| The exploration of natural resources such as metal ores and oil and gas are critical to energy security.As one of the important geophysical methods in resource exploration,controlled-source electromagnetic(CSEM)method can provide reliable subsurface electrical information,which can provide the basis and guidance for the subsequent exploration work.However,with the increasing complexity of the exploration environment,the CSEM exploration tends to be more and more in areas with undulating topography and complex geological structures,and relevant experiments show that there may be conductivity anisotropy in the subsurface media in some areas.The imaging and interpretation of the subsurface electrical structures in these real geological environments bring serious challenges to the development of CSEM forward modeling and inversion methods.If the inversion or interpretation of the field data obtained in these areas without considering the influence of undulating topography or conductivity anisotropy may lead to mistaken data interpretation results.Accordingly,this paper is dedicated to developing a large-scale 3D CSEM anisotropic inversion software for realistic and complex geological environments,and in this process,it can significantly reduce the consumption of computational resources such as memory and time in solving large-scale inversion problems,effectively reduce the number of inversion unknowns to a certain extent,improve its applicability and practicality,and thus provide reliable inversion results for relevant exploration or research work.To achieve the above-mentioned objectives,many studies has been carried out in this paper,and a series of solutions have been proposed for specific problems,and the main research contents and innovative work are as follows:(1)A goal-oriented adaptive CSEM forward modeling algorithm based on inherited tetrahedral mesh is proposed and implemented.To obtain accurate electromagnetic response from a realistic and complex geoelectric model,we developed a new three-dimensional CSEM anisotropy forward modeling software based on goal-oriented adaptive mesh refinement technique and vector finite-element method.The forward modeling software uses an unstructured tetrahedral mesh to discretize the model area,which can effectively handle arbitrary undulating terrain and complex geometries in the model.The accuracy of the forward modeling can be guaranteed by using the goal-oriented adaptive mesh refinement technique,but unlike the previous adaptive tetrahedral mesh refinement method based on the Delaunay triangular dissection algorithm,we use a new tetrahedral longest-edge bisection algorithm.Since the refined grid is nested into the coarse grid,this grid refinement technique can accurately map the physical parameters on the coarse grid into the refined grid,thus eliminating the additional numerical errors arising from the inconsistency of electrical parameters between different grids in the traditional refinement method,which makes the adaptive mesh refinement technique can be easily used in the inversion.(2)An efficient and robust iterative solution scheme for large-scale and ill-conditioned CSEM anisotropy forward problems is developed,and the robustness and convergence of the iterative solution scheme in solving anisotropic models with different types and intensities of anisotropic conductivity are systematically investigated.To reduce the memory and time consumption in solving large-scale and ill-conditioned anisotropic forward problems,we use the Auxiliary space Maxwell Solver(AMS)preconditioned FGMRES iterative solver to solve the system of linear system equations,and improve the computational efficiency by combining the domain-decomposition technique.Finally,we demonstrate the performance of our newly developed 3D CSEM anisotropy adaptive forward modeling software in terms of computational accuracy and efficiency using three isotropic and three anisotropic models,and also demonstrate the performance of the AMS preconditioned FGMRES iterative solver in handling different kinds of anisotropic models with different intensities through extensive tests,confirming that its convergence is almost independent of frequency,anisotropic angles and contrast,grid size,and problem size.In the scalability test,we use the high-performance computing platform to complete the CSEM forward problem containing billions of degrees of freedom in about three minutes.In addition,we have successfully applied the newly developed forward modeling software to the forward calculations of a real isotropic model of the Nihe iron ore mine and a real marine anisotropic model of the offshore region of Newfoundland,Canada,and obtained accurate EM response data for these two real complex models.(3)A triple parallel technique of "source-division-frequency-divisionarea" is proposed and implemented for CSEM large-scale inversion problems.According to the characteristics of CSEM inversion in frequency domain,this paper proposes a new triple MPI parallel strategy based on domain-decomposition technique,frequency and source,which can take full advantage of modern high-performance multi-core computing platforms to convert the multi-source and multi-frequency CSEM forward problem into several sub-problems to be solved simultaneously,and use domain decomposition method to accelerate the computation of each sub-problem on this basis.In addition,we also use the AMS pre-conditioned FGMRES iterative solver to solve the most computationally intensive original forward problem,the dual problem and the adjoint forward problem in the inversion,and adopt the LBFGS algorithm,which is most compatible with the iterative solver,as the inversion algorithm,thus significantly improving the efficiency of the large-scale CSEM 3D inversion and reducing the consumption of memory and other computational resources.For example,in the inversion test of the first theoretical synthetic data,we completed the 3D CSEM inversion problem with more than one million forward unknowns in about 20 minutes using more computational resources,and there is still much upgrading space for improvement of the computational speed with the increase of computational resources.(4)We construct an efficient and accurate grid decoupling strategy to meet the different grid requirements of forward modeling and inversion,and introduce the adaptive refinement of forward grid into inverse to ensure that the forward accuracy is not affected by the conductivity update during the inverse iteration.We develop three sets of forward and inverse grid schemes(forward and inverse with the same grid,forward grid based on the inverse grid for manual dichotomous refinement of measurement points and sources,and forward grid based on the inverse grid for adaptive refinement)to cope with different grid requirements for different inversion problems,thus effectively reducing the number of inversion unknowns.In addition,for the current problems of single type of EM inversion data from controlled sources in frequency domain and strong multi-solution inversion,and also to make full use of various types of observation data that may be collected in the field,we consider 20 types of observation data in the inversion,and test the inversion effect of different types of data on different models through a large number of comparison experiments,and find that the addition of z-component magnetic field can substantially improve the inversion effect.Finally,the inversion tests of four theoretical synthetic models(a simple isotropic model,a complex isotropic model with undulating topography,a triaxial conductivity anisotropic model,and a VTI medium model with undulating topography)fully confirm the reliability and efficiency of the newly developed 3D CSEM anisotropy inversion software and the correctness of its functions in handling large-scale complex models.The reliability,efficiency,and correctness of the functions of the newly developed 3D CSEM anisotropy inversion software for processing large-scale complex models are well established.(5)To test the application of the newly developed 3D CSEM anisotropy inversion procedure to the measured data,we successively invert the measured CSEM data collected from the Huaniushan Lead-Zinc Ore area in Gansu Province and the wide-area electromagnetic data collected from the Republican Basin area in Qinghai Province to obtain the subsurface 3D conductivity structures in these two areas.The inversion results of the two sets of measured data are in good agreement with the available borehole data,geological data and published results,which reveal the distribution of subsurface minerals and geothermal resources in these two areas,respectively,and fully confirm the capability of the newly developed 3D CSEM inversion program to handle field measured data. |
PDF Full Text Request