Three-dimensional digital core construction method and device, electronic equipment and storage medium

Abstract

The invention provides a three-dimensional digital core construction method and device, electronic equipment and a storage medium, and relates to the field of reservoir logging recognition. The method comprises the following steps: performing CT scanning on a typical sample of a research layer section to obtain CT data; reconstructing a CT three-dimensional rock core according to the CT data; selecting a representative rock core, and extracting a two-dimensional slice image from the representative rock core to construct a data set; training and parameter optimization are carried out on a constructed three-dimensional rock core construction model through the data set, the three-dimensional rock core construction model is a generative adversarial network model used for 3D reconstruction, and a model training process is constrained through a statistical characteristic function representing pore properties; and inputting the two-dimensional slice image of the target layer section into the three-dimensional rock core construction model after parameter optimization, and generating a three-dimensional digital rock core. According to the method, the statistical characteristic function representing the pore property is introduced into the training of the generative adversarial network to restrain the generation result, so that high-precision three-dimensional digital core simulation is realized, and the method is fast, simple and low in cost.

Description

Technical Field

[0001] The present invention belongs to the technical field of reservoir logging identification, and particularly relates to a method and device for constructing a three-dimensional digital core, an electronic device, and a storage medium. Background Art

[0002] The three-dimensional digital core technology is an important part of the reservoir rock characterization process and is widely used in engineering such as hydrogeology, mineral collection, and oil extraction. It is one of the important means in the field of oil and gas exploration and development. The key factor for accurate characterization of reservoir rock properties lies in the accuracy of three-dimensional digital core construction. Therefore, how to accurately and quickly construct a three-dimensional digital core using the pore distribution of the core will have an important impact on the research of the storage characteristics, transport characteristics, electromagnetic characteristics, and response characteristics of other logging signals of reservoir rocks.

[0003] Currently, the common methods for constructing a three-dimensional digital core are mainly divided into two categories: CT experiments and reconstruction methods based on two-dimensional images. CT experiments use a computer tomography instrument to scan the rock layer by layer. Since different components in the rock have different densities and different X-ray absorption coefficients, the rock skeleton and pore space can be distinguished. This method first requires sample preparation. However, it is difficult to obtain samples, and there are also deficiencies such as high experimental costs and long cycles, and the recognition accuracy of minerals is relatively low. In addition, when the reservoir heterogeneity is strong, it is difficult for CT scanning to obtain a three-dimensional core representing reservoir characteristics, thus affecting the judgment of the properties of the layer to be evaluated. In many cases, it is difficult to obtain a complete core in the layer where fractures are difficult to develop, and often only some two-dimensional image information is available, such as a surface slice or a part. Moreover, a considerable part of the main work of digital core physics is to characterize and analyze some mature oil reservoirs to improve the recovery rate. In many cases, the core materials of such reservoirs have been lost or are difficult to obtain, while the storage capacity of two-dimensional microscopic thin-section photos is considerable. The reconstruction methods based on two-dimensional images include the random method and the process method, etc. Relevant scholars at home and abroad have also proposed the multiple-point geostatistical method, which is an optimized random method. This method randomly reconstructs a statistically equivalent three-dimensional structure in a two-dimensional image, extracts the morphological characteristics of the two-dimensional thin slice from the given two-dimensional image, and reproduces them in the three-dimensional reconstructed core. The multiple-point geostatistical method expresses the correlation between multiple points in space and overcomes the deficiencies of traditional statistics in expressing complex reservoir structures and reproducing target geometric forms. However, the above reconstruction methods based on two-dimensional images are slow in constructing complex three-dimensional cores, and the construction accuracy is often low, which cannot meet the needs of physical property research in reservoir logging evaluation.

[0004] In recent years, the development of deep learning has become increasingly mature. Especially, the diversity and accuracy of the Generative Adversarial Network (GAN) in image construction. Since its development, the GAN algorithm has successfully generated highly realistic images in many fields such as face and natural image synthesis. Currently, the GAN algorithm has also been gradually introduced into the construction of three-dimensional porous media. For example, the Chinese patent document with the application number 202210508568.9 discloses a three-dimensional reconstruction method of shale digital core based on deep learning. This method learns the characteristics of core images through the generative adversarial network and combines the learning of motion information between adjacent frames by the optical flow method to construct a more continuous three-dimensional core model. Based on the above research, it is feasible to apply the GAN network to the three-dimensional digital core reconstruction work.

[0005] In summary, how to apply the GAN network to the three-dimensional digital core reconstruction work, how to reduce the dependence on CT scans, and how to improve the reconstruction method based on two-dimensional images are the key research contents in this field. Summary of the Invention

[0006] The purpose of the embodiments of the present invention is to provide a three-dimensional digital core construction method, device, electronic device, and storage medium to solve one or more defects existing in the traditional three-dimensional digital core construction method mentioned in the background technology.

[0007] To achieve the above purpose, the first aspect of the embodiments of the present invention provides a three-dimensional digital core construction method, which includes:

[0008] Obtain rock samples of the research interval and select typical samples for CT scan analysis to obtain CT data;

[0009] Reconstruct a CT three-dimensional core based on the CT data;

[0010] Select representative cores from the CT three-dimensional core, extract two-dimensional thin-section images from the representative cores, and construct a data set with these two-dimensional thin-section images;

[0011] Train and optimize the parameters of the constructed three-dimensional core construction model through the data set. When training, use the representative core as the ground truth;

[0012] Obtain two-dimensional thin-section images of the rocks in the target interval, input these two-dimensional thin-section images into the three-dimensional core construction model with optimized parameters to generate a three-dimensional digital core of the target interval;

[0013] Among them, the three-dimensional core construction model is a generative adversarial network model for 3D reconstruction, and the training process of the model is constrained by a statistical feature function representing pore properties.

[0014] Optionally, the method further includes:

[0015] Perform QEMSCAN (Quantitative Evaluation of Minerals by Scanning Electron Microscopy) scanning analysis on typical samples to obtain QEMSCAN data;

[0016] Extract mineral information from the QEMSCAN data, and establish the corresponding relationship between CT measurement gray levels and mineral components based on the CT data and mineral information;

[0017] Constrain the training process of the three-dimensional core construction model through the corresponding relationship between CT measurement gray levels and mineral components.

[0018] Optionally, the architecture of the generative adversarial network model is a DCGAN network.

[0019] Optionally, the generative adversarial network model includes a generator and a discriminator, and the ELU activation function is used for the non-output layers of both the generator and the discriminator.

[0020] Optionally, the RAdam optimizer is used during the training of the three-dimensional core construction model.

[0021] Optionally, two-dimensional thin-section images are extracted from representative cores, specifically:

[0022] Perform multiple thin-section analyses of representative cores along directions parallel and perpendicular to the Z-axis respectively, and extract two-dimensional thin-section images in three directions.

[0023] Optionally, the statistical characteristic functions representing pore properties include two-point probability function, local porosity distribution function, linear path function, and morphological characteristic function.

[0024] Optionally, the two-point probability function quantifies the pore distribution characteristics of the core in terms of probability by randomly selecting two points in the representative core and calculating the probability that the two points are in the same phase. The probability calculation formula is: S2(r) = P(x ∈ R, x + r ∈ R), where x and x + r represent two points separated by the lag distance r, and R represents the R phase.

[0025] Optionally, the local porosity distribution function quantifies the geometric characteristics of the porous medium of the representative core by selecting multiple measurement units in the representative core, calculating the local porosity of each measurement unit, and calculating the proportion of the measurement units with the local porosity equal to a specific porosity among all measurement units;

[0026] The formula for calculating the local porosity is:

[0027] The formula for calculating the proportion is:

[0028] Among them, V’ represents volume, and P represents the pore space of the porous medium. represents a measurement unit with side length L and centered on the lattice vector where N represents the total number of measurement units. represents a specific porosity, and δ is the Kronecker function, and the definition of the Kronecker function is:

[0029] Optionally, the linear path function quantifies the connectivity characteristics of the pore structure of the representative core in terms of probability by randomly selecting scalar line segments on the representative core and calculating the probability that the scalar line segment falls within the region occupied by phase i. The probability calculation formula is:

[0030] where r1 and r2 respectively represent the two endpoints of the scalar line segment, and r x represents any point on the scalar line segment, and v j represents the region occupied by phase j. When the region occupied by phase i is the pore space, the region occupied by phase j is the grain structure; when the region occupied by phase i is the grain structure, the region occupied by phase j is the pore space.

[0031] Optionally, the morphological feature function is characterized by the Minkowski functional to quantify the physical properties of the representative core. The function variables of the Minkowski functional include volume, specific surface area, and mean integral curvature.

[0032] Among them, the volume function is expressed as: φ = V pore / V, which reflects the size of the porosity of the representative core. V pore represents the pore space volume, and V represents the total volume of the rock sample.

[0033] The specific surface area function is expressed as: The value of the volume function characterizes the contact area between the rock and the pores. The larger the value of the volume function, the smaller the value of the specific surface area function.

[0034] The mean integral curvature function is expressed as: The simplified representation of this function is: χ = V″ - E + F - O. The magnitude of the mean integral curvature function value is related to the boundary and shape of the representative core and characterizes the grain morphological features. The grain morphological features include roundness, sphericity, and sorting. Among them, r1’ represents the maximum principal radius of curvature of the pore and skeleton interface, r2’ represents the minimum principal radius of curvature of the pore and skeleton interface, V″ represents the number of vertices, E represents the number of faces, F represents the number of edges, and O represents the number of objects.

[0035] The second aspect of the embodiments of the present invention provides a three-dimensional digital core construction device, which includes:

[0036] A CT three-dimensional core reconstruction module, configured to obtain rock samples of a research section, select typical samples for CT scan analysis to obtain CT data, and reconstruct a CT three-dimensional core based on the CT data;

[0037] A training dataset construction module, configured to select representative cores from the CT three-dimensional cores, extract two-dimensional thin section images from the representative cores, and construct a dataset with the two-dimensional thin section images;

[0038] A model training module, configured to train and optimize the parameters of a pre-constructed three-dimensional core construction model through the dataset, and use the representative core as the true value during training;

[0039] A three-dimensional digital core generation module, configured to obtain two-dimensional thin section images of rocks in a target section, input the two-dimensional thin section images into the three-dimensional core construction model with optimized parameters, and generate a three-dimensional digital core of the target section;

[0040] Wherein, the three-dimensional core construction model is a generative adversarial network model for 3D reconstruction, and the training process of the model is constrained by a statistical feature function characterizing pore properties.

[0041] Optionally, the device further includes:

[0042] A QEMSCAN data acquisition module, configured to perform QEMSCAN scan analysis on typical samples to obtain QEMSCAN data;

[0043] A first correspondence generation module, configured to extract mineral information from the QEMSCAN data, and establish a correspondence between CT measured gray level and mineral components based on the CT data and the mineral information, and the correspondence between CT measured gray level and mineral components is used to constrain the training process of the three-dimensional core construction model.

[0044] The third aspect of the embodiments of the present invention provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a three-dimensional digital core construction method as described in the first aspect of the embodiments of the present invention.

[0045] The fourth aspect of the embodiments of the present invention provides a storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements a three-dimensional digital core construction method as described in the first aspect of the embodiments of the present invention.

[0046] The main beneficial effects of the present invention are:

[0047] 1) Establish a three-dimensional core construction model with a GAN network for 3D reconstruction, such as a DCGAN network. This model generates three-dimensional images from two-dimensional thin section images representing the core, and introduces a statistical feature function characterizing pore properties during the training of the three-dimensional core construction model to constrain the model generation results, thereby ensuring that the finally obtained three-dimensional core construction model has a good generation effect on complex rocks. Compared with traditional two-dimensional image-based reconstruction methods, the generated three-dimensional digital core has higher accuracy, is simpler and more convenient, making the subsequent characterization of reservoir rock properties using this three-dimensional core construction model more accurate. At the same time, when using the constructed three-dimensional core construction model to generate three-dimensional digital cores of the layer to be evaluated, the dependence on CT scanning is avoided, thereby reducing the reconstruction cost;

[0048] 2) Obtain the mineral information of the rock through QEMSCAN scanning analysis, and then establish the corresponding relationship between CT measurement gray scale and mineral components. This corresponding relationship is used for model training constraints, so that the finally obtained three-dimensional core construction model has higher mineral recognition accuracy.

[0049] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific implementation part. Description of the Drawings

[0050] The drawings are used to provide a further understanding of the embodiments of the present invention, and constitute a part of the specification. Together with the following specific implementation, they are used to explain the embodiments of the present invention, but do not constitute a limitation to the embodiments of the present invention. In the drawings:

[0051] Figure 1 It is a schematic diagram of a three-dimensional digital core construction method implemented in Embodiment 1;

[0052] Figure 2 It is a schematic diagram of a three-dimensional digital core construction method implemented in Embodiment 2;

[0053] Figure 3 It is a schematic diagram of a binary CT three-dimensional core obtained after filtering and threshold segmentation;

[0054] Figure 4 It is a schematic diagram of extracting two-dimensional thin section images in three directions of a representative core;

[0055] Figure 5 It is a schematic diagram of the training process of a three-dimensional core construction model;

[0056] Figure 6 It is a schematic diagram of generating a three-dimensional digital core of the target layer segment;

[0057] Figure 7 It is a schematic diagram of the composition of a three-dimensional digital core construction device. Specific Embodiments

[0058] The following details the specific embodiments of the embodiments of the present invention in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining and illustrating the embodiments of the present invention, and are not used to limit the embodiments of the present invention.

[0059] For ease of understanding the following embodiments, the following is an explanation of the technical terms involved in the embodiments:

[0060] DCGAN network: Deep Convolutional GAN, that is, a deep convolutional generative adversarial network;

[0061] DCGAN network for 3D reconstruction: A 3D version of the DCGAN network, with 3D convolutional layers, etc., to achieve the conversion from two dimensions to three dimensions;

[0062] ELU function: Exponential Linear Unit, that is, an exponential linear unit function, and its mathematical expression is:

[0063] Conv layer: Convolutional layer;

[0064] BN layer: Batch Normalization, that is, batch normalization layer;

[0065] dropout layer: Skip connection layer.

[0066] The following Embodiment 1 and Embodiment 2 need to be referred to together Figures 1 to 6 .

[0067] Embodiment 1

[0068] This embodiment provides a method for constructing a three-dimensional digital core for constructing a three-dimensional digital core of a target interval.

[0069] Specifically, as Figure 1 shown, a specific implementation process of the three-dimensional digital core construction method is as follows:

[0070] S100. Obtain rock samples of the research section, select typical samples from the rock samples for CT scan analysis to obtain CT data. The typical samples refer to core samples that can represent the complex lithology of the reservoir.

[0071] S200. Reconstruct the CT three-dimensional core according to the CT data.

[0072] Combined with Figure 3As shown, when reconstructing a CT three-dimensional core from CT data, it can be carried out based on the CT scan reconstruction method in the general embodiment. For example, a specific process is as follows:

[0073] Preprocess the original CT data obtained by CT scanning. The preprocessing includes filtering, gray-scale transformation, equalization, etc., so as to obtain the CT data after image enhancement;

[0074] Perform threshold segmentation and binarization on the CT data after image enhancement to obtain the image information representing the pore space and particle structure of the core;

[0075] Reconstruct a binarized CT three-dimensional core according to this image information.

[0076] S300. Select a representative core from the CT three-dimensional core, extract a two-dimensional thin slice image from the representative core, and construct a data set with this two-dimensional thin slice image.

[0077] The training data collected and constructed in the traditional way often difficult to reasonably represent the target interval. In this embodiment, a representative core is reconstructed by scanning and analyzing typical samples in the research section, and then a two-dimensional thin slice image is obtained by slicing the representative core, which is used as training data to make it able to reasonably represent the target interval and improve the accuracy of the three-dimensional core construction model obtained by training.

[0078] In the general embodiment, a representative core is selected from the CT three-dimensional core through representative elementary volume analysis. By selecting the representative core, the core size that can best represent the overall core information is selected, balancing the memory and the size of the representative elementary volume.

[0079] Exemplarily, in combination with Figure 4 As shown, in one embodiment, the specific implementation process of extracting a two-dimensional thin slice image from a representative core is:

[0080] Carry out multiple thin section analyses of the representative core along the directions parallel and perpendicular to the Z-axis respectively, and extract two-dimensional thin slice images in three directions.

[0081] When training a three-dimensional core construction model with two-dimensional thin slice images, it can be trained first with multiple two-dimensional thin slice images in a single direction. When there are differences in the fineness or pore space between the three-dimensional structure actually generated by the three-dimensional core construction model and the representative core as the true value, then use the two-dimensional thin slice images obtained by slicing in multiple directions for training. For example, multiple slices in three directions perpendicular and parallel to the Z-axis can be arranged with four two-dimensional thin slice images in each direction as a group. It can be seen that using two-dimensional thin slice images in three directions to construct a training data set can make the evaluation work in model construction easier.

[0082] S400. Train and optimize the parameters of the constructed 3D core construction model using a dataset, with the representative core being used as the ground truth during training. Among them, the 3D core construction model is a generative adversarial network model for 3D reconstruction, and during the training of the 3D core construction model, the training process is constrained by a statistical feature function representing pore properties. The statistical feature function representing pore properties can also be referred to as the constraint condition for model training.

[0083] Exemplarily, in one embodiment, the architecture of the generative adversarial network model adopts a DCGAN network. Specifically, the DCGAN network for 3D reconstruction includes a generator G and a discriminator D. The generator G includes an input layer, multiple upsampling layers connected in sequence, and the network architecture behind the multiple upsampling layers consists of a cycle of a BN layer, an activation function, and a 3D Conv layer. The discriminator D includes a 2DConv layer, an activation function, a dropout layer, and a BN layer connected in sequence.

[0084] Exemplarily, in one embodiment, the ELU activation function is used for the non-output layers of both the generator and the discriminator in the DCGAN network. The use of the ELU activation function improves the learning ability and expressive ability of the DCGAN network, making the reconstruction accuracy of the 3D core construction model for complex rocks higher.

[0085] Exemplarily, in one embodiment, the RAdam optimizer is used during the training of the 3D core construction model. The learning rate of the RAdam optimizer can be 0.2×10 5 , thus making the convergence of the model learning process better.

[0086] Combined with Figure 5 As shown, during the training of the 3D core construction model, first train the discriminator to maximize the possibility of the discriminator correctly judging the given input. The training process of the discriminator is mainly as follows: construct real samples from the training set, that is, for each two-dimensional thin slice image generated by the generator G received by the discriminator, take a two-dimensional thin slice image representing the real core from the training set to ensure the balance between data, then forward pass and calculate the loss function, backward pass and calculate the gradient, and then input the two-dimensional thin slice image constructed by the generator into the discriminator. The discriminator calculates the loss function again, backward passes and accumulates the gradient, and repeats the above process to complete the training of the discriminator. Then perform generator training, forward pass and calculate the loss function, backward pass to calculate the gradient, and update the parameters of the generator through the optimizer. After each training process ends, evaluate each image generated by the generator through the constraint conditions to evaluate the generation performance of the 3D core construction model.

[0087] Exemplarily, in one embodiment, the statistical characteristic functions representing pore properties include two-point probability function, local porosity distribution function, linear path function, and morphological characteristic function. For those skilled in the art, in order to constrain the training results of the model for constructing a three-dimensional core, other statistical characteristic functions that can characterize pore properties other than the two-point probability function, local porosity distribution function, linear path function, and morphological characteristic function may also be introduced.

[0088] Among them, the two-point probability function is used to statistically represent the pore distribution characteristics of the core. The two-point probability function quantifies the pore distribution characteristics of the core in terms of probability by randomly selecting two points in the representative core and calculating the probability that the two points are in the same phase. Among them, the probability calculation formula is:

[0089] S2(r) = P(x ∈ R, x + r ∈ R) (Formula 1).

[0090] Among them, x and x + r represent two points separated by the lag distance r, and R represents the R phase.

[0091] The local porosity distribution function is used to statistically represent the geometric characteristics of the porous medium of the core. The local porosity distribution function quantifies the geometric characteristics of the porous medium of the representative core by selecting multiple measurement units in the representative core, calculating the local porosity of each measurement unit, and calculating the proportion of the measurement units with the local porosity equal to a specific porosity among all the measurement units.

[0092] Among them, the local porosity calculation formula is: The calculation formula for the proportion is:

[0093] In Formula 2 and Formula 3, V’ represents volume, P represents the pore space of the porous medium, represents a measurement unit with side length L and centered on the lattice vector , N represents the total number of measurement units, represents a specific porosity, δ is the Kronecker function, and the definition of the Kronecker function is:

[0094]

[0095] The linear path function is used to statistically represent the connectivity characteristics of the pore structure of the core. The linear path function quantifies the connectivity characteristics of the pore structure of the representative core in terms of probability by randomly selecting a scalar line segment on the representative core and calculating the probability that the scalar line segment falls in the region occupied by the i phase. The probability calculation formula is: In Formula 5, r1 and r2 respectively represent the two endpoints of the scalar line segment, r xrepresents any point on the scalar line segment, v j represents the region occupied by the j-phase. When the region occupied by the i-phase is the pore space, the region occupied by the j-phase is the granular structure; when the region occupied by the i-phase is the granular structure, the region occupied by the j-phase is the pore space.

[0096] The morphological characteristic function is used to statistically represent the physical properties of the core. The morphological characteristic function is characterized by the Minkowski functional, thereby quantifying the physical properties of the core. The function variables of the Minkowski functional include volume, specific surface area, and mean integral curvature.

[0097] Among them, the volume function is expressed as: φ = V pore / V (Formula VI), V pore represents the pore space volume, and V represents the total volume of the rock sample. The volume function reflects the size of the porosity of the core.

[0098] The specific surface area function is expressed as: The larger the value of the volume function, the smaller the value of the specific surface area function.

[0099] The mean integral curvature function is expressed as: For the representative core, since its core image is derived from CT scanning, it is equivalent to polyhedron analysis and can be further simplified to the Euler characteristic relationship of the polyhedron. Therefore, the simplified representation of this function is: χ = V″ - E + F - O (Formula IX). The magnitude of the mean integral curvature function value is related to the boundary and shape of the representative core and characterizes the particle morphological characteristics. The particle morphological characteristics include roundness, sphericity, sorting, etc. In Formulas VIII and IX, r1' represents the maximum principal radius of curvature of the pore and skeleton interface, r2' represents the minimum principal radius of curvature of the pore and skeleton interface, V″ represents the number of vertices, E represents the number of faces, F represents the number of edges, and O represents the number of objects.

[0100] S500. Obtain the two-dimensional thin section image of the rock in the target interval, and input this two-dimensional thin section image into the three-dimensional core construction model with optimized parameters to generate the three-dimensional digital core of the target interval.

[0101] In S500, to realize the construction of the three-dimensional digital core of the target interval, only a small number of two-dimensional thin section images need to be input into the three-dimensional core construction model to obtain a high-precision three-dimensional digital core of the target interval.

[0102] Example Two

[0103] The difference between this embodiment and the first embodiment is as follows: to improve the mineral identification accuracy of the three-dimensional digital core constructed in the first embodiment, in addition to performing CT scan analysis on typical samples, QEMSCAN scan analysis is also carried out. QEMSCAN scan analysis can obtain the mineral information of the rock. After obtaining the mineral information, the corresponding relationship between the CT measurement gray level and the mineral composition can be further constructed, and this corresponding relationship is applied to the training process of the three-dimensional core construction model to constrain the generation result of the three-dimensional core construction model.

[0104] Specifically, as Figure 2 shown, the method for constructing a three-dimensional digital core implemented in this embodiment specifically includes:

[0105] SS1. Obtain rock samples in the research section, select typical samples from the rock samples for CT scan analysis to obtain CT data, and perform QEMSCAN scan analysis to obtain QEMSCAN data.

[0106] SS2. Reconstruct the CT three-dimensional core according to the CT data.

[0107] SS3. Extract mineral information from the QEMSCAN data, and establish the corresponding relationship between the CT measurement gray level and the mineral composition based on the CT data and the mineral information. The mineral information includes the mineral distribution and content in the rock sample.

[0108] The CT data is a gray-scale image. The QEMSCAN data can provide a distribution map of mineral distribution, and can simultaneously output a two-dimensional gray-scale image of the corresponding slice. The gray-scale value in this two-dimensional gray-scale image is related to the atomic number of the mineral composition. The larger the atomic number, the larger the gray-scale value. The density of the mineral composition is positively correlated with the atomic number. Based on the above principle, the corresponding relationship between the gray level and the mineral composition is established, and then the corresponding relationship between the CT measurement gray level and the mineral composition is established. Visually, it is the corresponding relationship between the CT measurement gray-level value range and the mineral composition scan color.

[0109] SS4. Select representative cores from the CT three-dimensional core, extract two-dimensional thin-section images from the representative cores, and construct a data set with these two-dimensional thin-section images.

[0110] SS5. Train and optimize the parameters of the already constructed three-dimensional core construction model through the data set. When training, use the representative core as the true value. Among them, the three-dimensional core construction model is a generative adversarial network model for 3D reconstruction. And when training the three-dimensional core construction model, the training process is constrained by the statistical feature function characterizing the pore properties and the corresponding relationship between the CT measurement gray level and the mineral composition. The statistical feature function characterizing the pore properties and the corresponding relationship between the CT measurement gray level and the mineral composition can also be called the constraint conditions for model training.

[0111] SS6. Obtain the two-dimensional thin-section image of the rock in the target interval, and input the two-dimensional thin-section image into the three-dimensional core construction model with optimized parameters to generate the three-dimensional digital core of the target interval.

[0112] Optionally, the architecture of the generative adversarial network model is a DCGAN network.

[0113] Optionally, the generative adversarial network model includes a generator and a discriminator, and the ELU activation function is used for the non-output layers of the generator and the discriminator.

[0114] Optionally, the RAdam optimizer is used during the training of the three-dimensional core construction model.

[0115] Optionally, extract the two-dimensional thin-section image from the representative core, specifically: perform multiple thin-section analyses of the representative core along the directions parallel and perpendicular to the Z-axis respectively, and extract the two-dimensional thin-section images in three directions.

[0116] Optionally, the statistical feature functions representing pore properties include the two-point probability function, the local porosity distribution function, the linear path function, and the morphological feature function. The specific mathematical expressions of the two-point probability function, the local porosity distribution function, the linear path function, and the morphological feature function refer to the corresponding content in Embodiment 1, and this embodiment does not describe this part of the content in detail.

[0117] Device Embodiment

[0118] As Figure 7 shown, this embodiment provides a three-dimensional digital core construction device, which includes a CT three-dimensional core reconstruction module, a training dataset construction module, a model training module, and a three-dimensional digital core generation module connected in sequence.

[0119] The CT three-dimensional core reconstruction module is used to obtain the rock samples in the research interval, select typical samples for CT scan analysis to obtain CT data, and reconstruct the CT three-dimensional core according to the CT data.

[0120] The training dataset construction module is used to select representative cores from the CT three-dimensional core, extract two-dimensional thin-section images from the representative cores, and construct a dataset with the two-dimensional thin-section images.

[0121] The model training module is used to train and optimize the parameters of the constructed three-dimensional core construction model through the dataset, and use the representative core as the ground truth during training. Among them, the three-dimensional core construction model is a generative adversarial network model for 3D reconstruction, and the training process of the model is constrained by the statistical feature functions representing pore properties.

[0122] The three-dimensional digital core generation module is used to obtain two-dimensional thin-section images of the rocks in the target interval, and input the two-dimensional thin-section images into the three-dimensional core construction model with optimized parameters to generate the three-dimensional digital core of the target interval.

[0123] Optionally, the device further includes a QEMSCAN data acquisition module and a first correspondence generation module that are connected to each other, and the first correspondence generation module is also connected to the model training module.

[0124] The QEMSCAN data acquisition module is used to perform QEMSCAN scanning and analysis on typical samples to obtain QEMSCAN data.

[0125] The first correspondence generation module is used to extract mineral information from the QEMSCAN data, and establish the correspondence between the CT measurement gray level and the mineral components according to the CT data and the mineral information. The correspondence between the CT measurement gray level and the mineral components is used to constrain the training process of the three-dimensional core construction model.

[0126] Optionally, the architecture of the generative adversarial network model is a DCGAN network.

[0127] Optionally, the generative adversarial network model includes a generator and a discriminator, and the ELU activation function is used for the non-output layers of the generator and the discriminator.

[0128] Optionally, the RAdam optimizer is used during the training of the three-dimensional core construction model.

[0129] Optionally, the two-dimensional thin-section images are extracted from the representative core. Specifically, multiple thin-section analyses of the representative core are carried out along the directions parallel and perpendicular to the Z-axis respectively, and the two-dimensional thin-section images in three directions are extracted.

[0130] Optionally, the statistical feature functions representing pore properties include two-point probability function, local porosity distribution function, linear path function and morphological feature function. The specific mathematical expressions of the two-point probability function, local porosity distribution function, linear path function and morphological feature function refer to the corresponding content in Embodiment 1, and this embodiment does not describe this part of the content in detail.

[0131] It should be understood that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0132] On the other hand, the present invention also provides an electronic device, including a memory, a processor, and a computer program. The computer program is stored on the memory and can run on the processor. When the processor executes the computer program, it implements a three-dimensional digital core construction method as described in Embodiment 1 or Embodiment 2.

[0133] In another aspect, the present invention also provides a machine-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements a three-dimensional digital core construction method as described in Embodiment 1 or Embodiment 2.

[0134] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a three-dimensional digital core, characterized in that, The method includes: Obtain rock samples of the research interval, and select typical samples for CT scan analysis to obtain CT data; Reconstruct a CT three-dimensional core based on the CT data; Select a representative core from the CT three-dimensional core, extract two-dimensional thin-section images from the representative core, and construct a data set with the two-dimensional thin-section images; Train and optimize the parameters of the constructed three-dimensional core construction model through the data set. When training, use the representative core as the true value; Obtain two-dimensional thin-section images of the rock in the target interval, input the two-dimensional thin-section images into the three-dimensional core construction model with optimized parameters, and generate a three-dimensional digital core of the target interval; Among them, the three-dimensional core construction model is a generative adversarial network model for 3D reconstruction, and the training process of the model is constrained by a statistical feature function representing pore properties.

2. The 3D digital core construction method according to claim 1, characterized in that The method further includes: Perform QEMSCAN scan analysis on the typical samples to obtain QEMSCAN data; Extract mineral information from the QEMSCAN data, and establish the corresponding relationship between CT measurement gray level and mineral components according to the CT data and mineral information; Constrain the training process of the three-dimensional core construction model through the corresponding relationship between CT measurement gray level and mineral components.

3. A method for constructing a three-dimensional digital core according to claim 1, characterized in that, The architecture of the generative adversarial network model is a DCGAN network.

4. A method for constructing a three-dimensional digital core according to claim 3, wherein, The generative adversarial network model includes a generator and a discriminator. The non-output layers of the generator and the discriminator both use the ELU activation function.

5. A method for constructing a three-dimensional digital core according to claim 3, characterized in that The RAdam optimizer is used during the training of the three-dimensional core construction model.

6. The 3D digital core construction method according to claim 1, characterized in that Extract two-dimensional thin-section images from the representative core. Specifically: Conduct multiple thin-section analyses of the representative core along directions parallel and perpendicular to the Z-axis respectively, and extract two-dimensional thin-section images in three directions.

7. A method for constructing a three-dimensional digital core according to claim 1, characterized in that The statistical feature functions representing pore properties include two-point probability function, local porosity distribution function, linear path function, and morphological feature function.

8. A method for constructing a three-dimensional digital core according to claim 7, characterized in that, The two-point probability function quantifies the pore distribution characteristics of the core in terms of probability by randomly selecting two points in the representative core and calculating the probability that the two points are in the same phase. Among them, the probability calculation formula is: S2(r) = P(x ∈ R, x + r ∈ R), where x and x + r represent two points separated by the lag distance r, and R represents the R phase.

9. A method for constructing a three-dimensional digital core according to claim 7, characterized in that The local porosity distribution function quantifies the geometric characteristics of the porous medium of the representative core by selecting multiple measurement units in the representative core, calculating the local porosity of each measurement unit, and calculating the proportion of the measurement units with the local porosity equal to a specific porosity among all measurement units; The local porosity calculation formula is: The calculation formula for the proportion is: where V’ represents volume and P represents the pore space of the porous medium, represents a measurement unit with side length L centered on the lattice vector and N represents the total number of measurement units, represents a specific porosity, δ is the Kronecker function, and the definition of the Kronecker function is as follows:

10. A method for constructing a three-dimensional digital core according to claim 7, characterized in that, The linear path function quantifies the connectivity characteristics of the pore structure of the representative core in terms of probability by randomly selecting a scalar line segment on the representative core and calculating the probability that the scalar line segment falls in the area occupied by the i phase. The probability calculation formula is: Among them, r1 and r2 respectively represent the two endpoints of the scalar line segment, and r x represents any point on the scalar line segment, and v j represents the region occupied by the j-phase. When the region occupied by the i-phase is the pore space, the region occupied by the j-phase is the granular structure; when the region occupied by the i-phase is the granular structure, the region occupied by the j-phase is the pore space.

11. A method for constructing a three-dimensional digital core according to claim 7, characterized in that, The morphological feature function is characterized by the Minkowski functional, thereby quantifying the physical property parameters of the representative core. The function variables of the Minkowski functional include volume, specific surface area, and mean integral curvature; Among them, the volume function is expressed as: and reflects the size representing the core porosity, V pore represents the pore space volume, and V represents the total volume of the rock sample; The specific surface area function is expressed as: The volume function value characterizes the contact area between the rock and the pores. The larger the volume function value, the smaller the specific surface area function value. The average integral curvature function is expressed as: The simplified representation of this function is: χ = V″ - E + F - O. The magnitude of the average integral curvature function value is associated with the boundaries and shape representing the core, and characterizes the particle morphological features, where the particle morphological features include roundness, sphericity, and sorting. Among them, r1’ represents the maximum principal radius of curvature of the pore and skeleton interface, r2’ represents the minimum principal radius of curvature of the pore and skeleton interface, V″ represents the number of vertices, E represents the number of faces, F represents the number of edges, and O represents the number of objects.

12. A three-dimensional digital core construction device, characterized in that The device includes: A CT three-dimensional core reconstruction module, which is used to obtain rock samples of the research section, select typical samples for CT scan analysis to obtain CT data, and reconstruct a CT three-dimensional core based on the CT data; A training dataset construction module, which is used to select representative cores from the CT three-dimensional cores, extract two-dimensional thin section images from the representative cores, and construct a dataset with the two-dimensional thin section images; A model training module, which is used to train and optimize the parameters of the constructed three-dimensional core construction model through the dataset. When training, the representative core is used as the true value; A three-dimensional digital core generation module, which is used to obtain two-dimensional thin section images of the rocks in the target section, input the two-dimensional thin section images into the three-dimensional core construction model with optimized parameters, and generate a three-dimensional digital core of the target section; Among them, the three-dimensional core construction model is a generative adversarial network model for 3D reconstruction, and the training process of the model is constrained by a statistical feature function characterizing the pore properties.

13. A three-dimensional digital core construction device according to claim 12, characterized in that, The device further includes: A QEMSCAN data acquisition module, which is used to perform QEMSCAN scan analysis on typical samples to obtain QEMSCAN data; A first correspondence generation module, which is used to extract mineral information from the QEMSCAN data, and establish a correspondence between the CT measured gray level and the mineral composition based on the CT data and the mineral information. The correspondence between the CT measured gray level and the mineral composition is used to constrain the training process of the three-dimensional core construction model.

14. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a three-dimensional digital core construction method according to any one of claims 1 to 11.

15. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a three-dimensional digital core construction method according to any one of claims 1 to 11.

Images