A method and system for magnetotelluric multi-parameter inversion for mineral exploration

Abstract

The application is suitable for the field of geophysical technology, and provides a magnetotelluric multi-parameter inversion method for mineral exploration, comprising: constructing a synthetic training set integrating current-type distortion interference; constructing a joint inversion model; based on the synthetic training set, using a joint supervision training method of multi-parameter collaborative solution, and an error back propagation optimization method to train the joint inversion model, so that the joint inversion model synchronously outputs a resistivity model and a current-type distortion tensor. The inversion method can effectively suppress current-type distortion interference and accurately reconstruct the resistivity structure, has excellent robustness and generalization performance, can improve the resolution and reliability of MT deep electrical structure inversion imaging in complex areas, and provides key technical support for efficient and accurate exploration of deep mineral resources.

Description

Technical Field

[0001] This application belongs to the field of geophysical technology, and in particular relates to a magnetotelluric multi-parameter inversion method and system for mineral exploration. Background Technology

[0002] Magnetotelluric (MT) is an important method in geophysics used to infer the electrical structure distribution of subsurface media, such as resistivity or conductivity, based on magnetotelluric field data observed on the ground. It has advantages such as large detection depth, sensitivity to low-resistivity bodies, and low construction costs, making it a crucial tool for deep mineral resource exploration. However, as the geological structures studied become increasingly complex, the observational data are susceptible to interference from current-type distortions, which can introduce spurious electrical structures into subsequent inversion models.

[0003] In existing technologies, distortion correction methods based on impedance tensor decomposition, phase tensor decomposition methods, and static correction methods have been proposed to address the aforementioned current-type distortion problem. However, these methods typically rely on dimensionality assumptions about the target region's structure and often fail to obtain complete current-type distortion parameters. Furthermore, inversion methods that consider current-type distortion have been proposed, such as refining the shallow subsurface inversion grid or setting small-scale electrical anomalies in the shallow part of the initial model to simulate the impact of current-type distortion. However, the actual near-surface electrical inhomogeneities can reach centimeter scales, making realistic simulation difficult. Alternatively, phase tensor data unaffected by current-type distortion can be used for inversion. However, the effective observations of the phase tensor are reduced during the conversion process and do not contain amplitude information, leading to information loss, decreased inversion resolution, and strong dependence on the initial model. Another approach is to use complete current-type distortion parameters as inversion unknowns and solve them simultaneously with resistivity parameters. However, increasing the number of unknowns exacerbates the ill-posedness of the inversion, reduces reliability, and consumes memory and time when dealing with large-scale inversion problems, thus reducing inversion efficiency.

[0004] Therefore, there is an urgent need for an inversion method that can effectively suppress the interference of current-type distortion while accurately obtaining the inversion results of resistivity model, so as to improve the reliability of inversion interpretation. Summary of the Invention

[0005] This application provides a magnetotelluric multi-parameter inversion method and system for mineral exploration, which can solve one of the above-mentioned problems in the prior art.

[0006] In a first aspect, embodiments of this application provide a magnetotelluric multi-parameter inversion method for mineral exploration, including: Construct a synthetic training set incorporating current-mode distortion interference; Construct a joint inversion model; Based on the synthetic training set, a joint supervised training method with multi-parameter collaborative solution and an error backpropagation optimization method are used to train the joint inversion model, so that the joint inversion model can simultaneously output resistivity model and current-type distortion tensor.

[0007] Furthermore, the construction of the synthetic training set incorporating current-mode distortion interference includes: A two-dimensional resistivity model library was established using a random synthesis method. Based on the statistical distribution characteristics of the current-mode distortion tensor, current-mode distortion parameters are randomly generated, and a database of current-mode distortion parameters is established. A synthetic training set is generated based on the two-dimensional resistivity model library and the current-type distortion parameter database.

[0008] Furthermore, the generation of a synthetic training set based on the two-dimensional resistivity model library and the current-type distortion parameter database includes: For each two-dimensional resistivity model in the two-dimensional resistivity model library, the corresponding regional impedance tensor is generated through MT forward modeling. A two-dimensional resistivity model is randomly selected from the two-dimensional resistivity model library as the first synthesis parameter, and a current-type distortion parameter is randomly selected from the current-type distortion parameter database as the second synthesis parameter. Based on the region impedance tensor of the first synthesis parameters, and combined with the second synthesis parameters, the synthesized observation impedance tensor is calculated. A synthetic training set is generated by integrating multiple randomly generated synthetic observation impedance tensors and combining the first synthetic parameter, the second synthetic parameter, and the regional impedance tensor corresponding to each synthetic observation impedance tensor.

[0009] Furthermore, the joint inversion model includes a shared feature encoding module, a resistivity reconstruction module, and a current-type distortion parameter estimation module; The shared feature encoding module is used to extract deep common features of the input parameters and input them into the resistivity reconstruction module and the current-type distortion parameter estimation module; The resistivity reconstruction module is used to output a predicted resistivity model. The current-mode distortion parameter estimation module is used to output the predicted current-mode distortion tensor.

[0010] Furthermore, the training of the joint inversion model based on the synthetic training set, using a joint supervised training method with multi-parameter collaborative solution and an error backpropagation optimization method, includes: Using the synthetic observation impedance tensor in the synthetic training set as input parameters, deep common features are extracted through the shared feature encoding module. The deep common features are input into the resistivity reconstruction module and the current-type distortion parameter estimation module to obtain the predicted resistivity module and the predicted current-type distortion tensor, respectively. A joint supervised training method is used to optimize the parameters of the resistivity reconstruction module and the current-type distortion parameter estimation module, respectively, and the observation impedance tensor is constrained based on the consistency constraint of the observation impedance tensor. The parameter optimization results and constraint results are backpropagated to the shared feature encoding module to jointly optimize the parameters of the shared feature encoding module.

[0011] Furthermore, the input parameters are the real and imaginary parts of the four components of the synthetic observation impedance tensor.

[0012] Furthermore, the joint supervised training method is used to optimize the parameters of the resistivity reconstruction module and the current-mode distortion parameter estimation module, respectively, including: For the resistivity reconstruction module, the error between the predicted resistivity model and the expected resistivity model is used as the resistivity reconstruction supervision term to optimize the parameters of the resistivity reconstruction module. For the current-type distortion parameter estimation module, the error between the predicted current-type distortion tensor and the expected current-type distortion tensor is used as the distortion tensor supervision term to optimize the parameters of the current-type distortion parameter estimation module.

[0013] Furthermore, the constraint on the observation impedance tensor based on the consistency constraint includes: Based on the predicted current-type distortion tensor and the region impedance tensor, a predicted observation impedance tensor is constructed. The predicted observation impedance tensor and the expected observation impedance tensor are used as observation tensor supervision terms to constrain the observation impedance tensor.

[0014] Secondly, embodiments of this application provide a magnetotelluric multi-parameter inversion system for mineral exploration, including: First processing module: used to construct a synthetic training set incorporating current-type distortion interference; The second processing module is used to construct the joint inversion model. The third processing module is used to train the joint inversion model based on the synthetic training set using a joint supervised training method with multi-parameter collaborative solution and an error backpropagation optimization method, so that the joint inversion model can simultaneously output resistivity model and current-type distortion tensor.

[0015] The beneficial effects of the embodiments in this application compared with the prior art are: This application discloses a magnetotelluric (MT) multi-parameter inversion method for mineral exploration. Based on the statistical distribution characteristics of the current-type distortion tensor, a synthetic training set incorporating current-type distortion interference is constructed to train a joint inversion model. Simultaneously, for the joint inversion model, which addresses both the current-type distortion tensor and resistivity model, a joint supervised training method with multi-parameter collaborative solution and an inverse error propagation optimization method are used to train the joint inversion model, enabling it to simultaneously solve for the resistivity model and current-type distortion parameters. This inversion method effectively suppresses current-type distortion interference and accurately reconstructs the resistivity structure, exhibiting excellent robustness and generalization performance. It can improve the resolution and reliability of magnetotelluric (MT) deep electrical structure inversion imaging in complex areas, providing key technical support for efficient and accurate exploration of deep mineral resources. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a magnetotelluric multi-parameter inversion method for mineral exploration provided in an embodiment of the present invention. Figure 2 yes Figure 1 A schematic diagram of the joint inversion model in the embodiment shown; Figure 3 yes Figure 1 Another schematic diagram of the joint inversion model in the illustrated embodiment; Figure 4 The resistivity model results are obtained by inverting out-of-distribution synthetic data through GDCNet; among them, (a) is the true synthetic resistivity model, (b) is the inversion result of data with 5% noise level distortion, and (c) is the inversion result of data with 10% noise level distortion. Figure 5 The results are the current-type distortion tensor ellipse and intensity distribution obtained by inverting out-of-distribution synthetic data through GDCNet; among them, (a) is the true distortion ellipse and intensity distribution, (b) is the distortion ellipse and intensity distribution obtained by GDCNet inversion at a noise level of 5%, and (c) is the distortion ellipse and intensity distribution obtained by GDCNet inversion at a noise level of 10%. Figure 6 It is the true distribution of the four components of the current-type distortion tensor in the out-of-distribution synthetic test set and the GDCNet inversion distribution under noisy conditions; among them, (a) Components of the current-mode distortion tensor at a 5% noise level Comparison of the true distribution and the inverted distribution; (b) Components of the current-mode distortion tensor at a 5% noise level Comparison of the true distribution and the inverted distribution; (c) Components of the current-mode distortion tensor at a 5% noise level Comparison of the true distribution and the inverted distribution; (d) represents the components of the current-mode distortion tensor at a 5% noise level. Comparison of the true distribution and the inverted distribution; (e) represents the components of the current-mode distortion tensor at a 10% noise level. Comparison of the true distribution and the inverted distribution; (f) represents the components of the current-mode distortion tensor at a 10% noise level. Comparison of the true distribution and the inverted distribution; (g) represents the components of the current-mode distortion tensor at a noise level of 10%. Comparison of the true distribution and the inverted distribution; (h) represents the components of the current-mode distortion tensor at a 10% noise level. Comparison of the true distribution and the inverted distribution; Figure 7 This is a schematic diagram of the structure of a magnetotelluric multi-parameter inversion system for mineral exploration provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0018] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0019] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0020] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0021] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0022] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0023] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0024] Please see Figure 1 As shown, this invention is a magnetotelluric multi-parameter inversion method for mineral exploration, comprising the following steps: S100. Construct a synthetic training set incorporating current-mode distortion interference; In some embodiments, step S100 above includes: A two-dimensional resistivity model library was established using a random synthesis method. Based on the statistical distribution characteristics of the current-mode distortion tensor, current-mode distortion parameters are randomly generated, and a database of current-mode distortion parameters is established. A synthetic training set is generated based on the two-dimensional resistivity model library and the current-type distortion parameter database.

[0025] Specifically, the resistivity model divides the underground space into several units, each of which is assigned a specific resistivity value to approximate the resistivity distribution of different underground geological bodies. In this embodiment, a two-dimensional resistivity model library is constructed using a random synthesis method to build a training set. In this two-dimensional resistivity model library, the resistivity distribution, geological body shape, size, and location parameters of each two-dimensional resistivity model are randomly set to simulate the complexity and diversity of real underground geological conditions.

[0026] Furthermore, publicly available research indicates that the diagonal components of the current-type distortion tensor approximately follow a Gaussian distribution with a mean of 1, while the off-diagonal components approximately follow a Gaussian distribution with a mean of 0. For example, in a study published in the Chinese Journal of Geophysics in 2016 by Li Xin et al., the phase tensor method was used to conduct statistical analysis on the current-type distortion, and the conclusion clearly pointed out that "the off-diagonal elements of the distortion tensor are quasi-symmetrically distributed near the zero point, while the diagonal elements are concentrated near 1." Therefore, based on the statistical distribution characteristics of the current-type distortion tensor, this application constructs a current-type distortion parameter database for the construction of a synthetic training set.

[0027] Specifically, for each current-mode distortion observation point, a random variable following a Gaussian distribution is generated, and the corresponding current-mode distortion parameter is assigned, as follows: ; in, , , and These are the current-mode distortion tensors. The four components, and Indicates that it follows a Gaussian distribution random variables, , Let mean and standard deviation be represented respectively. Thus, multiple current-type distortion tensors and their four components that follow a Gaussian distribution can be randomly generated to form a current-type distortion parameter database.

[0028] In some embodiments, generating a synthetic training set based on the two-dimensional resistivity model library and the current-mode distortion parameter database includes: For each two-dimensional resistivity model in the two-dimensional resistivity model library, the corresponding regional impedance tensor is generated through MT forward modeling. A two-dimensional resistivity model is randomly selected from the two-dimensional resistivity model library as the first synthesis parameter, and a current-type distortion parameter is randomly selected from the current-type distortion parameter database as the second synthesis parameter. Based on the region impedance tensor of the first synthesis parameters, and combined with the second synthesis parameters, the synthesized observation impedance tensor is calculated. A synthetic training set is generated by integrating multiple randomly generated synthetic observation impedance tensors and combining the first synthetic parameter, the second synthetic parameter, and the regional impedance tensor corresponding to each synthetic observation impedance tensor.

[0029] Furthermore, for MT observation data, when small-scale electrically inhomogeneous bodies exist near the Earth's surface with geometric scales much smaller than the skin depth of the observed electromagnetic waves, they will be affected by current-type distortion. Therefore, their observation impedance tensor can be expressed by the following formula: ; in, The observation impedance tensor represents the interference caused by current-mode distortion. The current-mode distortion tensor is a frequency-independent second-order real matrix. This represents the idealized region impedance tensor, while when the region electrical structure is two-dimensional, then... Then, the formula for calculating the observation impedance tensor in the two-dimensional case can be expressed as: ; Therefore, based on the above formula for calculating the observation impedance tensor in the two-dimensional case, and combined with the two-dimensional resistivity model and the current-type distortion tensor, a synthetic observation impedance tensor incorporating current-type distortion interference can be obtained.

[0030] Specifically, MT forward modeling refers to calculating the response of the magnetotelluric field under a known resistivity model of the subsurface medium using certain mathematical and physical methods, that is, simulating the impedance tensor observed on the ground.

[0031] It is worth noting that the two-dimensional resistivity models in the two-dimensional resistivity model library constructed in this application are randomly generated idealized models used to characterize the subsurface medium. They do not consider current-type distortion interference caused by local inhomogeneities on the surface, that is, they assume that the subsurface medium is uniform or distributed according to a certain pattern, and there are no anomalous bodies that can cause additional current. Therefore, the MT forward modeling method is used. Based on the idealized two-dimensional resistivity model, the impedance tensor obtained by theoretical calculation can be regarded as the regional impedance tensor that is not affected by current-type distortion interference. It reflects the electromagnetic response characteristics of the real subsurface geological structure under ideal conditions. It can be used as a reference standard for subsequent model training and interpretation, and can be used to compare and analyze with the impedance tensor that is actually observed to be affected by distortion interference, thereby identifying and correcting current-type distortion interference and improving the interpretation accuracy of magnetotelluric sounding data.

[0032] Therefore, for each two-dimensional resistivity model in the two-dimensional resistivity model library, the corresponding impedance tensor can be calculated using MT forward modeling. , Furthermore, this impedance tensor is an idealized region impedance tensor that is not affected by current-type distortion interference. When a two-dimensional resistivity model is randomly selected from the two-dimensional resistivity model library as the first synthesis parameter, the corresponding region impedance tensor can be obtained simultaneously. Then, a current-type distortion parameter is randomly selected from the current-type distortion parameter database as the second synthesis parameter. Combining the calculation formula of the observation impedance tensor in the two-dimensional case, the synthetic observation impedance tensor incorporating current-type distortion interference can be calculated. The randomly generated synthetic observation impedance tensors are then integrated into a synthetic training set. Each synthetic observation impedance tensor in this synthetic training set incorporates current-type distortion interference and can be used to train the inversion model constructed under current-type distortion interference.

[0033] Understandably, the synthetic training set in this application is a dataset that incorporates current-type distortion interference, constructed based on the statistical distribution characteristics of the current-type distortion tensor. In model training, it can effectively balance current-type distortion interference and resistivity model reconstruction.

[0034] In addition, each synthetic observation impedance tensor in the synthetic training set needs to be recorded in correspondence with its randomly selected two-dimensional resistivity model and its generated regional impedance tensor, as well as the current-type distortion tensor, to facilitate data tracing and retrieval for subsequent model training.

[0035] S200, Construct a joint inversion model; In this application, a deep learning inversion method is used to solve the problem of magnetotelluric multi-parameter inversion. Specifically, the deep learning inversion method is to pre-establish a synthetic training set that incorporates current-type distortion interference and perform "offline" learning and training. This enables the deep neural network to automatically acquire the nonlinear mapping function relationship between the input and output data, thereby achieving "online" real-time inversion. Compared with conventional deterministic inversion methods, it has advantages such as strong global optimization ability, high inversion efficiency, and easy incorporation of prior information.

[0036] In some embodiments, the joint inversion model includes a shared feature encoding module, a resistivity reconstruction module, and a current-mode distortion parameter estimation module; The shared feature encoding module is used to extract deep common features of the input parameters and input them into the resistivity reconstruction module and the current-type distortion parameter estimation module; The resistivity reconstruction module is used to output a predicted resistivity model. The current-mode distortion parameter estimation module is used to output the predicted current-mode distortion tensor.

[0037] This application constructs a joint inversion model for current-type distortion tensor and resistivity model to achieve rapid correction of MT observation data affected by current-type distortion and reconstruction of the resistivity model. The joint inversion model includes at least a resistivity reconstruction module and a current-type distortion parameter estimation module, which are used for resistivity model reconstruction and current-type distortion parameter estimation, respectively. In addition, a shared feature encoding module is used to process the input parameters and extract their deep common features for training the resistivity reconstruction module and the current-type distortion parameter estimation module.

[0038] In a preferred embodiment, the joint inversion model is a shared-encoding dual-task deep neural network model based on the Swin Transformer. The two branches of this model correspond to the resistivity reconstruction module and the current-mode distortion parameter estimation module, respectively. Specifically, as shown... Figure 2 As shown, in this shared-encoding dual-task deep neural network model, the shared feature encoding module includes an input block processing layer, a linear feature map, a SwinTransformer layer, and a block feature downsampling layer. The input block processing layer is used to divide the input parameters into smaller, fixed-size blocks or subsequences. The linear feature map is used to perform linear transformations on the block-processed data, which is beneficial for subsequent extraction and representation of data features. The SwinTransformer layer is used to capture long-distance dependencies and local features in the data. The block feature downsampling layer is used to perform downsampling and fusion operations on the block features processed by the SwinTransformer layer. Downsampling can further reduce the dimensionality of features, reduce the amount of computation, and retain important feature information. The fusion operation integrates the features of different blocks, summarizes the local feature information into a more global feature representation, and finally forms a deep common feature that can represent the input parameters. Both the resistivity reconstruction module and the current-type distortion tensor estimation module include a block feature upsampling extension layer, a Swin Transformer layer, and a linear feature mapping layer. In these modules, the block feature upsampling extension layer restores the feature details of deep common features and increases the feature dimension. Then, the Swin Transformer layer captures long-distance dependencies and multi-scale features. Finally, the linear feature mapping layer converts the features into the final resistivity reconstruction result and the current-type distortion tensor estimation result. In these modules, the layers work together to achieve accurate and effective resistivity reconstruction and current-type distortion tensor estimation.

[0039] In other embodiments, the joint inversion model can also be a shared-coding dual-task deep neural network model based on a residual network, where the two branches of this model correspond to a resistivity reconstruction module and a current-type distortion tensor estimation module, respectively. Specifically, as shown... Figure 3 As shown, in this shared-encoding dual-task deep neural network model, the shared feature encoding module includes a linear mapping layer, a residual unit layer, and a max-pooling layer. The linear mapping layer performs initial feature mapping on the input current-distorted observation impedance tensor. The residual unit layer extracts and enhances deep features from the input data. The max-pooling layer downsamples the features, reducing feature dimensionality and computational cost while preserving key feature information, thus gradually forming deep common features that characterize the input parameters. Both the resistivity reconstruction module and the current-distortion tensor estimation module include deconvolution layers, residual unit layers, and linear mapping layers. The deconvolution layer upsamples the deep common features to gradually restore feature resolution and detail information. The residual unit layer further extracts and optimizes the feature representation during the decoding process. The linear mapping layer maps the decoded features to resistivity reconstruction results and current-distortion tensor estimation results, respectively. Furthermore, bottleneck layers are set between the shared feature encoding module and the resistivity reconstruction module, and between the shared feature encoding module and the current-type distortion tensor estimation module, to facilitate the transition and integration of the encoded deep features. At the same time, skip connections are set between the shared feature encoding module and the corresponding layers of the two task branches to transfer the shallow detailed features from the encoding stage to the decoding stage, thereby achieving effective fusion of shallow information and deep semantic features and improving the accuracy and stability of resistivity reconstruction and current-type distortion tensor estimation.

[0040] S300. Based on the synthetic training set, the joint inversion model is trained using a multi-parameter collaborative solution method and an error backpropagation optimization method, so that the joint inversion model can simultaneously output resistivity model and current-type distortion tensor.

[0041] In some embodiments, training the joint inversion model based on the synthetic training set using a joint supervised training method with multi-parameter collaborative solution and an error backpropagation optimization method includes: Using the synthetic observation impedance tensor in the synthetic training set as input parameters, deep common features are extracted through the shared feature encoding module. The deep common features are input into the resistivity reconstruction module and the current-type distortion parameter estimation module to obtain the predicted resistivity module and the predicted current-type distortion tensor, respectively. A joint supervised training method is used to optimize the parameters of the resistivity reconstruction module and the current-type distortion parameter estimation module, respectively, and the observation impedance tensor is constrained based on the consistency constraint of the observation impedance tensor. The parameter optimization results and constraint results are backpropagated to the shared feature encoding module to jointly optimize the parameters of the shared feature encoding module.

[0042] In this application, a joint supervised training method with multi-parameter collaborative solution is adopted, which enables the joint inversion model to solve the resistivity model and the current-type distortion tensor simultaneously. Specifically, during the training phase, parameter optimization is performed on the resistivity reconstruction module and the current-type distortion parameter estimation module, and the observation impedance tensor consistency constraint is executed on the observation impedance tensor generated during the parameter optimization process that is disturbed by the current-type distortion, thereby improving the stability and reliability of joint inversion through the joint inversion model.

[0043] In some embodiments, the input parameters are the real and imaginary parts of the four components of the synthetic observation impedance tensor.

[0044] In this application, the joint inversion model is trained using the synthetic training set incorporating current-type distortion interference generated in step S100 above. The input data specifically consists of the synthetic observation impedance tensors in the synthetic training set. Furthermore, to preserve the amplitude and phase information in the distortion observation data, this embodiment uses the real and imaginary parts of the four components of the synthetic observation impedance tensor. A total of eight feature parameters were used as eight-channel input data for model training to characterize the coupling features between regional electrical structure and near-surface distortion disturbance.

[0045] Therefore, the shared feature encoding module extracts deep common features from the eight-channel input data of the synthetic observation impedance tensor, which are used for parameter optimization of the resistivity reconstruction module and the current-type distortion parameter estimation module.

[0046] Specifically, during a training iteration, the eight parameters—the real and imaginary parts of each component of the synthetic observation impedance tensor—are transmitted as eight-channel input data to the shared feature encoding module of the joint inversion model. Deep common features are obtained through forward propagation of the shared feature encoding module. ,in, To synthesize the observation impedance tensor, Indicates that the shared feature encoding module supports... Network coding, These are the deep common features output by the shared feature encoding module.

[0047] In some embodiments, the joint supervised training method is used to optimize the parameters of the resistivity reconstruction module and the current-mode distortion parameter estimation module, respectively, including: For the resistivity reconstruction module, the error between the predicted resistivity model and the expected resistivity model is used as the resistivity reconstruction supervision term to optimize the parameters of the resistivity reconstruction module. For the current-type distortion parameter estimation module, the error between the predicted current-type distortion tensor and the expected current-type distortion tensor is used as the distortion tensor supervision term to optimize the parameters of the current-type distortion parameter estimation module.

[0048] In this embodiment, during a single training iteration, the deep common features output by the shared feature encoding module are... The input resistivity reconstruction module and the current-mode distortion parameter estimation module respectively complete the parameter prediction of the resistivity model and the current-mode distortion tensor, thereby obtaining the predicted resistivity model. and predicting current-type distortion tensor , specifically ,in, This indicates that the resistivity reconstruction module has deep common features. Network coding, This indicates that the current-mode distortion parameter estimation module has deep common features. Network coding.

[0049] Furthermore, for the resistivity reconstruction module, a resistivity reconstruction supervision term is constructed based on the predicted resistivity model and the expected resistivity model. Specifically, the resistivity reconstruction supervision term represents the error between the predicted resistivity model and the expected resistivity model. For the current-type distortion parameter estimation module, a distortion tensor supervision term is constructed based on the predicted current-type distortion tensor and the expected current-type distortion tensor. The distortion tensor supervision term represents the error between the predicted current-type distortion tensor and the expected current-type distortion tensor.

[0050] Therefore, during model training, resistivity reconstruction supervision terms and distortion tensor supervision terms are introduced to update the network parameters of the corresponding modules, thereby constraining the solution accuracy of the corresponding parameters. Specifically, resistivity reconstruction supervision terms are used to constrain the accuracy of electrical structure reconstruction, and distortion tensor supervision terms are used to constrain the solution accuracy of distortion parameters.

[0051] It is worth noting that the expected resistivity model and the expected current-type distortion tensor are the two-dimensional resistivity model and current-type distortion parameters corresponding to the synthetic observation impedance tensor in the synthetic training set, respectively. Specifically, the predicted resistivity model output by the resistivity reconstruction module based on the synthetic observation impedance tensor should be close to the resistivity model in the synthetic training library corresponding to the synthetic observation impedance tensor in order to output the reconstructed resistivity model towards an ideal state. Therefore, the expected resistivity model corresponds to the resistivity model in the synthetic training set. Similarly, the expected current-type distortion tensor corresponds to the current-type distortion parameters in the synthetic training set.

[0052] In some embodiments, constraining the observation impedance tensor based on the observation impedance tensor consistency constraint includes: Based on the predicted current-type distortion tensor and the region impedance tensor, a predicted observation impedance tensor is constructed. The predicted observation impedance tensor and the expected observation impedance tensor are used as observation tensor supervision terms to constrain the observation impedance tensor.

[0053] In this application, during model training, an observation tensor supervision term is introduced to impose consistency constraints on the observation impedance tensor. Specifically, this is done after obtaining the predicted current-type distortion tensor. Under these conditions, the region impedance tensor corresponding to the synthetic observation impedance tensor used for training is combined. Obtain the predicted observation impedance tensor under current-mode distortion interference. , specifically Then it is compared with the desired observation impedance tensor. By comparing the results, the observation impedance tensor reconstruction error is formed. Using this as a monitoring term for the observation tensor to constrain the observation impedance tensor, it is understandable that this monitoring term directly acts on the predicted observation impedance tensor affected by current-type distortion. It can constrain the physical rationality of the predicted current-type distortion tensor from the perspective of data consistency and suppress inversion spurious anomalies caused by distortion parameter estimation bias.

[0054] Furthermore, for the desired observation impedance tensor, which corresponds to the synthetic observation impedance tensor in the synthetic training set, the observation impedance tensor can output in an idealized state.

[0055] Furthermore, during a training iteration, after parameter optimization based on resistivity reconstruction supervision and distortion tensor supervision, and observation impedance tensor consistency constraints based on observation tensor supervision, when the error is backpropagated to the shared feature encoding module, the gradient information from the parameter optimization process needs to be fused to jointly optimize the parameters of the shared feature encoding module. The gradient information fusion process is as follows: ; in, This represents the loss in the resistivity model, while This represents the desired resistivity model. This represents the data loss between the distortion tensor and the observation impedance tensor, while This represents the desired current-type distortion tensor. Represents the desired observation impedance tensor; express and The joint loss during backpropagation to the shared feature encoding module, and Corresponding to and The gradient information from the resistivity reconstruction module and the current-type distortion parameter estimation module is backpropagated to the shared feature encoding module.

[0056] Furthermore, regarding the data loss of the distortion tensor and the observation impedance tensor... To balance the inversion prediction performance between the resistivity model and the current-type distortion tensor, the average of the distortion tensor error and the observation impedance tensor error is taken as the data loss of the distortion tensor and the observation impedance tensor.

[0057] Therefore, by iteratively training the joint inversion model multiple times to meet the preset iteration conditions, such as reaching the preset number of iterations or each supervision term reaching the preset threshold, the joint inversion model can achieve synchronous and rapid solution of resistivity model and current-type distortion parameters under the same network framework, that is, synchronously output resistivity model and current-type distortion tensor.

[0058] In some embodiments, to reduce the difficulty of network training and help the network more accurately and efficiently mine and learn the nonlinear mapping relationship between input and output data, reasonable data standardization processing is required for the synthetic observation impedance tensor and its corresponding current-type distortion parameters and resistivity model during model training. Due to the order-of-magnitude differences between the real and imaginary parts of the synthetic observation impedance tensor, the current-type distortion parameters, and the resistivity model, the input and output data are standardized in a corresponding manner. Specifically, since the resistivity model value varies from 1 to 10000... Therefore, the logarithmic resistivity with base 10 is standardized and used as the output of the expected resistivity model of the resistivity reconstruction module. This achieves order-of-magnitude consistency with the expected current-type distortion tensor output of the current-type distortion parameter estimation module. Specifically, ,in, Represents the resistivity model. This represents the standardized resistivity model. Furthermore, Z-score standardization is performed on the eight characteristic parameters (real and imaginary parts) of each component of the synthesized impedance tensor, as follows: ,in, This represents the real or imaginary part of a component of the standardized synthetic observation impedance tensor. , These are the mean and standard deviation of the real or imaginary part of the corresponding component of the synthetic observation impedance tensor, respectively.

[0059] In one embodiment, a shared-encoding dual-task deep neural network model GDCWNet is built on the Tensorflow platform as a joint inversion model. The network is trained and predicted in CPU mode. The hardware configuration used is: Intel(R) Core(TM) i5-13600K CPU @3.50GHz, 64GB RAM. The generalization performance of the joint inversion model GDCWNet is tested in this way.

[0060] Specifically, to test the generalization performance of the joint inversion model GDCWNet, the Groom-Bailey (GB) decomposition method was used to generate the current-mode distortion tensor. ,in, This represents the synthesized current-type distortion tensor generated by the GB decomposition method. This represents the station gain, which is a scalar factor. Represents a twisted matrix. Represents the shearing matrix. This represents anisotropic matrices, each with values ​​randomly selected within a corresponding range according to a Gaussian distribution. Further, this can be achieved using the formula: The synthetic observation impedance tensor incorporating current-type distortion interference is calculated. Based on this, an out-of-distribution synthetic test set containing 1000 samples and incorporating synthetic current-type distortion interference generated by GB decomposition is constructed using the above method, and the inversion test of out-of-distribution synthetic data is carried out.

[0061] like Figure 4-5 The image shows the inversion results of a randomly selected set of sample data from an out-of-distribution synthetic test set. Performance is measured by the root mean square error (RMSE) between the inversion results and the true values. Figure 4As shown, the synthetic training set constructed based on the statistical distribution characteristics of current-type distortion still exhibits excellent generalization performance when facing current-type distortion interference generated by GB decomposition. Under the noise levels of 5% and 10%, the resistivity model inverted by GDCNet accurately recovers the electrical structure. The spatial location, scale, morphology, and resistivity value of the main anomaly are basically consistent with the real model, and no false anomalies induced by distortion-noise coupling appear.

[0062] like Figure 5 As shown, the results of the current-mode distortion tensor ellipse and intensity distribution retrieved from GDCNet are presented. The current-mode distortion intensity is used to characterize the severity of distortion at each measurement point, and its specific definition is as follows: ,in, The distortion tensor represents the intensity of current-mode distortion; it is a scalar, and a larger value indicates stronger distortion. Since normal data does not contain current-mode distortion interference, the distortion tensor... It is an identity matrix with zero distortion. From Figure 5 As can be seen, under the interference of noise, the distortion intensity of some measuring points is slightly overestimated or underestimated. However, overall, the distorted ellipse shape, the direction of the major axis of the ellipse, and the distortion intensity obtained by inversion are relatively consistent with the actual situation, and there is no obvious distortion of the ellipse shape or deviation of the intensity value.

[0063] like Figure 6 As shown, these are the four components of the current-type distortion tensor from 1000 samples in the out-of-distribution synthetic test set. , , , The true distribution and the inverted distribution under noisy conditions are plotted, and performance is measured by the root mean square error (RMSE) between the true and inverted distributions. The figure shows that the inverted distribution matches the true distribution well, with no obvious systematic shift. Furthermore, the red dashed lines x=0 and x=1 represent the statistical distribution means of the off-diagonal and diagonal components of the current-type distortion tensor, respectively. The off-diagonal component represents... as well as Therefore, it can be seen that the inversion distribution of the off-diagonal components approximately follows a Gaussian distribution with a mean of 0, while the diagonal components represent... as well as Therefore, it can be seen that the inversion distribution of the diagonal component approximately follows a Gaussian distribution with a mean of 1, just like the true distribution.

[0064] In summary, the inversion test of the out-of-distribution synthetic test set demonstrates, on the one hand, that the synthetic training set constructed by the statistical distribution characteristics of the current-type distortion tensor in this application, which incorporates current-type distortion interference, is effective and exhibits excellent generalization ability when dealing with distortion interference data generated across mechanisms. On the other hand, it also shows that the magnetotelluric multi-parameter inversion method for mineral exploration provided in this application has high stability and reliability, and can robustly solve for resistivity models and current-type distortion parameters simultaneously.

[0065] Please see Figure 7 As shown, the present invention also provides a magnetotelluric multi-parameter inversion system for mineral exploration, the system comprising: First processing module 701: used to construct a synthetic training set incorporating current-type distortion interference; Second processing module 702: Used to construct the joint inversion model; The third processing module 703 is used to train the joint inversion model based on the synthetic training set using a joint supervised training method with multi-parameter collaborative solution and an error backpropagation optimization method, so that the joint inversion model can simultaneously output resistivity model and current-type distortion tensor.

[0066] It is understandable that, such as Figure 1 The content of the magnetotelluric multi-parameter inversion method embodiment for mineral exploration shown herein is applicable to the magnetotelluric multi-parameter inversion system embodiment for mineral exploration. The specific functions implemented by the magnetotelluric multi-parameter inversion system embodiment for mineral exploration are as follows: Figure 1 The illustrated embodiment of the magnetotelluric multi-parameter inversion method for mineral exploration is the same, and the beneficial effects achieved are the same as those shown. Figure 1 The beneficial effects achieved by the illustrated embodiment of the magnetotelluric multi-parameter inversion method for mineral exploration are also the same.

[0067] It should be noted that the information interaction and execution process between the above systems are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0068] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0069] Please see Figure 8 As shown, this embodiment of the invention also provides a computer device 8, including: a memory 802 and a processor 801, and a computer program 803 stored in the memory 802. When the computer program 803 is executed on the processor 801, it implements the magnetotelluric multi-parameter inversion method for mineral exploration as described in any of the above methods.

[0070] The computer device 8 may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device 8 may include, but is not limited to, a processor 801 and a memory 802. Those skilled in the art will understand that... Figure 8 The computer device 8 is merely an example and does not constitute a limitation on the computer device 8. It may include more or fewer components than shown, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0071] The processor 801 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0072] In some embodiments, the memory 802 may be an internal storage unit of the computer device 8, such as a hard disk or memory of the computer device 8. In other embodiments, the memory 802 may be an external storage device of the computer device 8, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 8. Furthermore, the memory 802 may include both internal and external storage units of the computer device 8. The memory 802 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory 802 can also be used to temporarily store data that has been output or will be output.

[0073] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the magnetotelluric multi-parameter inversion method for mineral exploration as described in any of the above methods.

[0074] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographic device / computer device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0075] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A magnetotelluric multi-parameter inversion method for mineral exploration, characterized in that, include: Construct a synthetic training set incorporating current-mode distortion interference; Construct a joint inversion model; Based on the synthetic training set, a joint supervised training method with multi-parameter collaborative solution and an error backpropagation optimization method are used to train the joint inversion model, so that the joint inversion model can simultaneously output resistivity model and current-type distortion tensor.

2. The method as described in claim 1, characterized in that, The construction of the synthetic training set incorporating current-mode distortion interference includes: A two-dimensional resistivity model library was established using a random synthesis method. Based on the statistical distribution characteristics of the current-mode distortion tensor, current-mode distortion parameters are randomly generated, and a database of current-mode distortion parameters is established. A synthetic training set is generated based on the two-dimensional resistivity model library and the current-type distortion parameter database.

3. The method as described in claim 2, characterized in that, The process of generating a synthetic training set based on the two-dimensional resistivity model library and the current-type distortion parameter database includes: For each two-dimensional resistivity model in the two-dimensional resistivity model library, the corresponding regional impedance tensor is generated through MT forward modeling. A two-dimensional resistivity model is randomly selected from the two-dimensional resistivity model library as the first synthesis parameter, and a current-type distortion parameter is randomly selected from the current-type distortion parameter database as the second synthesis parameter. Based on the region impedance tensor of the first synthesis parameters, and combined with the second synthesis parameters, the synthesized observation impedance tensor is calculated. A synthetic training set is generated by integrating multiple randomly generated synthetic observation impedance tensors and combining the first synthetic parameter, the second synthetic parameter, and the regional impedance tensor corresponding to each synthetic observation impedance tensor.

4. The method as described in claim 1, characterized in that, The joint inversion model includes a shared feature encoding module, a resistivity reconstruction module, and a current-type distortion parameter estimation module. The shared feature encoding module is used to extract deep common features of the input parameters and input them into the resistivity reconstruction module and the current-type distortion parameter estimation module; The resistivity reconstruction module is used to output a predicted resistivity model. The current-mode distortion parameter estimation module is used to output the predicted current-mode distortion tensor.

5. The method as described in claim 4, characterized in that, The training of the joint inversion model based on the synthetic training set, using a joint supervised training method with multi-parameter collaborative solution and an error backpropagation optimization method, includes: Using the synthetic observation impedance tensor in the synthetic training set as input parameters, deep common features are extracted through the shared feature encoding module. The deep common features are input into the resistivity reconstruction module and the current-type distortion parameter estimation module to obtain the predicted resistivity module and the predicted current-type distortion tensor, respectively. A joint supervised training method is used to optimize the parameters of the resistivity reconstruction module and the current-type distortion parameter estimation module, respectively, and the observation impedance tensor is constrained based on the consistency constraint of the observation impedance tensor. The parameter optimization results and constraint results are backpropagated to the shared feature encoding module to jointly optimize the parameters of the shared feature encoding module.

6. The method as described in claim 5, characterized in that, The input parameters are the real and imaginary parts of the four components of the synthetic observation impedance tensor.

7. The method as described in claim 5, characterized in that, The method employs a joint supervised training approach to optimize the parameters of both the resistivity reconstruction module and the current-mode distortion parameter estimation module, including: For the resistivity reconstruction module, the error between the predicted resistivity model and the expected resistivity model is used as the resistivity reconstruction supervision term to optimize the parameters of the resistivity reconstruction module. For the current-type distortion parameter estimation module, the error between the predicted current-type distortion tensor and the expected current-type distortion tensor is used as the distortion tensor supervision term to optimize the parameters of the current-type distortion parameter estimation module.

8. The method as described in claim 7, characterized in that, The constraint on the observation impedance tensor based on the consistency constraint of the observation impedance tensor includes: Based on the predicted current-type distortion tensor and the region impedance tensor, a predicted observation impedance tensor is constructed. The predicted observation impedance tensor and the expected observation impedance tensor are used as observation tensor supervision terms to constrain the observation impedance tensor.

9. A magnetotelluric multi-parameter inversion system for mineral exploration, characterized in that, include: First processing module: used to construct a synthetic training set incorporating current-type distortion interference; The second processing module is used to construct the joint inversion model. The third processing module is used to train the joint inversion model based on the synthetic training set using a joint supervised training method with multi-parameter collaborative solution and an error backpropagation optimization method, so that the joint inversion model can simultaneously output resistivity model and current-type distortion tensor.

Images