Three-dimensional geological body modeling system based on sparse borehole data and generative adversarial network

Abstract

The application discloses a kind of sparse borehole data three-dimensional geologic body modeling systems based on generative adversarial network.The modeling system includes: data acquisition module, for obtaining the sparse distribution of borehole data in monitoring area;Feature extraction module is used to obtain multi-scale geological feature vector;Adversarial network construction module is used to construct the generative adversarial network model containing generator and discriminator;Prior constraint training module is used to introduce geological prior constraint condition to train generative adversarial network model;Geological model generation module is used to filter out target model satisfying preset geological rationality threshold;Optimization module is used to obtain geological attribute uncertainty distribution map and optimize target model, obtain final three-dimensional geologic body model.The application solves the problem of sparse borehole data geological modeling, realizes the accurate modeling and uncertainty quantification of complex geologic body, improves the accuracy and reliability of geological modeling.

Description

Technical Field

[0001] This invention relates to the field of geological modeling technology, and more specifically, to a three-dimensional geological body modeling system based on sparse borehole data from generative adversarial networks. Background Technology

[0002] In the process of geological exploration, due to the high cost and time-consuming nature of borehole sampling, only sparse borehole data can usually be obtained, which makes it extremely difficult to accurately construct a three-dimensional geological model.

[0003] Traditional geological modeling methods, such as Kriging interpolation and deterministic modeling, often fail to capture complex geological structures and nonlinear features when processing sparse borehole data, leading to significant deviations between the models and actual geological conditions. This is particularly true in areas with blurred orebody boundaries and complex geological structures, where these methods struggle to accurately predict the geological characteristics of intermediate areas, thus affecting the accuracy of mineral resource assessment and the rationality of mining planning. Although deep learning methods have been introduced into the field of geological modeling in recent years, existing models still lack sufficient geological rationality and diversity in the results generated when dealing with highly uncertain geological data and limited sample sizes, failing to fully reflect the possible structural changes of geological bodies.

[0004] In view of this, the present invention proposes a three-dimensional geological body modeling system based on sparse borehole data using generative adversarial networks to solve the above problems. Summary of the Invention

[0005] The main objective of this invention is to provide a three-dimensional geological body modeling system based on sparse borehole data using generative adversarial networks, in order to overcome the shortcomings of existing technologies.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solution.

[0007] Some embodiments of the present invention provide a three-dimensional geological body modeling system based on sparse borehole data using generative adversarial networks, comprising:

[0008] The data acquisition module is used to acquire sparsely distributed borehole data within the monitoring area. The borehole data includes the spatial coordinates, geological attribute values, and stratigraphic interface depth of each borehole point.

[0009] The feature extraction module is used to construct an initial three-dimensional geological grid based on the spatial distribution characteristics of the borehole data, and to perform multi-scale stereomicroscopic feature extraction on the initial three-dimensional geological grid to obtain the multi-scale geological feature vector of each grid cell.

[0010] An adversarial network construction module is used to construct a generative adversarial network model based on the multi-scale geological feature vectors. The generative adversarial network model includes a generator and a discriminator, wherein the generator is used to generate candidate three-dimensional geological body models, and the discriminator is used to evaluate the geological rationality of the candidate three-dimensional geological body models.

[0011] The prior constraint training module is used to train the generative adversarial network model by introducing geological prior constraints, which include stratigraphic continuity constraints, geological attribute gradient constraints, and boundary ambiguity constraints.

[0012] The geological model generation module is used to generate multiple candidate 3D geological body models based on the trained generative adversarial network model, and to select the target 3D geological body model that meets the preset geological rationality threshold according to the evaluation results of the discriminator.

[0013] The optimization module is used to perform uncertainty quantification analysis on the target three-dimensional geological body model, obtain the uncertainty distribution map of the geological attributes of each grid cell, and optimize the target three-dimensional geological body model based on the uncertainty distribution map to obtain the final three-dimensional geological body model.

[0014] In one embodiment, the step of performing multi-scale stereomicroscopic feature extraction on the initial three-dimensional geological grid to obtain a multi-scale geological feature vector for each grid cell includes:

[0015] The initial three-dimensional geological grid is divided into multiple scale levels, with each scale level corresponding to a different grid resolution;

[0016] At each scale level, based on the spatial coordinates and geological attribute values ​​of the borehole data, the initial three-dimensional geological grid is filled with attributes using a three-dimensional anisotropic interpolation method to obtain the geological attribute field of the corresponding scale level.

[0017] For the geological property field at each scale level, stereomicroscopic features are extracted, including local geological property gradients, local curvature, and local connectivity.

[0018] Based on the stereomicroscope features of each grid cell at different scale levels, a multi-scale feature matrix is ​​constructed for the corresponding grid cell.

[0019] Principal component analysis is performed on the multi-scale feature matrix to extract principal component features, and the principal component features are normalized to the multi-scale geological feature vector.

[0020] Furthermore, the implementation of the three-dimensional anisotropic interpolation method includes:

[0021] Based on the spatial coordinates of the borehole data, a three-dimensional anisotropic distance field is constructed; based on the three-dimensional anisotropic distance field, an anisotropic interpolation kernel function is designed; using the anisotropic interpolation kernel function, geological attribute interpolation is performed on each grid cell in the initial three-dimensional geological grid to obtain the geological attribute field.

[0022] In one embodiment, the generative adversarial network model is constructed in the following ways:

[0023] The generator is constructed, which includes an input layer, a multi-scale feature fusion layer, and a 3D decoding layer. The input layer is used to receive the multi-scale geological feature vectors, the multi-scale feature fusion layer is used to perform weighted fusion of features at different scale levels, and the 3D decoding layer is used to generate the candidate 3D geological body model.

[0024] The discriminator is constructed, which includes a three-dimensional convolutional layer, a geological rule embedding layer, and a classification layer. The three-dimensional convolutional layer is used to extract the spatial features of the candidate three-dimensional geological body model, the geological rule embedding layer is used to introduce the geological prior constraints, and the classification layer is used to output the geological rationality score of the candidate three-dimensional geological body model.

[0025] The loss function of the generative adversarial network model is designed, which includes adversarial loss, geological attribute matching loss and boundary ambiguity loss. The adversarial loss is used to measure the adversarial effect between the generator and the discriminator. The geological attribute matching loss is used to measure the degree of agreement between the candidate 3D geological body model and the borehole data. The boundary ambiguity loss is used to measure the smoothness of the geological boundaries in the candidate 3D geological body model.

[0026] Furthermore, the construction method of the multi-scale feature fusion layer includes:

[0027] In the multi-scale feature fusion layer, a scale attention mechanism is designed. The scale attention mechanism obtains the attention weight of the corresponding scale level by calculating the global correlation of the multi-scale geological feature vectors at each scale level.

[0028] Based on the attention weights, the multi-scale geological feature vectors at different scale levels are weighted and summed to obtain a fused feature vector;

[0029] The fused feature vector is subjected to nonlinear activation processing, and then the fused feature vector is added to the multi-scale geological feature vector of the input layer through a residual connection structure to obtain the final fused feature output.

[0030] In one embodiment, the geological prior constraints are obtained by means of:

[0031] Based on the stratigraphic interface depth of the borehole data, stratigraphic continuity constraints are constructed; based on the geological attribute values ​​of the borehole data, geological attribute gradient constraints are constructed; based on the spatial distribution characteristics of the borehole data, boundary ambiguity constraints are constructed.

[0032] Furthermore, the step of training the generative adversarial network model by introducing geological prior constraints includes:

[0033] The geological prior constraints are transformed into regularization terms and embedded into the loss function of the generative adversarial network model;

[0034] During training, an adaptive weight adjustment strategy is adopted to dynamically adjust the weight of the regularization term based on the degree of agreement between the candidate 3D geological body model generated by the generator and the borehole data.

[0035] In each training iteration, a portion of the borehole data is randomly selected as a validation set. The geological rationality score of the candidate 3D geological body model on the validation set is calculated, and the parameters of the generator and the discriminator are updated according to the geological rationality score until the geological rationality score converges.

[0036] In one embodiment, generating multiple candidate 3D geological body models and selecting a target 3D geological body model that meets a preset geological rationality threshold based on the evaluation results of the discriminator includes:

[0037] By introducing a random noise vector into the input layer of the generator, multiple different candidate 3D geological body models are generated.

[0038] For each candidate 3D geological body model, the discriminator is used to calculate its geological rationality score, and the probability distribution of geological attributes of each grid cell in the candidate 3D geological body model is extracted.

[0039] Based on the probability distribution of the geological attributes, calculate the diversity index of each candidate three-dimensional geological body model;

[0040] The candidate 3D geological body model with the highest geological rationality score and the highest diversity index is determined as the target 3D geological body model.

[0041] In one embodiment, the uncertainty quantification analysis of the target three-dimensional geological body model to obtain the uncertainty distribution map of the geological attributes of each grid cell includes:

[0042] Based on the target 3D geological body model, the Monte Carlo dropout method is used to perform multiple forward propagations on the generator to generate multiple perturbed 3D geological body models.

[0043] For each grid cell, the distribution of geological attribute values ​​of the corresponding grid cell in the disturbed three-dimensional geological body model is statistically analyzed, and the mean and variance of the geological attribute values ​​are obtained.

[0044] Based on the variance of the geological attribute values, an uncertainty metric is constructed for each grid cell, and the uncertainty metric is normalized into an uncertainty distribution map of the geological attributes.

[0045] In one embodiment, optimizing the target 3D geological body model based on the uncertainty distribution map to obtain the final 3D geological body model includes:

[0046] Based on the geological attribute uncertainty distribution map, identify grid cells in the target three-dimensional geological body model whose uncertainty measure is greater than a preset uncertainty threshold, and record them as high uncertainty regions;

[0047] For the high uncertainty region, a local geological attribute interpolation field is constructed based on the spatial distribution characteristics of the borehole data;

[0048] The local geological attribute interpolation field is weighted and fused with the geological attribute values ​​of the corresponding grid cells in the target three-dimensional geological body model to obtain the optimized three-dimensional geological body model.

[0049] The optimized three-dimensional geological model is subjected to geological rationality verification. The geological rationality verification is carried out by calculating the degree of consistency between the optimized three-dimensional geological model and the borehole data. If the degree of consistency is greater than a preset consistency threshold, the optimized three-dimensional geological model is determined as the final three-dimensional geological model.

[0050] Compared with existing technologies, the 3D geological body modeling system based on sparse borehole data of this invention effectively solves the problems of discontinuous geological body structure, blurred boundaries, and loss of details caused by sparse borehole data in traditional geological modeling methods by acquiring sparsely distributed borehole data and combining it with multi-scale stereomicroscopy feature extraction technology. By constructing a generative adversarial network model and introducing an adversarial mechanism between the generator and discriminator, the system can achieve accurate modeling of geological bodies under limited borehole data conditions, greatly improving the accuracy and reliability of the model. The introduced geological prior constraints make the generated geological body model more consistent with actual geological laws, effectively overcoming the limitations of traditional interpolation methods in expressing complex geological features. Through the generation and screening mechanism of multiple candidate 3D geological body models, the system can ensure the geological rationality of the final model and avoid the bias and uncertainty that may be caused by a single model. The innovative uncertainty quantification analysis method can accurately identify high uncertainty areas and perform targeted optimization, greatly improving the applicability and accuracy of the model under complex geological conditions. The multi-scale feature fusion and attention mechanism adopted by the system can adaptively integrate geological feature information at different scales, enabling the model to better capture the multi-scale structural features of geological bodies. This system enables an intelligent modeling process from sparse borehole data to high-precision three-dimensional geological models, providing more accurate and reliable geological information support for mineral resource exploration, oil and gas reservoir evaluation, and groundwater resource management, with significant economic and social benefits. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of a three-dimensional geological body modeling system based on sparse borehole data according to an embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram of a three-dimensional geological body modeling method based on sparse borehole data using generative adversarial networks, according to an embodiment of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Example 1

[0055] Please see Figure 1 As shown in the figure, the 3D geological body modeling system based on sparse borehole data provided in this embodiment includes:

[0056] The data acquisition module is used to acquire sparsely distributed borehole data within the monitoring area. The borehole data includes the spatial coordinates, geological attribute values, and stratigraphic interface depth of each borehole point.

[0057] The feature extraction module is used to construct an initial three-dimensional geological grid based on the spatial distribution characteristics of the borehole data, and to perform multi-scale stereomicroscopic feature extraction on the initial three-dimensional geological grid to obtain the multi-scale geological feature vector of each grid cell.

[0058] An adversarial network construction module is used to construct a generative adversarial network model based on the multi-scale geological feature vectors. The generative adversarial network model includes a generator and a discriminator, wherein the generator is used to generate candidate three-dimensional geological body models, and the discriminator is used to evaluate the geological rationality of the candidate three-dimensional geological body models.

[0059] The prior constraint training module is used to train the generative adversarial network model by introducing geological prior constraints, which include stratigraphic continuity constraints, geological attribute gradient constraints, and boundary ambiguity constraints.

[0060] The geological model generation module is used to generate multiple candidate 3D geological body models based on the trained generative adversarial network model, and to select the target 3D geological body model that meets the preset geological rationality threshold according to the evaluation results of the discriminator.

[0061] The optimization module is used to perform uncertainty quantification analysis on the target three-dimensional geological body model, obtain the uncertainty distribution map of the geological attributes of each grid cell, and optimize the target three-dimensional geological body model based on the uncertainty distribution map to obtain the final three-dimensional geological body model.

[0062] The data acquisition module is responsible for acquiring sparsely distributed borehole data within the monitoring area. The borehole data includes the spatial coordinates, geological attribute values, and stratigraphic interface depth of each borehole point.

[0063] Within the geological area requiring modeling, borehole data distributed at different locations is obtained through drilling. The spatial coordinates of each borehole point are typically represented by three-dimensional coordinates (x, y, z), where x and y represent the horizontal position and z represents the vertical depth. Geological attribute values ​​include physical property parameters such as lithology, porosity, and permeability. The stratigraphic interface depth records the interface location between different geological strata.

[0064] It should be noted that due to the high cost of drilling, boreholes are typically sparsely distributed within the study area, making it difficult to directly reflect the complete underground geological structure. This system is designed specifically for this sparse data condition, effectively utilizing limited borehole information to construct a reasonable three-dimensional geological model.

[0065] In one implementation of this embodiment, the borehole data undergoes preprocessing, including outlier detection and correction, coordinate system unification, and data format standardization, to ensure that the data quality meets the requirements of subsequent modeling.

[0066] Based on the spatial distribution characteristics of the borehole data, the feature extraction module constructs an initial three-dimensional geological grid and performs multi-scale stereomicroscopic feature extraction on the initial three-dimensional geological grid to obtain the multi-scale geological feature vector of each grid cell.

[0067] Subsurface geological structures exhibit different structural features at different spatial scales, and feature extraction at a single scale is insufficient to fully capture the structural information of geological bodies. Therefore, this invention employs a multi-scale stereomicroscopy feature extraction method, which can extract geological structural features at different spatial resolutions, thereby improving the model's ability to represent complex geological structures.

[0068] Preferably, in some possible implementations of this embodiment, the method for multi-scale stereomicroscope feature extraction includes: dividing the initial three-dimensional geological grid into multiple scale levels, each scale level corresponding to a different grid resolution; at each scale level, based on the spatial coordinates and geological attribute values ​​of the borehole data, using a three-dimensional anisotropic interpolation method to fill the initial three-dimensional geological grid with attributes to obtain a geological attribute field for the corresponding scale level; extracting stereomicroscope features from the geological attribute field at each scale level, the stereomicroscope features including local geological attribute gradients, local curvature, and local connectivity; constructing a multi-scale feature matrix for each grid unit based on the stereomicroscope features of each grid unit at different scale levels; performing principal component analysis on the multi-scale feature matrix to extract principal component features, and normalizing the principal component features into the multi-scale geological feature vector.

[0069] Different scale levels can capture the structural features of geological bodies at different spatial resolutions. Coarse-scale levels reflect large-scale geological structures such as faults and folds, while fine-scale levels reflect small-scale geological details such as sedimentary structures and small fractures. Therefore, through multi-scale feature extraction, the structural information of geological bodies can be comprehensively captured.

[0070] In one implementation of this embodiment, the initial three-dimensional geological grid is divided into three scale levels, corresponding to grid resolutions of 100 meters, 50 meters, and 25 meters, respectively. The stereomicroscope feature vector of each grid cell has a dimension of 15 (5 features per scale level), which is reduced to an 8-dimensional feature vector after principal component analysis.

[0071] The implementation of the three-dimensional anisotropic interpolation method includes: constructing a three-dimensional anisotropic distance field based on the spatial coordinates of the borehole data, wherein the three-dimensional anisotropic distance field is implemented by calculating the distance from each grid cell to the nearest borehole point and introducing a weighting factor of the geological attribute gradient direction; designing anisotropic interpolation kernel function based on the three-dimensional anisotropic distance field, wherein the anisotropic interpolation kernel function is implemented by introducing a direction-dependent ellipsoidal model in the distance calculation; and using the anisotropic interpolation kernel function to perform geological attribute interpolation on each grid cell in the initial three-dimensional geological grid to obtain the geological attribute field.

[0072] Geological bodies typically exhibit anisotropic characteristics, meaning that the patterns of geological property variation differ in different directions. For example, sedimentary strata often show good continuity in the horizontal direction but greater variation in the vertical direction. Therefore, when interpolating geological properties, spatial directionality needs to be considered, and anisotropic interpolation methods should be used.

[0073] In one implementation of this embodiment, the anisotropic interpolation kernel function adopts a Gaussian kernel function based on an ellipsoid model. The major axis of the ellipsoid is aligned with the direction of the main geological structure. The ratio of the major and minor axes of the ellipsoid is determined according to the degree of anisotropy of the geological properties. Typically, the ratio of the horizontal to the vertical direction is set to 5:1 to 10:1.

[0074] The adversarial network construction module constructs a generative adversarial network model based on the multi-scale geological feature vectors. The generative adversarial network model includes a generator and a discriminator, wherein the generator is used to generate candidate three-dimensional geological body models, and the discriminator is used to evaluate the geological rationality of the candidate three-dimensional geological body models.

[0075] Generative Adversarial Networks (GANs) are deep learning models consisting of a generator and a discriminator, which optimize the model through adversarial training. In this system, the generator is responsible for generating candidate 3D geological body models based on the input multi-scale geological feature vectors, while the discriminator is responsible for evaluating the geological plausibility of the generated models. The two interact and work together to improve the model quality.

[0076] Preferably, in some possible implementations of this embodiment, the construction of the generative adversarial network model includes: constructing the generator, which includes an input layer, a multi-scale feature fusion layer, and a 3D decoding layer, wherein the input layer is used to receive the multi-scale geological feature vectors, the multi-scale feature fusion layer is used to perform weighted fusion of features at different scale levels, and the 3D decoding layer is used to generate the candidate 3D geological body model; constructing the discriminator, which includes a 3D convolutional layer, a geological rule embedding layer, and a classification layer, wherein the 3D convolutional layer is used to extract the spatial features of the candidate 3D geological body model, the geological rule embedding layer is used to introduce the geological prior constraints, and the classification layer is used to output the geological rationality score of the candidate 3D geological body model; and designing the loss function of the generative adversarial network model, which includes adversarial loss, geological attribute matching loss, and boundary ambiguity loss, wherein the adversarial loss is used to measure the adversarial effect between the generator and the discriminator, the geological attribute matching loss is used to measure the consistency between the candidate 3D geological body model and the borehole data, and the boundary ambiguity loss is used to measure the smoothness of the geological boundaries in the candidate 3D geological body model.

[0077] The construction method of the multi-scale feature fusion layer includes: designing a scale attention mechanism in the multi-scale feature fusion layer; the scale attention mechanism obtains the attention weight of the corresponding scale level by calculating the global correlation of the multi-scale geological feature vectors at each scale level; according to the attention weight, the multi-scale geological feature vectors at different scale levels are weighted and summed to obtain the fused feature vector; the fused feature vector is subjected to nonlinear activation processing, and the fused feature vector is added to the multi-scale geological feature vector of the input layer through a residual connection structure to obtain the final fused feature output.

[0078] Geological features at different scales have varying importance in the modeling process. Scale attention mechanisms can adaptively adjust the weights of features at different scales, highlighting the role of key features and improving model generation performance. Residual connection structures can effectively alleviate the vanishing gradient problem in deep network training, improving model training stability.

[0079] In one implementation of this embodiment, the generator adopts the U-Net architecture, and the 3D decoding layer contains 5 3D transposed convolutional layers, each followed by a batch normalization layer and a ReLU activation function; the discriminator adopts the PatchGAN architecture, and the 3D convolutional layer contains 4 3D convolutional layers, each followed by a batch normalization layer and a LeakyReLU activation function.

[0080] In one implementation of this embodiment, the loss function adopts a weighted sum form, the adversarial loss adopts a least squares loss function, the geological attribute matching loss adopts the L1 norm, and the boundary ambiguity loss adopts the total variation regularization term. The weight ratio of the three is set to 1:5:2.

[0081] The prior constraint training module trains the generative adversarial network model by introducing geological prior constraints, which include stratigraphic continuity constraints, geological attribute gradient constraints, and boundary ambiguity constraints.

[0082] Existing generative adversarial networks (GANs) face challenges in 3D geological modeling, including unstable training and a lack of geological plausibility in generated results. By introducing prior geological constraints and integrating geological knowledge into the model training process, it is possible to effectively guide the model to generate 3D geological models that conform to geological laws, thereby improving the geological plausibility and reliability of the generated results.

[0083] Preferably, in some possible implementations of this embodiment, the method for obtaining the geological prior constraints includes: constructing a stratigraphic continuity constraint based on the stratigraphic interface depth of the borehole data, wherein the stratigraphic continuity constraint is achieved by calculating the variance of the depth difference between the stratigraphic interfaces of adjacent grid cells; constructing a geological attribute gradient constraint based on the geological attribute values ​​of the borehole data, wherein the geological attribute gradient constraint is achieved by calculating the deviation between the geological attribute gradient of each grid cell in the initial three-dimensional geological grid and the global geological attribute gradient distribution; and constructing a boundary ambiguity constraint based on the spatial distribution characteristics of the borehole data, wherein the boundary ambiguity constraint is achieved by introducing a distance-weighted fuzzy kernel function into the initial three-dimensional geological grid to smooth the attribute values ​​of the geological boundary region.

[0084] Stratigraphic continuity constraints ensure the smooth continuity of stratigraphic interfaces in the generated geological model, conforming to the laws of geological sedimentation; geological attribute gradient constraints ensure that the spatial variation of geological attributes in the generated model conforms to the variation law of real geological bodies; and boundary ambiguity constraints take into account that real geological interfaces are often not absolutely clear boundaries, but have certain transition zones.

[0085] The prior constraint training module trains the generative adversarial network model by introducing geological prior constraints, including: converting the geological prior constraints into regularization terms and embedding them into the loss function of the generative adversarial network model; during training, an adaptive weight adjustment strategy is adopted to dynamically adjust the weights of the regularization terms based on the degree of fit between the candidate 3D geological body model generated by the generator and the borehole data; in each training iteration, a portion of the borehole data is randomly selected as a validation set, the geological rationality score of the candidate 3D geological body model on the validation set is calculated, and the parameters of the generator and the discriminator are updated based on the geological rationality score until the geological rationality score converges.

[0086] The adaptive weight adjustment strategy can dynamically adjust the weights of constraints based on the quality of the model's generated results during training. In the early stages of training, it focuses more on data fitting, while in the later stages, it focuses more on geological rationality, thus achieving a balanced optimization of model performance.

[0087] In one implementation of this embodiment, the Adam optimizer is used for training, with an initial learning rate of 0.0002, which is halved every 50 training epochs; the training batch size is set to 16, and the number of training epochs is set to 300; the validation set ratio is set to 20% of the borehole data.

[0088] The geological model generation module generates multiple candidate 3D geological body models based on the trained generative adversarial network model, and selects the target 3D geological body model that meets the preset geological rationality threshold according to the evaluation results of the discriminator.

[0089] The stochastic nature of generative adversarial networks (GANs) allows them to generate multiple candidate models. While these models are all based on the same borehole data, they differ in their detailed representations. This difference can be used to express the uncertainties in geological modeling. By generating and filtering multiple candidate models, a more reliable 3D geological model can be obtained.

[0090] Preferably, in some possible implementations of this embodiment, generating multiple candidate 3D geological body models and selecting a target 3D geological body model that meets a preset geological rationality threshold based on the evaluation results of the discriminator includes: generating multiple different candidate 3D geological body models by introducing random noise vectors into the input layer of the generator; calculating the geological rationality score of each candidate 3D geological body model using the discriminator and extracting the probability distribution of geological attributes for each grid cell in the candidate 3D geological body model; calculating a diversity index for each candidate 3D geological body model based on the probability distribution of geological attributes, wherein the diversity index is realized through the entropy value of the probability distribution of geological attributes; and determining the candidate 3D geological body model with a geological rationality score greater than the preset geological rationality threshold and the highest diversity index as the target 3D geological body model.

[0091] Introducing random noise vectors can increase the diversity of model generation results, enabling the generated candidate models to explore possible geological structural spaces more comprehensively; geological rationality scores ensure that the selected models conform to geological laws; and diversity indices ensure that the models can fully express geological uncertainties and avoid overfitting to a single explanation.

[0092] In one implementation of this embodiment, the random noise vector is obtained by random sampling using a standard normal distribution, with a dimension set to 128; the preset geological rationality threshold is set to 0.8; and the number of generated candidate three-dimensional geological body models is set to 100.

[0093] The optimization module performs uncertainty quantification analysis on the target three-dimensional geological body model, obtains the uncertainty distribution map of the geological attributes of each grid cell, and optimizes the target three-dimensional geological body model based on the uncertainty distribution map to obtain the final three-dimensional geological body model.

[0094] Uncertainty quantification in geological modeling is a crucial step, providing essential data for subsequent resource assessment and risk analysis. By performing uncertainty quantification analysis on the target 3D geological body model, identifying high-uncertainty areas, and then optimizing them accordingly, the reliability of the final model can be improved.

[0095] Preferably, in some possible implementations of this embodiment, performing uncertainty quantification analysis on the target three-dimensional geological body model to obtain the geological attribute uncertainty distribution map of each grid cell includes: based on the target three-dimensional geological body model, using the Monte Carlo dropout method to perform multiple forward propagations on the generator to generate multiple perturbed three-dimensional geological body models; for each grid cell, statistically analyzing the distribution of geological attribute values ​​of the corresponding grid cell in the perturbed three-dimensional geological body model, and obtaining the mean and variance of the geological attribute values; constructing an uncertainty metric for each grid cell based on the variance of the geological attribute values, and normalizing the uncertainty metric to the geological attribute uncertainty distribution map.

[0096] Monte Carlo dropout is an efficient Bayesian approximation method that can estimate model uncertainty through multiple samplings without increasing model parameters. By keeping the dropout layer active during the inference phase, each forward propagation yields slightly different results, and the statistical distribution of these results can be used to quantify the uncertainty of the model's predictions.

[0097] Optimizing the target 3D geological body model based on the uncertainty distribution map to obtain the final 3D geological body model includes: identifying grid cells in the target 3D geological body model whose uncertainty measure is greater than a preset uncertainty threshold, and recording them as high uncertainty regions, based on the geological attribute uncertainty distribution map; constructing a local geological attribute interpolation field for the high uncertainty regions based on the spatial distribution characteristics of the borehole data, wherein the local geological attribute interpolation field is implemented by introducing a distance-weighted anisotropic kernel function; weighting and fusing the local geological attribute interpolation field with the geological attribute values ​​of the corresponding grid cells in the target 3D geological body model to obtain the optimized 3D geological body model; and performing geological rationality verification on the optimized 3D geological body model, wherein the geological rationality verification is achieved by calculating the degree of consistency between the optimized 3D geological body model and the borehole data, and if the degree of consistency is greater than a preset consistency threshold, then the optimized 3D geological body model is determined as the final 3D geological body model.

[0098] Targeted optimization in regions of high uncertainty can effectively improve the accuracy and reliability of the model in these areas. By combining the results of the generative model with traditional interpolation methods, the advantages of both can be fully utilized to achieve a comprehensive improvement in model performance.

[0099] In one implementation of this embodiment, the dropout probability is set to 0.3 and the number of forward propagations is set to 50 in the Monte Carlo dropout method; the preset uncertainty threshold is set to the 85th percentile of the uncertainty distribution; and the preset matching threshold is set to 0.9.

[0100] This embodiment effectively solves the problems of discontinuous geological body structure, blurred boundaries, and loss of details caused by sparse borehole data in traditional geological modeling methods by acquiring sparsely distributed borehole data and combining it with multi-scale stereomicroscopy feature extraction technology. By constructing a generative adversarial network model and introducing an adversarial mechanism between the generator and discriminator, the system can achieve accurate modeling of geological bodies under limited borehole data conditions, greatly improving the accuracy and reliability of the model. The introduced geological prior constraints make the generated geological body model more consistent with actual geological laws, effectively overcoming the limitations of traditional interpolation methods in expressing complex geological features. Through the generation and screening mechanism of multiple candidate 3D geological body models, the system can ensure the geological rationality of the final model and avoid the bias and uncertainty that may be caused by a single model. The innovative uncertainty quantification analysis method can accurately identify high uncertainty areas and perform targeted optimization, greatly improving the applicability and accuracy of the model under complex geological conditions. The multi-scale feature fusion and attention mechanism adopted by the system can adaptively integrate geological feature information at different scales, enabling the model to better capture the multi-scale structural features of geological bodies. This system enables an intelligent modeling process from sparse borehole data to high-precision three-dimensional geological models, providing more accurate and reliable geological information support for mineral resource exploration, oil and gas reservoir evaluation, and groundwater resource management, with significant economic and social benefits.

[0101] Example 2

[0102] Please see Figure 2 As shown, this embodiment provides a method for modeling three-dimensional geological bodies based on sparse borehole data using generative adversarial networks. Details not described in detail are provided in Embodiment 1. This modeling method includes:

[0103] Step S1: Obtain sparsely distributed borehole data within the monitoring area. The borehole data includes the spatial coordinates, geological attribute values, and stratigraphic interface depth of each borehole point.

[0104] Step S2: Based on the spatial distribution characteristics of the borehole data, construct an initial three-dimensional geological grid, and perform multi-scale stereomicroscopic feature extraction on the initial three-dimensional geological grid to obtain the multi-scale geological feature vector of each grid unit;

[0105] Step S3: Based on the multi-scale geological feature vector, construct a generative adversarial network model, which includes a generator and a discriminator, wherein the generator is used to generate candidate three-dimensional geological body models, and the discriminator is used to evaluate the geological rationality of the candidate three-dimensional geological body models;

[0106] Step S4: Train the generative adversarial network model by introducing geological prior constraints, which include stratigraphic continuity constraints, geological attribute gradient constraints, and boundary ambiguity constraints.

[0107] Step S5: Based on the trained generative adversarial network model, generate multiple candidate 3D geological body models, and select the target 3D geological body model that meets the preset geological rationality threshold according to the evaluation results of the discriminator.

[0108] Step S6: Perform uncertainty quantification analysis on the target three-dimensional geological body model, obtain the uncertainty distribution map of geological attributes of each grid cell, and optimize the target three-dimensional geological body model based on the uncertainty distribution map to obtain the final three-dimensional geological body model.

[0109] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, 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. When the processor executes the program, it implements the three-dimensional geological body modeling method based on sparse borehole data according to any of the above embodiments.

[0110] For example, the hardware structure of the electronic device may include a processor, a memory, an input / output interface, a communication interface, and a bus. The processor, memory, input / output interface, and communication interface are interconnected internally via the bus.

[0111] Furthermore, in specific implementations, the electronic device may also include other components necessary for normal operation. Additionally, those skilled in the art will understand that the device may include only the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0112] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, the present invention also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the three-dimensional geological body modeling method based on sparse borehole data as described in any of the above embodiments.

[0113] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0114] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0115] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0116] In the description of this invention, it should be understood that the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0117] In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0118] In the description of this invention, "several" means one or more, and "a large number" means two or more.

[0119] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0120] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0121] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A sparse borehole data 3D geobody modeling system based on a generative adversarial network, characterized in that, The method comprises the following steps: a data acquisition module is configured to acquire sparse drilling data in a monitoring area, wherein the drilling data comprises spatial coordinates, geological attribute values and formation interface depths of each drilling point; a feature extraction module is configured to construct an initial three-dimensional geological grid according to spatial distribution characteristics of the drilling data, and perform multi-scale stereomicroscope feature extraction on the initial three-dimensional geological grid to obtain a multi-scale geological feature vector of each grid cell; an adversarial network construction module is configured to construct a generative adversarial network model based on the multi-scale geological feature vector, wherein the generative adversarial network model comprises a generator and a discriminator, the generator is configured to generate a candidate three-dimensional geological body model, and the discriminator is configured to evaluate the geological rationality of the candidate three-dimensional geological body model; a prior constraint training module is configured to train the generative adversarial network model by introducing geological prior constraint conditions, wherein the geological prior constraint conditions comprise a stratigraphic continuity constraint, a geological attribute gradient constraint and a boundary ambiguity constraint; a geological model generation module is configured to generate a plurality of candidate three-dimensional geological body models based on the trained generative adversarial network model, and select a target three-dimensional geological body model that satisfies a preset geological rationality threshold according to an evaluation result of the discriminator; an optimization module is configured to perform uncertainty quantification analysis on the target three-dimensional geological body model to obtain a geological attribute uncertainty distribution map of each grid cell, and optimize the target three-dimensional geological body model based on the uncertainty distribution map to obtain a final three-dimensional geological body model.

2. The generative adversarial network-based sparse borehole data 3D geobody modeling system of claim 1, wherein, The multi-scale stereomicroscope feature extraction on the initial three-dimensional geological grid to obtain a multi-scale geological feature vector of each grid cell comprises: dividing the initial three-dimensional geological grid into a plurality of scale levels, each scale level corresponding to a different grid resolution; at each scale level, filling the initial three-dimensional geological grid with attributes based on the spatial coordinates and geological attribute values of the drilling data using a three-dimensional anisotropic interpolation method to obtain a geological attribute field of the corresponding scale level; extracting stereomicroscope features from the geological attribute field of each scale level, wherein the stereomicroscope features comprise local geological attribute gradients, local curvatures and local connectivity; constructing a multi-scale feature matrix of each grid cell according to the stereomicroscope features of the grid cell at different scale levels; performing principal component analysis on the multi-scale feature matrix to extract principal component features, and normalizing the principal component features into the multi-scale geological feature vector.

3. The generative adversarial network-based sparse borehole data 3D geobody modeling system of claim 2, wherein, The implementation of the three-dimensional anisotropic interpolation method comprises: constructing a three-dimensional anisotropic distance field according to the spatial coordinates of the drilling data, designing an anisotropic interpolation kernel function based on the three-dimensional anisotropic distance field, and performing geological attribute interpolation on each grid cell in the initial three-dimensional geological grid using the anisotropic interpolation kernel function to obtain the geological attribute field.

4. The generative adversarial network-based sparse borehole data 3D geobody modeling system of claim 1, wherein, The construction method of the generative adversarial network model comprises: constructing the generator, the generator comprising an input layer, a multi-scale feature fusion layer, and a three-dimensional decoding layer, wherein the input layer is configured to receive the multi-scale geological feature vector, the multi-scale feature fusion layer is configured to perform weighted fusion on features of different scale levels, and the three-dimensional decoding layer is configured to generate the candidate three-dimensional geological body model; constructing the discriminator, the discriminator comprising a three-dimensional convolutional layer, a geological rule embedding layer, and a classification layer, wherein the three-dimensional convolutional layer is configured to extract spatial features of the candidate three-dimensional geological body model, the geological rule embedding layer is configured to introduce the geological prior constraint condition, and the classification layer is configured to output a geological rationality score of the candidate three-dimensional geological body model; designing a loss function of the generative adversarial network model, the loss function comprising an adversarial loss, a geological attribute matching loss, and a boundary fuzziness loss, wherein the adversarial loss is configured to measure an adversarial effect between the generator and the discriminator, the geological attribute matching loss is configured to measure a degree of fit between the candidate three-dimensional geological body model and the drilling data, and the boundary fuzziness loss is configured to measure a smoothness of a geological boundary in the candidate three-dimensional geological body model.

5. The generative adversarial network-based sparse borehole data 3D geobody modeling system of claim 4, wherein, The construction method of the multi-scale feature fusion layer comprises: in the multi-scale feature fusion layer, a scale attention mechanism is designed, which obtains attention weights of corresponding scale levels by calculating global correlations of the multi-scale geological feature vectors of each scale level; the multi-scale geological feature vectors of different scale levels are weighted and summed according to the attention weights to obtain a fused feature vector; the fused feature vector is subjected to nonlinear activation processing, and the fused feature vector is added to the multi-scale geological feature vector of the input layer through a residual connection structure to obtain a final fused feature output.

6. The generative adversarial network-based sparse borehole data 3D geobody modeling system of claim 1, wherein, The acquisition method of the geological prior constraint condition comprises: a stratigraphic continuity constraint is constructed according to a stratigraphic interface depth of the drilling data, a geological attribute gradient constraint is constructed according to a geological attribute value of the drilling data, and a boundary fuzziness constraint is constructed according to a spatial distribution characteristic of the drilling data.

7. The generative adversarial network-based sparse borehole data 3D geobody modeling system of claim 1, wherein, The training of the generative adversarial network model by introducing the geological prior constraint condition comprises: the geological prior constraint condition is converted into a regularization term and embedded in a loss function of the generative adversarial network model; in the training process, an adaptive weight adjustment strategy is adopted to dynamically adjust a weight of the regularization term according to a degree of fit between the candidate three-dimensional geological body model generated by the generator and the drilling data; in each training iteration, part of the drilling data is randomly extracted as a validation set, a geological rationality score of the candidate three-dimensional geological body model on the validation set is calculated, and parameters of the generator and the discriminator are updated according to the geological rationality score until the geological rationality score converges.

8. The generative adversarial network-based sparse borehole data 3D geobody modeling system of claim 1, wherein, The generation of multiple candidate three-dimensional geological body models and the screening of a target three-dimensional geological body model satisfying a preset geological rationality threshold according to an evaluation result of the discriminator comprise: introducing a random noise vector in an input layer of the generator, a plurality of different candidate three-dimensional geological body models are generated; for each candidate three-dimensional geological body model, the discriminator is used to calculate a geological rationality score thereof, and a geological property probability distribution of each grid cell in the candidate three-dimensional geological body model is extracted; a diversity index of each candidate three-dimensional geological body model is calculated according to the geological property probability distribution; the candidate three-dimensional geological body model with the highest geological rationality score and the highest diversity index is determined as the target three-dimensional geological body model.

9. The generative adversarial network-based sparse borehole data 3D geobody modeling system of claim 1, wherein, the uncertainty quantification analysis on the target three-dimensional geological body model is performed to obtain a geological property uncertainty distribution map of each grid cell, including: based on the target three-dimensional geological body model, the generator is subjected to multiple forward propagations by using a Monte Carlo dropout method to generate a plurality of perturbed three-dimensional geological body models; for each grid cell, the distribution of the geological property values of the corresponding grid cell in the perturbed three-dimensional geological body models is counted to obtain the mean and variance of the geological property values; according to the variance of the geological property values, an uncertainty measure of each grid cell is constructed, and the uncertainty measure is normalized to the geological property uncertainty distribution map.

10. The generative adversarial network-based sparse borehole data 3D geobody modeling system of claim 1, wherein, the target three-dimensional geological body model is optimized based on the uncertainty distribution map to obtain a final three-dimensional geological body model, including: according to the geological property uncertainty distribution map, grid cells in the target three-dimensional geological body model with an uncertainty measure greater than a preset uncertainty threshold are identified and recorded as high-uncertainty areas; for the high-uncertainty areas, a local geological property interpolation field is constructed based on the spatial distribution characteristics of the drilling data; the local geological property interpolation field and the geological property values of the corresponding grid cells in the target three-dimensional geological body model are weighted and fused to obtain an optimized three-dimensional geological body model; the optimized three-dimensional geological body model is subjected to a geological rationality verification, and the geological rationality verification is passed by calculating the goodness of fit between the optimized three-dimensional geological body model and the drilling data, and if the goodness of fit is greater than a preset goodness threshold, the optimized three-dimensional geological body model is determined as the final three-dimensional geological body model.

Images