Drill hole equalization upsampling method and device based on generative adversarial network

Abstract

The invention discloses a borehole equalization upsampling method and device based on a generative adversarial network. The method comprises the following steps: (1) reading an unbalanced borehole data set; (2) performing data balance analysis on the unbalanced drilling data set to obtain a sparse stratum and a sparse drilling area; (3) sampling each drill hole in the unbalanced drill hole data set to obtain drill hole sampling point data including sampling point positions and stratum type attributes, and forming a training sample data set; (4) constructing a drilling up-sampling model based on the generative adversarial network, and training the drilling up-sampling model by adopting the training sample data set; (5) inputting the stratum type attribute of the sparse stratum into the trained generator for up-sampling to obtain up-sampled drilling sampling point data; and (6) combining the drilling sampling point data with the data of the unbalanced drilling data set to obtain balanced drilling sampling point data. The method is wide in application scene and higher in accuracy.

Description

Technical Field

[0001] The present invention relates to geographic information technology, and particularly to a borehole equalization upsampling method and device based on a generative adversarial network. Background Art

[0002] Borehole data, as the core data source for geological modeling, is mainly obtained through methods such as drilling, acoustic wave testing, and electromagnetic wave detection. Among them, the drilling method has become the dominant method due to its intuitiveness and wide applicability. Borehole data records key information such as the demarcation points between rock layers and soil layers, the elevation of the layer top, the category, and the thickness, providing a basic support for revealing the distribution law of strata. However, such data faces the dual challenges of uneven spatial distribution and imbalance in the number of strata categories in application, directly affecting the quality of subsequent modeling.

[0003] Three-dimensional geological modeling methods based on borehole data can be divided into two categories: direct modeling and intelligent modeling based on neural networks. The former constructs a model relying on the spatial topological relationship of the original boreholes, and the latter simulates the distribution law of strata through feature learning. Both methods are significantly affected by the number of boreholes, spatial density, and strata balance: uneven spatial distribution easily leads to a decrease in the accuracy of the model in sparse areas, while imbalance in strata categories causes weight deviation problems in neural network training. Although geological profile extraction technology can expand the amount of borehole data, it is difficult to fundamentally solve the problems of data distribution and category balance.

[0004] Traditional upsampling techniques improve data density through interpolation algorithms, but their applications have significant objectivity and adaptability defects. First, the interpolation method relies on expert experience to select parameters, and subjective factors lead to insufficient stability of the results. Second, traditional algorithms are based on the assumptions of isotropy and stationarity, and it is difficult to adapt to the discontinuity and complexity of actual geological structures. In addition, existing methods cannot generate data specifically for a particular strata category, resulting in the problem of imbalance in the number of strata not being effectively alleviated (Marinoni, 2003). In summary, traditional upsampling techniques have a narrow range of applicable scenarios and inaccurate calculation results. Summary of the Invention

[0005] Aiming at the problems existing in the prior art, the purpose of the present invention is to provide a borehole equalization upsampling method and device based on a generative adversarial network with a wide range of applicable scenarios and higher accuracy of calculation results.

[0006] In order to achieve the above-mentioned invention purpose, the present invention provides the following technical solutions:

[0007] A borehole equalization upsampling method based on a generative adversarial network, the method comprising the following steps:

[0008] (1) Read the borehole data that needs to be upsampled to form an unbalanced borehole data set;

[0009] (2) Perform data balance analysis on the unbalanced borehole dataset to obtain sparse strata and sparse borehole areas that can equalize the dataset categories and spatial positions;

[0010] (3) Sample each borehole in the unbalanced borehole dataset to obtain borehole sampling point data including sampling point positions and strata type attributes, forming a training sample dataset;

[0011] (4) Construct a borehole upsampling model based on a generative adversarial network, and use the training sample dataset to train the borehole upsampling model; wherein, the generator of the borehole upsampling model is used to generate pseudo-borehole sampling point data using random noise according to the input strata type attributes, and the pseudo-borehole sampling point data includes the positions and strata type attributes of the pseudo-sampling points; the discriminator of the borehole upsampling model is used to identify whether the input borehole sampling point data or pseudo-borehole sampling point data is a real sample;

[0012] (5) Input the strata type attributes of the sparse strata obtained in step (2) into the generator of the trained borehole upsampling model for upsampling, and perform boundary constraints on the upsampling results based on the sparse borehole areas to obtain upsampled borehole sampling point data;

[0013] (6) Combine the borehole sampling point data in step (5) with the data in the unbalanced borehole dataset to obtain balanced borehole sampling point data.

[0014] Further, step (2) specifically includes:

[0015] (2-1) Traverse the unbalanced borehole dataset and count the total number of boreholes nd;

[0016] (2-2) Statistically analyze the thicknesses of the strata where the boreholes are located to form a strata thickness set thickness = {t i |i = 1, 2,..., nc}, and calculate the average strata thickness mean_thick; where t i represents the thickness of the i-th strata type, and nc represents the number of strata types;

[0017] (2-3) Divide the difference between mean_thick and each strata thickness in thickness by mean_thick to obtain the sparsity of each strata, and determine the strata with positive sparsity as sparse strata;

[0018] (2-4) Grid the drilling coordinate area to obtain the maximum longitude max_lon, maximum latitude max_lat, minimum longitude min_lon, and minimum latitude min_lat of the drilling coordinates, and calculate the drilling area area;

[0019] (2-5) Divide the total number of drill holes nd by the drilling area area to obtain the average density of the drilling area;

[0020] (2-6) Divide the drilling area into several blocks, calculate the average density of the sub-block drilling area of each block in turn, and subtract the average density of the drilling area from the average density of the sub-block drilling area of each block. The sub-block with a positive difference is determined as the sparse drilling area.

[0021] Further, the calculation formula for the drilling area area in step (2-3) is:

[0022] area = (max_lon - min_lon) * (max_lat - max_lat) * piexl_width 2

[0023] In the formula, piexl_width is the actual distance converted from the unit coordinate value to the geospatial.

[0024] Further, step (3) specifically includes:

[0025] (3-1) Sequentially read the drill hole p in the unbalanced drill hole dataset i ;

[0026] (3-2) According to the top elevation of each layer of the formation of the drill hole p i , calculate the sampling point spacing according to the following formula:

[0027] H i,j = (d i,j - d i,j+1 ) / nj

[0028] Among them, H i,j represents the sampling point spacing of the formation j of the drill hole p i , d i,j , d i,j+1 are the top elevations of the formations j and j + 1 of the drill hole p i respectively, and nj is the number of sampling points of the formation j;

[0029] (3-3) For each formation j of the drill hole p i , starting from the top elevation of the formation, vertically downward every sampling point spacing H i,jPerform a sampling, and store the longitude, latitude, elevation, and the stratum category attribute of the sampling point as the borehole sampling point data.

[0030] (3-4) Traverse the boreholes in the unbalanced borehole dataset, and use the set of all borehole sampling point data as training samples to form a training sample dataset.

[0031] Further, the network architecture of the generator is as follows: The first layer is a feature embedding layer, which is used to perform feature embedding on the stratum type attribute to obtain a feature embedding matrix; the second layer is a splicing layer, which splices the preset random noise matrix and the feature embedding matrix; the third layer is a fully connected layer, the fourth layer is a LeakReLU activation layer and a BatchNorm1d layer; the fifth layer is a fully connected layer; the sixth layer is a LeakReLU activation layer and a BatchNorm1d layer; the fifth layer is a fully connected layer; the seventh layer is a LeakReLU activation layer and a BatchNorm1d layer; the eighth layer is a fully connected layer.

[0032] Further, the network architecture of the discriminator is as follows: The first layer is a fully connected layer; the second layer is a LeakReLU activation layer; the third layer is a fully connected layer; the fourth layer is a LeakReLU activation layer; the fifth layer is a feature embedding layer; the sixth layer is a fully connected layer; the seventh layer is a LeakReLU activation layer; the eighth layer is a fully connected layer; the ninth layer is a Dropout layer; the tenth layer is a fully connected layer; the last layer is a Sigmoid activation layer.

[0033] Further, step (5) specifically includes:

[0034] (5-1) Use the stratum type attribute of the sparse stratum obtained in step (2) as the input to train the generator to generate the coordinate positions corresponding to the stratum type.

[0035] (5-2) Use the surface elevation model DEM of the sparse borehole area, the upper surface of the bedrock, the outcropping bedrock area, and the boundary of the sparse borehole area as constraint conditions to constrain the coordinate positions generated in (5-1), and use the qualified coordinate positions and the corresponding stratum type attributes as the upsampled borehole sampling point data.

[0036] (5-3) Loop (5-1) and (5-2) until the sparsity of the sparse stratum is less than the preset threshold.

[0037] A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the above method.

[0038] A computer-readable storage medium has a computer program / instructions stored thereon, and the computer program / instructions, when executed by a processor, implement the above method.

[0039] A computer program product includes a computer program / instructions, and the computer program / instructions, when executed by a processor, implement the above method.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention performs upsampling based on the Generative Adversarial Network (GAN). This model dynamically optimizes the generator and discriminator through an adversarial training mechanism, can synchronously solve the problems of sparse spatial distribution and formation category imbalance, significantly improve the accuracy and applicable scenarios of the upsampling results, realizes the full-process automation of borehole data upsampling, and provides a high-quality data basis for geological modeling. Description of the Drawings

[0041] Figure 1 is a flowchart of the borehole equalization upsampling method based on the generative adversarial network provided by the present invention;

[0042] Figure 2 is a spatial distribution map of Nanjing borehole data;

[0043] Figure 3 is a model structure diagram of the generator;

[0044] Figure 4 is a model structure diagram of the discriminator;

[0045] Figure 5 is a line graph of the model training Loss value;

[0046] Figure 6 is a tif image of the Nanjing ground elevation model;

[0047] Figure 7 is a tif image of the elevation of the upper surface of the unexposed bedrock in Nanjing;

[0048] Figure 8 is the distribution of the exposed bedrock area in Nanjing;

[0049] Figure 9 is a two-dimensional spatial distribution map of the upsampled borehole data;

[0050] Figure 10 is a three-dimensional spatial distribution map of the upsampled borehole data;

[0051] Figure 11 is an analysis diagram of the balance degree of the borehole formation thickness after upsampling; Detailed Embodiments

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention.

[0053] The embodiment of the present invention provides a drilling equalization upsampling method based on a generative adversarial network, as Figure 1 shown, including the following steps:

[0054] (1) Read the drilling data that needs to be upsampled to form an unbalanced drilling data set.

[0055] The Nanjing drilling data used in this embodiment after standardization is shown in Table 1, and the spatial distribution is as Figure 2 shown:

[0056] Table 1 Nanjing loose layer drilling data table (partial)

[0057]

[0058] (2) Perform data balance analysis on the unbalanced drilling data set to obtain sparse strata and sparse drilling areas that can equalize the data set category and spatial position.

[0059] This step specifically includes:

[0060] (2-1) Traverse the unbalanced drilling data set and count the total number of drill holes nd = 3858;

[0061] (2-2) Statistically analyze the thickness of each formation where the drill holes are located to form a formation thickness set thickness = {t i |i = 1, 2,..., nc}, and calculate the average formation thickness mean_thick; where t i represents the thickness of the i-th formation type, and nc represents the number of formation types;

[0062]

[0063] In this embodiment, thickness = {2101, 18939, 10139, 3317, 62, 360, 10520}, nc = 7, mean_thick = 6391.143;

[0064] (2-3) Divide the difference between mean_thick and each formation thickness in thickness by mean_thick to obtain the sparsity of each formation. Determine the formation with a positive sparsity as a sparse formation;

[0065] parsity = (mean_thick - t i ) / mean_thick

[0066] In this embodiment:

[0067] sparsity = {3114, -12447.857, -3647.857, 3174.143, 6492.143, 6131.143, -4028.857}, and the corresponding sparse formations are Formation 1, 4, 5, and 6.

[0068] (2 - 4) Rasterize the drilling coordinate area to obtain the maximum longitude max_lon, maximum latitude max_lat, minimum longitude min_lon, and minimum latitude min_lat of the drilling coordinates, and calculate the drilling area area;

[0069] area = (max_lon - min_lon) * (max_lat - min_lat) * piexl_width 2

[0070] Where piexl_width is the actual distance when the unit coordinate value is converted to the geographical space;

[0071] In this embodiment, max_lon = 210091.386719, max_lat = 173038.148438, min_lon = 89348.1484375, min_lat = 56491.3867188, piexl_width = 5 (m), and calculate area = 3.51805860525 * 10 5 (km 2 )

[0072] (2 - 5) Divide the total number of drill holes nd by the drilling area area to obtain the average density ρ of the drilling area;

[0073]

[0074] In this embodiment, ρ = 0.010966 (num / km 2 )

[0075] (2 - 6) Divide the drilling area into several blocks, calculate the average density of the sub - drilling area of each block in turn, and subtract the average density of the drilling area from the average density of the sub - drilling area of each block. The blocks with positive differences are determined as drilling sparse areas.

[0076] (3) Sample each drill hole in the unbalanced drill hole dataset to obtain drill hole sampling point data including the positions of sampling points and formation type attributes, and form a training sample dataset.

[0077] This step specifically includes:

[0078] (3-1) Sequentially read borehole p in the unbalanced borehole dataset i ;

[0079] (3-2) According to borehole p i The elevation of the top of each layer of the formation, calculate the sampling point spacing according to the following formula:

[0080] H i,j =(d i,j -d i,j+1 ) / nj

[0081] Where, H i,j represents the sampling point spacing of formation j of borehole p i , d i,j , d i,j+1 are respectively the elevation of the top of formation j and j+1 of borehole p i , and nj is the number of sampling points of formation j; in this embodiment, nj of each formation is 10;

[0082] (3-3) For each formation j of borehole p i , starting from the elevation of the top of the formation, sample once every sampling point spacing H i,j vertically downward, and store the position (longitude, latitude, elevation) of the sampling point and the formation category attribute where it is located as the borehole sampling point data;

[0083] (3-4) Traverse the boreholes in the unbalanced borehole dataset, and use the set of all borehole sampling point data as the training samples to form a training sample dataset.

[0084] (4) Construct a borehole upsampling model based on a generative adversarial network, and train the borehole upsampling model using the training sample dataset.

[0085] Among them, the generator of the borehole upsampling model is used to generate pseudo-borehole sampling point data according to the input formation type attribute c. The pseudo-borehole sampling point data includes the position of the pseudo-sampling point and the formation type attribute c, specifically (x ′ , y ′ , z ′ , c); x ′ , y ′ , z ′ respectively represent the longitude, latitude, and elevation of the pseudo-sampling point. As Figure 3As shown in the figure, the network architecture of the generator is as follows: The first layer is the label_embedding layer with a dimension of (num_litho_type, label_emb_dim), which completes the feature embedding of the formation categories; the second layer is the concatenation layer that concatenates the features of the noise matrix and the embedding matrix, with a dimension of (label_emb_dim + latent_dim, 64); the third layer is the fully connected layer with a dimension of (64, 256), the fourth layer is the LeakReLU activation layer and the BatchNorm1d layer; the fifth layer is the fully connected layer with a dimension of (256, 512); the sixth layer is the LeakReLU activation layer and the BatchNorm1d layer; the fifth layer is the fully connected layer with a dimension of (512, 1024); the seventh layer is the LeakReLU activation layer and the BatchNorm1d layer; the eighth layer is the fully connected layer with a dimension of (1024, 3). Among them, num_litho_type is the number of formation categories, label_emb_dim is the dimension of the formation category feature embedding, and latent_dim is the dimension of the random noise. In this embodiment, the number of formation categories num_litho_type is 7, the dimension of the formation category feature embedding label_emb_dim is 10, and the dimension of the random noise latent_dim is 64;

[0086] The discriminator of the borehole upsampling model is used to identify whether the input borehole sampling point data or pseudo-borehole sampling point data is a real sample. Specifically, taking a sampling point and formation attribute features (x, y, z, c) as the input, it outputs a binary classification probability value indicating whether the point is a real sample. As Figure 4 shown in the figure, the network architecture of the discriminator is as follows: The first layer is the fully connected layer with a dimension of (3, 8); the second layer is the LeakReLU activation layer; the third layer is the fully connected layer with a dimension of (8, 16); the fourth layer is the LeakReLU activation layer; the fifth layer is the label_embedding layer with a dimension of (num_litho_type = 7, label_emb_dim = 10); the sixth layer is the fully connected layer with a dimension of (16 + label_emb_dim = 10, 128); the seventh layer is the LeakReLU activation layer; the eighth layer is the fully connected layer with a dimension of (128, 128); the ninth layer is the Dropout layer with a parameter of 0.2; the tenth layer is the fully connected layer with a dimension of (128, 1); the last layer is the Sigmoid activation layer.

[0087] The training process adopts a dynamic learning rate adjustment strategy. When the Loss value stabilizes within the range of ±0.05 for each cycle, the learning rate is adjusted to 50%, and the initial learning rate is 0.1. The batch_size for model training is 128, the training cycle is 500, and the Drop_out regularization technique is used. The splitting ratio of the training set to the test set is 4:1. The optimizers for both the generator and the discriminator are selected as the Adam optimizer, and the loss function is BCEWithLogitsLoss(); the generator and the discriminator are alternately adversarially trained in each batch, and their loss values are recorded. When the training is completed, the loss values of the trainer and the discriminator should converge to complete the model training. In this embodiment, the loss value during model training is as Figure 5 shown.

[0088] (5) Input the formation type attributes of the sparse formation obtained in step (2) into the generator of the trained borehole upsampling model for upsampling, and perform boundary constraints on the upsampling results based on the sparse borehole area to obtain the upsampled borehole sampling point data.

[0089] This step specifically includes:

[0090] The ground elevation model in this embodiment is as Figure 6 shown, the upper surface of the bedrock is as Figure 7 shown, and the outcropping bedrock area is as Figure 8 shown;

[0091] (5-1) Use the formation type attributes of the sparse formation obtained in step (2) as the input to the trained generator to generate the coordinate positions corresponding to the formation types;

[0092] (5-2) Use the DEM of the surface elevation model, the upper surface of the bedrock, the outcropping bedrock area, and the boundary of the sparse borehole area in the sparse borehole area as constraint conditions to constrain the coordinate positions generated by the random noise matrix in (5-1), and use the qualified coordinate positions and the corresponding formation type attributes as the upsampled borehole sampling point data;

[0093] (5-3) Loop (5-1) and (5-2) until the sparsity of the sparse formation is less than the preset threshold of 0.5. The two-dimensional spatial distribution of the upsampled borehole data is as Figure 9 shown, and the three-dimensional spatial distribution is as Figure 10 shown.

[0094] (6) Combine the borehole sampling point data in step (5) with the data in the unbalanced borehole dataset to obtain the balanced borehole sampling point data.

[0095] The analysis of the balance degree of the borehole formation thickness after upsampling in this embodiment is as Figure 11 shown.

[0096] An embodiment of the present invention further provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the above method.

[0097] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program / instructions are stored, and the computer program / instructions implement the above method when executed by a processor.

[0098] An embodiment of the present invention further provides a computer program product, including a computer program / instructions, and the computer program / instructions implement the above method when executed by a processor.

[0099] It should be understood that the above embodiments and the descriptions in the specification only describe the principles, main features, and advantages of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the protection scope of the present invention.

Claims

1. A drilling equalization upsampling method based on a generative adversarial network, characterized in that, The method includes the following steps: (1) Read the borehole data that needs to be upsampled to form an unbalanced borehole dataset; (2) Perform data balance analysis on the unbalanced borehole dataset to obtain sparse strata and sparse borehole areas that can equalize the dataset category and spatial position; (3) Sample each borehole in the unbalanced borehole dataset to obtain borehole sampling point data including the sampling point position and formation type attribute, and form a training sample dataset; (4) Construct a borehole upsampling model based on a generative adversarial network, and use the training sample dataset to train the borehole upsampling model; wherein, the generator of the borehole upsampling model is used to generate pseudo-borehole sampling point data using random noise according to the input formation type attribute, and the pseudo-borehole sampling point data includes the position and formation type attribute of the pseudo-sampling point; the discriminator of the borehole upsampling model is used to identify whether the input borehole sampling point data or pseudo-borehole sampling point data is a real sample; (5) Input the formation type attribute of the sparse strata obtained in step (2) into the generator of the trained borehole upsampling model for upsampling, and perform boundary constraint on the upsampling result based on the sparse borehole area to obtain upsampled borehole sampling point data; (6) Combine the borehole sampling point data obtained in step (5) with the data of the unbalanced borehole dataset to obtain balanced borehole sampling point data.

2. The method for borehole equalization upsampling based on a generative adversarial network according to claim 1, wherein (7) Step (2) specifically includes: (2-1) Traverse the unbalanced borehole dataset and count the total number of boreholes \(n_d\); (2-2) Statistically calculate the thickness of each stratum where the boreholes are located to form a stratum thickness set thickness = {t i | i = 1, 2, …, nc}, and calculate the average stratum thickness mean_thick; where t i represents the thickness of the i-th stratum type, and nc represents the number of stratum types; (2-3) Divide the difference between the mean thickness in \(thickness\) and each formation thickness by \(mean\_thick\) to obtain the sparsity of each formation \(sparsity\), and determine the formation with \(sparsity\) being a positive number as a sparse formation; (2-4) Grid the borehole coordinate area, obtain the maximum longitude \(max\_lon\), maximum latitude \(max\_lat\), minimum longitude \(min\_lon\), and minimum latitude \(min\_lat\) of the borehole coordinates, and calculate the borehole area \(area\); (2-5) Divide the total number of boreholes \(n_d\) by the borehole area \(area\) to obtain the average density of the borehole area; (2-6) Divide the borehole area into several blocks, calculate the average density of each sub-block borehole area in turn, and subtract the average density of the borehole area from the average density of each sub-block borehole area. The sub-block with a positive difference is determined as a sparse borehole area.

3. The method for drilling equalization upsampling based on a generative adversarial network according to claim 2, wherein (13) The calculation formula for the borehole area \(area\) in step (2-3) is: area=(max_lon - min_lon)*(max_lat - max_lat) * piexl_width 2 where \(piexl\_width\) is the actual distance when converting the unit coordinate value to the geospatial.

4. The method for hole drilling equalization upsampling based on a generative adversarial network according to claim 1, characterized in that (15) Step (3) specifically includes: (3-1) Sequentially read borehole p in the unbalanced borehole dataset i ; (3-2) According to borehole p i The elevation of the top of each stratum is used to calculate the sampling point spacing according to the following formula: H i,j = (d i,j - d i,j+1 ) / nj Among them, H i,j represents the sampling point spacing of formation j of borehole p i , d i,j and d i,j+1 are respectively the elevation of the top of formation j and j + 1 of borehole p i , and nj is the number of sampling points of formation j; (3-3) For borehole p i For each formation j, starting from the elevation of the formation top, sample vertically downward every sampling point spacing H i,j once, and store the longitude, latitude, elevation, and formation category attribute of the sampling point as borehole sampling point data; (3-4) Traverse the boreholes in the unbalanced borehole dataset, and use the set of all borehole sampling point data as training samples to form a training sample dataset.

5. The method for drill hole equalization upsampling based on a generative adversarial network according to claim 1, wherein, (17) The network architecture of the generator is as follows: The first layer is a feature embedding layer, which is used to perform feature embedding on the formation type attribute to obtain a feature embedding matrix; the second layer is a concatenation layer, which concatenates the preset random noise matrix and the feature embedding matrix; The third layer is a fully connected layer, and the fourth layer is a LeakyReLU activation layer and a BatchNorm1d layer; the fifth layer is a fully connected layer; The sixth layer is a LeakyReLU activation layer and a BatchNorm1d layer; the fifth layer is a fully connected layer; the seventh layer is a LeakyReLU activation layer and a BatchNorm1d layer; the eighth layer is a fully connected layer.

6. The method for hole drilling equalization upsampling based on a generative adversarial network according to claim 1, wherein, The network architecture of the discriminator is as follows: the first layer is a fully connected layer; the second layer is a LeakyReLU activation layer; the third layer is a fully connected layer; the fourth layer is a LeakyReLU activation layer; the fifth layer is a feature embedding layer; The sixth layer is a fully connected layer; The seventh layer is a LeakyReLU activation layer; the eighth layer is a fully connected layer; The ninth layer is a Dropout layer; the tenth layer is a fully connected layer; the last layer is a Sigmoid activation layer.

7. The method for drill hole equalization upsampling based on a generative adversarial network according to claim 1, characterized in that Step (5) specifically includes: (5-1) Use the formation type attribute of the sparse formation obtained in step (2) as the input to the trained generator to generate the coordinate positions corresponding to the formation type; (5-2) Use the digital elevation model (DEM) of the ground surface, the upper surface of the bedrock, the outcropping bedrock area, and the boundary of the sparse borehole area in the sparse borehole area as the constraint conditions to constrain the coordinate positions generated by the random noise matrix in (5-1), and use the qualified coordinate positions and the corresponding formation type attributes as the upsampled borehole sampling point data; (5-3) Loop (5-1) and (5-2) until the sparsity of the sparse formation is less than the preset threshold.

8. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that: The processor executes the computer program to implement the method according to any one of claims 1-7.

9. A computer-readable storage medium having computer programs / instructions stored thereon, characterized in that, The computer program / instructions implement the method according to any one of claims 1-7 when executed by the processor.

10. A computer program product, comprising a computer program / instructions, characterized in that, The computer program / instructions, when executed by the processor, implement the method according to any one of claims 1-7.

Images