Deep learning based method and device for magnetotelluric data inversion based on sub-region coding
By dividing the magnetotelluric data inversion area into sub-regions and using a deep learning model to generate codes, the problem of difficulty in embedding prior information in existing technologies is solved, improving the accuracy and resolution of the inversion, and reducing the complexity and computational cost of the training set.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TSINGHUA UNIVERSITY
- Filing Date
- 2023-01-10
- Publication Date
- 2026-06-23
AI Technical Summary
Existing magnetotelluric data inversion methods suffer from problems such as high ill-conditioning when incorporating prior information, difficulty in embedding complex prior information, and high requirements for training set diversity and scale, resulting in insufficient inversion accuracy and resolution.
The inversion area is divided into multiple sub-regions. A deep learning model is used to generate codes for regions with and without prior knowledge. The geoelectric model is reparameterized through direct mapping, embedding multiple prior information to improve the accuracy and resolution of the inversion.
It enables flexible embedding of prior information with different complexities and spatial ranges, improving the accuracy and resolution of inversion while reducing the complexity and computational cost of the training set.
Smart Images
Figure CN116186538B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of geophysical inversion imaging technology, and in particular to a deep learning magnetotelluric data inversion method and apparatus based on sub-region coding. Background Technology
[0002] Magnetotelluric (MT) is a geophysical electromagnetic detection method. It infers the electrical conductivity of underground structures by measuring the electric and magnetic fields generated by natural sources, and is widely used in oil and gas exploration, mineral exploration, and the study of crustal and upper mantle structures.
[0003] Magnetotelluric (MT) data inversion typically discretizes the model into a grid, assigns a conductivity parameter to each grid point, and reconstructs the conductivity at each grid point during the inversion. Under the Gaussian assumption, a deterministic inversion framework is employed, iteratively optimizing the objective function of the inverse problem. The objective function generally includes data residuals and a suitable regularization term, typically based on L2 or L1 norm regularization. This pixel-based method suffers from somewhat vague boundary characterization and lacks fine-structure resolution. Geophysicists possess various prior knowledge about the survey area, but this prior knowledge has not been used to constrain the reconstruction of the magnetotelluric model.
[0004] Model reparameterization methods rationally construct the parameter space describing the model, thereby embedding prior information into the inversion. Existing inversion methods based on model reparameterization mainly include model-based inversion and parameter transformation-based inversion. In model-based inversion, layers and blocks in the survey area model are characterized by open and closed polygons, respectively, and model reconstruction is completed by inverting the coordinates of the center points and vertices of the polygons. This method incorporates prior information by constructing an initial model, reducing the number of parameters and flexibly constraining the topological structure of the inverted model. However, this method has a high degree of uncertainty and is difficult to incorporate prior information with different degrees of uncertainty. In parameter transformation-based inversion, the coefficients of the model in the transform domain are obtained through parameter transformations such as PCA, SVD, or wavelet transform, and the model is reconstructed by inverting these coefficients. This method uses specific basis functions to represent prior information. However, the basis functions designed in this way have a weak ability to represent prior information and are difficult to embed prior information with complex patterns.
[0005] Deep learning-based reparameterization methods can automatically select the parameter space based on the training set incorporating priors, flexibly enabling model reparameterization. Deep generative models, including variational autoencoders (VAEs) and generative adversarial networks (GANs), can construct a parameter latent space through self-supervised training. In inversion, the model is reconstructed by encoding the inverted model in the latent space. However, current encoding methods generally encode the entire inversion region. In high-dimensional and complex backgrounds, this approach places high demands on the diversity and scale of the training set, and the flexibility in training set generation is insufficient, making it difficult to embed prior information with varying degrees of uncertainty and limited scope. Summary of the Invention
[0006] The present invention aims to at least partially solve one of the technical problems in the related art.
[0007] To address this, this invention proposes a deep learning-based magnetotelluric data inversion method based on sub-region coding. The inversion area is divided into several sub-regions according to a certain criterion. Within each sub-region, a training set is generated using prior information. Codings with and without prior regions are generated using a deep learning model and direct mapping, respectively, achieving model reparameterization. The codings of all sub-regions are inverted simultaneously, and the geoelectric model is recovered from each coding. This allows for the flexible integration of prior information into magnetotelluric data inversion, improving inversion accuracy and resolution.
[0008] A second aspect of the present invention is to propose a deep learning magnetotelluric data inversion device based on sub-region coding.
[0009] To achieve the above objectives, this invention proposes a deep learning-based magnetotelluric data inversion method based on sub-region coding, comprising:
[0010] The inversion area of the magnetotelluric data is divided into multiple coded sub-regions;
[0011] A training dataset is constructed based on the prior information of the multiple encoded sub-regions;
[0012] The training dataset is used to train a deep learning model to generate codes with prior regions, and direct mapping is used to generate codes without prior regions, so as to reparameterize the geoelectric model and obtain the reparameterization result.
[0013] The encoding is optimized based on the reparameterization results to achieve magnetotelluric data inversion.
[0014] The deep learning-based magnetotelluric data inversion method based on sub-region coding implemented in this invention may also have the following additional technical features:
[0015] Furthermore, the objective function for the magnetotelluric data inversion is:
[0016]
[0017]
[0018]
[0019] Among them, {v i} represents the set of latent variables for all subregions, and v represents the set of latent variables for the entire region, composed of the latent variables of the subregions; {D i} represents the family of mappings from latent variables to geoelectric models across all subregions, {Di (v i Let} be the set of all sub-region geoelectric models, D(v) be the whole-region geoelectric model composed of sub-region geoelectric models, F be the forward operator for solving the magnetotelluric forward problem, R be the regularization term describing the smoothness of the geoelectric model, S be the regularization term for the constraint coding norm, and d be the set of all sub-region geoelectric models. obs The data represents magnetotelluric observation data, where α, β, and γ are parameters used to adjust various parameters.
[0020] Furthermore, the mapping family includes neural networks and direct mappings, wherein the neural networks include networks with various structures.
[0021] Furthermore, the regularization term R includes a first plurality of forms, and the regularization term S includes a second plurality of forms.
[0022] Furthermore, the objective function is minimized, including various minimization methods.
[0023] To achieve the above objectives, another aspect of the present invention proposes a deep learning magnetotelluric data inversion device based on sub-region coding, comprising:
[0024] The geoelectric model partitioning module is used to divide the inversion area of magnetotelluric data into multiple coded sub-regions;
[0025] The training set construction module is used to construct a training dataset based on the prior information of the multiple encoded sub-regions;
[0026] The reparameterization module is used to train a deep learning model using the training dataset to generate codes with prior regions and to generate codes without prior regions using direct mapping, so as to reparameterize the geoelectric model and obtain the reparameterization result.
[0027] The inversion module is used to optimize the encoding based on the reparameterization results in order to achieve magnetotelluric data inversion.
[0028] The deep learning magnetotelluric data inversion method and apparatus based on sub-region coding of the present invention can flexibly embed a variety of ubiquitous prior knowledge, ensuring low cost of deep learning training and improving the accuracy and resolution of inversion.
[0029] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0030] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:
[0031] Figure 1This is a flowchart of a deep learning magnetotelluric data inversion method based on sub-region coding according to an embodiment of the present invention;
[0032] Figure 2 This is an architecture diagram of a deep learning magnetotelluric data inversion method based on sub-region coding according to an embodiment of the present invention.
[0033] Figure 3 This is a schematic diagram of the geoelectric model partitioning method according to an embodiment of the present invention;
[0034] Figure 4 This is a schematic diagram of the dataset used in neural network training according to an embodiment of the present invention;
[0035] Figure 5 This is a schematic diagram of a simulation test according to an embodiment of the present invention;
[0036] Figure 6 This is a schematic diagram of the structure of a variational autoencoder (VAE) according to an embodiment of the present invention;
[0037] Figure 7 This is a structural diagram of a deep learning magnetotelluric data inversion device based on sub-region coding according to an embodiment of the present invention. Detailed Implementation
[0038] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0039] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0040] The following description, with reference to the accompanying drawings, describes a deep learning magnetotelluric data inversion method and apparatus based on sub-region coding, according to embodiments of the present invention.
[0041] Figure 1 This is a flowchart of a deep learning magnetotelluric data inversion method based on sub-region coding, according to an embodiment of the present invention.
[0042] like Figure 1 As shown, the method includes, but is not limited to, the following steps:
[0043] S1 divides the inversion area of magnetotelluric data into multiple coded sub-regions;
[0044] S2, construct the training dataset based on prior information from multiple coding sub-regions;
[0045] S3, use the training dataset to train a deep learning model to generate codes with prior regions, and use direct mapping to generate codes without prior regions, so as to reparameterize the geoelectric model and obtain the reparameterization result.
[0046] S4 optimizes the encoding based on the reparameterization results to achieve magnetotelluric data inversion.
[0047] Specifically, such as Figure 2 As shown, this invention divides the inversion region into several coded sub-regions, constructs a training set based on the prior information of each sub-region, and uses deep learning methods and direct mapping to construct the latent variable space of each coded sub-region. By simultaneously reconstructing the latent variable set corresponding to the geoelectric model, magnetotelluric data inversion is achieved.
[0048] Furthermore, the objective function has the following form:
[0049]
[0050]
[0051]
[0052] Among them, {v i} represents the set of latent variables for all subregions, and v represents the set of latent variables for the entire region, composed of the latent variables of the subregions; {D i} represents the family of mappings from latent variables to geoelectric models across all subregions, {D i (v i Let} be the set of all sub-region geoelectric models, D(v) be the whole-region geoelectric model composed of sub-region geoelectric models, F be the forward operator for solving the magnetotelluric forward problem, R be the regularization term describing the smoothness of the geoelectric model, and S be the regularization term for the constraint coding norm. obs This is magnetotelluric observation data. α, β, and γ are parameters used to adjust various parameters.
[0053] Furthermore, the objective function is minimized through an iterative method.
[0054] Furthermore, the geoelectric model sub-regions include various division methods.
[0055] Furthermore, the family of mappings includes a combination of neural networks and direct mappings, wherein the neural networks include various structures.
[0056] Furthermore, the regular term R includes various forms.
[0057] Furthermore, the regular term S includes various forms.
[0058] Furthermore, methods for minimizing the objective functional include various minimization techniques.
[0059] As an example, the present invention requires dividing the resistivity model corresponding to the measurement area to be inverted into several sub-regions, such as... Figure 3 As shown, the boundaries of each sub-region are delineated by black lines. All sub-regions include several parallel regions in the middle of the inversion region, as well as the complement of the former in the entire region. The division method is determined based on prior knowledge of the survey area and the characteristics of the inversion problem. For sub-regions with prior information, a training set is generated based on the prior information to complete the deep learning model training, such as... Figure 4 The image shows the dataset used in neural network training. Assuming... Figure 3 The central rectangular sub-region contains prior information, and a training set conforming to this prior is generated using computer simulation. A latent variable space conforming to the prior information is then constructed. For sub-regions lacking prior information, the model's resistivity attribute values are directly used to construct the encoding.
[0060] In one embodiment of this invention, there are many choices for the deep learning model, one of which is a variational autoencoder. A variational autoencoder is a deep neural network structure consisting of a cascaded encoder E and a decoder D, as shown in the schematic diagram below. Figure 6 As shown. The objective function during its training process is:
[0061] L b =L+D KL
[0062] in
[0063]
[0064]
[0065] In the formula, m represents the label of the geoelectric model in the training set. For the ground power model output by the decoder, μ ε and Let be the outputs of the encoder, and c be a constant. The objective function is optimized using the Adam algorithm until convergence. The decoder D is then taken, which represents the latent variable-geoelectric model mapping based on the deep neural network.
[0066] Furthermore, when R uses the smoothest constraint and S uses the reference encoding for constraint, the objective function for optimization is specifically written as:
[0067]
[0068]
[0069]
[0070]
[0071] in Gradient operators in the depth and horizontal directions, respectively, β h ,β v These are the regularization coefficients for the depth and horizontal directions, respectively, v. i, Let v be the reference encoding for the i-th sub-region. The encodings of all sub-regions v are updated iteratively using the Gauss-Newton method. In the k-th iteration, the gradient vector and Hessian matrix of the objective function are written as:
[0072]
[0073]
[0074] Among them, J D and J F These are the sensitivity matrices for the coding-geoelectric model mapping and the magnetotelluric forward problem, respectively. D The sensitivity matrix J mapped by the coding-geoelectric model of all sub-regions Di composition:
[0075]
[0076] The update direction p at step k k Solve the following system of linear equations:
[0077] H(v k )p k =-g(v k )
[0078] The inversion code is updated according to the following formula:
[0079] v k+1 =v k +a k p k
[0080] In the formula a k The step size is obtained by line search to update the step size.
[0081] Repeat the above iterations until the termination condition is met, then output the final code v. The reconstructed geoelectric model m is then:
[0082]
[0083] m i =D i(v i )
[0084] In summary, the simulation test results of this invention are as follows: Figure 5 As shown. Among them, Figure 5 (a) in the figure is the real resistivity model used for simulation testing, and simulated magnetotelluric observation data is generated based on this resistivity model. Figure 5 (b) in the figure represents the traditional pixel-based inversion reconstruction result. Figure 5 In the diagram, (c) represents the deep learning inversion reconstruction result based on sub-region coding. The method proposed in this invention outperforms traditional algorithms in terms of reconstruction resolution and accuracy.
[0085] The deep learning magnetotelluric data inversion method based on sub-region coding according to embodiments of the present invention can flexibly embed prior information with different complexities, certainties and spatial ranges; can adaptively improve the inversion effect; can enhance the reconstruction resolution in specific regions or directions; and has low training set complexity and training computation.
[0086] To achieve the above embodiments, such as Figure 7 As shown, this embodiment also provides a deep learning magnetotelluric data inversion device 10 based on sub-region coding. The device 10 includes a geoelectric model partitioning module 100, a training set construction module 200, a reparameter module 300, and an inversion module 400.
[0087] The geoelectric model partitioning module 100 is used to divide the inversion area of magnetoelectric data into multiple coded sub-regions.
[0088] Training set construction module 200 is used to construct a training dataset based on prior information from multiple encoded sub-regions;
[0089] The reparameter module 300 is used to train a deep learning model using the training dataset to generate codes with prior regions and to generate codes without prior regions using direct mapping, so as to reparameterize the geoelectric model and obtain the reparameterization result.
[0090] The inversion module 400 is used to optimize the encoding based on the reparameterization results in order to achieve magnetotelluric data inversion.
[0091] Furthermore, the objective function for magnetotelluric data inversion is:
[0092]
[0093]
[0094]
[0095] Among them, {v i} represents the set of latent variables for all subregions, and v represents the set of latent variables for the entire region, composed of the latent variables of the subregions; {D i} represents the family of mappings from latent variables to geoelectric models across all subregions, {D i (v i Let} be the set of all sub-region geoelectric models, D(v) be the whole-region geoelectric model composed of sub-region geoelectric models, F be the forward operator for solving the magnetotelluric forward problem, R be the regularization term describing the smoothness of the geoelectric model, S be the regularization term for the constraint coding norm, and d be the set of all sub-region geoelectric models. obs The data represents magnetotelluric observation data, where α, β, and γ are parameters used to adjust various parameters.
[0096] Furthermore, the mapping family includes neural networks and direct mappings, where neural networks include networks with various structures.
[0097] Furthermore, the regularization term R includes a first set of multiple forms, and the regularization term S includes a second set of multiple forms.
[0098] Furthermore, the objective function is minimized, including various minimization methods.
[0099] The deep learning magnetotelluric data inversion device based on sub-region coding according to embodiments of the present invention can flexibly embed prior information with different complexities, certainties and spatial ranges; can adaptively improve the inversion effect; can enhance the reconstruction resolution in specific regions or directions; and has low training set complexity and training computation.
[0100] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.
[0101] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform a method for determining the sediment content in flowing water according to the above embodiments.
[0102] Computer program code for performing the operations of this disclosure can be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, and C++, and conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0103] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0104] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.
[0105] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.
[0106] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.
[0107] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0108] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.
[0109] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.
[0110] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.
Claims
1. A deep learning-based magnetotelluric data inversion method based on sub-region coding, characterized in that, Includes the following steps: The inversion area of the magnetotelluric data is divided into several coded sub-regions, including coded sub-regions with prior knowledge and coded sub-regions without prior knowledge. The coded sub-regions with prior knowledge include several rectangular coded sub-regions arranged side by side within the inversion area. The coded sub-regions without prior knowledge are the complement of the former in the whole area. The division method is determined based on the prior knowledge of the survey area and the characteristics of the inversion problem. A training dataset is constructed based on prior information from multiple prior encoded sub-regions. The deep learning model is trained using the training dataset to generate the encoding of the prior encoded sub-region, thus completing the reparameterization of the model with the prior encoded sub-region. The encoding of the non-prior encoded sub-region is directly constructed using the resistivity attribute value of the model, thus completing the reparameterization of the model without the prior encoded sub-region. In this way, the reparameterization of the entire geoelectric model is completed and the reparameterization result is obtained. The encoding is optimized based on the reparameterization results to achieve magnetotelluric data inversion; The objective function for magnetotelluric data inversion is: in, The set of latent variables for all subregions. The hidden variables of the entire region are composed of the hidden variables of the subregions; This represents the family of mappings from latent variables to geoelectric models across all sub-regions. This is the set of geoelectric models for all sub-regions. A full-region geoelectric model composed of sub-regional geoelectric models. To find the forward operator for solving the magnetotelluric forward problem, The regularization term describing the smoothness of the geoelectric model, To constrain the regularization term of the encoding norm, For magnetotelluric observation data, and To adjust the parameters of each item.
2. The method according to claim 1, characterized in that, The mapping family includes neural networks and direct mappings, wherein the neural networks include networks with various structures.
3. The method according to claim 1, characterized in that, The regularization term The regularization term S includes a first set of multiple forms, and the regularization term S includes a second set of multiple forms.
4. The method according to claim 1, characterized in that, Minimize the objective function, including various minimization methods.
5. A deep learning magnetotelluric data inversion device based on sub-region coding using the method described in claim 1, characterized in that, include: The geoelectric model partitioning module is used to divide the inversion area of magnetotelluric data into several coded sub-regions, including coded sub-regions with prior knowledge and coded sub-regions without prior knowledge. The coded sub-regions with prior knowledge include several rectangular coded sub-regions arranged side by side within the inversion area, while the coded sub-regions without prior knowledge are the complement of the former in the whole area. The partitioning method is determined based on the prior knowledge of the survey area and the characteristics of the inversion problem. The training set construction module is used to construct a training dataset based on prior information from multiple prior encoded sub-regions. The reparameterization module is used to train a deep learning model using the training dataset to generate codes for sub-regions with prior codes, and to complete the reparameterization of the model with prior codes. It also directly uses the resistivity attribute values of the model to construct codes for sub-regions without prior codes, and completes the reparameterization of the model without prior codes, thereby completing the reparameterization of the entire geoelectric model and obtaining the reparameterization result. The inversion module is used to optimize the encoding based on the reparameterization results in order to achieve magnetotelluric data inversion. The objective function for magnetotelluric data inversion is: in, The set of latent variables for all subregions. The hidden variables of the entire region are composed of the hidden variables of the subregions; This represents the family of mappings from latent variables to geoelectric models across all sub-regions. This is the set of geoelectric models for all sub-regions. A full-region geoelectric model composed of sub-regional geoelectric models. To find the forward operator for solving the magnetotelluric forward problem, The regularization term describing the smoothness of the geoelectric model, To constrain the regularization term of the encoding norm, For magnetotelluric observation data, and To adjust the parameters of each item.
6. The apparatus according to claim 5, characterized in that, The mapping family includes neural networks and direct mappings, wherein the neural networks include networks with various structures.
7. The apparatus according to claim 5, characterized in that, The regularization term The regularization term S includes a first set of multiple forms, and the regularization term S includes a second set of multiple forms.
8. The apparatus according to claim 5, characterized in that, Minimize the objective function, including various minimization methods.