Three-dimensional geological model establishment method, storage medium and equipment

Abstract

The invention discloses a three-dimensional geologic model establishment method, a storage medium and equipment, and relates to the field of three-dimensional geologic body modeling, and the method comprises the steps: carrying out the discretization sampling of continuous geologic feature data, and dividing the sampling data into a training set and a test set; constructing a lithology category prediction model which comprises an encoder and a decoder, and training and testing the lithology category prediction model by using the training set and the test set; generating a regular grid point array by using a DEM model of a research area, and predicting lithology category labels of regular grid points by using the trained lithology category prediction model; a regular grid point is used as an angular point of a voxel grid, the step length of a regular grid point array is used as the size of the voxel grid, a regular grid voxel model is constructed, and a gradient function is set according to lithology category labels of the angular point of the voxel grid to simulate lithology composition of each voxel unit. The model is more in line with the basic rule of stratigraphic distribution, and the precision of the model is improved.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional geological body modeling, and in particular to a three-dimensional geological model establishment method, storage medium, and equipment. Background Art

[0002] Due to the diversity of geological environments and data preprocessing, the automatic 3D modeling and visualization of complex geological bodies present many difficulties. In the traditional modeling process, the modeling results still need to be manually modified. With the development of deep learning technology, 3D geological modeling technology has undergone significant innovation in recent years. An increasing number of deep learning neural network models are being applied to the field of 3D geological modeling, and 3D modeling is moving towards automation and intelligence. However, the technology of intelligent modeling methods based on deep learning is not yet mature. The shortcomings of deep learning network models in the field of 3D geological modeling are reflected in the following aspects:

[0003] In terms of data source organization, the amount of data used for modeling is limited and unevenly distributed, making it ineffective for building complex geological models. Therefore, when building models, it is important to integrate as many sources of information as possible. In practical applications, it is necessary to integrate multi-source spatial data representing the same region, or to update previous spatial data with current data. In addition to discrete sample data such as borehole data, sample data also contains a large amount of two-dimensional or high-dimensional continuous data. This data is often not directly usable for model training, making matrixing and integrating continuous sample data an essential step in the model building process.

[0004] When it comes to sampling methods within deep learning models, few consider the topological structure of geological entities. Some studies focus solely on individual geological entities, ignoring the correctness of the topological relationships between them. Deep learning models based on point cloud segmentation also lack consideration of the specific stratigraphic structure when performing 3D geological modeling. Strata have different horizontal and vertical distribution characteristics. In the actual sampling process of most deep learning-based geological models, the actual layered distribution characteristics of the geological layers are not considered. Instead, nearest neighbor sampling or random sampling methods are used to extract sample points within a certain spatial range to generate each batch of training data.

[0005] In terms of the model structure of deep learning models, complex geological phenomena such as faults and folds make it extremely complex to construct models that accurately simulate real-world 3D geological objects and incorporate interactive and analytical capabilities. 3D geological models constructed using traditional point cloud segmentation algorithms lack the application of relevant geological knowledge and fail to incorporate correlation mechanisms between strata or boreholes. Instead, they encode each sample point as an independent unit. Consequently, the resulting 3D geological models do not conform to the basic laws of stratum distribution.

[0006] Regarding the representation of 3D geological models, 3D geological model visualization has gradually become a common area of ​​concern for various teams working in structural modeling or geographic information systems. With the increasing complexity of models and modeling technologies, new issues related to model visualization and information representation are constantly emerging. Visualization and data conversion, along with the sharing and exchange of modeling data, are key technical issues. Current 3D geological models constructed using point cloud segmentation algorithms often require the introduction of an air layer and the construction of a grid point model before further model application. The lack of vector boundary constraints results in models that cannot simulate realistic surface relief, resulting in low model accuracy and poor visualization. Adopting appropriate methods to construct 3D geological voxel models that can simulate surface relief remains a major challenge. Summary of the Invention

[0007] The purpose of the present invention is to solve the problem that the three-dimensional geological body model in the prior art does not conform to the basic law of stratum distribution and has low accuracy, and to propose a three-dimensional geological model establishment method, comprising the following steps:

[0008] S1. Obtain continuous geological characteristic data, perform discretization sampling on the continuous geological characteristic data, and divide the sampled data into a training set and a test set;

[0009] S2. Constructing a lithology category prediction model, including an encoder and a decoder, and using a training set and a test set to train and test the lithology category prediction model to obtain a trained lithology category prediction model;

[0010] S3. Generate a regular grid point array using the DEM model of the study area, and use the trained lithology category prediction model to predict the lithology category labels of the regular grid points;

[0011] S4. Use the regular grid points as the corner points of each voxel grid and the step size of the regular grid as the size of the voxel grid to construct a regular grid voxel model. Set a gradient function according to the lithologic category label of the voxel grid corner point to simulate the lithologic composition of each voxel unit.

[0012] Furthermore, continuous geological characteristic data include: profile data and geological age data.

[0013] Furthermore, S1 is specifically:

[0014] S11. Obtain profile data and geological age data of the study area, use the geological age data as an additional feature of the profile data, calculate the minimum thickness of the stratum in the profile data, determine the sampling step length of the sample points of the continuous geological feature data, the sampling step lengths in the X and Y directions are equal, and the sampling step length in the Z direction is less than the minimum thickness of the stratum, and perform point sampling of the continuous geological feature data;

[0015] S12. Arrange the continuous geological feature data of point sampling in descending order of Z coordinates, divide all points with the same Z coordinate into a sample layer, and normalize them to obtain a feature matrix, including a spatial feature matrix and an additional feature matrix, wherein the feature matrix of the profile data is the spatial feature matrix, and the feature matrix of the geological age data is the additional feature matrix.

[0016] Furthermore, the encoder includes: an intra-layer feature aggregation module, an inter-layer attention score calculation module, and a global feature and local feature connection module;

[0017] The intra-layer feature aggregation module aggregates the local feature vectors of the sample points in each sample layer of the feature matrix to obtain the feature matrix of the aggregated features;

[0018] The inter-layer attention score calculation module is used to calculate the self-attention score between each layer of the feature matrix of the aggregated feature, and obtain the overall feature vector based on the self-attention score;

[0019] The global feature and local feature connection module is used to connect the global feature vector with the local feature vector of each sample point, perform linear feature transformation, and obtain a feature matrix that aggregates the global features and local features.

[0020] Furthermore, the intra-layer feature aggregation module includes a position encoding unit and an attentive pooling unit;

[0021] The position encoding unit calculates the spatial relationship between the central sample point of each layer of the spatial feature matrix and other sample points in the same layer to obtain a spatial relationship feature matrix. The central sample point is obtained by random sampling in each layer. The additional features of the broadcasted central sample point are concatenated with the additional features of other sample points in the same layer to obtain an additional feature matrix. The spatial relationship feature matrix is ​​concatenated with the additional feature matrix to obtain an intra-layer feature matrix.

[0022] The attentive pooling unit normalizes the intra-layer feature matrix, uses the normalized intra-layer feature matrix and the intra-layer feature matrix to obtain the attention matrix, adds the eigenvalues ​​of the attention matrix row by row to obtain the attentive pooling result, and aggregates the attentive pooling results of each sample layer to obtain the feature matrix of the aggregated features.

[0023] Furthermore, the spatial relationship is expressed by the following formula:

[0024]

[0025] in, Represents the central sample point p of the i-th layer i and the kth sample point of the i-th layer The spatial relationship between , ⊕ represents connection.

[0026] Furthermore, the decoder includes an upsampling unit, a feature dimension reduction unit, and a lithology category prediction unit;

[0027] The output of the encoder is sampled by neighboring point interpolation through the upsampling unit of the decoder, and then reduced in dimension using a multi-layer shared perceptron through the feature dimensionality reduction unit to obtain the lithology category features. The category feature score calculation unit converts the lithology category features into a probability distribution, and the lithology category with the largest probability value is used as the lithology category label.

[0028] Furthermore, the DEM model of the study area is used to generate a regular grid lattice. Specifically, a regular plane grid is established in the study area based on the principle that the step lengths in the X-axis and Y-axis directions are equal. The DEM model is used to determine the elevation range of the sample point column established at each grid point. A regular grid lattice is constructed based on the principle that the step length in the Z-axis direction is less than the minimum stratum thickness and the number of sample points in each sample point column is the same.

[0029] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the three-dimensional geological model establishment method is implemented.

[0030] The present invention also proposes an electronic device, comprising a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program includes computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the above-mentioned three-dimensional geological model establishment method.

[0031] The beneficial effects brought about by the technical solution provided by the present invention are:

[0032] The geological modeling method provided by the present invention matrixes continuous sample data, effectively increasing the data sources of samples; utilizes the geological characteristics of the layered distribution of geological bodies, performs stratified random sampling on the data set, retains the overall characteristics of each layer, and increases the model training efficiency; according to the layered distribution characteristics of the geological body, the characteristics of sample points in the same layer are aggregated, and an attention mechanism is adopted between layers to explore the potential correlation between the sampling layers. While modeling, the overall characteristics of the geological body, the overall characteristics of each sampling layer, and the local characteristics of the sample points in each sampling layer are effectively utilized, so that the training results are more in line with the distribution law of the geological body; uses DEM data to constrain the scope of the geological model, simulates the undulating characteristics of the surface, converts the final grid point label prediction results into the corner point attributes of the grid voxel model, constructs a three-dimensional geological voxel model, improves the accuracy of the model, and is beneficial to the application and maintenance of the geological model. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a flow chart of a method for establishing a three-dimensional geological model according to an embodiment of the present invention;

[0034] Figure 2 is a diagram showing the working principle of the lithology category prediction model according to an embodiment of the present invention;

[0035] Figure 3 It is a block diagram of an electronic device in an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0036] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0037] Example 1: A flow chart of a method for establishing a three-dimensional geological model according to an embodiment of the present invention is shown in FIG. Figure 1 , specifically including the following steps:

[0038] S1. Obtain continuous geological characteristic data, discretize and sample the continuous geological characteristic data, and divide the sampled data into a training set and a test set; the continuous geological characteristic data includes: profile data and geological age data.

[0039] In a preferred embodiment of the present invention, specifically,

[0040] S11. Obtain the profile data and geological age data of the study area. Use the geological age data as an additional feature of the profile data. Calculate the minimum thickness of the strata in the profile data. Determine the sampling step length of the sample points of the continuous geological feature data. The sampling step lengths in the X and Y directions are equal, and the sampling step length in the Z direction is less than the minimum thickness of the strata. Perform point sampling of the continuous geological feature data and determine the geological age characteristics of the sample points based on the distribution range of the geological age data. In addition to geological age data, other additional features can also be selected according to actual needs. Continuous profile data can also be replaced by discrete borehole data. Discrete borehole data does not require discretization sampling.

[0041] S12. Arrange the continuous geological feature data of the point sampling in descending order of Z coordinates, and number the geological age data, starting from 0, to ensure that each number represents a geological age. Arrange all sample point data in descending order of Z coordinates, divide all points with the same Z coordinate into a sample layer, and normalize them to obtain a feature matrix. The feature matrix includes a spatial feature matrix A and an additional feature matrix B, where the feature matrix of the profile data is the spatial feature matrix, and the feature matrix of the geological age data is the additional feature matrix. The additional feature matrix B is connected to the spatial feature matrix A in the form of an augmented matrix to form a feature matrix (A, B).

[0042] S2. Construct a lithology category prediction model, including an encoder and a decoder, and use a training set and a test set to train and test the lithology category prediction model to obtain a trained lithology category prediction model.

[0043] In a preferred embodiment of the present invention, the encoder includes: an intra-layer feature aggregation module, an inter-layer attention score calculation module, and a global feature and local feature connection module. The intra-layer feature aggregation module performs feature aggregation on the local feature vectors of the sample points in each sample layer of the feature matrix to obtain a feature matrix of the aggregated features. The inter-layer attention score calculation module is used to calculate the self-attention scores between each layer of the feature matrix of the aggregated features, and obtain the global feature vector based on the self-attention scores. The global feature and local feature connection module is used to connect the global feature vector with the local feature vector of each sample point, perform linear feature transformation, and obtain a feature matrix of the aggregated global features and local features.

[0044] The intra-layer feature aggregation module includes a position encoding unit and a careful pooling unit. The position encoding unit calculates the spatial relationship between the center sample point of each layer of the spatial feature matrix and other sample points in the same layer. The center sample point is obtained by random sampling in each layer. For the center sample point p of the i-th layer, i Sample points on the same layer Spatial relationships can be expressed as:

[0045]

[0046] Where, Represents the central sample point p of the i-th layer i and the kth sample point of the i-th layer The spatial relationship of p i and It is expressed in the form of spatial coordinates or vectors, ⊕ represents connection, ‖ ‖ represents the modulus of the vector, and the calculation method is: (1,2)⊕(2,1)=(1,2,2,1), calculating the central sample point p of the i-th layer i The spatial relationship feature matrix of the i-th layer is obtained by the spatial relationship with the sample points in the same layer n represents the number of sample points in the i-th layer, and the number of sample points in each layer is the same, which is n. The additional features of the central sample point in the i-th layer after broadcasting and additional features of other sample points in the same layer Connection, additional features and additional features Including geological age features, other features can also be selected according to actual needs to obtain the additional feature matrix of the i-th layer The spatial feature matrix C of each layer i and the additional feature matrix D iThrough linear transformation, they are increased to the same dimension respectively, and the spatial relationship feature matrix and the additional feature matrix are connected to obtain the intra-layer feature matrix A i =(C i ,D i ), A i Represents the intra-layer feature matrix of the i-th layer.

[0047] The careful pooling unit uses the normalized exponential function to transform the feature matrix A within the layer i All elements corresponding to each feature in are mapped to the interval (0,1), and the sum of the elements corresponding to each feature is equal to 1, and the probability distribution matrix A is obtained. ′ i . Calculate the attention score within the layer and get the attention matrix T of the i-th layer i =A i T A ′ i , add the eigenvalues ​​of the attention matrix row by row to get the careful pooling result d i , aggregate the careful pooling results of each sample layer to obtain the feature matrix D of the aggregated features.

[0048] In a preferred embodiment of the present invention, the decoder includes an upsampling unit, a feature dimensionality reduction unit, and a lithology category prediction unit. The encoder output is sampled by neighboring point interpolation in the decoder's upsampling unit. The feature dimensionality reduction unit then performs dimensionality reduction using a multi-layer shared perceptron to obtain lithology category features. The category feature score calculation unit converts the lithology category features into a probability distribution, with the lithology category with the highest probability value being used as the lithology category label.

[0049] refer to Figure 2 , Figure 2 This is a working principle diagram of the lithologic category prediction model according to an embodiment of the present invention. In the figure, m represents the number of sample layers, n represents the number of sample points in a single sample layer, and d represents the additional feature dimension. The encoder part uses the position encoding unit in the intra-layer feature aggregation module to convert the features (m, n, 3+d) of the layered sample data into k-dimensional features (m, n, k). The layer features are extracted by the careful pooling unit to obtain k-dimensional features. ′ dimensional layer feature vectors, and aggregate the features of all points in each layer to obtain the features (m,1,k ′ ); use the inter-layer attention score calculation module to calculate the inter-layer attention score, and pool it into a k-dimensional feature vector (1,1,k) containing global features. The decoder part combines the broadcasted k-dimensional global feature vector with the k ′ The k-dimensional layer feature vector is connected to obtain the integration layer feature and the global feature (k ′ +k")-dimensional layer feature vector (m,1,k ′ +k"), similarly, the broadcasted (k′ +k") dimensional layer feature vector is connected with the k-dimensional layer feature matrix obtained by feature transformation to obtain (k+k ′ +k")-dimensional sample point features (m,n,k+k ′ +k"), after feature conversion using the fully connected layer, the point features are reduced to c dimensions through linear transformation to obtain features (m, n, c), where c is the number of lithology categories of the drilling point. Finally, the normalized exponential function is used to convert the sample point features into probability distribution, and the maximum probability category of each point is taken as the lithology label of the point.

[0050] S3. Use the DEM model of the study area to generate a regular grid point array, and use the trained lithology category prediction model to predict the lithology category labels of the regular grid points.

[0051] In a preferred embodiment of the present invention, a DEM model of the study area is used to generate a regular grid lattice. Specifically, a regular plane grid is established in the study area based on the principle that the step lengths in the X-axis direction and the Y-axis direction are equal. The DEM model is used to determine the elevation range of the sample point column established at each grid point. The regular grid lattice is constructed based on the principle that the step length in the Z-axis direction is less than the minimum stratum thickness and the number of sample points in each sample point column is the same.

[0052] S4. Use the regular grid points as the corner points of each voxel grid and the step size of the regular grid as the size of the voxel grid to construct a regular grid voxel model. Set a gradient function according to the lithologic category label of the voxel grid corner point to simulate the lithologic composition of each voxel unit.

[0053] Example 2: In an exemplary embodiment, a computer-readable storage medium is included, which stores a computer program. When the computer program is executed by a processor, the above-mentioned three-dimensional geological model establishment method is implemented.

[0054] Example 3: Please refer to Figure 3 In an exemplary embodiment, an electronic device is also included, including at least one processor, at least one memory, and at least one communication bus.

[0055] The memory stores a computer program including computer-readable instructions. The processor calls the computer-readable instructions stored in the memory via a communication bus to execute the above-mentioned three-dimensional geological model establishment method.

[0056] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for establishing a three-dimensional geological model, characterized in that: The following steps are involved: S1. Obtain continuous geological characteristic data, discretize and sample the continuous geological characteristic data, and divide the sampled data into a training set and a test set; S2, constructing a lithology category prediction model, including an encoder and a decoder, and using a training set and a test set to train and test the lithology category prediction model to obtain a trained lithology category prediction model; S3, using the DEM model of the study area to generate a regular grid point array, and using the trained lithology category prediction model to predict the lithology category labels of the regular grid points; S4. Use the regular grid points as the corner points of each voxel grid, the step size of the regular grid dot matrix as the size of the voxel grid, construct a regular grid voxel model, and set a gradient function according to the lithology category label of the voxel grid corner point to simulate the lithology composition of each voxel unit.

2. A three-dimensional geological model establishment method according to claim 1, characterized in that: Continuous geological characteristic data include: profile data and geological age data.

3. A three-dimensional geological model establishment method according to claim 2, characterized in that: S1 is specifically: S11, obtaining the profile data and geological age data of the study area, using the geological age data as an additional feature of the profile data, calculating the minimum thickness of the strata in the profile data, determining the sampling step length of the sample points of the continuous geological feature data, the sampling step lengths in the X and Y directions are equal, and the sampling step length in the Z direction is less than the minimum thickness of the strata, and performing point sampling of the continuous geological feature data; S12, arranging the continuous geological feature data of point sampling in descending order of Z coordinates, dividing all points with the same Z coordinate into a sample layer, and normalizing them to obtain a feature matrix, including a spatial feature matrix and an additional feature matrix, wherein the feature matrix of the profile data is the spatial feature matrix, and the feature matrix of the geological age data is the additional feature matrix; S13. Divide the data set of the feature matrix into a training set and a test set.

4. A three-dimensional geological model establishment method according to claim 3, characterized in that: The encoder includes: intra-layer feature aggregation module, inter-layer attention score calculation module, and global feature and local feature connection module; The intra-layer feature aggregation module aggregates the local feature vectors of the sample points in each sample layer of the feature matrix to obtain a feature matrix of aggregated features; The inter-layer attention score calculation module is used to calculate the self-attention scores between each layer of the feature matrix of the aggregated features, and obtain the overall feature vector based on the self-attention scores; The global feature and local feature connection module is used to connect the global feature vector with the local feature vector of each sample point, perform linear feature transformation, and obtain a feature matrix that aggregates the global features and local features.

5. A three-dimensional geological model establishment method according to claim 4, characterized in that: The intra-layer feature aggregation module includes a position encoding unit and an attentive pooling unit; The position encoding unit calculates the spatial relationship between the central sample point of each layer of the spatial feature matrix and other sample points in the same layer to obtain a spatial relationship feature matrix. The central sample point is obtained by random sampling in each layer. The additional features of the central sample point after broadcasting are connected with the additional features of other sample points in the same layer to obtain an additional feature matrix. The spatial relationship feature matrix is ​​connected with the additional feature matrix to obtain an intra-layer feature matrix. The attentive pooling unit normalizes the intra-layer feature matrix, uses the normalized intra-layer feature matrix and the intra-layer feature matrix to obtain the attention matrix, adds the eigenvalues ​​of the attention matrix row by row to obtain the attentive pooling result, and aggregates the attentive pooling results of each sample layer to obtain the feature matrix of the aggregated features.

6. A three-dimensional geological model establishment method according to claim 5, characterized in that: The spatial relationship is expressed by the following formula: in, Represents the central sample point p of the i-th layer i and the kth sample point of the i-th layer The spatial relationship Indicates a connection.

7. A three-dimensional geological model establishment method according to claim 1, characterized in that: The decoder includes an upsampling unit, a feature dimension reduction unit, and a lithology category prediction unit; The output of the encoder is sampled by neighboring point interpolation through the upsampling unit of the decoder, and then reduced in dimension by a multi-layer shared perceptron through the feature dimension reduction unit to obtain the lithology category features. The category feature score calculation unit converts the lithology category features into probability distribution, and the lithology category with the largest probability value is used as the lithology category label.

8. A three-dimensional geological model establishment method according to claim 3, characterized in that: The DEM model of the study area is used to generate a regular grid lattice. Specifically, a regular plane grid is established in the study area based on the principle of equal step lengths in the X-axis and Y-axis directions. The DEM model is used to determine the elevation range of the sample point column established at each grid point. A regular grid lattice is constructed based on the principle that the step length in the Z-axis direction is less than the minimum stratum thickness and the number of sample points in each sample point column is the same.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the method according to any one of claims 1 to 8.

Images