A mesh generation method and system for transient electromagnetic 3D forward modeling of undulating terrain

By employing a constrained Delaunay triangulation and an adaptive mesh generation method driven by a local residual error indicator, combined with a GPU parallel AMG solver, the problem of low mesh generation and solving efficiency in transient electromagnetic 3D forward modeling of undulating terrain is solved, achieving efficient and accurate 3D forward modeling calculations.

CN123118959APending Publication Date: 2026-11-17水利部河湖保护中心 +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611124744.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-28
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies for generating transient electromagnetic 3D forward modeling meshes for undulating terrain suffer from several problems, including difficulty in automatically generating high-quality unstructured meshes, high cost of solving sparse linear systems, low time stepping efficiency, and poor GPU parallelization adaptability.

Method used

An initial tetrahedral mesh is generated using a constrained Delaunay triangulation algorithm, and adaptive mesh refinement is driven by a local residual error indicator. The solution is obtained by combining a second-order backward difference scheme and an algebraic multigrid preconditioned conjugate gradient method. A GPU-parallel AMG preconditioned iterative solver adapted to edge element sparse systems is designed to achieve efficient adaptive mesh generation.

Benefits of technology

It achieves high-precision and high-efficiency 3D forward modeling of transient electromagnetic undulation terrain, significantly reducing computing resource requirements and time costs, improving GPU parallel solution efficiency, and is highly adaptable to large-scale engineering exploration data processing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118959A_ABST
    Figure CN123118959A_ABST
Patent Text Reader

Abstract

This application discloses a mesh generation method and system for transient electromagnetic 3D forward modeling of undulating terrain. The method includes: generating an unstructured tetrahedral mesh for undulating terrain; first generating triangular prism elements in the near-surface region and then splitting them into tetrahedrons according to the shortest diagonal rule to accurately fit the terrain and strata; calculating a local residual error indicator at each time step, thereby simultaneously driving the dynamic refinement / coarsening of the mesh and the adaptive adjustment of the time step; employing an unconditionally stable second-order backward differential implicit time-progression scheme; and using a physical-driven algebraic multigrid preconditioned conjugate gradient method based on conductivity gradient weighted node connection strength to solve the linear equation system in parallel on a GPU. This application reduces the terrain-introduced error to less than 3%, reduces the mesh size by 60% to 80% compared to uniform refinement, reduces the total number of time steps to 1 / 10 to 1 / 50 of the explicit scheme, and achieves a forward modeling error of less than 2%, significantly improving the accuracy and efficiency of magnetotelluric 3D forward modeling under complex terrain.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the fields of geophysical forward modeling and finite element method, and in particular to a mesh generation method and system for transient electromagnetic three-dimensional forward modeling of undulating terrain. Background Technology

[0002] The Nedelec Whitney element, also known as the H(curl) harmonic element, is a finite element basis function specifically designed for electromagnetic field calculations. Unlike traditional nodal elements, the Whitney element defines its degrees of freedom on the edges of the mesh element. The basis function has continuous tangential components and discontinuous normal components on each edge, naturally satisfying the tangential continuity condition of the electromagnetic field and effectively avoiding the spurious solution problem that easily occurs with nodal elements. A first-order Nedelec Whitney element has 6 edges on a single tetrahedral element, corresponding to 6 computational degrees of freedom.

[0003] Generating meshes for undulating terrain is difficult: Transient electromagnetic field exploration often involves undulating surfaces such as mountains, valleys, and dams. Edge elements rely on unstructured tetrahedral meshes to fit the real terrain, and it is difficult to automatically generate high-quality unstructured meshes.

[0004] Solving large sparse linear systems is costly: edge element forward modeling eventually forms complex symmetric sparse linear systems of orders of hundreds of thousands to millions. Direct solution incurs huge memory and time overhead, while iterative solution lacks efficient fitting preconditioners.

[0005] Low time stepping efficiency: Time-domain forward modeling requires solving the linear system once at each time step, resulting in a large number of time steps and a significant increase in overall computation time.

[0006] Lack of adaptive mesh refinement strategy: Traditional edge elements use a fixed uniform mesh, which cannot automatically refine the mesh according to local calculation errors, resulting in insufficient accuracy or mesh redundancy waste.

[0007] GPU parallelization has poor adaptability: edge element sparse matrices have irregular sparse structures, and traditional GPU sparse storage and operation formats are inefficient, making it difficult to take advantage of CUDA's parallel acceleration. Summary of the Invention

[0008] This application provides a mesh generation method and system for transient electromagnetic three-dimensional forward modeling of undulating terrain, which can realize the automatic generation of high-quality unstructured tetrahedral meshes for transient electromagnetic undulating terrain.

[0009] This application provides a mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain, including: Obtain GPS elevation data of the surface of the survey area, construct a digital elevation model, and generate a surface triangular mesh on the digital elevation model using a constrained Delaunay triangulation algorithm. During the triangulation process, the extracted terrain feature lines are used as constraint edges to generate an initial tetrahedral mesh. After the electric field solution is completed at the current time step, all tetrahedral elements are traversed, the local residual error indicator of each element is calculated, and the tetrahedral elements are refined based on the local residual error indicator. According to the updated mesh, the finite element system matrix is ​​recalculated and updated only for the mesh regions where the topology has changed, and the system matrix of the unchanged regions is directly reused. A second-order backward difference scheme is used to construct the linear equation system for solving the electric field at the next time step. For each time step, the linear equation system is solved using the algebraic multigrid preconditioned conjugate gradient method running on a graphics processor, and the final grid is generated based on the solution results.

[0010] This application provides a mesh generation system for transient electromagnetic three-dimensional forward modeling of undulating terrain, including a processor and a memory. The memory stores a computer program, which, when executed by the processor, implements the steps of the aforementioned mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain.

[0011] This application's embodiments design a GPU-parallel AMG preconditioning iterative solver adapted to edge-element sparse systems; it employs posterior error estimation to drive adaptive mesh refinement, controlling mesh size and computational load while ensuring computational accuracy. The method of this application can achieve automatic generation of high-quality unstructured tetrahedral meshes for transient electromagnetic undulation terrain.

[0012] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0013] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a basic flowchart illustrating the mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain according to an embodiment of this application. Detailed Implementation

[0014] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0015] This application provides a mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain, such as... Figure 1 As shown, it includes the following steps: In step S101, GPS elevation data of the survey area surface is acquired, and a digital elevation model (DEM) is constructed. On the DEM, a constrained Delaunay triangulation algorithm is used to generate a surface triangular mesh. During the triangulation process, extracted terrain feature lines are used as constraint edges to generate an initial tetrahedral mesh. In a specific example, GPS elevation data from the survey line is used to reconstruct a three-dimensional surface DEM with a resolution of 1m to 5m through bilinear interpolation. The constrained Delaunay triangulation algorithm is then used to perform surface meshing on the surface DEM, accurately preserving terrain feature lines such as ridges and valleys.

[0016] This step describes the extension method from the 2D surface mesh to the 3D volume mesh. Specifically, the volume mesh is generated downwards from the triangular face mesh of the surface. For the near-surface region from the surface to a predetermined depth underground (such as the first electrical interface), tetrahedrons are not generated directly; instead, vertical triangular prisms (Prism) elements are generated first. The top and bottom faces of each prism are triangles.

[0017] In step S102, after the electric field solution is completed in the current time step, all tetrahedral elements are traversed, the local residual error indicator of each element is calculated, and the tetrahedral elements are refined based on the local residual error indicator. According to the updated mesh, the finite element system matrix is ​​recalculated and updated only for the mesh regions where the topology has changed, and the system matrix of the unchanged regions is directly reused.

[0018] In other words, after the electric field calculation is completed at each time step, the following operations are performed: First, a physically driven local residual error indicator is calculated. This indicator essentially measures the magnitude of the residuals in the electromagnetic field governing equations (i.e., the Helmholtz equations for the electric field) within each cell. Its calculation formula is as follows: In the formula: For unit error indicator; The element feature size; Electric field strength; The dielectric conductivity; The source current density; For unit Norm. This indicator integrates the cell's geometry, field variations, and source distribution, accurately identifying regions with significant numerical errors.

[0019] In step S103, a second-order backward difference scheme is used to construct the linear equation system for solving the electric field in the next time step. For the linear equation system constructed in each time step, the algebraic multigrid preconditioned conjugate gradient method running on a graphics processor is used to solve it, so as to generate the final grid based on the solution results.

[0020] This application's embodiments design a GPU-parallel AMG preconditioning iterative solver adapted to edge-element sparse systems; it employs posterior error estimation to drive adaptive mesh refinement, controlling mesh size and computational load while ensuring computational accuracy. The method of this application can achieve automatic generation of high-quality unstructured tetrahedral meshes for transient electromagnetic undulation terrain.

[0021] In some embodiments, a constrained Delaunay triangulation algorithm is used to generate a surface triangular mesh, and the extracted terrain feature lines are used as constraint edges during the triangulation process to generate an initial tetrahedral mesh, including: Using the surface triangular mesh as the top surface, an unstructured tetrahedral mesh for three-dimensional forward modeling is generated by extending downwards; wherein, from the surface to the near-surface region at a predetermined depth, a triangular prism mesh is first generated, and then each triangular prism unit is split into multiple tetrahedral units according to a preset rule to fit the stratigraphic layer interface; below the near-surface region, tetrahedral units are generated by expansion.

[0022] In a specific example, splitting a triangular prism involves: uniformly selecting the two vertices of the shortest diagonal on its three quadrilateral faces, connecting them, and then splitting each triangular prism element into three tetrahedral elements along the face formed by this connection and the corresponding sides of the top and bottom triangles. The advantages of this approach are: the triangular prism perfectly conforms to undulating surfaces and horizontal or slightly inclined subsurface layers, ensuring the geometric accuracy of stratigraphic stratification within the mesh; and uniformly splitting along the shortest diagonal effectively avoids generating overly elongated or deformed tetrahedrons, ensuring the quality of the near-surface mesh. Below the near-surface region, a conventional expansion method is used to gradually increase the size of the tetrahedral elements to reduce the total mesh size. After mesh generation, quality control and optimization are required. Inferior elements with a minimum dihedral angle less than 15 degrees are corrected using Laplacian smoothing or edge flipping operations.

[0023] In some embodiments, refining the tetrahedral elements based on a local residual error indicator includes: Cells whose error indicator values ​​exceed the maximum value of all current cell error indicators by a first preset ratio are marked as cells to be refined, and cells whose error indicator values ​​are lower than the maximum value of all current cell error indicators by a second preset ratio are marked as cells to be coarsened. For example, cells whose error indicator values ​​exceed 80% of the maximum value of all current cells are marked as refined cells, and red refinement is used to split them into 8 sub-tetrahedrons. To solve the dangling node problem caused by refinement, green refinement is performed on adjacent cells.

[0024] Furthermore, this application introduces a reverse process—coarsening. If the error indicator value of some elements remains below a low threshold (e.g., 10% of the global maximum), it indicates that the field change in that region is gradual, and excessively fine meshes are a waste. In this case, edge folding and other operations are performed to restore them to coarser elements. This bidirectional dynamic strategy of "refinable and coarsable" ensures that the total number of meshes is maintained at an optimal level throughout the simulation process.

[0025] Perform mesh topology modification operations on the marked cells to be refined and cells to be coarsened to obtain the updated mesh; Based on the statistical results of the local residual error indicator, it is determined that the preset time step adjustment conditions are met, so as to adjust the time step of the next time step.

[0026] In some implementations, based on the statistical results of the local residual error indicator, determining whether the preset time step adjustment conditions are met includes: If the maximum value of the local residual error indicator calculated at the current time step is lower than the third preset threshold, the time step length will be increased by a preset multiple in the next time step. If the maximum value of the local residual error indicator is higher than the fourth preset threshold, the calculation result of the current time step is rejected, the time step size is reduced by a preset multiple, and the calculation of the current time step is performed again.

[0027] In a specific example, this application directly uses the statistical results of the error indicators to control the time step. The specific mechanism is as follows: if, after the current step is completed, the maximum value of the error indicators for all elements is below a lower threshold (e.g., 0.1), indicating that the current solution is very smooth, the time step size Δt is increased by a preset ratio (e.g., 1.2 times) in the next time step. Conversely, if the maximum value exceeds an upper threshold (e.g., 1.0), indicating that the error is growing too rapidly, the calculation result of the current step is rejected, the time step size Δt is reduced to half its original size, and the current step is recalculated. This mechanism unifies mesh refinement and time step adjustment within the same error control framework, forming a closed-loop feedback control that ensures the accuracy and efficiency of the entire simulation process.

[0028] After the mesh topology is modified, this application only reassembles the finite element stiffness matrix and mass matrix for the local mesh regions that have changed, while the matrix elements of the unchanged regions are fully reused, which greatly reduces the amount of computation per time step.

[0029] Implementation steps of the unit local residual error indicator: After each time step is completed, the local residual error indicator is calculated for each tetrahedral element. Mark cells whose error indicators exceed 80% of the global maximum value as refined cells; Red refinement is applied to the marked elements, splitting a single tetrahedron into 8 sub-tetrahedrons, and the node coordinates are updated by midpoint interpolation; green refinement is applied to adjacent elements to repair dangling nodes and ensure mesh topology continuity. After mesh refinement, the stiffness matrix of the refined region is updated locally, while the calculation results of the unrefined region are directly reused to continue the time step calculation.

[0030] In some embodiments, a second-order backward difference scheme is used to construct the linear equation system for solving the electric field at the next time step. The linear equation system constructed for each time step includes: The electric field at the next time step is constructed using a second-order backdifference BDF2 implicit time-progression scheme to solve the linear equations. The second-order backdated difference BDF2 implicit time-progression scheme is as follows: in The curl-curl stiffness matrix is ​​related to the mesh edges. The mass matrix is ​​related to the conductivity of the dielectric. f Let E be the source term, E be the electric field vector to be solved, and the superscript n indicates the nth time step.

[0031] By changing the time step Δt in the linear equation system, the time step can be adaptively adjusted until the preset total simulation time is reached. The time step Δt is not limited by CFL and can be adaptively adjusted according to accuracy requirements. For example, by setting a small time step in the early stage and a large time step in the later stage, the total number of time steps can be reduced to 1 / 10 to 1 / 50 of the explicit scheme.

[0032] In some embodiments, the solution is obtained using the algebraic multigrid preconditioned conjugate gradient method running on a graphics processor, including: For the linear equation system constructed at each time step, the algebraic multigrid preconditional conjugate gradient method running on a graphics processor is used for solution. The coarsening strategy of the algebraic multigrid is as follows: the connection strength between finite element grid nodes is calculated based on the conductivity gradient value of each element, and the nodes are aggregated according to the connection strength to construct a multi-layer coarsened grid, so that the coarsening process is adaptive to the electrical distribution of the underground medium.

[0033] This application addresses the construction of large-scale sparse linear equation systems by employing an efficient algebraic multigrid preconditioned conjugate gradient method executed on a GPU. This application proposes a physicalized AMG coarsening strategy. Traditional AMG methods determine the connection strength between nodes solely based on the numerical values ​​of matrix elements. This invention, however, associates this with a physical model: for the edges connecting nodes i and j, the connection strength is calculated, ensuring that in regions with large conductivity gradients (such as orebody boundaries), the connection strength is weak, and the coarsening process naturally skips these nodes, reserving them for fine-grained solutions; while in surrounding rock regions with uniform conductivity, the connection strength is strong, and coarsening proceeds smoothly. This physically driven coarsening strategy generates a mesh hierarchy that better matches the characteristics of the equations, significantly improving the efficiency and robustness of AMG as a preconditioner.

[0034] In some embodiments, the solution using the algebraic multigrid preconditional conjugate gradient method running on a graphics processor further includes: The sparse matrix in the process of solving sparse linear equations adopts the ELL+COO hybrid HYB storage format, which is adapted to the irregular sparse structure of edge elements. It can efficiently process the irregular sparse mode generated by the discretization of edge elements, and its sparse matrix vector multiplication computation throughput is improved by about 30% compared with the traditional CSR format.

[0035] AMG Gaussian-Seidel smoothing employs red-black coloring for parallel GPU execution, with constraint and continuation operations performed entirely within the GPU. The entire solution process, including the red-black sorted Gaussian-Seidel smoothing iteration, is executed within the GPU, avoiding the bottleneck of massive data exchange between the CPU and GPU. For systems with millions of unknowns, the GPU-AMG+PCG solver is 5 to 20 times faster than traditional direct methods on the CPU.

[0036] Since the mesh topology only changes locally, but the AMG hierarchy can be corrected by local updates rather than global reconstruction after the change, the matrix topology of adjacent time steps is consistent in this embodiment, and the AMG hierarchy is constructed only once.

[0037] Compared with the prior art, the method of this application has the following specific and verifiable beneficial effects: It exhibits strong terrain adaptability and high accuracy: Through a mesh generation strategy of "constrained Delaunay triangulation + triangular prism splitting," the geometric fitting error between the surface and shallow strata is minimized. Validated by an international standard model, the numerical error introduced by terrain in this method is less than 3%, significantly better than the 5%–10% error of the traditional stepped mesh FDTD method.

[0038] Extremely efficient allocation of computational resources with significant capacity reduction: A closed-loop optimization system is formed through a bidirectional adaptive mesh strategy that allows for both fine and coarse meshing, along with error-based time step adaptation. Compared to globally uniform meshing that maintains the same accuracy, this method reduces the mesh size by 60% to 80%, while the total number of time steps is only 1 / 10 to 1 / 50 of the explicit scheme, resulting in an order-of-magnitude reduction in computational load and memory consumption.

[0039] The GPU solver offers significant speedup: the physics-driven AMG preconditioner can more effectively handle ill-conditioned systems caused by drastic electrical differences. Combined with the HYB sparse format and full GPU parallelism, it greatly reduces the total forward modeling time.

[0040] High degree of engineering practicality: The entire process, from importing raw field GPS data to outputting final results, is fully automated, requiring no manual grid division or parameter adjustment intervention, laying a solid technical foundation for large-scale, batch engineering exploration data processing.

[0041] This application also proposes a mesh generation system for transient electromagnetic three-dimensional forward modeling of undulating terrain, including a processor and a memory. The memory stores a computer program, which, when executed by the processor, implements the steps of the aforementioned mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain.

[0042] It should be noted that, in the embodiments of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0043] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0044] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0045] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims. All of these forms are within the protection scope of this application.

Claims

1. A mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain, characterized in that, include: Obtain GPS elevation data of the surface of the survey area, construct a digital elevation model, and generate a surface triangular mesh on the digital elevation model using a constrained Delaunay triangulation algorithm. During the triangulation process, the extracted terrain feature lines are used as constraint edges to generate an initial tetrahedral mesh. After the electric field solution is completed at the current time step, all tetrahedral elements are traversed, the local residual error indicator of each element is calculated, and the tetrahedral elements are refined based on the local residual error indicator. According to the updated mesh, the finite element system matrix is ​​recalculated and updated only for the mesh regions where the topology has changed, and the system matrix of the unchanged regions is directly reused. A second-order backward difference scheme is used to construct the linear equation system for solving the electric field at the next time step. For each time step, the linear equation system is solved using the algebraic multigrid preconditioned conjugate gradient method running on a graphics processor, and the final grid is generated based on the solution results.

2. The mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain as described in claim 1, characterized in that, A constrained Delaunay triangulation algorithm is used to generate a surface triangular mesh, and the extracted terrain feature lines are used as constraint edges during the triangulation process to generate an initial tetrahedral mesh, including: Using the aforementioned surface triangular mesh as the top surface, an unstructured tetrahedral mesh for 3D forward modeling is generated by extending downwards; wherein, From the surface to the near-surface region at a predetermined depth, a triangular prism grid is first generated, and then each triangular prism unit is split into multiple tetrahedral units according to a preset rule to fit the stratigraphic layer interface. Below the near-surface region, tetrahedral units are generated using an expansion method.

3. The mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain as described in claim 1, characterized in that, Calculate the local residual error indicator for each cell, where the local residual error indicator is: In the formula: For unit error indicator; The element feature size; Electric field strength; The dielectric conductivity; The source current density; For unit Norm; Refinement of tetrahedral elements based on local residual error indicators includes: Units whose error indicator values ​​exceed the maximum value of all current unit error indicators up to a first preset ratio are marked as units to be refined, and units whose error indicator values ​​are lower than the maximum value of all current unit error indicators up to a second preset ratio are marked as units to be coarsened. Perform mesh topology modification operations on the marked cells to be refined and cells to be coarsened to obtain the updated mesh; Based on the statistical results of the local residual error indicator, it is determined that the preset time step adjustment conditions are met, so as to adjust the time step of the next time step.

4. The mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain as described in claim 3, characterized in that, Based on the statistical results of the local residual error indicator, the following conditions are considered to be met for time step adjustment: If the maximum value of the local residual error indicator calculated at the current time step is lower than the third preset threshold, the time step length will be increased by a preset multiple in the next time step. If the maximum value of the local residual error indicator is higher than the fourth preset threshold, the calculation result of the current time step is rejected, the time step size is reduced by a preset multiple, and the calculation of the current time step is performed again.

5. The mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain as described in claim 3, characterized in that, A second-order backward difference scheme is used to construct the linear equation system for solving the electric field at the next time step. The linear equation system constructed for each time step includes: The electric field at the next time step is constructed using a second-order backdifference BDF2 implicit time-progression scheme to solve the linear equations. The second-order backdated difference BDF2 implicit time-progression scheme is as follows: in The curl-curl stiffness matrix is ​​related to the mesh edges. The mass matrix is ​​related to the conductivity of the dielectric. Here, E is the electric field vector to be solved, and the superscript n indicates the nth time step. By changing the time step Δt in the linear equation system, the time step is adaptively adjusted until the preset total simulation duration is reached.

6. The mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain as described in claim 5, characterized in that, The solution is obtained using the algebraic multigrid preconditioned conjugate gradient method running on a graphics processing unit, including: For the linear equation system constructed at each time step, the algebraic multigrid preconditional conjugate gradient method running on a graphics processor is used for solution. The coarsening strategy of the algebraic multigrid is as follows: the connection strength between finite element grid nodes is calculated based on the conductivity gradient value of each element, and the nodes are aggregated according to the connection strength to construct a multi-layer coarsened grid, so that the coarsening process is adaptive to the electrical distribution of the underground medium.

7. The mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain as described in claim 6, characterized in that, Solving using the algebraic multigrid preconditional conjugate gradient method running on a graphics processing unit also includes: The sparse matrix in the process of solving the sparse linear equation system adopts the ELL+COO hybrid HYB storage format, which is adapted to the irregular sparse structure of edge elements. AMG Gaussian-Seidel smoothing uses red-black shading for parallel GPU execution, with constraint and extension operations completed entirely within the GPU. The matrix topology is consistent between adjacent time steps, and the AMG hierarchy is constructed only once.

8. A mesh generation system for transient electromagnetic three-dimensional forward modeling of undulating terrain, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, implements the steps of the mesh generation method for transient electromagnetic three-dimensional forward modeling of undulating terrain as described in any one of claims 1 to 7.