A three-dimensional multi-mineral digital core modeling method

Abstract

This invention belongs to the field of oil and gas reservoir rock data identification technology, and particularly relates to a three-dimensional multi-mineral digital core modeling method. This application proposes a 2D-3DGAN algorithm, which can generate a three-dimensional data volume using only a single two-dimensional image. By combining the advantages of the 2D-3DGAN and AMICS methods, a "AMICS+2D-3DGAN" modeling method is proposed to achieve the establishment of multi-mineral, high-precision, and efficient three-dimensional multi-mineral digital cores. The former is used to obtain representative two-dimensional mineral scanning images, while the latter uses these two-dimensional scanning images as training images to generate three-dimensional multi-mineral digital images. The established model not only includes multiple minerals but also considers accuracy and economy, and can represent reservoir rocks, providing a solid foundation for subsequent diagenetic simulation based on multi-mineral digital cores, as well as research on reservoir electrical, acoustic, and mechanical properties.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas reservoir rock data identification technology, and in particular relates to a three-dimensional multi-mineral digital core modeling method. Background Technology

[0002] Digital core technology has long been an important method for studying rock properties. Traditional rock physics experiments struggle to quantitatively measure and control pore structure and fluid distribution, and for complex oil and gas reservoirs, core sampling success rates are low and displacement is difficult. Therefore, relying solely on rock physics experiments is insufficient to accurately study the influence of various microscopic factors on macroscopic physical properties. Digital core technology, on the other hand, offers advantages such as low cost, short processing time, high accuracy, repeatability, and the ability to quantitatively examine the influence of various factors on core physical parameters at the microscopic scale. Consequently, digital core technology has been widely applied in rock property research. The accuracy and applicability of numerical simulations using digital cores largely depend on the established microscopic rock model. Only when the model's pore structure accurately reflects the pore structure of the actual rock sample will the simulation results have theoretical and practical value. Furthermore, current digital core models are mostly based on single minerals, which cannot be used to study the electrical, acoustic, and mechanical properties of reservoirs. Therefore, it is essential to conduct multi-mineral digital core studies that accurately reflect real geological conditions.

[0003] Digital core modeling methods mainly fall into two categories: physical experimental methods and numerical reconstruction methods. Physical experimental methods utilize high-precision experimental equipment such as scanning electron microscopes, focused ion beam microscopes, high-magnification optical microscopes, or X-ray CT scanners to acquire two-dimensional images of different locations within the core, and then perform three-dimensional reconstruction from these images to obtain a three-dimensional digital core. This method can effectively characterize pore channel features, but the scanning accuracy is affected by the scanning resolution. X-ray nano-CT cannot accurately identify the connectivity and porosity of unconventional reservoirs, while nano-CT can reflect rock porosity information at the nanoscale, but the sample size is very small, sample preparation is difficult, and scanning costs are high. Because common minerals in reservoir rocks (such as quartz, potassium feldspar, and sodium feldspar) have similar densities, it is difficult to accurately determine the grayscale range of their lattice in CT images, making accurate mineral classification challenging. Numerical reconstruction algorithms, based on a small number of two-dimensional images of thin core sections, construct three-dimensional digital cores under constraints such as morphology and grain size using various algorithms. While this method may not be as accurate as physical experimental methods, it is less expensive, faster, and some algorithms can create rocks with multi-scale porosity or multiple minerals, giving it a greater advantage. This method includes many modeling algorithms, such as simulated annealing, Markov chain Monte Carlo methods, sequential indicator simulation, multi-point statistical methods, machine learning methods, and process modeling methods.

[0004] Xiao (2022) optimized the simulated annealing method and established multi-mineral digital cores. However, in multi-mineral modeling, this method has a large computational cost and is highly sensitive to initial conditions and parameter selection. Furthermore, the simulated annealing method is not designed for multi-mineral modeling and has limited effectiveness in handling complex mineral distributions. Nie et al. (2016) used the Markov chain method (MCMC) to create digital cores for each component to create multi-component rocks. They then combined the individual component models into a final model containing all corresponding components through a specific workflow. However, this method failed to fully consider the correlation between geology and minerals. Liu et al. (2023) used multi-point statistics for multi-mineral core modeling. By capturing the spatial correlation of multi-point statistics, they were able to reconstruct complex mineral distributions relatively accurately. However, the computational cost of multi-point statistics is extremely large, especially in multi-mineral modeling. Due to limitations in sample size and training time, the generated digital cores are often affected by data quality and selection. Furthermore, multi-point statistical methods still require a certain degree of idealization. Sun (2023) combined CT scans with the Res-UNet network to create multi-mineral 3D digital cores, but this method requires CT scans, is costly and time-consuming, and cannot quickly create multi-mineral digital cores in batches. Chi et al. (2024) used the GAN algorithm to generate large-scale anisotropic 3D digital cores from 2D images. However, the above methods do not distinguish between mineral types and are only applicable to single mineral types. Liu et al. (2009) used the sequential indicator simulation method to reconstruct 3D digital cores from 2D images of the cores, but the reconstructed digital cores had a small average pore radius, poor pore connectivity, and did not distinguish between mineral types. Pang et al. (2017) proposed a 3D reconstruction method for digital shale cores based on nanoscale volumetric data and multi-point statistics, but due to the strong heterogeneity of shale, the permeability of the reconstructed cores at different locations may differ by a factor of 100.

[0005] The problems and shortcomings of the existing technology are as follows:

[0006] (1) Existing physical experimental methods for creating digital cores require high costs and long time. The experimental results are grayscale images, which make it difficult to accurately distinguish different minerals. Due to the complexity of the images, they cannot effectively present the spatial distribution of multiple minerals, and the processing is cumbersome, which limits their efficiency and accuracy in practical applications.

[0007] (2) Existing numerical simulation methods often adopt a single mineral assumption when constructing three-dimensional digital cores that conform to actual geological characteristics. Such methods cannot simultaneously take into account the authenticity of the core structure and the diversity of mineral composition, resulting in a single mineral composition in the constructed three-dimensional digital cores, which cannot simulate reservoir electrical, acoustic and other properties.

[0008] (3) Existing numerical simulation techniques for constructing multi-mineral three-dimensional digital cores are mostly based on idealized model assumptions. The mineral distribution morphology and spatial combination are simplified, which differs greatly from the heterogeneity and complex structure of minerals in the real geological environment, resulting in limited geological applicability and engineering reference value of the models. Summary of the Invention

[0009] To overcome the problems existing in the prior art, the present invention provides a three-dimensional multi-mineral digital core modeling method.

[0010] To achieve the above objectives, the present invention includes the following steps:

[0011] Step 1: Obtain the training images required for the 2D-3DGAN algorithm through AMICS experiment: Obtain the initial two-dimensional mineral scan image through AMICS. Under the premise of ensuring the representativeness of mineral types and the integrity of structure, select the region with typical mineral combination characteristics and representative texture structure from the initial two-dimensional mineral scan image as the training image TI1 of the 2D-3DGAN algorithm.

[0012] Step 2, training image processing: Convert the selected training image from RGB type to grayscale type, encode the mineral phases in sequence in the grayscale image, and scale the image without loss of quality to obtain the final training image TI2 (Training image 2).

[0013] Step 3, 2D-3DGAN algorithm to obtain three-dimensional multi-mineral digital cores: Input the training image TI2 into the 2D-3DGAN algorithm to generate multiple three-dimensional multi-mineral digital cores.

[0014] Step 4, generate digital core accuracy verification: the most accurate three-dimensional multi-mineral digital core S3D1 (Sample 3D 1) is obtained by analyzing the diagenetic characteristics of two-dimensional slices, comparing mineral content, comparing pore diameter, and comparing two-point correlation functions.

[0015] Step 5, Post-processing: Re-encode the mineral phases of S3D1 and color-code the minerals to obtain the final three-dimensional multi-mineral digital core S3D2 (Sample 3D 2).

[0016] In step 1, an initial two-dimensional mineral scan image is obtained using AMICS, with N1*N2 pixels. A representative region is selected from this image using the REV method as a training image. If the porosity and mineral content of this region differ from those of the original mineral scan image by less than 10%, then this region is identified as a REV image with N32 pixels.

[0017] In step 2, the selected training image is converted from RGB to grayscale in MATLAB. Then, the resulting grayscale image is encoded in MATLAB according to the mineral phase sequence starting from 0. Finally, the image is scaled in MATLAB without distortion to obtain the final training image TI2 with N42 pixels.

[0018] In step 3, the training image TI2 is input into the 2D-3DGAN algorithm to generate multiple three-dimensional multi-mineral digital cores. The steps include generating three-dimensional volume and extracting slices, comparing the judge and optimizing the generator, preserving multi-mineral features and outputting high-fidelity data.

[0019] ① Generating 3D Volume and Slice Extraction: The generator G maps the input latent vector to a 3D cube volume. To achieve effective comparison of 2D training images, 2D slices are extracted along the X, Y, and Z directions of the generated volume. This operation ensures the structural consistency of the generated volume in the three directions, making the information distribution between different slices uniform and avoiding image artifacts or non-uniformity caused by insufficient edge information.

[0020] ② Discriminator comparison and generator optimization: The extracted 2D slices are input into the discriminator (D) and compared with random cropping of real 2D training images to update the discriminator parameters. The feedback of the discriminator is passed to the generator through backpropagation to guide the generator to continuously optimize the 3D output.

[0021] ③ Preservation and Fidelity of Multi-Mineral Features: The trained generator can quickly generate high-resolution 3D digital cores, with each voxel precisely corresponding to the mineral type and pore size. The generated cores not only preserve the shape and combination patterns of the minerals but also maintain the connectivity and distribution patterns of the pore spaces, thus accurately reflecting the microstructural characteristics of the reservoir.

[0022] In step 4, the mineral distribution of each 3D volume on the planar slice is observed and judged in conjunction with typical diagenetic characteristics to ensure that the core not only has the correct mineral composition, but also that the pore structure, grain arrangement, and diagenetic processes are consistent with the real reservoir. The volume fraction of each mineral in the generated 3D image is statistically analyzed and compared with the percentage of each mineral in the training image to ensure consistency in mineral composition of the generated volume. Furthermore, the pore diameter in the training image is compared with the pore diameter of the generated 3D digital core to verify the consistency of the generated results in terms of pore structure. The two-point correlation function between the selected 3D digital core and the 2D training image is calculated separately, and their spatial statistical correlation is compared. Through quantified spatial statistical indicators, the fidelity of the generated 3D digital core in terms of microstructure is further verified.

[0023] In step 5, due to the disordered mineral numbering of the three-dimensional volume obtained using the 2D-3DGAN algorithm, the three-dimensional voxel data are renumbered one by one according to the grayscale labeling rules set in the preprocessing stage, thereby restoring the correct mineral category mapping relationship. This step ensures the consistency of mineral categories from the input image to the final digital core, and also provides a reliable data foundation for subsequent pore structure analysis, diagenetic simulation, and physical property prediction.

[0024] Based on the RGB color values ​​of each mineral in the mine scan image, the minerals in the generated result are assigned the corresponding colors in the mine scan image, thereby ensuring the consistency of visual representation of different images. This not only facilitates intuitive identification of mineral types but also makes it easier to conduct subsequent visualization analysis and model display.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] (1) The advantages of the present invention are that it overcomes the disadvantages of traditional physical experimental methods, such as high cost, poor economy, inability to identify multiple minerals, and difficulty in balancing multiple minerals and accuracy in current numerical simulation methods. It integrates the advantages of AMICS and 2D-3DGAN algorithms. The model not only includes multiple minerals but also takes into account accuracy and economy. It can represent reservoir rocks and provides a solid foundation for subsequent simulation of diagenesis based on multi-mineral digital cores, as well as the study of reservoir electrical, acoustic and mechanical properties.

[0027] (2) This invention uses AMICS experiments to obtain mineral scanning images of the sample surface, extracts typical areas from these images, converts them into grayscale images, assigns mineral numbers, and scales the images to obtain the two-dimensional training images required for the final 2D-3DGAN algorithm. This step not only ensures the authenticity and effectiveness of the training images but also minimizes the computational load of the 2D-3DGAN algorithm. This invention integrates the advantages of AMICS and the 2D-3DGAN algorithm, using the former to obtain two-dimensional training images and then using the latter to construct three-dimensional multi-mineral digital cores.

[0028] (3) Compared with ideal models and single-mineral digital core models, the accuracy and multi-mineral known cores can more accurately predict the porosity, permeability, resistivity, elastic parameters of rocks, as well as the electrical, acoustic and mechanical properties of reservoirs, and lay a solid foundation for subsequent numerical simulation of diagenesis.

[0029] (4) The technical solution of the present invention makes up for the shortcomings of existing modeling techniques: traditional physical experimental methods are costly, time-consuming and unable to distinguish mineral types. Existing numerical simulation methods cannot simultaneously take into account accuracy and multiple minerals, while this solution combines the accuracy of AMICS experiments as output data with the advantages of the economy and efficiency of the 2D-3DGAN algorithm. It uses AMICS to obtain two-dimensional training images and uses the 2D-3DGAN algorithm to generate three-dimensional data. Therefore, the present invention can construct a three-dimensional digital core model that simultaneously takes into account economy, accuracy and multiple minerals.

[0030] (5) The technical solution of the present invention can generate multiple three-dimensional multi-mineral digital core models at the same time, which greatly saves time and economic costs. Attached Figure Description

[0031] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0032] Figure 1 This is a flowchart of the process for constructing a three-dimensional multi-mineral digital core model using the "AMICS+2D-3DGAN" method provided in this embodiment of the invention.

[0033] Figure 2a This is a flowchart of the 2D-3DGAN algorithm provided in the embodiments of the present invention;

[0034] Figure 2b This is a generator structure diagram of the 2D-3DGAN algorithm provided in the embodiments of the present invention;

[0035] Figure 2c This is a diagram of the discriminator structure of the 2D-3DGAN algorithm provided in this embodiment of the invention;

[0036] Figure 3 This invention provides a flowchart of the "AMICS+2D-3DGAN" method for constructing a three-dimensional multi-mineral digital core. AMICS is used to acquire high-resolution, high-fidelity two-dimensional training images, and the 2D-3DGAN algorithm is used to generate a three-dimensional multi-mineral digital core.

[0037] Figure 4a The two-dimensional training image of sample S1 is obtained by filtering, cropping, and scaling a typical region from a two-dimensional high-precision mining scan image obtained through AMICS experiment provided in the embodiments of the present invention.

[0038] Figure 4b This is a two-dimensional slice of the three-dimensional multi-mineral digital core of sample S1 generated by the 2D-3DGAN algorithm, provided in this embodiment of the invention.

[0039] Figure 4cThis is a three-dimensional view of the S1 sample three-dimensional multi-mineral digital core generated by the 2D-3DGAN algorithm, provided in an embodiment of the present invention.

[0040] Figure 4d This invention provides a comparison of the correlation functions between two points of the three-dimensional multi-mineral digital core of sample S1 generated by the 2D-3DGAN algorithm and the two-dimensional training image.

[0041] Figure 5a The two-dimensional training image of sample S1 is obtained by filtering, cropping, and scaling a typical region from a two-dimensional high-precision mining scan image obtained through AMICS experiment provided in the embodiments of the present invention.

[0042] Figure 5b This is a two-dimensional slice of the three-dimensional multi-mineral digital core of sample S2 generated by the 2D-3DGAN algorithm, provided in this embodiment of the invention.

[0043] Figure 5c This is a three-dimensional view of the S2 sample three-dimensional multi-mineral digital core generated by the 2D-3DGAN algorithm, provided in an embodiment of the present invention.

[0044] Figure 5d This invention provides a comparison of the correlation functions between two points of the three-dimensional multi-mineral digital core of the S2 sample generated by the 2D-3DGAN algorithm and the two-dimensional training image. Detailed Implementation

[0045] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0046] like Figure 1 As shown, this invention provides a three-dimensional multi-mineral digital core modeling method. The specific implementation steps of this method will be described in detail below in logical order:

[0047] Step 1: Obtaining 2D-3DGAN Algorithm Training Images via AMICS Experiment: Samples are scanned using AMICS to obtain high-resolution initial two-dimensional mineral distribution images. These images accurately reflect the spatial distribution characteristics, grain morphology, and contact relationships between different minerals. Considering the large amount of data and redundant information in the initial scan images, directly using them for 2D-3DGAN algorithm training would significantly increase computational cost and training time. To reduce the training load, a typical region is selected from the obtained initial two-dimensional mineral scan images as training images TI1. This method effectively reduces the number of training samples while fully preserving the key features of mineral distribution, laying the foundation for subsequent model training and improved recognition accuracy.

[0048] Step 2: Training Image Processing: The TI1 image selected in Step S1 is preprocessed. First, the original RGB training image is converted to a grayscale image to facilitate mineral phase numbering. Then, based on the category information of different mineral phases, the mineral phases in the grayscale image are sequentially encoded, allowing each mineral phase to be distinguished by a different numerical label in the image, ensuring the uniqueness and identifiability of the mineral phase information. On this basis, without introducing distortion or destroying the boundaries and spatial structure features of the mineral phases, the encoded image is scaled to meet the input requirements of the 2D-3DGAN algorithm model, ultimately obtaining the standardized training image TI2, providing a unified and stable data input for model training.

[0049] Step 3: Obtaining 3D Multi-Mineral Digital Cores using the 2D-3D GAN Algorithm: Using the standardized training image TI2 obtained in Step S2 as input data, the 2D-3D GAN algorithm is introduced for model training and generation. Figure 2a To generate a three-dimensional multi-mineral digital core diagram using 2D-3DGAN, Figure 2b This is a diagram of the generator structure for the 2D-3DGAN algorithm. Figure 2c This is a diagram of the discriminator structure of the 2D-3DGAN algorithm. 2D-3DGAN learns the spatial distribution and morphological features of mineral phases in two-dimensional training images, as well as the relationships between different mineral phases, to achieve the mapping and reconstruction of two-dimensional slices into three-dimensional structures within a generative network framework. After sufficient training, the trained 2D-3DGAN model generates multiple three-dimensional multi-mineral digital core models with statistical consistency and spatial coherence.

[0050] Step 4: Accuracy Verification of Digital Core Generation: For the multiple 3D multi-mineral digital core models generated by the 2D-3DGAN algorithm in Step S3, a systematic accuracy and reliability verification was conducted. First, the generated 3D digital cores were processed into 2D slices, and the slice results were analyzed from the perspective of diagenetic characteristics, focusing on comparing whether features such as mineral grain morphology, arrangement, and mineral phase contact relationships remained consistent with the original training images. Second, a quantitative comparative analysis of mineral content was performed between the generated digital cores and experimentally obtained real samples to assess the consistency of the volume fraction of each mineral phase. Furthermore, the ability of the generated digital cores to represent pore structure scale characteristics was verified by comparing pore diameter distribution. Simultaneously, a two-point correlation function was introduced to conduct statistical comparative analysis of the generated results to examine their similarity to real rocks in terms of spatial correlation and structural continuity. Based on the above evaluation indicators, the generated 3D multi-mineral digital cores were screened, and the 3D multi-mineral digital core S3D1 with the highest accuracy was finally selected.

[0051] Step 5: Post-processing: Based on mineral types and their numbering rules, each mineral phase in the digital core is re-encoded to ensure the uniqueness and consistency of the identification of different mineral phases in three-dimensional space. Subsequently, the re-encoded mineral phases are color-coded, assigning corresponding standard colors to different mineral types, thereby achieving an intuitive visualization of the three-dimensional multi-mineral digital core. Through the above processing, a three-dimensional multi-mineral digital core model S3D2 with clear mineral phase identification, distinct colors, and complete structural information is finally obtained.

[0052] In the experiment of this invention, taking two mudstone samples S1 and S2 as examples, the accuracy of the "AMICS+2D-3DGAN" modeling method is verified by comparing the diagenetic characteristics, mineral content, pore diameter and two-point correlation function of the two-dimensional slice of the model built by the "AMICS+2D-3DGAN" method with the two-dimensional training image.

[0053] Figure 4a This is a two-dimensional training image of sample S1 provided in this embodiment of the invention, with a resolution of 0.057 μm / voxel and a size of 1000×1000 pixels. The percentages of porosity, area of ​​albite, potassium feldspar, quartz, carbonate minerals, chlorite, and illite are 4.4%, 32.9%, 8.1%, 27.5%, 3.8%, 16.2%, and 7.1%, respectively. This region contains albite grain-edge dissolution pores and numerous intragranular dissolution pores, a small amount of potassium feldspar intragranular dissolution pores and grain-edge dissolution pores, as well as intragranular dissolution pores and carbonate cement formed by the dissolution of carbonate minerals. The percentages of pore diameters of 0-1 μm, 1-2 μm, 2-3 μm, and >3 μm are 76.5%, 14.4%, 6.6%, and 2.5%, respectively, with an average pore diameter of 0.758 μm. Figure 4bObservations show that the generated multi-mineral digital cores accurately reproduce the diagenetic characteristics of real reservoirs in terms of microstructure. In the two-dimensional slices of the generated samples, albite commonly exhibits numerous perigranular and intragranular dissolution pores, with mostly irregular morphologies, reflecting typical albite dissolution characteristics. Potassium feldspar minerals in the slices show some perigranular dissolution pores and a small number of internal dissolution pores, indicating relatively weak dissolution, consistent with the dissolution phenomena of potassium feldspar in actual reservoirs. Furthermore, carbonate minerals clearly show the distribution of internal dissolution pores and cement in the slices, demonstrating not only the model's ability to reproduce the multi-stage diagenetic process involving mineral dissolution and cementation, but also indicating that the generated multi-mineral digital cores possess high structural realism and rationality. Figure 4c The sample S1 provided in this embodiment of the invention uses "AMICS+2D-3DGAN" to construct a digital core 3D view. The porosity, volume percentage of albite, potassium feldspar, quartz, carbonate minerals, chlorite, and illite in S1_3D_1, S1_3D_2, S1_3D_3, and S1_3D_4 are 2.6%, 35.8%, 12.6%, 22.2%, 4.2%, 15.5%, 7.1%; 4.6%, 29.7%, 10.3%, 30.1%, 2.9%, 14.3%, 8.1%; 4.9%, 34%, 9.8%, 23.3%, 4.5%, 15.1%, 8.4%; 5%, 30.1%, 10%, 24%, 4%, 17.9%, and 9%. Table 1 shows that the average relative errors are 0.22, 0.1255, 0.1198, and 0.1225, respectively. Except for S1_3D_1, the average relative errors of the other three generated cores are all around 0.1, indicating that the three-dimensional multi-mineral digital cores generated by 2D-3DGAN are relatively close to the two-dimensional images in terms of mineral content, and the differences between the mineral components are small, indicating that the generated results have good compositional consistency and reliability. The proportions of pore diameters of 0-1μm, 1-2μm, 2-3μm, and >3μm in S1_3D_1, S1_3D_2, S1_3D_3, and S1_3D_4 are 69%, 24%, 5%, and 2%; 68%, 25.2%, 3.6%, and 3.2%; 81%, 13.8%, 3.1%, and 2.1%; and 92%, 6%, 1%, and 1%, respectively. Table 2 shows that the average pore diameters are 0.89μm, 0.94μm, 0.74μm, and 0.53μm, respectively. The average relative errors were 0.1775, 0.2337, 0.0223, and 0.3065, respectively. The average relative error of the average pore diameter of S1_3D_3 was larger than that of the other three.

[0054] Figure 4dThe results of the two-point correlation function comparison for sample S1 are shown. The S2 value (two-point correlation function) curves of the two-dimensional and three-dimensional samples of sample S1 have a high degree of fit, indicating that the three-dimensional reconstructed digital core inherits the spatial correlation of the two-dimensional training image well in terms of the overall pore structure characteristics. Among them, the S1_3D_1 multi-mineral digital core almost overlaps with the two-dimensional training image, indicating that this sample has the highest similarity to the training image in terms of microstructure and the best reconstruction accuracy.

[0055] Figure 5a This is a two-dimensional training image of sample S1 provided in this embodiment of the invention, with a resolution of 0.057 μm / voxel and a size of 2000×2000 pixels. The percentages of porosity, area of ​​albite, potassium feldspar, quartz, carbonate minerals, chlorite, and illite are 3.9%, 16%, 3.2%, 41.6%, 4.7%, 13.3%, and 17.3%, respectively. This region contains albite edge dissolution pores and numerous intragranular dissolution pores, a small amount of potassium feldspar intragranular dissolution pores and edge dissolution pores, as well as intragranular dissolution pores and carbonate cement formed by the dissolution of carbonate minerals. The percentages of pore diameters of 0-1 μm, 1-2 μm, 2-3 μm, 3-4 μm, 4-5 μm, and >5 μm are 38.51%, 16.34%, 15.34%, 11.49%, 7.6%, and 10.72%, respectively, with an average pore diameter of 2.29 μm. Figure 5b The sample S2 provided in this embodiment of the invention uses four two-dimensional slices of a three-dimensional multi-mineral digital core constructed using "AMICS+2D-3DGAN". Observation results show that the generated multi-mineral digital core reproduces the diagenetic characteristics of the actual reservoir in terms of microstructure. In the two-dimensional slices of the generated sample, albite generally exhibits numerous perigranular and intragranular dissolution pores, with irregular pore morphologies, reflecting typical albite dissolution characteristics. Potassium feldspar minerals in the slices show some perigranular dissolution pores and a small number of internal dissolution pores, indicating relatively weak dissolution, consistent with the dissolution phenomenon of potassium feldspar in actual reservoirs. Furthermore, carbonate minerals clearly show the distribution of internal dissolution pores and cement in the slices, demonstrating not only the model's ability to reproduce the multi-stage diagenetic process of mineral dissolution and cementation, but also indicating that the generated multi-mineral digital core has high structural realism and rationality. Figure 5cThe sample S2 provided in this embodiment of the invention uses "AMICS+2D-3DGAN" to construct a digital core 3D view. The porosity, volume percentage of albite, potassium feldspar, quartz, carbonate minerals, chlorite, and illite in S2_3D_1, S2_3D_2, S2_3D_3, and S2_3D_4 are 3.4%, 9%, 3.9%, 47%, 4.2%, 15.1%, 17.4%; 4%, 12.9%, 2%, 39%, 2.9%, 19.1%, 20.1%; 4.3%, 15.9%, 3.6%, 37%, 5%, 16.5%, 17.7%; and 3.5%, 14.3%, 3.9%, 39.4%, 4%, 16.8%, 18.1%. The average relative errors of mineral content were 0.1681, 0.211, 0.091, and 0.178, respectively. Except for S1_3D_2, the average relative errors of the other three generated bodies were all within 0.2, indicating that the three-dimensional multi-mineral digital cores generated by 2D-3DGAN are close to the two-dimensional images in terms of mineral content, and the differences between the mineral components are small. The generated results have good compositional consistency and reliability. The percentages of pore diameters (0-1μm, 1-2μm, 2-3μm, 3-4μm, 4-5μm, >5μm) for S1_3D_1, S1_3D_2, S1_3D_3, and S1_3D_4 are 33.95%, 25.33%, 17.9%, 8%, 3.8%, and 11.01%; 32.75%, 26.533%, 14.83%, 10.31%, 6.66%, and 8.88%; 41.93%, 22.93%, 12.61%, 8.668%, 5.14%, and 8.81%; and 39.35%, 23.42%, 14.79%, 8.45%, 6.074%, and 7.835%, respectively. The average pore diameters are 2.4μm, 2.32μm, 2.12μm, and 2.14μm, respectively. As shown in Table 3, the average relative errors are 0.456, 0.109, 0.0729, and 0.0658, respectively. The average relative error of the average pore diameter of S1_3D_1 is larger than that of the other three, while the average relative errors of the other three are smaller. Figure 5d The comparison results of the two-point correlation function for sample S2 are presented. It can be seen that the S2 value of the multi-mineral digital core generated from sample S2 is slightly higher than that of the two-dimensional training image in the short distance range (r < 20 voxels), and matches well in the medium to long distance range (r > 20 voxels), showing good overall consistency. Among them, the generated S2_3D_1 multi-mineral digital core is closest to the two-dimensional training image, indicating that its three-dimensional reconstruction structure has the highest degree of agreement with the original two-dimensional image in terms of spatial correlation.

[0056] Table 1. Average relative error of mineral content in generated samples

[0057] sample 3D-1 3D-2 3D-3 3D-4 Sample S1 0.22 0.1255 0.1198 0.1225 Sample S2 0.1681 0.211 0.091 0.178

[0058] Table 2. Average pore diameter (μm) of training images and generated samples

[0059] sample 2D 3D-1 3D-2 3D-3 3D-4 Sample S1 0.758 0.8925 0.9352 0.7411 0.5257 Sample S2 2.2907 2.3951 2.3157 2.1238 2.140

[0060] Table 3. Relative error of average pore diameter in training images

[0061] sample 3D-1 3D-2 3D-3 3D-4 Sample S1 0.1775 0.2337 0.0223 0.3065 Sample S2 0.456 0.109 0.0729 0.0658

[0062] This invention utilizes AMICS to obtain high-resolution, high-fidelity two-dimensional training images and employs a 2D-3DGAN algorithm to generate three-dimensional multi-mineral digital cores. Unlike conventional GAN ​​methods, 2D-3DGAN uses a 3D generator network to generate a complete voxel volume, which is then sliced ​​and checked by a 2D discriminator. Feedback from the discriminator is passed to the generator via backpropagation, guiding the generator to continuously optimize the 3D output. It is this "3D generation + 2D discrimination" mechanism that enables 2D-3DGAN to generate high-fidelity 3D data based on a single 2D training image. This invention proposes a method of "2D mine scan image + 2D-3DGAN" for constructing multi-mineral 3D digital cores. This method requires only one mine scan image as input to efficiently generate high-fidelity 3D digital cores. After training and generating multiple 3D multi-mineral digital cores, their accuracy was verified through the following methods: (1) analyzing the diagenetic characteristics of the generated 3D multi-mineral digital cores; (2) comparing the mineral content and pore diameter distribution of the generated results with those of the training images; and (3) calculating and comparing the two-point correlation function between the 3D multi-mineral digital cores and the training images to evaluate their pore space correlation. The "2D mineral scanning image + 2D-3DGAN" method proposed in this study significantly improves the realism, accuracy, economy, and efficiency of generating 3D multi-mineral digital cores. This method not only provides an efficient path for the construction of multi-mineral digital cores, but also provides a solid foundation for subsequent diagenetic simulation based on multi-mineral digital cores, as well as for the study of reservoir electrical, acoustic, and mechanical properties.

[0063] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the protection scope of the claims and specification of the present invention.

Claims

1. A three-dimensional multi-mineral digital core modeling method, characterized in that, Includes the following steps: Step 1: Obtain the training images required for the 2D-3DGAN algorithm through AMICS experiment: Obtain the initial two-dimensional mineral scan image through AMICS. Under the premise of ensuring the representativeness of mineral types and the integrity of structure, select the region with typical mineral combination characteristics and representative texture structure from the initial two-dimensional mineral scan image as the training image TI1 of the 2D-3DGAN algorithm. Step 2, Training Image Processing: Convert the selected training image from RGB to grayscale, encode the mineral phases in sequence in the grayscale image, and scale the image without loss of quality to obtain the final training image TI2; Step 3: Use the 2D-3DGAN algorithm to obtain three-dimensional multi-mineral digital cores. Input the training image TI2 into the 2D-3DGAN algorithm to generate multiple three-dimensional multi-mineral digital cores. The specific steps are as follows: ① Generate three-dimensional volume and extract slices. The generator G maps the input latent vector to a three-dimensional cube volume. To achieve effective comparison of the two-dimensional training image, two-dimensional slices are extracted along the X, Y, and Z directions of the generated volume. ② Discriminator comparison and generator optimization. The extracted two-dimensional slices are input into the discriminator D and compared with random cropping of the real two-dimensional training image to update the discriminator parameters. The feedback of the discriminator is passed to the generator through backpropagation to guide the generator to continuously optimize the three-dimensional output. ③ Multi-mineral feature preservation and fidelity output. The trained generator quickly generates three-dimensional digital cores, with each voxel accurately corresponding to the mineral type and porosity. Step 4: Verify the accuracy of the digital core: The most accurate three-dimensional multi-mineral digital core S3D1 is obtained by analyzing the diagenetic characteristics of two-dimensional slices, comparing mineral content, comparing pore diameter, and comparing two-point correlation functions. Step 5, Post-processing: Re-encode the mineral phases of S3D1 and color-code the minerals to obtain the final three-dimensional multi-mineral digital core S3D2.

2. The three-dimensional multi-mineral digital core modeling method according to claim 1, characterized in that: In step 1, the initial two-dimensional mineral scanning image has N1*N2 pixels. The REV method is used to select a representative region from the initial two-dimensional mineral scanning image as a training image. If the porosity and mineral content of the region differ from those of the mineral scanning image by less than 10%, the region is identified as REV, and the number of pixels is N32.

3. The three-dimensional multi-mineral digital core modeling method according to claim 1, characterized in that: For step 2, the sequential encoding is performed in order starting from 0, and the image is scaled to obtain the final training image TI2 with N42 pixels.

4. The three-dimensional multi-mineral digital core modeling method according to claim 1, characterized in that, Step 4 involves performing two-dimensional slicing on the generated three-dimensional digital core, analyzing the slicing results from the perspective of diagenetic characteristics, and comparing whether the mineral grain morphology, arrangement, and mineral phase contact relationship characteristics are consistent with the original training image; quantitatively comparing the mineral content of the generated digital core with that of the experimentally obtained real sample to assess the consistency of the volume fraction of each mineral phase; verifying the ability of the generated digital core to represent the pore structure scale characteristics by comparing the pore diameter distribution; and introducing a two-point correlation function to conduct statistical comparative analysis of the generated results to examine its similarity to real rocks in terms of spatial correlation and structural continuity. Based on the above evaluation indicators, the generated three-dimensional multi-mineral digital cores were screened, and the three-dimensional multi-mineral digital core S3D1 with the highest accuracy was finally selected.

5. The three-dimensional multi-mineral digital core modeling method according to claim 1, characterized in that, Step 5 involves re-encoding each mineral phase in the digital core according to the mineral type and its numbering rules to ensure that the identification of different mineral phases in three-dimensional space is unique and consistent. Then, the re-encoded mineral phases are color-coded, and corresponding standard colors are assigned to different mineral types, thereby realizing an intuitive visualization of three-dimensional multi-mineral digital cores.

Images