三维大地电磁自动建模与快速反演算法

The accurate, high-resolution and reliable three-dimensional(3D) resistivity image of earth can be gotten by using 3D inversion of magnetotelluric(MT) data. It is imperative that the detailed interpretation of MT data for the deep earth be supported by 3D inversion techniques. Normal 3D MT inversion algorithm needs great amount of CPU-time and computer memory, and has poor result for real data. These shortcomings limit the application of 3D inversion for real MT data. This proposal is about to introduce two techniques. Firstly, an initial-model auto-generation method is proposed by using data fusion and interpolation algorithm from the results of 1D Bostick inversion, 2D rapid relaxation inversion(RRI), 1D/2D Occam's inversion of 3D MT data set. The built approximate model can be used as the initial-model of non-linear conjugate gradient(NLCG) inversion algorithm which is very sensitive to the start model. Secondly, quasi-3D inversion algorithm and hybrid 3D inversion algorithm are proposed to establish a fast 3D MT inversion software. The LU decomposition algorithm will be used to solve the system equation of 3D MT forward problem. The sensitivity matrix will be optimized by combining the Footprint technique and wavelet compression algorithm. In the hybrid algorithm, the Occam's inversion routine which is always reliable will be first called to get a rough recovered model. After a few iteration, the NLCG inversion routine which is more efficient and less computation cost will be invoked. The NLCG algorithm will fit the observed data to a expected level and add more detail information to the recovered model. The whole algorithm will be parallelized on MPI platform. An efficient and stable inversion software will be established by using the above techniques. The inversion software will be verified by applying to real data. The goal is developing an efficient and practical inversion software of 3D MT data with full independent intellectual property rights. The above studies will provide well technical supports for practical application of 3D MT inversion and effective means for exploring the 3D electrical conductivity structure of deep earth.

三维(3D)大地电磁(MT)反演可以获得高精度、高分辨率、高可靠度的三维地下电性分布。因此，地球深部探测MT资料的精细解释迫切需要3D反演的支持。通常的3D反演算法面临着计算时间长、所需内存大、对实测数据反演效果差等问题，严重影响了3D反演在实测数据解释中的应用。课题拟将多种一维和二维反演(Bostick，RRI，Occam)的结果，通过融合插值，自动建立3D反演的初始模型。通过使用LU分解法直接求解正演问题的线性方程组、Footprint技术和小波变换算法压缩灵敏度矩阵、拟3D反演算法和混合反演算法等手段，提高3D MT反演算法的效率和稳定性。反演算法将基于MPI平台并行化。通过实测3D MT数据的反演解释，检验算法的实际应用效果。最终形成一套高效、可靠的3D MT反演算法。这些研究工作的开展，将为3D MT反演的实用化提供技术支撑，为深化对地球内部三维电性结构的认识，提供有效的手段。

为了满足在大地电磁探测领域对实测数据越来越紧迫的三维反演的需求，本项目拟研究开发一套从初始模型自动建立到快速反演算法的实用化程序系统。经过三年的的持续努力，通过与本领域国际顶尖学者合作，顺利完成了研究方案中设定的研究目标。通过对多个具有代表性的实测数据的反演计算和地质解释，验证了实现的三维大地电磁数据自动建模与快速反演算法具有良好的实用性和灵活性，可广泛用于各种不同探测目标的成像反演。在算法研究方面，针对灵敏度矩阵计算与存储的多个优化措施达到预期的目的，有效提高了算法对计算资源的消耗。此外，在实测数据的反演与解释过程中，提出了新的模型约束方案，以利于在地质解释过程中考虑多种不同的可能性。在实测数据的反演方面，项目取得较丰硕的成果。截止到结题时，已发表SCI期刊论文2篇，其中TOP期刊论文1篇，以合作者发表SCI TOP期刊论文2篇，后续还有多篇论文正在撰写之中；以口头报告和展板报告形式，先后4次参加美国AGU年会和2次EMIW研讨会，发表会议论文和摘要6篇；在国际学术交流与合作方面，在本项目经费资助和部分资助下，先后有三人次赴美国、德国、澳大利亚进修和短期访问；两次邀请外国专家赴华讲学和合作研究。总体来说，本项目圆满完成了各项科学研究目标，取得了较丰硕的成果，同时也大力推进了项目组成员的国际学术研究合作与交流活动。