Geological structure interpretation model construction method, interpretation method and equipment
By combining conditional generative adversarial networks with geological rule-based constraint loss, a geological structure interpretation model is trained, which solves the problems of low efficiency and low accuracy in existing geological structure interpretation technologies, and achieves efficient and highly accurate geological structure interpretation.
Patent Information
- Application Number
- CN202510982912.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-16
- Publication Date
- 2025-12-09
AI Technical Summary
Existing geological structure interpretation methods are computationally inefficient and inaccurate, making it difficult to meet the needs of practical applications.
A conditional generative adversarial network is adopted, combined with geological rule-constrained losses, including fault loss, fold loss and tectonic structure loss. The geological structure interpretation model is trained through generator network and discriminator network to ensure that the generated results conform to geological rules and physical laws.
It improves the accuracy and reliability of geological structure interpretation, while also increasing computational efficiency. The generated results conform to geological and physical laws, reducing reliance on manual annotation.
Smart Images
Figure CN121092984A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of geological interpretation technology, and in particular to a method for constructing a geological structure interpretation model, an interpretation method, and equipment. Background Technology
[0002] Currently, intelligent interpretation of geological structures mainly relies on traditional machine learning methods, such as Support Vector Machines (SVM), Random Forest, and Convolutional Neural Networks (CNN) in deep learning. These methods typically identify and classify geological structures by training on large amounts of geological data. For example, some existing technologies utilize seismic profile data for automatic identification of faults and folds, constructing feature vectors by extracting seismic attributes (such as amplitude, frequency, and phase), and then using machine learning algorithms for classification. However, these methods are computationally inefficient when processing large-scale geological data, resulting in low accuracy of interpretation results and failing to meet the needs of practical applications. Summary of the Invention
[0003] This invention provides a method for constructing a geological structure interpretation model, an interpretation method, and an apparatus to address the shortcomings of low efficiency and low accuracy in existing geological structure interpretation technologies, thereby achieving efficient and highly accurate geological structure interpretation.
[0004] This invention provides a method for constructing a geological structure interpretation model, comprising the following steps.
[0005] Obtain geological profile maps and corresponding geological structural interpretation data; Using the geological profile map as input to the conditional generative adversarial network and the geological structure interpretation map corresponding to the geological profile map as a label, the conditional generative adversarial network is trained to obtain the geological structure interpretation model. The conditional generative adversarial network includes a generator network, and the comprehensive loss of the generator network includes adversarial loss, reconstruction loss and geological constraint loss. The geological constraint loss is used to constrain the consistency between the geological structures generated by the generator network and geological rules.
[0006] According to the geological structure interpretation model construction method provided by the present invention, the comprehensive loss further includes: physical constraint loss; The physical constraint loss is constructed based on the gravity anomaly simulation parameters and the measured gravity anomaly parameters, and is used to constrain the consistency between the gravity anomaly simulation data and the measured gravity anomaly data. The gravity anomaly simulation data is obtained by converting the geological structures generated by the generator network through differentiable rendering technology.
[0007] According to the geological structure interpretation model construction method provided by the present invention, the geological constraint loss includes at least one of fault loss, fold loss and structural structure loss, wherein the fault loss is used to constrain the consistency between the fault geometric parameters generated by the generator network and geological rules, the fold loss is used to constrain the consistency between the fold morphology generated by the generator network and geological rules, and the structural structure loss is used to constrain the consistency between the geological structures generated by the generator network and the distribution of real geological structures in multi-scale spatial features.
[0008] According to the geological structure interpretation model construction method provided by the present invention, the geological constraint loss is as follows: in, Indicates geological constraint loss. Indicates fault loss, Indicates wrinkle loss. Indicates structural loss. The weighting coefficient represents the fault loss. This represents the weighting coefficient of the wrinkle loss. The weighting coefficient represents the loss of the constructed structure.
[0009] According to the geological structure interpretation model construction method provided by the present invention, the fault loss is as follows: in, This indicates the number of faults generated by the generator network. The generator network generates the first... i The strike angle of the fault, Indicates the reference strike angle of the fault. This indicates the dip angle of the fault generated by the generator network. Indicates the reference dip angle of the fault. This represents the tolerance threshold.
[0010] According to the geological structure interpretation model construction method provided by the present invention, the fold loss is as follows: in, This indicates the wavelength at which the generator network generates the wrinkles. Indicates the thickness of the rock strata. Indicates viscosity ratio.
[0011] According to the geological structure interpretation model construction method provided by the present invention, the structural loss is as follows: in, This indicates a differentiable renderer. Represents a generator network. Represents a geological profile. This indicates measured gravity anomaly data. This represents the density gradient.
[0012] According to the geological structure interpretation model construction method provided by the present invention, the conditional generative adversarial network further includes a discriminator network, the comprehensive loss of which is as follows: in, This represents the overall loss of the discriminator network. Indicating resistance to loss, Indicates structural loss. This represents the gradient penalty. The weighting coefficients represent the structural loss. This represents the weight coefficient of the gradient penalty.
[0013] This invention also provides a method for interpreting geological structures, comprising: Obtain the geological profile to be interpreted; The geological profile to be interpreted is input into a geological structure interpretation model constructed by any of the above-mentioned geological structure interpretation model construction methods to obtain the geological structure interpretation data corresponding to the geological profile to be interpreted.
[0014] The present invention also provides a geological structure interpretation model construction device, comprising the following modules: The sample data acquisition module is used to acquire geological profile maps and the geological structural interpretation data corresponding to the geological profile maps; The model training module is used to train the conditional generative adversarial network (GAN) using the geological profile map as input and the geological structure interpretation map corresponding to the geological profile map as labels, to obtain the geological structure interpretation model. The conditional GAN includes a generator network, and the generator network's comprehensive loss includes adversarial loss, reconstruction loss, and geological constraint loss. The geological constraint loss is used to constrain the consistency between the geological structures generated by the generator network and geological rules.
[0015] The present invention also provides a geological structure interpretation device, comprising the following modules: The module for acquiring geological profile maps to be interpreted is used to acquire geological profile maps to be interpreted. The interpretation module is used to input the geological profile to be interpreted into a geological structure interpretation model constructed by any of the above-mentioned geological structure interpretation model construction methods, and obtain the geological structure interpretation data corresponding to the geological profile to be interpreted.
[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the geological structure interpretation model construction method or geological structure interpretation method as described above.
[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the geological structure interpretation model construction method or geological structure interpretation method as described above.
[0018] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the geological structure interpretation model construction method or geological structure interpretation method as described above.
[0019] The geological structure interpretation model construction method, interpretation method, and equipment provided by this invention effectively improve the accuracy and reliability of the interpretation results and increase computational efficiency by incorporating geological rule constraint loss into the loss of the geological structure interpretation model. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0021] Figure 1 This is a flowchart illustrating the geological structure interpretation model construction method provided in this embodiment of the invention.
[0022] Figure 2 This is a flowchart illustrating the geological structure interpretation method provided in this embodiment of the invention.
[0023] Figure 3 This is a schematic diagram of the geological structure interpretation model construction device provided in the embodiment of the present invention.
[0024] Figure 4 This is a schematic diagram of the geological structure interpretation device provided in an embodiment of the present invention.
[0025] Figure 5 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0027] Figure 1 This is a flowchart illustrating the geological structure interpretation model construction method provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps: Step 101: Obtain the geological profile map and the corresponding geological structure interpretation data.
[0028] Step 102: Using the geological profile map as input to the conditional generative adversarial network (GAN), and the corresponding geological structure interpretation map as labels, the GAN is trained to obtain a geological structure interpretation model. The conditional GAN includes a generator network, whose overall loss includes adversarial loss, reconstruction loss, and geological constraint loss. The geological constraint loss is used to constrain the consistency between the geological structures generated by the generator network and geological rules.
[0029] This invention incorporates structural geological rules, such as fault strike and fold morphology, as soft constraints into the generation process. In this way, the generated geological structures not only conform to the data distribution but also to geological laws, improving the accuracy of geological structure interpretation. Furthermore, compared to simply using machine learning algorithms for geological structure interpretation without incorporating structural geological rules, this invention, by introducing constraints, saves training time for the machine learning model to some extent, thus improving efficiency.
[0030] In one example embodiment, geological rules are transformed into computable soft constraints in three forms.
[0031] (1) Mathematical formula constraints Fault strike constraint: The fault strike must conform to the direction of the regional tectonic stress field. Its mathematical expression is: in, Indicates the strike angle of the fault. Indicates the dominant trend of a region (e.g., extraction from a regional tectonic map). This represents the weighting coefficient (controlling the strength of the constraint).
[0032] Wrinkle morphology constraints: Wrinkle wavelength ( Controlled by the viscosity ratio of the rock strata (Ramberg dominant wavelength theory): in, Indicates the wavelength at which wrinkles are generated; Indicates the thickness of the rock strata. This indicates the viscosity ratio between the capable and incapable layers (preset geological parameters).
[0033] (2) Constraints of the probabilistic model Fault dip angle distribution constraints: reverse fault dip probability model The generator loss function is supplemented with a KL divergence penalty for deviations from the distribution.
[0034] Fold symmetry constraint: Difference in inclination angle between the two wings of the fold It must satisfy a Gaussian distribution: (3) Rule base constraints Fault-fold spatial relationship rules: The rule base is embedded in the generator forward propagation process in the form of logical judgments, and a high loss value is returned when a violation occurs.
[0035] The specific forms of regularity in fault strike, fold morphology, etc., include: (1) Regular fault strike The orientation is consistent with the regional tectonics: The fault strike must be parallel to the direction of the regional principal stress (such as the strike of an orogenic belt): Fault plane attitude coordination normal fault dip angle reverse fault dip angle Strike-slip fault, nearly vertical: (2) Regular fold shape Wavelength-thickness relationship: Initial wavelength of the fold: Where d represents the thickness of the hardening layer; Indicates the viscosity ratio of the hardening layer to the matrix; wavelength at which wrinkles are generated. A deviation of more than 20% from the theoretical value is considered a violation.
[0036] Curvature constraint at the turning point: Curvature at the turning point with wing angle Positive correlation: (3) Construction of combination rules: Fault cutting fold rules: When a transverse fault cuts through an anticline, the core of the uplifted disk widens.
[0037] Joint-fold association rules: radial tension joints develop at the turning point of an anticline, and conjugate shear joints develop on the wing.
[0038] Among them, the fault orientation rule uses a rotatable Gabor filter to construct an orientation-sensitive convolution kernel, and is implemented using a coordinate rotation formula. To achieve alignment of the principal stress directions in the region, and by combining gradient-based calculation-based attitude compatibility constraints (dip verification function) and over-threshold gradient amplification mechanism, we can ensure that the fault strike is consistent with the regional structure and that the attitude conforms to geological laws.
[0039] A viscosity-adaptive dynamic pooling layer is designed based on the regularity of the fold morphology, and its pooling kernel size is... Real-time response to changes in rock strata parameters, combined with curvature correction activation function A real-time wavelength monitoring process based on Fourier transform is implemented to achieve physical constraints on the wavelength-thickness relationship and the curvature at the turning point.
[0040] A spatial relationship inference engine was developed based on construction combination rules. It identifies fault / fold types using a construction element detection head and applies a joint orientation field data model. (Radial directional field) and (Conjugate shear joint field) enables a differentiable expression of the spatial relationship between faults and folds and the development law of joints.
[0041] In one example embodiment, differentiable rendering technology can be used to convert geological profiles into gravity anomaly simulation data, thereby indirectly constraining geological structures during the training process.
[0042] (1) The conversion principle from geological profile map to gravity anomaly simulation The gravity anomaly calculation based on the physical model uses a polygonal cross-section column model, and the formulas used are as follows: in, Represents the gravitational constant. Indicates the density difference of the formation. Represents the coordinates of the polygon vertices. , .
[0043] The model discretizes the geological profile into polygonal cylinders with uniform density and obtains gravity anomalies through superposition calculations.
[0044] (2) Differentiable rendering algorithm flow Input: Geological profile (including stratigraphic boundary coordinates), lithological density parameter table, and observation point coordinate sequence.
[0045] The processing includes: converting the stratigraphic boundary into a closed polygon; and applying the depth information. Formula; apply the Talwani formula to each polygon; integrate the contributions of all polygons along the observation line.
[0046] The output includes: gravity anomaly curves Gradient tensor field ( Conditional generative adversarial networks include generator networks and discriminator networks.
[0047] The above embodiments introduce some geological rules. The embodiments of the present invention integrate some of the above geological rules as constraints into the geological structure interpretation provided by the embodiments of the present invention. Specifically, the integration of geological rules into the geological structure interpretation is achieved by incorporating some geological rules into the loss of the conditional generative adversarial network.
[0048] Conditional generative adversarial networks (GANs) consist of a generator network and a discriminator network, with a core structure that can be a hybrid architecture of U-Net and ResNet. The generator network's loss (which we call the comprehensive loss) includes adversarial loss, reconstruction loss, and geological constraint loss. The geological constraint loss is used to constrain the consistency between the geological structures generated by the generator network and geological rules.
[0049] (1) Countering losses ( ): Where D represents the multi-scale discriminator, G represents the generator, x represents the real geological profile, and z represents the input noise vector.
[0050] (2) Reconstruction losses ( ): Where y represents the actual geological structure label, and SSIM represents the structural similarity index.
[0051] (3) Geological constraint loss ( This includes fault loss, folding loss, and structural loss, as shown in the following formulas: 1) Fault loss: in, Indicates the strike angle (in degrees) of the generated fault. Indicates the regional tectonic reference strike. Indicates the dip angle (in degrees) of the generated fault. Indicates the reference dip angle for fault type. This represents the tolerance threshold.
[0052] Fault loss is a loss term in geological structural interpretation models used to constrain the consistency of generated fault geometry parameters (strike, dip) with geological rules. Its core objective is to ensure that the generated faults conform to actual geological laws (such as the direction of regional tectonic stress fields and the rationality of fault attitude) and avoid results that violate geological common sense (such as faults whose strike is perpendicular to the principal stress field).
[0053] Fault loss: The strike of the fault is forcibly generated to match the direction of the regional principal stress field (e.g., θ extracted from seismic data or tectonic maps). ref The fault dip angle is limited to a geologically reasonable range (e.g., the dip angle of a reverse fault is typically 30°–60°). A certain deviation is allowed by ϵ to avoid excessively penalizing minor errors. By constraining the fault strike to align with the regional tectonic stress field, faults that "violate mechanical principles" (e.g., faults whose strike is perpendicular to the principal compressive stress) are avoided, improving interpretation accuracy. Even with limited labeled data, reasonable fault parameters can still be generated through geological rules, reducing reliance on manual annotation.
[0054] Fault loss is addressed by explicitly embedding geological rules into a generative adversarial network through directional and attitude constraints, ensuring that the generated faults conform to both data distribution patterns and actual geological conditions. This design significantly improves the reliability and practicality of geological structure interpretation.
[0055] 2) Fold Loss: in, The wavelength representing the fold formation is d, and the thickness of the rock layer is d. Indicates viscosity ratio.
[0056] Folding loss is a key loss term in geological structural interpretation models used to constrain the consistency of generated fold morphology (such as wavelength and symmetry) with geological rules. Its core objective is to ensure that the generated folds conform to natural geological laws (such as wavelength controlled by rock layer viscosity ratio and fold symmetry) and avoid results that violate geological common sense (such as folds where the wavelength does not match the rock layer thickness).
[0057] Folding loss: By constraining the physical relationship between fold wavelength and rock layer thickness, it avoids generating abnormal folds that do not conform to geological laws (such as long-wavelength folds generated in thin rock layers), thereby improving the interpretation accuracy; by explicitly embedding geological theories (such as Ramberg theory) into the loss function, it reduces the dependence on pure data-driven approaches and improves the model's generalization ability in complex geological scenarios.
[0058] Folding loss: Combined with adversarial loss, adversarial loss ensures that the overall distribution of generated folds is consistent with the real data, while folding loss constrains the rationality of local morphology. The combination of the two avoids generating folds that "seem reasonable but violate physical laws". It complements reconstruction loss, which ensures the fidelity of fold geometric details, while folding loss constrains the geological rationality of macroscopic morphology, forming a "detail-macro" dual constraint. It is linked with physical constraint loss, which verifies the physical feasibility of fold generation through gravity anomaly simulation, while folding loss further constrains rationality from the perspective of geometric morphology, forming a multi-dimensional verification.
[0059] Fold loss, through physical and geometric constraints, transforms geological rules into a computable mathematical form, ensuring that the generated folds conform to both data distribution patterns and actual geological conditions. This design significantly enhances the scientific rigor and engineering applicability of geological structure interpretation.
[0060] 3) Structural losses ( ): Single-scale structural loss: Where s represents the scale index. Indicates the width of the feature map. Indicates the feature map height. This represents the feature map at the s-th scale. Indicates a geological rule reference sample. This represents the Huber loss function.
[0061] Structural loss is used to constrain the consistency of generated geological structures with the distribution of real data across multiple scales of spatial features. Its core objective is to ensure that the generated results conform to the statistical regularities of real geological data in both the global structural pattern (such as the consistency of regional fault strike) and local structural details (such as the dip angle of fold limbs) through multi-resolution feature alignment.
[0062] The structural loss is assessed using a multi-scale feature alignment mechanism: a low-resolution branch (global features) to capture regional structural patterns (such as fault strike consistency and fold axial distribution); a medium-resolution branch (regional features) to assess the spatial correlation of structural units (such as the combination of anticlines and synclines); and a high-resolution branch (local features) to constrain the detail fidelity of geological structures (such as fault line continuity and fold inflection curvature).
[0063] The structural loss, adversarial loss (global distribution matching), reconstruction loss (detail fidelity), and physical constraint loss (physical laws) are combined to form a synergistic optimization, avoiding the bias caused by a single loss dominating the process.
[0064] The structural loss mechanism transforms the spatial statistical regularities of geological structures into optimizable mathematical objectives through a multi-scale feature alignment mechanism, ensuring that the generated results conform to the characteristics of real geological data in both global pattern and local details.
[0065] In one example embodiment, the overall loss of the generator network also includes a physical constraint loss. This physical constraint loss is a key loss term in the geological structure interpretation model used to ensure that the generated results conform to geophysical laws (such as gravity field distribution and density distribution). It is constructed based on gravity anomaly simulation parameters and measured gravity anomaly parameters, and is used to constrain the consistency between the gravity anomaly simulation data and the measured gravity anomaly data. The gravity anomaly simulation data is obtained by converting the geological structures generated by the generator network using differentiable rendering technology. Its core objective is to ensure, through physical simulation technology, that the generated geological structures are physically reasonable (such as the spatial relationship between faults and folds, and the matching of stratigraphic density distribution with gravity anomalies), avoiding results that violate physical laws (such as no corresponding structure in areas of abrupt density changes).
[0066] Physical constraint loss ( ): in, This indicates a differentiable renderer. This indicates an anomaly in the measured gravity. This represents the density gradient.
[0067] Differentiable rendering (Render) transforms the geological profile (strata boundaries, lithology) output by the generator into simulated gravity anomaly data (such as a gravity gradient tensor field). By comparing the simulated gravity anomaly (Render(G(z))) with the measured gravity anomaly, the physical plausibility of the generated results is constrained. Density gradient penalty. The density field is penalized for discontinuities (such as the need for a smooth transition between density abrupt changes on both sides of a fault).
[0068] Physical constraint loss verifies the physical rationality of the generated results through physical simulation, avoiding interpretations that are "geometrically reasonable but physically infeasible" (such as areas with abrupt density changes lacking structural explanation), thus improving interpretation reliability. Adversarial loss ensures the generated results conform to the data distribution, while physical constraint loss constrains physical laws; the combination of the two avoids generating results that are "seemingly reasonable but physically infeasible." Reconstruction loss guarantees the fidelity of geometric details, while physical constraint loss constrains macroscopic physical rationality (such as fault dip angles needing to match gravity anomalies). Geological constraint loss (such as regular fault strike) provides prior knowledge, while physical constraint loss verifies the rationality of the priors through physical simulation (such as whether the strike matches the regional stress field).
[0069] The physical constraint loss, through differentiable rendering technology and geophysical simulation, transforms the physical rationality of geological structures into an optimizable mathematical objective, ensuring that the generated results conform to both geological rules and physical laws. This design significantly enhances the reliability and engineering application value of geological interpretation results.
[0070] In one example embodiment, the overall loss of the discriminator network is as follows: in, This represents the overall loss of the discriminator network. Indicating resistance to loss, Indicates structural loss. This represents the gradient penalty. The weighting coefficients represent the structural loss. This represents the weight coefficient of the gradient penalty.
[0071] The following is a specific example illustrating the geological structure interpretation model construction method provided in this embodiment of the invention.
[0072] Data preprocessing: Converting seismic profile data into geological profile maps.
[0073] Generator training: A generator architecture with geological prior constraints is used in conjunction with differentiable rendering techniques to generate geological structures.
[0074] (1) The generator network architecture is as follows: the network type is conditional generative adversarial network (cGAN); the core structure is a hybrid architecture of U-Net and ResNet.
[0075] (2) Loss function: Total loss function: Combating loss ( ): Where D represents the multi-scale discriminator, G represents the generator, x represents the real geological profile, and z represents the input noise vector.
[0076] Reconstruction losses ( ): Where y represents the actual geological structure label, and SSIM represents the structural similarity index.
[0077] Geological constraint loss ( ): Fault loss: in, Indicates the strike angle (in degrees) of the generated fault. Indicates the regional tectonic reference strike. Indicates the dip angle (in degrees) of the generated fault. Indicates the reference dip angle for fault type. This represents the tolerance threshold.
[0078] Wrinkle loss: in, The wavelength representing the fold formation is d, and the thickness of the rock layer is d. Indicates viscosity ratio.
[0079] Physical constraint loss ( ): in, This indicates a differentiable renderer. This indicates an anomaly in the measured gravity. This represents the density gradient.
[0080] Discriminator training: Use a multi-scale discriminator network to evaluate the reasonableness of the generated results.
[0081] The loss function is as follows: Discriminator total loss function: Combating loss ( ): Where D represents the discriminator network, G represents the generator network, x represents the real geological profile sample, and z represents the generator input noise vector. Represents the true data distribution. This represents the noise vector distribution.
[0082] Structural loss ( ): Single-scale structural loss: Where s represents the scale index. Indicates the width of the feature map. Indicates the feature map height. This represents the feature map at the s-th scale. Indicates a geological rule reference sample. This represents the Huber loss function.
[0083] Gradient penalty ( ): in, This represents the linear interpolation between the real sample and the generated sample. Random interpolation coefficients, This represents the gradient over the interpolated sample.
[0084] Uncertainty quantification: Multiple interpretation schemes are generated through the Monte Carlo Dropout mechanism, and the probability distribution of the orientation of the tectonic surface is calculated.
[0085] (Uncertainty quantification is achieved through the Monte Carlo Dropout mechanism: During the testing phase, the Dropout rate set during training (usually 0.2-0.5) is enabled to sample the same input data multiple times, generating multiple interpretation schemes; based on the above results, the probability distribution of the attitude (such as dip angle and strike) of the structural surface is calculated using the kernel density estimation method, and finally, key statistics and 95% confidence intervals are output, thereby quantifying the uncertainty of the model for the interpretation results.) The detailed flowchart is as follows: [Data Preprocessing] -> [Generator Training] -> [Discriminator Training] -> [Uncertainty Quantization] This invention also provides a method for interpreting geological structures, see [link to relevant documentation]. Figure 2 The method includes the following steps: Step 201: Obtain the geological profile map to be interpreted; Step 202: Input the geological profile to be interpreted into the geological structure interpretation model constructed by the geological structure interpretation model construction method described in the above embodiment to obtain the geological structure interpretation data corresponding to the geological profile to be interpreted.
[0086] In practical applications, the technical solution of this invention can be implemented through the following steps: Data input: Input seismic profile data and perform preprocessing.
[0087] Generate geological structures: Use the generator to generate geological structures.
[0088] Reasonableness assessment: The reasonableness of the generated results is assessed using a multi-scale discriminator network.
[0089] Uncertainty quantification: Multiple interpretation schemes are generated through the Monte Carlo Dropout mechanism, and the probability distribution of the orientation of the tectonic surface is calculated.
[0090] Output results: Output the final interpretation results, including geological structure maps and uncertainty quantification results.
[0091] The following describes the geological structure interpretation model construction device and the geological structure interpretation device provided by the present invention. The geological structure interpretation model construction device described below and the geological structure interpretation model construction method described above can be referred to in correspondence with each other.
[0092] See Figure 3 The geological structure interpretation model construction device includes the following modules: The sample data acquisition module 301 is used to acquire geological profile maps and geological structural interpretation data corresponding to the geological profile maps; The model training module 302 is used to train the conditional generative adversarial network (GAN) using the geological profile map as input and the geological structure interpretation map corresponding to the geological profile map as a label, to obtain the geological structure interpretation model. The conditional GAN includes a generator network, and the comprehensive loss of the generator network includes adversarial loss, reconstruction loss, and geological constraint loss. The geological constraint loss is used to constrain the consistency between the geological structures generated by the generator network and geological rules.
[0093] See Figure 4 The geological structure interpretation device includes the following modules: The geological profile acquisition module 401 is used to acquire the geological profile to be interpreted. The interpretation module 402 is used to input the geological profile to be interpreted into the geological structure interpretation model constructed by the geological structure interpretation model construction method, and obtain the geological structure interpretation data corresponding to the geological profile to be interpreted.
[0094] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540. The processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a geological structure interpretation model construction method or a geological structure interpretation method. The geological structure interpretation model construction method includes: acquiring a geological profile map and the corresponding geological structure interpretation data; using the geological profile map as input to the conditional generative adversarial network (GAN), and using the corresponding geological structure interpretation map as a label, training the GAN to obtain the geological structure interpretation model; wherein the conditional GAN includes a generator network, and the comprehensive loss of the generator network includes adversarial loss, reconstruction loss, and geological constraint loss, the geological constraint loss being used to constrain the consistency between the geological structures generated by the generator network and geological rules. The geological structure interpretation method includes: obtaining a geological profile map to be interpreted; inputting the geological profile map to be interpreted into a geological structure interpretation model constructed by the above-mentioned geological structure interpretation model construction method to obtain geological structure interpretation data corresponding to the geological profile map to be interpreted.
[0095] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0096] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the geological structure interpretation model construction method or geological structure interpretation method provided by the above methods. The geological structure interpretation model construction method includes: acquiring a geological profile map and the geological structure interpretation data corresponding to the geological profile map; using the geological profile map as input to the conditional generative adversarial network, and using the geological structure interpretation map corresponding to the geological profile map as a label, training the conditional generative adversarial network to obtain the geological structure interpretation model; wherein, the conditional generative adversarial network includes a generator network, and the comprehensive loss of the generator network includes adversarial loss, reconstruction loss, and geological constraint loss, wherein the geological constraint loss is used to constrain the consistency between the geological structures generated by the generator network and geological rules. The geological structure interpretation method includes: obtaining a geological profile map to be interpreted; inputting the geological profile map to be interpreted into a geological structure interpretation model constructed by the above-mentioned geological structure interpretation model construction method to obtain geological structure interpretation data corresponding to the geological profile map to be interpreted.
[0097] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the geological structure interpretation model construction method or geological structure interpretation method provided by the above methods. The geological structure interpretation model construction method includes: acquiring a geological profile map and corresponding geological structure interpretation data; using the geological profile map as input to the conditional generative adversarial network (GAN), and using the corresponding geological structure interpretation map as a label, training the GAN to obtain the geological structure interpretation model; wherein the conditional GAN includes a generator network, and the overall loss of the generator network includes adversarial loss, reconstruction loss, and geological constraint loss, the geological constraint loss being used to constrain the consistency between the geological structures generated by the generator network and geological rules. The geological structure interpretation method includes: acquiring a geological profile map to be interpreted; inputting the geological profile map to be interpreted into the geological structure interpretation model constructed by the above geological structure interpretation model construction method to obtain the geological structure interpretation data corresponding to the geological profile map to be interpreted.
[0098] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0099] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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; and these 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 the present invention.
Claims
1. A method for constructing a geological structure interpretation model, characterized in that, include: Obtain geological profile maps and corresponding geological structural interpretation data; Using the geological profile map as input to the conditional generative adversarial network and the geological structure interpretation map corresponding to the geological profile map as a label, the conditional generative adversarial network is trained to obtain the geological structure interpretation model. The conditional generative adversarial network includes a generator network, and the comprehensive loss of the generator network includes adversarial loss, reconstruction loss and geological constraint loss. The geological constraint loss is used to constrain the consistency between the geological structures generated by the generator network and geological rules.
2. The method for constructing a geological structure interpretation model according to claim 1, characterized in that, The overall loss also includes: physical constraint loss; The physical constraint loss is constructed based on the gravity anomaly simulation parameters and the measured gravity anomaly parameters, and is used to constrain the consistency between the gravity anomaly simulation data and the measured gravity anomaly data. The gravity anomaly simulation data is obtained by converting the geological structures generated by the generator network through differentiable rendering technology.
3. The method for constructing a geological structure interpretation model according to claim 1, characterized in that, The geological constraint loss includes at least one of fault loss, fold loss, and tectonic structure loss, wherein the fault loss is used to constrain the consistency between the fault geometry parameters generated by the generator network and geological rules, the fold loss is used to constrain the consistency between the fold morphology generated by the generator network and geological rules, and the tectonic structure loss is used to constrain the consistency between the distribution of geological structures generated by the generator network and the distribution of real geological structures in multi-scale spatial features.
4. The method for constructing a geological structure interpretation model according to claim 1, characterized in that, The geological constraint losses are as follows: in, Indicates geological constraint loss. Indicates fault loss, Indicates wrinkle loss. Indicates structural loss. The weighting coefficient represents the fault loss. This represents the weighting coefficient of the wrinkle loss. The weighting coefficient represents the loss of the constructed structure.
5. The method for constructing a geological structure interpretation model according to claim 3 or 4, characterized in that, The fault loss is as follows: in, This indicates the number of faults generated by the generator network. The generator network generates the first... i The strike angle of the fault, Indicates the reference strike angle of the fault. This indicates the dip angle of the fault generated by the generator network. Indicates the reference dip angle of the fault. This represents the tolerance threshold.
6. The method for constructing a geological structure interpretation model according to claim 3 or 4, characterized in that, The wrinkle loss is as follows: in, This indicates the wavelength at which the generator network generates the wrinkles. Indicates the thickness of the rock strata. Indicates viscosity ratio.
7. The method for constructing a geological structure interpretation model according to claim 3 or 4, characterized in that, The structural loss is as follows: in, This indicates a differentiable renderer. Represents a generator network. Represents a geological profile. This indicates measured gravity anomaly data. This represents the density gradient.
8. The method for constructing a geological structure interpretation model according to claim 1, characterized in that, The conditional generative adversarial network also includes a discriminator network, the comprehensive loss of which is as follows: in, This represents the overall loss of the discriminator network. Indicating resistance to loss, Indicates structural loss. This represents the gradient penalty. The weighting coefficients represent the structural loss. This represents the weight coefficient of the gradient penalty.
9. A method for interpreting geological structures, characterized in that, include: Obtain the geological profile to be interpreted; The geological profile to be interpreted is input into the geological structure interpretation model constructed by the geological structure interpretation model construction method according to any one of claims 1-8 to obtain the geological structure interpretation data corresponding to the geological profile to be interpreted.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the geological structure interpretation model construction method as described in any one of claims 1 to 8 or the geological structure interpretation method as described in claim 9.