可控源电磁法（CSEM）是矿产和油气勘探的重要手段。随着勘探程度加大，要求的探测深度更大、分辨率更高，同时面临着复杂地形、构造环境和各向异性等实际问题。本项目针对复杂勘探环境和各向异性介质，综合利用地面数据横向分辨率高和井中数据探测深度大、纵向分辨率高的优点，利用非结构四面体有限元法和L-BFGS算法进行三维地面和井中CSEM数据联合反演研究。在联合反演中，基于地面大功率张量源和多分量数据进一步增大对大深度复杂构造的反演能力。本项目利用动量法改进联合反演的最优化过程，提高反演收敛速度和避免反演易陷入局部极小。此外，为充分利用非地球物理信息进行融合反演解释和压制各向异性反演的非唯一性，本项目在建立的标准联合反演中，构造包含岩石物理和地质信息的高斯混合模型，以替代传统的高斯先验模型，在每次迭代中根据期望最大算法更新高斯混合模型，实现岩石物理和地质信息融合的三维地面和井中CSEM数据联合反演。

The controlled-source electromagnetic method (CSEM) is one kind of an important method for oil and gas exploration. As the rapid development of resource exploration, large depth and high resolution are required for CSEM, and at the same time, it faces practical problems such as complex terrain, tectonic environment, and anisotropy. To overcome the above practical problems, we take the advantages of the complementary nature of the surface and borehole CSEM data, which are high horizontal resolution for the former and large depth and vertical resolution for the latter. In this project, joint inversion of 3D surface and borehole CSEM data will be developed with the unstructured tetrahedral finite element method for anisotropic media. To further increase the inversion capability for full-depth and multi-scale structures, multi-component data from the surface and borehole will be used by a high power tensor source. We will use the momentum method to improve the optimization process in the inversion, which will improve the inversion convergence speed and avoid the objective function falling into the local minima. In order to make full use of non-geophysical information to suppress the non-uniqueness of anisotropic inversion, we propose a new framework for incorporating petrophysical and geological information into the joint inversion of surface and borehole CSEM data. To achieve our goal we replace the Gaussian prior, used in the standard joint inversion approach, by a Gaussian mixture model which represents both the petrophysical and geological knowledge and can be updated according to the Expectation-Maximization algorithm on each iteration.