大地电磁法各向异性介质三维Occam-非线性共轭梯度反演研究

In our country, Magnetotellurics (MT) plays an important role in the field of energy exploration, deep geological detection and so on. MT 3D modeling and 2D inversion techniques have been practical. However, the real geological structure underground is complicated, 3D and anisotropic in most cases. Modeling and inversion for anisotropic conductivities have been the hot topics in the field of MT. Currently, reports on the MT inversion with 2D anisotropic conductivities are increasing, while there are no public reports on MT inversion with 3D anisotropic conductivities. Finite difference method is used in the MT 3D anisotropic forward modeling in this project. Based on the pseudo-forward problem of conjugate gradient method and regularization factor searching idea of Occam method, a novel Occam-NLCG inversion method which can reduce computer memory requirement, promote the stability of the inversion convergence and improve computation efficiency is developed. A new searching approach for two regularization factors is put forward in the inversion scheme. This inversion method is used in the 3D anisotropic inversion of MT data in which many inversion model parameters are used and an anisotropic constraint is introduced into the objective function. MPI and GPU are used to implement 3D forward and inversion parallel algorithm. This effective and practical 3D inversion algorithm can be used to invert anisotropic MT field data and promote the application of MT in complicated geological detection.

大地电磁法(MT)在我国资源勘探和深部构造探测等领域具有重要的作用。大地电磁各向同性介质的三维正演和二维反演技术已经成熟。然而，实际地下介质电性参数的分布往往是复杂的、三维的和各向异性的，各向异性正反演的研究已成为当前大地电磁领域研究的热点与前沿课题。近年来，大地电磁二维各向异性介质反演的成果逐渐增多，三维各向异性介质反演的成果尚未见到公开文献报道。本项目采用有限差分法实现三维正演。基于共轭梯度“拟正演”思想和Occam最优权重因子搜索思路构建出一种节约内存、提高反演收敛稳定性和计算效率的Occam-非线性共轭梯度反演方法。该反演方法将提出一套新的2个最优正则化权重因子的搜索方案并应用于实现多种模型参数和多目标函数项的各向异性三维反演。并且把MPI和GPU运用于三维正反演并行算法中，形成高效率、实用化的大地电磁各向异性三维正反演技术，提高大地电磁法在复杂地质情况下的应用效果。

大地电磁法(MT)在我国资源勘探和深部地质探测等领域具有重要的作用。各向异性正反演的研究成为当前大地电磁领域研究的难点与前沿课题。近年来，大地电磁二维各向异性介质反演的成果逐渐增多，三维各向异性介质反演的成果尚未见到公开文献报道。本项目采用有限差分法实现了三维正演。与成熟的二维各向异性代码的计算结果对比验证了所实现三维正演算法的精度。基于共轭梯度“拟正演”思想和Occam最优权重因子搜索思路构建出一种节约内存、提高反演收敛稳定性和计算效率的Occam-非线性共轭梯度反演方法。分别使用Occam-非线性共轭梯度和有限内存拟牛顿（LBFGS）2种反演方法实现了多反演模型参数和多目标函数项的各向异性三维反演。理论模型合成算例的计算结果验证了所实现三维反演算法的可靠性和稳定性。采用MPI语言实现了三维正反演的并行计算。理论模型合成算例表明并行算法的计算效率明显提高。项目在西藏南迦巴瓦地区补充采集了23个测点的大地电磁数据。用三维各向异性反演算法并行程序对该地区所有实测资料（原有测点+新测点）进行反演，反演结果验证了所实现的三维反演算法的有效性和实用性。该高效率、实用化的大地电磁各向异性三维正反演技术，能够提高大地电磁法在能源勘探和深部地质探测等领域的应用效果。