is employed to solve for the 2D EM field distribution within the internal domain. Concurrently, a boundary integral equation (BIE) is applied on the boundary, serving as an exact radiation condition to truncate the SEM domain. This hybrid SEBI method is used to compute the equivalent electric/magnetic currents on the boundary for both transverse electric (TE) and transverse magnetic (TM) modes. Finally, they are combined with 2D layered-medium dyadic Green's functions to compute the scattered fields at the receiver array. In the full-wave inversion (FWI), to reconstruct the model parameters of isotropic scatterers in the TE mode and anisotropic scatterers in the TM mode, respectively, we computes the first-order derivatives of the system's mass matrix with respect to the isotropic parameters and the first-order derivatives of the system's stiffness matrix with respect to the anisotropic parameters. These derivatives are then used to construct the sensitivity matrix. Subsequently, the LM optimization algorithm iteratively calls the SEBI forward solver to complete the inversion process. Finally, the computation accuracy and efficiency of the proposed SEBI-LM algorithm in both forward modeling and FWI are validated through several numerical experiments.

       

    /

    返回文章
    返回