DC / IP data efficient parallel inversion method and system

Abstract

The present disclosure provides a DC / IP data efficient parallel inversion method and system, relating to the technical field of geophysical exploration, comprising: obtaining DC / IP actual observation data of a target area, and obtaining synthetic observation data by adding noise; constructing an initial underground model; based on the synthetic observation data, iteratively updating the initial underground model by the Gauss-Newton method, linearizing the optimization problem of DC / IP inversion to obtain a normal equation; according to the synthetic observation data and the current model parameters, implicitly calculating the Jacobian matrix by the pseudo-forward technology, solving the linear equation set of pseudo-forward by the direct solver, and performing parallel calculation by the hybrid MPI+OpenMP technology; according to the Jacobian matrix and the normal equation, solving the model parameter increment, and updating the underground model by using the model parameter increment; the present disclosure realizes accurate inversion of DC / IP data, and is convenient for popularization and use.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of geophysical exploration, in particular to a DC / IP data efficient parallel inversion method and system. BACKGROUND

[0002] The direct current resistivity method (DC) and the induced polarization method (IP) have been widely used in mineral exploration, geophysical exploration, environmental research and geotechnical engineering. Among them, the distributed array DC / IP exploration can collect more full-azimuth data sets in a large area, provide more intensive data coverage and smaller directional deviation, and through DC / IP inversion, a more accurate underground model can be obtained. The traditional DC / IP inversion algorithm explicitly calculates the Jacobian matrix, and distributes the calculation of the iterative solver to a large number of CPUs to solve the normal equation of the Gauss-Newton method in parallel. However, in the distributed array DC / IP survey, it is not uncommon to use hundreds of transmitting sources to collect hundreds of thousands of data. Using the traditional algorithm for forward and inversion calculation requires a large amount of memory and time.

[0003] Therefore, the existing large-scale distributed array DC / IP inversion method is difficult to balance the contradiction between inversion accuracy and inversion time and space cost, affecting the inversion speed and calculation efficiency. SUMMARY

[0004] The present disclosure proposes a DC / IP data efficient parallel inversion method and system to solve the above problems, which realizes accurate and efficient inversion of DC / IP data and is convenient for popularization and use.

[0005] According to some embodiments, the present disclosure adopts the following technical solutions:

[0006] A DC / IP data efficient parallel inversion method, comprising:

[0007] Obtaining DC / IP actual observation data of a target area, and obtaining synthetic observation data by adding noise;

[0008] According to the existing geological information of the target area, an initial underground model is constructed;

[0009] Based on the synthetic observation data, the initial underground model is iteratively updated by the Gauss-Newton method until the iteration stopping condition is met, and the final underground model is obtained;

[0010] The iterative inversion update of the initial subsurface model using the Gauss-Newton method involves linearizing the optimization problem of DC / IP inversion to obtain the normal equations. Based on synthetic observation data and current model parameters, the Jacobian matrix is ​​implicitly calculated using pseudo-forward modeling technology. A direct solver is used to solve the linear equations of pseudo-forward modeling, and parallel computation is performed using a hybrid MPI+OpenMP technique. Based on the Jacobian matrix and the normal equations, the model parameter increments are solved, and the subsurface model is updated using these increments.

[0011] According to some embodiments, the present disclosure adopts the following technical solutions:

[0012] A high-efficiency parallel inversion system for DC / IP data includes:

[0013] The observation data acquisition module is configured to: acquire actual DC / IP observation data of the target area, and obtain synthetic observation data by adding noise;

[0014] The initial model building module is configured to build an initial underground model based on the existing geological information of the target area.

[0015] The model inversion update module is configured to: based on synthetic observation data, iteratively invert and update the initial subsurface model using the Gauss-Newton method until the iteration stops, thus obtaining the final subsurface model;

[0016] The iterative inversion update of the initial subsurface model using the Gauss-Newton method involves linearizing the optimization problem of DC / IP inversion to obtain the normal equations. Based on synthetic observation data and current model parameters, the Jacobian matrix is ​​implicitly calculated using pseudo-forward modeling technology. A direct solver is used to solve the linear equations of pseudo-forward modeling, and parallel computation is performed using a hybrid MPI+OpenMP technique. Based on the Jacobian matrix and the normal equations, the model parameter increments are solved, and the subsurface model is updated using these increments.

[0017] According to some embodiments, the present disclosure adopts the following technical solutions:

[0018] A computer program product includes a computer program that, when executed by a processor, implements the aforementioned efficient parallel inversion method for DC / IP data.

[0019] According to some embodiments, the present disclosure adopts the following technical solutions:

[0020] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned efficient parallel inversion method for DC / IP data.

[0021] According to some embodiments, the present disclosure adopts the following technical solutions:

[0022] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the aforementioned DC / IP data efficient parallelization inversion method.

[0023] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0024] This disclosure presents a highly efficient parallel inversion algorithm for large-scale DC / IP data. By using an unstructured tetrahedral mesh for initial modeling, it can adapt to complex geometric conditions and improve inversion accuracy. The use of a pseudo-finite difference method for forward modeling lays the foundation for future adaptive mesh refinement techniques. The algorithm implicitly calculates the Jacobian matrix using pseudo-forward modeling, solves the linear equations of the pseudo-forward modeling using a direct solver, and employs a hybrid MPI (Message Passing Interface) + OpenMP (Open Multi-Processing) technology for parallel computation, significantly improving computational efficiency. This disclosure enables accurate and efficient inversion of large-scale DC / IP data, greatly improving data processing efficiency in electrical resistivity tomography and providing valuable experience for the rapid inversion of geophysical exploration data. Attached Figure Description

[0025] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0026] Figure 1 This is a flowchart of the method in Example 1;

[0027] Figure 2 This is a schematic diagram of the vertex-based finite difference method in Example 1;

[0028] Figure 3 This is a schematic diagram of the parallelization of the hybrid MPI+OpenMP scheme in Example 1;

[0029] Figure 4 This is a schematic diagram of the initial synthesis model in Example 1;

[0030] Figure 5 The image shows a comparison of observed and predicted data in Example 1, where (a) is the observed data from three different vertical profiles and (b) is the predicted data from three different vertical profiles.

[0031] Figure 6The above are the apparent resistivity profiles of the tetrahedral mesh model obtained by inversion in Example 1, where (a) is the vertical apparent resistivity profile, (b) is the shallow horizontal apparent resistivity profile, and (c) is the deep horizontal apparent resistivity profile. Detailed Implementation

[0032] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0033] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0034] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0035] Example 1

[0036] One embodiment of this disclosure provides a method for efficient parallel inversion of DC / IP data, including:

[0037] Acquire actual DC / IP observation data of the target area, and obtain synthetic observation data by adding noise;

[0038] Based on the existing geological information of the target area, an initial underground model is constructed;

[0039] Based on synthetic observation data, the initial subsurface model is iteratively inverted and updated using the Gauss-Newton method until the iteration stops, thus obtaining the final subsurface model.

[0040] The iterative inversion update of the initial subsurface model using the Gauss-Newton method involves linearizing the optimization problem of DC / IP inversion to obtain the normal equations. Based on synthetic observation data and current model parameters, the Jacobian matrix is ​​implicitly calculated using pseudo-forward modeling technology. A direct solver is used to solve the linear equations of pseudo-forward modeling, and parallel computation is performed using a hybrid MPI+OpenMP technique. Based on the Jacobian matrix and the normal equations, the model parameter increments are solved, and the subsurface model is updated using these increments.

[0041] As one embodiment, this disclosure provides a method for efficient parallel inversion of DC / IP data, achieving accurate and efficient inversion of DC / IP data, which is easy to promote and use. Figure 1As shown, the specific implementation process is as follows:

[0042] Step 1: Conduct geological exploration using the direct current method or induced polarization method to obtain actual DC / IP observation data. Add Gaussian noise to the actual observation data to obtain synthetic observation data d. obs .

[0043] Specifically, select Figure 4 The data values ​​obtained from the forward modeling of the artificial synthetic model are used as the observation data for inversion. The conductivity and polarizability of the model are known. Gaussian noise with a standard deviation of 3% is added to the data values ​​to obtain the synthetic observation data.

[0044] Synthetic models in x Electrodes are arranged on seven cross-sections along the direction, with y-coordinates ranging from 200 to 800 m. There are a total of 133 source electrodes and 693 data points. When performing forward modeling, the forward modeling model is discretized into 542,152 tetrahedral elements and 666,030 edges.

[0045] Step 2: Construct an initial inversion model based on existing geological information. Use an unstructured tetrahedral mesh to mesh the initial model and generate the tetrahedral mesh model required for inversion.

[0046] The initial model for inversion is divided using an unstructured grid. The area near the survey line is a densified area with a small grid size, while the area outside the densified area is an undensified area with a larger grid size. In addition, secondary densification is performed around each electrode. The inversion model needs to interpolate the terrain data onto a two-dimensional grid, and use the interpolated grid as the terrain surface of the three-dimensional unstructured grid.

[0047] Specifically, in step two, a uniform half-space model with a conductivity of 0.001 S / m is used as the initial model. The selection of the initial model does not consider the synthetic model. A top center position is set as (500 m, 500 m, 0), and the size is 1100 m × 1100 m × 500 m. x , y and z Blocks with dimensions of 1100 m, 1100 m, and 500 m in different directions were used as the active region for inversion. During mesh generation, the tetrahedral elements within the active region were subject to a maximum volume constraint. An unstructured tetrahedral mesh containing 350,058 elements and 443,810 edges was generated using the TetGen (Tetrahedral Mesh Generator). The mesh was refined on the ground surface of the survey area by forcing the use of smaller element sizes to cover all electrodes. Regular tetrahedral elements with a side length of 1.5 meters were inserted into the mesh around the electrodes.

[0048] The iterative process from steps three to eight was calculated, with the initial regularization parameter set to 0.1 and the initial data fit difference to 386.134. The inversion calculation used 20 MPI processes, each with two OpenMP threads. The inversion process took 12.8 minutes to complete, with a maximum memory consumption of 44.74 GB. The inversion converged after three iterations. The final data fit difference in step eight was 1.035. A comparison of the predicted and observed data calculated in step three is shown in the figure. Figure 5 , Figure 5 (a) shows the model's observational data at 300m, 500m, and 700m north. Figure 5 (b) shows the model’s predicted data at 300m, 500m and 700m north, and the observed data fit the predicted data well.

[0049] Step 3: Perform a forward modeling calculation on the latest tetrahedral mesh model. Taking the Wenner device as an example, first calculate the underground potential distribution when each source electrode supplies power individually, and then obtain the potential difference between all receiving electrode pairs by arranging them. φ MN That is, the predicted data, denoted by d pre express;

[0050] Specifically, forward modeling uses the vertex-based mimetic finite-difference (MFD) method to discretize the Poisson equation, such as... Figure 2 As shown, the MFD method discretizes the potential of each element in a tetrahedral mesh onto the four vertices of the element. The forward-modeled Poisson equation and the discretized equation are expressed as equations (1) and (2), respectively.

[0051] (1)

[0052] (2)

[0053] in, For Hamiltonian operators, σ For conductivity distribution, For electric potential, I The source current amplitude, δ Let r be the Dirac function and r be the coefficients of mass. s Let K be the coordinates of a point and the coordinates of the source point, respectively, and K be the coefficient matrix. Let be the potential distribution vector composed of the electric potentials of all vertices in the grid. For source terms.

[0054] If it is DC forward modeling, then perform a forward modeling of the conductivity model; if it is IP forward modeling, then the background conductivity needs to be used separately. σb and effective conductivity σ η Perform two DC forward models and calculate the apparent polarizability using the results of the two models. The calculation equation is as follows:

[0055] (3)

[0056] in, For apparent polarizability, To account for the potential distribution when the polarization effect is excited, This represents the potential distribution without considering the excitation polarization effect.

[0057] Step 4: Construct the objective function for DC / IP inversion using regularization methods;

[0058] The objective function is:

[0059] (4)

[0060] The Gauss-Newton algorithm is used to compute a nonlinear optimization problem with an objective function where λ is a regularization parameter. It is the predicted data d pre and observation data d obs The fit difference between them The regularization terms are represented as follows:

[0061] (5)

[0062] (6)

[0063] in, N d It refers to the amount of observation data, W. d It is a data weighting matrix, W s W is the weighting matrix of the constraint model. z It is the constraint model smoothness matrix. α s and α z It is the penalty factor, and m is the current model. It is a reference model.

[0064] Step 5: Linearize the optimization problem of the objective function to obtain the normal equation:

[0065] (7)

[0066] in, n It is the number of iterations. λ Here, J is the regularization parameter, and J is the Jacobian matrix. δ mIt is the model update amount between two adjacent iterations, d n-1 It is the predicted data obtained in the (n-1)th iteration.

[0067] Furthermore, the linearization of the objective function described in step five first involves transforming the model m after the nth iteration... n The corresponding predicted data d n Expanded according to the linear Taylor approximation, it is as follows: Substituting into equation (4) yields a new objective function expression. Then, the new objective function expression is applied to the unknowns. δ m Find the partial derivative and set it to 0 to obtain the normal equation (7).

[0068] Furthermore, the normal equations described in step five are solved iteratively using a CG (Conjugate Gradient) solver to obtain the model parameter increments. δ m The CG solver uses the conjugate gradient method for iterative solution. In each iteration, the left-hand side of equation (7) is... It needs to be multiplied by a vector v, which is the conjugate direction of the conjugate gradient method in each iteration.

[0069] Step Six: Based on the latest forecast data d n and the current model parameter m j The Jacobian matrix and its transpose are implicitly computed using pseudo-forward modeling.

[0070] Furthermore, if the Jacobian matrix is ​​used for DC inversion, the equation for calculating the Jacobian matrix is:

[0071] (8)

[0072] in, It is the Jacobian matrix in the DC inversion process, and Q is the mapping matrix. It is the coefficient matrix in the DC inversion process. , , j =1, 2, …, N m , It is the first j The conductivity of each unit It is the potential distribution during the DC inversion process, S= diag {( σ 1, σ 2,…, σ Nm ) T}

[0073] If used for IP inversion, the equation for calculating the Jacobian matrix is:

[0074] (9)

[0075] in, It is the Jacobian matrix in the IP inversion process. , , Q is the mapping matrix. It is the coefficient matrix in the IP inversion process. , , j =1, 2, …, N m , It is the potential distribution during the IP inversion process, S b = diag {( σ b1 , σ b2 ,…, σ bNm ) T}, It is the first j Background conductivity of each unit .

[0076] Furthermore, the pseudo-forward modeling technique does not directly obtain the Jacobian matrix, but instead calculates the product of the Jacobian matrix and a vector. To simplify the calculation, the Jacobian matrix is ​​implicitly obtained. Taking DC inversion as an example, The value y can be obtained by multiplying the mapping matrix Q by the solution x of a system of linear equations, i.e., y = Qx. The system of linear equations is as follows:

[0077] (10)

[0078] in, , The vector v is the conjugate direction from step five.

[0079] Similarly, the transpose of the Jacobian matrix also employs a pseudo-forward modeling technique, through J... T u is implicitly computed. Vector u is related to Jv, and according to the left-hand side of equation (7), vector u is expressed as .

[0080] Furthermore, the linear equation system (10) is solved using the Multifrontal Massively Parallel Sparse Direct Solver (MUMPS) to obtain the solution vector x. Solving the linear equation system (10) involves three steps: analysis, factorization, and solution.Figure 3 As shown, a hybrid MPI+OpenMP scheme is used for parallel computation. First, the total number of sources, the number of MPI processes, and the number of OpenMP threads associated with each MPI process are determined. One MPI process is selected to perform the communication task, and the different right-hand side terms are evenly distributed among the remaining MPI processes. Each MPI process independently forms the linear equation system required for the forward and pseudo-forward problems; this process is the analysis. Then, MUMPS is used to factorize the linear equation system. The linear equation system is divided into blocks based on the number of rows in the coefficient matrix K and the number of associated OpenMP threads. The computation of different blocks is evenly distributed among the associated OpenMP threads; this process is the factorization. If multiple OpenMP threads are allocated to each MPI process, the assembly and decomposition of the coefficient matrices for the forward and inverse problems can also be parallelized using OpenMP. Finally, after the above analysis and factorization process, the linear equation system is subjected to hybrid MPI+OpenMP parallel computation.

[0081] Step 7: Solve for the model parameter increments based on the Jacobian matrix and the normal equation. δ m The model is updated incrementally, and the fit difference after the model update is calculated. .

[0082] Step 8: Determine the fit difference If the model update is small enough and the target data fit is within a predefined threshold, then return to step four for iterative calculation; otherwise, terminate the iteration and the inversion ends.

[0083] Furthermore, the iteration described in step eight is achieved by controlling the magnitude of the regularization parameter λ. The initial value of λ is very small, and its value doubles with each iteration. When the data fit difference approaches the target data fit difference, the magnitude of the change in λ decreases, in order to obtain a fit difference as close as possible to the target data fit difference. .

[0084] By inverting the measured data through the above process, a tetrahedral mesh model containing the apparent resistivity parameter can be obtained. The vertical slice diagram of the apparent conductivity of the model is shown below. Figure 6 As shown, Figure 6 (a) shows the apparent conductivity distribution of the model on a vertical profile at 475m north. Figure 6 (b) shows the apparent conductivity distribution of the model on a horizontal profile at a depth of 20m underground. Figure 6 (c) shows the apparent conductivity distribution of the model on a horizontal profile at a depth of 180m underground, which matches well with the initial synthetic model.

[0085] Example 2

[0086] One embodiment of this disclosure provides a high-efficiency parallel inversion system for DC / IP data, comprising:

[0087] The observation data acquisition module is configured to: acquire actual DC / IP observation data of the target area, and obtain synthetic observation data by adding noise;

[0088] The initial model building module is configured to build an initial underground model based on the existing geological information of the target area.

[0089] The model inversion update module is configured to: based on synthetic observation data, iteratively invert and update the initial subsurface model using the Gauss-Newton method until the iteration stops, thus obtaining the final subsurface model;

[0090] The iterative inversion update of the initial subsurface model using the Gauss-Newton method involves linearizing the optimization problem of DC / IP inversion to obtain the normal equations. Based on synthetic observation data and current model parameters, the Jacobian matrix is ​​implicitly calculated using pseudo-forward modeling technology. A direct solver is used to solve the linear equations of pseudo-forward modeling, and parallel computation is performed using a hybrid MPI+OpenMP technique. Based on the Jacobian matrix and the normal equations, the model parameter increments are solved, and the subsurface model is updated using these increments.

[0091] Example 3

[0092] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned efficient parallel inversion method for DC / IP data.

[0093] Example 4

[0094] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the aforementioned efficient parallel inversion method for DC / IP data.

[0095] Example 5

[0096] One embodiment of this disclosure provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the aforementioned DC / IP data efficient parallelization inversion method.

[0097] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure One One or more processes and / or boxes Figure One A device that provides the functions specified in one or more boxes.

[0098] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure One One or more processes and / or boxes Figure One The steps of the function specified in one or more boxes.

[0099] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A DC / IP data efficient parallelization inversion method, characterized in that, The method comprises the following steps: Obtaining DC / IP actual observation data of a target area, and obtaining synthetic observation data by adding noise; Constructing an initial underground model according to existing geological information of the target area; Iteratively updating the initial underground model by Gauss-Newton method based on the synthetic observation data until a stop condition is met, and obtaining a final underground model; The iterative updating of the initial underground model by the Gauss-Newton method comprises linearizing an optimization problem of DC / IP inversion to obtain a normal equation; implicitly calculating a Jacobian matrix by pseudo-forward technology according to the synthetic observation data and current model parameters, solving a linear equation set of the pseudo-forward by a direct solver, and performing parallel computation by hybrid MPI+OpenMP technology; and solving a model parameter increment according to the Jacobian matrix and the normal equation, and updating the underground model by using the model parameter increment. The iterative updating comprises the following steps: Performing forward calculation on the latest underground model to obtain predicted data; Constructing a target function of DC / IP inversion by a regularization method based on the predicted data and the synthetic observation data; Linearizing an optimization problem of the target function to obtain a normal equation; Implicitly calculating a Jacobian matrix by pseudo-forward technology according to the synthetic observation data and current model parameters, and calculating a linear equation set of the pseudo-forward by a MUMPS direct solver and hybrid MPI+OpenMP technology; Solving a model parameter increment by a conjugate gradient method according to the Jacobian matrix and the normal equation, and updating the underground model by using the model parameter increment. The target function is expressed by a formula as follows: where λ is a regularization parameter, is the fit difference between the predicted data and the synthetic observation data, is a regularization term; The linear equation set of the pseudo-forward is expressed by a formula as follows: wherein , , is a coefficient matrix, , , v is a vector used in the pseudo-forward calculation to multiply with the Jacobian matrix; DC is a direct current resistivity method, and IP is an induced polarization method.

2. The DC / IP data high efficiency parallel inversion method of claim 1, wherein, The initial underground model is constructed by constructing an initial model for inversion according to existing geological information, and performing grid division on the initial model by using an unstructured tetrahedral mesh to generate a tetrahedral mesh model required for inversion.

3. The DC / IP data high efficiency parallel inversion method of claim 1, wherein, The parallel computation by hybrid MPI+OpenMP technology is performed by determining the number of MPI processes and the number of OpenMP threads associated with each MPI process according to the total number of sources, performing parallel computation by hybrid MPI+OpenMP scheme, independently forming a linear equation set required for forward and pseudo-forward problems by each process, and then factorizing the linear equation set by MUMPS.

4. A DC / IP data high-efficiency parallel inversion system, characterized in that, The method comprises the following steps: An observation data acquisition module is configured to acquire DC / IP actual observation data of a target area, and obtain synthetic observation data by adding noise; An initial model construction module is configured to construct an initial underground model according to existing geological information of the target area; A model inversion updating module is configured to iteratively update the initial underground model by Gauss-Newton method based on the synthetic observation data until a stop condition is met, and obtain a final underground model. The inversion updating by the Gauss-Newton method to the initial subsurface model is linearization of an optimization problem of DC / IP inversion to obtain a method equation; the Jacobian matrix is implicitly calculated by pseudo-forward technology according to the synthetic observation data and the current model parameter, a direct solver is used to solve the linear equation set of pseudo-forward, and hybrid MPI+OpenMP technology is used for parallel calculation; the model parameter increment is solved according to the Jacobian matrix and the method equation, and the subsurface model is updated by using the model parameter increment; The inversion updating specifically comprises the following iterative steps: forward calculation is performed on the latest subsurface model to obtain predicted data; a regularization method is used to construct an objective function of DC / IP inversion based on the predicted data and the synthetic observation data; the optimization problem of the objective function is linearized to obtain a method equation; the Jacobian matrix is implicitly calculated by pseudo-forward technology according to the synthetic observation data and the current model parameter, and the linear equation set of pseudo-forward is calculated by using a MUMPS direct solver and hybrid MPI+OpenMP technology; the model parameter increment is solved by using a conjugate gradient method according to the Jacobian matrix and the method equation, and the subsurface model is updated by using the model parameter increment; the objective function is expressed by a formula as follows: where λ is a regularization parameter, is the fit difference between the predicted data and the synthetic observation data, is the regularization term; the linear equation set of pseudo-forward is expressed by a formula as follows: wherein , , is a coefficient matrix, , , v is a vector used in the pseudo-forward calculation to multiply with the Jacobian matrix; DC is a direct current resistivity method, and IP is an induced polarization method.

5. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to realize the DC / IP data efficient parallel inversion method of any one of claims 1-3.

6. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium is used to store computer instructions, and the computer instructions are executed by the processor to realize the DC / IP data efficient parallel inversion method of any one of claims 1-3.

7. An electronic device, comprising: It comprises: a processor, a memory and a computer program; wherein the processor is connected with the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to make the electronic device execute the DC / IP data efficient parallel inversion method of any one of claims 1-3.

Images