Continuous depth stratum parameter simulation method, device and equipment and storage medium
By reconstructing the initial three-dimensional digital core using rock scanning data, and using neural network models to generate and simulate multiple stratigraphic parameters, the problem of continuous stratigraphic parameters simulation in the existing technology is solved, and efficient and accurate reservoir parameter simulation is achieved.
Patent Information
- Application Number
- CN202311800933.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-25
- Publication Date
- 2025-06-27
AI Technical Summary
The existing three-dimensional digital core construction methods have problems such as difficult to obtain samples, high experimental costs, long cycles, and the inability to simulate the parameters of the continuous formation of wellbores, which affects the comprehensive understanding and evaluation of reservoirs.
By reconstructing the initial three-dimensional digital core of the centering section of the research area using rock scanning data, superimposing the condition information, using the first neural network model to generate multiple first digital cores, stratigraphic parameters are simulated, and finally finding the most similar stratigraphic parameters through machine learning for simulation.
It realizes rapid and simple simulation of the parameters of the continuous depth formation of the wellbore, reduces the experimental cost and time, and improves the comprehensive understanding of reservoirs and the accuracy of evaluation.
Smart Images
Figure CN120211754A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of logging identification, especially the logging identification field of carbonate reservoirs, and specifically relates to a continuous-depth formation parameter simulation method, device, equipment and storage medium. Background Art
[0002] Unconventional oil and gas reservoirs have become an important potential battlefield for oil and gas exploration. However, compared with conventional oil and gas reservoirs, there are many difficult problems, and we also face huge challenges in petrophysics. At present, two main methods are adopted at home and abroad to study formation parameters such as the electrical properties and elastic properties of rocks. The first method is petrophysical experiments, and the second method is a simulation method based on digital cores. Among them, petrophysical experiments have disadvantages such as high costs and long cycles. Moreover, the rock formations of some unconventional oil and gas reservoirs have small porosities, and it is difficult to obtain representative cores through petrophysical experiments for the study of corresponding formation parameters. However, the simulation method based on digital cores can well avoid the above problems.
[0003] Digital core technology uses numerical simulation and data analysis methods to establish a digital pore structure model with the same pore structure as that of the formation, and has become a commonly used means for studying reservoir properties. However, the construction of existing three-dimensional digital cores mainly relies on CT image scanning. This method has great difficulties in sample acquisition, high experimental costs, and a relatively long experimental cycle. Moreover, this method can only construct digital core models for cored rock samples and cannot simulate wellbore continuous formation parameters, affecting the comprehensive understanding and evaluation of reservoirs. Summary of the Invention
[0004] The purpose of the embodiments of the present invention is to provide a continuous-depth formation parameter simulation method, device, equipment and storage medium to overcome one or more defects existing in the background art, such as petrophysical experiments and simulation methods based on digital cores.
[0005] To achieve the above purpose, the first aspect of the embodiments of the present invention provides a continuous-depth formation parameter simulation method, and the method includes:
[0006] Reconstruct an initial three-dimensional digital core of the cored section in the study area by using rock scanning data, and a corresponding relationship between the gray level of the rock sample image and the mineral composition is established in the initial three-dimensional digital core;
[0007] After respectively superimposing different condition information on the initial three-dimensional digital core, input it into the constructed continuous-depth digital core generation model to generate different first digital cores one by one. Among them, the continuous-depth digital core generation model is a first neural network model, and the first neural network model converts the received condition information into a condition constraint for its own generation result. The condition information includes mineral composition, pore size and pore distribution;
[0008] For each of the first digital cores, formation parameter simulation is performed to obtain the formation parameter simulation results of each first digital core;
[0009] According to the logging mineral components, logging porosity sizes, and logging pore distributions at each depth point of the continuous formation, the most similar first digital core is searched for each depth point through machine learning, and the formation parameter simulation results of the found first digital cores are used as the formation parameters of each depth point one by one.
[0010] Optionally, the first neural network model is CDCGAN.
[0011] Optionally, the initial three-dimensional digital core of the cored section in the study area is reconstructed using rock scan data. The initial three-dimensional digital core has established a correspondence between the gray level of the rock sample image and the mineral components, specifically:
[0012] Perform multi-scale CT scanning and SEM scanning analysis on the rock samples in the cored section of the study area to obtain core images at different scales;
[0013] Perform multi-scale QEMSCAN scanning analysis on the rock samples in the cored section of the study area to obtain QEMSCAN data, and determine the mineral components of the rock samples according to the QEMSCAN data;
[0014] Register the core images at different scales;
[0015] Segment the high-resolution core image, and establish a correspondence between the gray level and mineral components of the high-resolution core image;
[0016] According to the image registration relationship, map the segmentation result to the low-resolution core image, and apply the correspondence to the low-resolution core image;
[0017] After the mapping and application are completed, the initial three-dimensional digital core of the rock sample is reconstructed.
[0018] Optionally, the formation parameters include rock electrical parameters and elastic parameters. When performing rock electrical parameter and elastic parameter simulation on the first digital core, the finite element method is used.
[0019] Optionally, the searching for the most similar first digital core for each depth point through machine learning according to the logging mineral components, logging porosity sizes, and logging pore distributions at each depth point of the continuous formation is specifically:
[0020] Select each depth point of the continuous formation respectively, input its corresponding logging mineral components, logging porosity sizes, and logging pore distributions into the constructed second neural network model, and search for the most similar first digital core;
[0021] An index between the first digital core and the core features is created in the second neural network model. The core features are obtained by extracting features from the first digital core, and the core features include mineral composition, porosity size, and pore distribution.
[0022] A second aspect of the embodiments of the present invention provides a continuous depth formation parameter simulation device, and the device includes:
[0023] An initial three-dimensional digital core reconstruction module, configured to reconstruct an initial three-dimensional digital core of the cored section of the study area by using rock scan data. A correspondence relationship between the gray level of the rock sample image and the mineral composition is established in the initial three-dimensional digital core;
[0024] A digital core library construction module, configured to input the initial three-dimensional digital core superimposed with different condition information into the constructed continuous depth digital core generation model respectively, and generate different first digital cores one by one. Among them, the continuous depth digital core generation model is a first neural network model, and the first neural network model converts the received condition information into a condition constraint for its own generation result. The condition information includes mineral composition, porosity size, and pore distribution;
[0025] A formation parameter simulation chart library generation module, configured to perform formation parameter simulation on each first digital core to obtain the formation parameter simulation results of each first digital core;
[0026] A continuous depth formation parameter simulation module, configured to search for the most similar first digital core for each depth point of the continuous formation according to the logging mineral composition, logging porosity size, and logging pore distribution at each depth point of the continuous formation, and use the formation parameter simulation results of the found first digital cores corresponding to each other as the formation parameters of each depth point.
[0027] Optionally, the first neural network model is a CDCGAN.
[0028] Optionally, in the initial three-dimensional digital core reconstruction module, the initial three-dimensional digital core of the cored section of the study area is reconstructed by using rock scan data, and a correspondence relationship between the gray level of the rock sample image and the mineral composition is established in the initial three-dimensional digital core. The specific process includes:
[0029] Perform multi-scale CT scanning and SEM scanning analysis on the rock samples of the cored section of the study area to obtain core images at different scales;
[0030] Perform multi-scale QEMSCAN scanning analysis on the rock samples of the study cored section to obtain QEMSCAN data, and determine the mineral composition of the rock samples according to the QEMSCAN data;
[0031] Register core images of different scales;
[0032] Segment high-resolution core images and establish the correspondence between the gray level and mineral components of high-resolution core images;
[0033] According to the image registration relationship, map the segmentation result to the low-resolution core image and apply the correspondence to the low-resolution core image;
[0034] After the mapping and application are completed, the initial three-dimensional digital core of the rock sample is reconstructed.
[0035] Optionally, the formation parameters include rock electrical parameters and elastic parameters. When simulating the rock electrical parameters and elastic parameters of the first digital core, the finite element method is used.
[0036] Optionally, in the continuous-depth formation parameter simulation module, according to the logging mineral components, logging porosity size, and logging pore distribution at each depth point of the continuous formation, the most similar first digital core is searched for each depth point through machine learning. The specific process is as follows:
[0037] Select each depth point of the continuous formation respectively, input the corresponding logging mineral components, logging porosity size, and logging pore distribution into the constructed second neural network model, and search for the first digital core most similar to it;
[0038] An index between the first digital core and the core features is created in the second neural network model. The core features are obtained by extracting features from the first digital core, and the core features include mineral components, porosity size, and pore distribution.
[0039] A third aspect of the embodiments of the present invention further provides a device, including 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 continuous-depth formation parameter simulation method as described in the first aspect of the embodiments of the present invention.
[0040] A fourth aspect of the embodiments of the present invention further provides a storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements a continuous-depth formation parameter simulation method as described in the first aspect of the embodiments of the present invention.
[0041] After reconstructing the three-dimensional digital core (initial three-dimensional digital core) of the cored section in the study area, the above technical solution randomly varies the porosity size, pore distribution, and mineral composition to generate a variety of new three-dimensional digital cores (first digital cores). For each first digital core, formation parameter simulation is carried out. Based on the formation parameter simulation results, a formation parameter simulation chart library is constructed. In the formation parameter simulation chart library, each chart corresponds to each first digital core one by one. Finally, to realize the formation parameter simulation of the section to be evaluated, according to the logging mineral composition, logging porosity, and logging pore distribution at each depth point in the continuous formation of the section to be evaluated, the most similar first digital core corresponding to each depth point is searched through machine learning, and then the formation parameter data corresponding to this depth point is associated from the formation parameter simulation chart library, realizing the simulation of continuous depth formation parameters quickly and simply, and having strong practicability.
[0042] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific implementation part. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] 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 manners, 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:
[0044] Figure 1 It is a schematic flowchart of a method for simulating formation parameters of continuous depth;
[0045] Figure 2 It is a schematic diagram of training data of a generation model of continuous depth digital cores;
[0046] Figure 3 It is a schematic diagram of a first digital core with similar pore distribution and different porosities;
[0047] Figure 4 It is a schematic diagram of some first digital cores in the digital core library;
[0048] Figure 5 It is a schematic diagram of a resistivity-water saturation change curve obtained by simulating formation parameters of a first digital core;
[0049] Figure 6 It is a schematic diagram of a result of simulating formation parameters of continuous depth;
[0050] Figure 7 It is a schematic diagram of the composition of a device for simulating formation parameters of continuous depth. DETAILED DESCRIPTION OF THE INVENTION
[0051] The following will describe in detail the specific implementation manners of the embodiments of the present invention with reference to the accompanying drawings. It should be understood that the specific implementation manners described herein are only used to illustrate and explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention.
[0052] For ease of understanding the following embodiments, the following is an explanatory description of some technical terms involved in the following embodiments.
[0053] CDCGAN refers to Conditional Deep Convolution Generative Adversarial Networks, that is, conditional deep convolutional generative adversarial networks. This network is derived from the generative adversarial network (GAN). Since the traditional generative adversarial network is relatively simple and performs poorly in data generation, a large number of scholars have improved it in subsequent research, mainly including improvements in three aspects: network architecture, loss function, and training method. DCGAN proposed by Radford et al. optimized the traditional GAN in terms of network architecture, combined CNN and GAN, and both the generator and discriminator in the generative adversarial network used convolutional neural networks, replacing the original fully connected network. This improvement laid the basic architecture of almost all subsequent generative adversarial networks. The emergence of DCGAN greatly improved the training stability and the quality of generated data, especially in the generation of image data. The images generated by DCGAN have the characteristics of the source images and can be used as the original training set. However, a major problem with DCGAN is that it cannot guide the GAN network to generate specific types of data. For example, in the experiment of digit generation, it is impossible to determine which specific digit from 0 to 9 the GAN network generates. To solve this problem, the improvement idea is to draw on the idea of CGAN. Compared with the classical generative adversarial network, CGAN can add a constraint condition y when inputting data, so as to optimize the shortcoming that the traditional GAN cannot control the generation mode of data. If both the generator and discriminator are conditioned on some additional information y, the generative adversarial network can be extended to a conditional model, and y can be any type of auxiliary information, such as class labels or data from other modalities. We can perform the adjustment by taking y as an additional input layer and inputting it into both the discriminator and the generator simultaneously. Therefore, borrowing the design idea of CGAN, CDCGAN was produced.
[0054] Method Embodiment
[0055] An embodiment of the present invention provides a continuous deep formation parameter simulation method for simulating formation parameters at continuous depths in a section to be evaluated, where the formation parameters include rock electrical parameters and rock elastic parameters.
[0056] Specifically, as Figures 1 to 6As shown, the continuous deep formation parameter simulation method includes four steps, specifically:
[0057] S100. Construct an initial three-dimensional digital core for the cored section of the study area;
[0058] S200. Generate corresponding first digital cores by randomly changing mineral components, porosity sizes, and pore distributions, and construct a digital core library;
[0059] S300. Construct a formation parameter simulation chart library;
[0060] S400. Continuously simulate deep formation parameters.
[0061] Among them, in S100, a specific implementation process for constructing the initial three-dimensional digital core of the cored section of the study area is as follows:
[0062] S101. Use rock scan data to reconstruct the initial three-dimensional digital core of the cored section of the study area. The initial three-dimensional digital core establishes a correspondence between the gray level of the core sample image and the mineral components.
[0063] Exemplarily, in one embodiment, S101 specifically includes:
[0064] S1011. Perform multi-scale CT scans and electron microscope scans on the core samples of the cored section of the study area to obtain core images at different scales, and perform multi-scale QEMSCAN scans to obtain QEMSCAN data;
[0065] S1012. Determine the mineral components of the core samples according to the QEMSCAN data;
[0066] S1013. Perform image registration on the core images at different scales;
[0067] S1014. Perform image segmentation on the high-resolution core images, and establish a correspondence between the gray level and mineral components of the high-resolution core images;
[0068] S1015. According to the image registration relationship, map the segmentation results to the low-resolution core images, and apply the correspondence between the gray level and mineral components of the high-resolution core images to the low-resolution core images to achieve image segmentation from high resolution to low resolution, which is an associated segmentation at multiple scales;
[0069] S1016. After mapping and application, reconstruct the initial three-dimensional digital core of the core sample. At this time, the initial three-dimensional digital core contains pores of different sizes and scales.
[0070] Optionally, in order to better perform image fusion on core images scanned at different scales, it is necessary to register the images so that the core image scanned at high resolution is included in the core image scanned at low resolution, realizing the one-to-one correspondence of core images at different scales in terms of spatial position. Generally, image registration includes processes such as image feature extraction and feature matching. Specifically, the image registration can adopt the registration method in the general embodiment, and this embodiment does not describe this part of the content in detail.
[0071] Optionally, the image segmentation can be binary threshold segmentation, etc., to realize the segmentation of the pore space and the skeleton, so as to determine the porosity size and the specific pore distribution of the rock sample. The binary threshold segmentation can select the segmentation method in the general embodiment, and this embodiment does not describe this part of the content in detail either.
[0072] When simulating formation parameters on a three-dimensional digital core, a key factor determining the accuracy of the formation parameter simulation results is the resolution of CT scanning. In the above embodiment, through multi-scale CT scanning and electron microscopy scanning, high-resolution core images and low-resolution core images are obtained, and then the high-resolution core images and low-resolution core images are fused through means such as image registration and associated segmentation, so that the pores in the initially reconstructed three-dimensional digital core are multi-scale pores. The refinement of the initial three-dimensional digital core makes the formation parameter simulation results carried out on the basis of the three-dimensional digital core more accurate later.
[0073] In S200, a specific implementation process for generating a digital core library is as follows:
[0074] S201. After respectively superimposing different condition information on the initial three-dimensional digital core, input it into the constructed continuous-depth digital core generation model to generate different first digital cores one by one. Among them, the continuous-depth digital core generation model is a first neural network model, and the first neural network model converts the received condition information into a conditional constraint of its own generation result. The condition information includes mineral composition, porosity size, and pore distribution;
[0075] S202. The digital core library is composed of all the first digital cores.
[0076] The main factors affecting the pore structure of the core include the pore radius, porosity, throat radius, and pore-throat coordination number, etc. Therefore, to generate a digital core library, that is, to generate multiple different first digital cores, at least one of the porosity, pore distribution, and mineral composition is changed, that is, different condition information is generated. When changing the porosity, pore distribution, and mineral composition, the specific adjusted values can be determined according to the scanning results of CT, electron microscopy, and QEMSCAN. For example: taking the change of pores as an example, let the value range of porosity be 0-30%, with a change of 0.1% each time; let the value range of the average pore diameter be 1-20 voxels, with a change of 0.1 voxel each time; let the value range of the pore size distribution variance be 1-10 voxels, with a change of 0.1 voxel each time. For details, please refer to Figure 3 and Figure 4 shown Figure 3 shows a series of schematic diagrams of the first digital cores with similar pore distributions and different porosities. Figure 3 The condition information of the second three-dimensional model from the left in it is: the porosity value is 0.021, the average pore diameter is 2.7, and the pore size distribution variance is 6.9. Figure 3 The condition information of the third three-dimensional model from the left in it is: the porosity value is 0.067, the average pore diameter is 3.3, and the pore size distribution variance is 5.2. Figure 3 The condition information of the rightmost three-dimensional model in it is: the porosity value is 0.253, the average pore diameter is 4.4, and the pore size distribution variance is 6.7. Figure 4 shows a series of first digital cores obtained by changing the mineral composition and porosity.
[0077] Exemplarily, in one embodiment, the first neural network model selects the CDCGAN network. The CDCGAN network can adopt the architecture in the ordinary embodiment, for example, it consists of a generator and a discriminator. The role of the generator is to convert the random noise with condition information into a conditional constraint on the generated first digital core through a series of convolutional operations, and the role of the discriminator is to judge whether the generated first digital core is real compared with the real rock sample.
[0078] Before training, multiple typical rock samples in the cored section of the study area can be selected for multi-scale CT scanning, electron microscopy scanning, and QEMSCAN scanning analysis, and a variety of initial three-dimensional digital cores are reconstructed correspondingly, and these initial three-dimensional digital cores form the training data set of the CDCGAN network. For example, when the reservoir in the study area is carbonate rock, as Figure 2 shown, 10 carbonate rock samples from a certain well can be selected to make the training data set, and a total of 640 cubic training data with a side length of 128 voxels are made. In addition, it should be noted that Figures 3 to 5 the illustrated content is all generated under the conditions of the aforementioned 10 carbonate rock samples.
[0079] During the training process, when training the generator, the parameters of the discriminator are kept unchanged. The real three-dimensional digital core and the first digital core reconstructed by the generator are input into the discriminator. The discriminator makes a judgment by extracting image information. If the discriminator can determine that the first digital core generated by the generator is fake, the parameters of the generator are adjusted and the generator is continuously trained until the discriminator cannot determine that the first digital core generated by the generator is fake. When training the discriminator, the parameters of the generator are kept unchanged. The real three-dimensional digital core and the reconstructed first digital core are input into the discriminator. The discriminator makes a judgment by extracting image information. If the discriminator cannot determine that the first digital core generated by the generator is fake, the parameters of the discriminator are adjusted and the discriminator is continuously trained until the discriminator can determine that the first digital core generated by the generator is fake.
[0080] The loss function of the CDCGAN network includes the loss function of the generator and the loss function of the discriminator. If both the discriminator loss function and the generator loss function reach the Nash equilibrium, the training of the CDCGAN network is completed.
[0081] Among them, the loss function of the generator is defined as: (Equation 1); the loss function of the discriminator is defined as: (Equation 2). In Equation 1 and Equation 2, and are both mathematical expectations, z is the random noise, y is the conditional vector, x is the real sample, p is the distribution of the sample or noise, G(z|y) is the first digital core generated according to the condition, D(x|y) is the discriminant result of the discriminator for the real sample, D(G(z|y)) is the discriminant result of the discriminator for the reconstructed first digital core, λ is the gradient penalty coefficient, is the gradient penalty term.
[0082] In S300, a specific implementation process for constructing the formation parameter simulation chart library is as follows:
[0083] S301. Perform formation parameter simulation on each first digital core to obtain the formation parameter simulation results of each first digital core;
[0084] S302. The formation parameter simulation chart library is composed of all formation parameter simulation results. Each formation parameter chart in the formation parameter simulation chart library corresponds one-to-one with each first digital core.
[0085] Optionally, when performing formation parameter simulation on each first digital core, the simulation method in the general embodiment can be used.
[0086] For example, the main methods for simulating rock electrical parameters based on digital cores include: Kirchhoff circuit node method, random walk method, lattice Boltzmann method, and finite element method.
[0087] Among them, the basic principle of Kirchhoff's circuit node method is that at any instant, the sum of the currents flowing out of a certain node or closed region is equal to the sum of the currents flowing out of this node or closed region. The random walk method is relatively easy to implement and can well repeat the simulation of formation factors. The basic idea of this method for simulating the electrical properties of rocks is to use the fact that the same Laplace equations are solved between diffusion and current conduction in the steady state, and it is deduced that electrical parameters can be used to represent the spatial tortuosity. Arbitrarily take a pixel in the pore space and place a particle that can move in any direction on this pixel. The particle moves to 19 adjacent pixels in the unit time step. Which one of the above 19 adjacent pixels the moving particle appears in at the next moment is randomly determined. If the particle hits the rock skeleton, it returns to the pore space along the original path. Record the spatial position of the moving particle in the three-dimensional digital core at each moment, and calculate the spatial distance between the current position and the starting position. In equal time, the farther the moving particle is from the starting position, the smaller the tortuosity of the pore space and the lower the formation factor. By changing the fluid distribution pattern in the pore space, different water saturation values can be obtained, and the resistance increase factor can also be calculated. The lattice Boltzmann method uses an artificial microscopic model to abstract the fluid in the pore structure into a large number of microscopic particles, and they expand and migrate on discrete grids according to a certain simple rule. By statistically analyzing the motion of the microscopic particles, the macroscopic characteristics of fluid motion can be simulated. Based on this principle, this method is usually used to simulate the electrical parameters and permeability of rocks. The basic principle of the finite element method is simply piecewise approximation, that is, a suitable approximate function is selected in each pixel to replace the solution function. For rocks, they can be regarded as two-phase composite materials composed of pore fluid and rock skeleton, and the three-dimensional digital core is a special three-dimensional digital image. Based on the three-dimensional digital image, the finite element method can be used to calculate the physical properties of the composite material. For example, each pixel in the three-dimensional digital image of the composite material is regarded as a unit in the finite element mesh division, and each unit contains 8 nodes. An external electric field is applied on the boundary of the three-dimensional digital image. According to the principle of minimum energy in the three-dimensional digital core image, the voltage distribution at each node is determined, and then physical parameters such as the effective conductivity of the composite material are calculated.
[0088] For another example, the main methods for simulating rock elastic parameters based on digital cores include: the boundary method, the self-consistent method, the differential method, the Eshelby equivalent inclusion method, the Mori-Tanaka method, the finite element method, etc. Among them, the finite element method is not affected by the shape and properties of the inclusions, and a three-dimensional digital core closer to the actual core can be established, and then the elastic parameters of the rock can be simulated.
[0089] Exemplarily, in one embodiment, when simulating the electrical parameters and elastic parameters of the first digital core, the finite element method is used for both.Figure 5 The resistivity-water saturation variation curve obtained by simulating formation parameters of the first digital core is shown, where the mathematical expression of the variation curve is y = 1.1957x -2.224 , where y represents the resistivity RI and x represents the water saturation Sw.
[0090] In S400, a specific implementation process of continuous-depth formation parameter simulation is as follows:
[0091] S401. According to the logging data of the section to be evaluated, obtain the logging mineral components, logging porosity sizes, and logging pore distributions at each depth point of the continuous formation, and perform the search for the most similar first digital core at each depth point through machine learning;
[0092] S402. Correspondingly use the formation parameter simulation results of each found first digital core as the formation parameters at each depth point.
[0093] Exemplarily, in one embodiment, S401 specifically includes:
[0094] S4011. Select each depth point of the continuous formation respectively, input the corresponding logging mineral components, logging porosity sizes, and logging pore distributions into the constructed second neural network model, and search for the most similar first digital core. Among them, an index between the first digital core and the core characteristics is created in the second neural network model, and the core characteristics are obtained by extracting the characteristics of the first digital core. The core characteristics include mineral components, porosity sizes, and pore distributions. It can be seen that the input of the second neural network model is text content, and the output is the similarity ranking result. The one ranked first is the most similar first digital core corresponding to the depth point.
[0095] It is known that regression and convolutional neural networks can be used for core feature extraction. When training this regression and convolutional neural network, the first digital cores generated by the continuous-depth digital core model can be used as training data. It should be understood that the regression and convolutional neural networks for feature extraction can be designed with the feature extraction network in a general embodiment as the baseline. Those skilled in the art should know how to perform network transformation on this baseline to achieve core feature extraction. There can be multiple network transformation schemes, and the method of network transformation is not the innovation point of this embodiment. Therefore, this embodiment does not describe this part of the content in detail.
[0096] To verify the effectiveness of the continuous-depth formation parameter simulation method implemented in this embodiment, Figure 6 the specific application result diagram of this method in the section to be evaluated is shown. In Figure 6 :
[0097] 1) The first to fourth channels are the depth channel, the lithology curves (natural gamma, caliper, and photoelectric absorption cross-section), the triple porosity curves (density, neutron, and acoustic), and the resistivity curves (deep and shallow lateral resistivity), respectively;
[0098] 2) The fifth channel is the borehole wall porosity image generated based on the electrical imaging logging image;
[0099] 3) The sixth channel is the comparison of the porosity simulated in this embodiment with the porosity calculated by logging according to the formation lithology and porosity changes. The comparison shows that the trends and value distributions of the two are basically the same, confirming that the most similar first digital core found through step S400 conforms to the formation changes;
[0100] 4) The seventh channel is the rock resistivity under 100% water saturation simulated in this embodiment. Of course, the resistivity of the formation rock under different water saturation conditions and in the case of filling with special minerals (such as pyrite, asphalt, etc.) can also be constructed as needed. According to the resistivity changes, the main controlling factors affecting the formation resistivity can be clarified to guide the fine evaluation of the reservoir.
[0101] Device Embodiment
[0102] As Figure 7 shown, the embodiment of the present invention also provides a continuous depth formation parameter simulation device, including an initial three-dimensional digital core reconstruction module, a digital core library construction module, a formation parameter simulation chart library generation module, and a continuous depth formation parameter simulation module that are connected in sequence.
[0103] The initial three-dimensional digital core reconstruction module is used to reconstruct the initial three-dimensional digital core of the cored section in the study area using the rock scan data. The initial three-dimensional digital core establishes the corresponding relationship between the gray level of the rock sample image and the mineral components.
[0104] The digital core library construction module is used to input the initial three-dimensional digital core after superimposing different condition information into the constructed continuous depth digital core generation model, and generate different first digital cores one by one. Among them, the continuous depth digital core generation model is the first neural network model, and the first neural network model converts the received condition information into the condition constraints of its own generation results. The condition information includes mineral components, porosity size, and pore distribution.
[0105] The formation parameter simulation chart library generation module is used to perform formation parameter simulation on each first digital core to obtain the formation parameter simulation results of each first digital core.
[0106] The continuous-depth formation parameter simulation module is used to perform a search for the most similar first digital core for each depth point through machine learning based on the logging mineral components, logging porosity, and logging pore distribution at each depth point of the continuous formation, and use the formation parameter simulation results of the found first digital cores corresponding to each depth point as the formation parameters for each depth point.
[0107] Preferably, the first neural network model is CDCGAN.
[0108] Preferably, in the initial three-dimensional digital core reconstruction module, the initial three-dimensional digital core of the cored section in the study area is reconstructed using rock scan data, and a correspondence between the gray level of the rock sample image and the mineral components is established. The specific process includes:
[0109] Perform multi-scale CT scans and SEM scans on the rock samples in the cored section of the study area to obtain core images at different scales;
[0110] Perform multi-scale QEMSCAN scans on the rock samples in the cored section of the study area to obtain QEMSCAN data, and determine the mineral components of the rock samples based on the QEMSCAN data;
[0111] Register the core images at different scales;
[0112] Segment the high-resolution core images and establish a correspondence between the gray level and mineral components of the high-resolution core images;
[0113] According to the image registration relationship, map the segmentation results to the low-resolution core images and apply the correspondence to the low-resolution core images;
[0114] After the mapping and application are completed, the initial three-dimensional digital core of the rock sample is reconstructed.
[0115] Preferably, the formation parameters include rock electrical parameters and elastic parameters. When simulating the rock electrical parameters and elastic parameters of the first digital core, the finite element method is used.
[0116] Preferably, in the continuous-depth formation parameter simulation module, based on the logging mineral components, logging porosity, and logging pore distribution at each depth point of the continuous formation, a search for the most similar first digital core is performed for each depth point through machine learning. The specific process is as follows:
[0117] Select each depth point of the continuous formation respectively, input its corresponding logging mineral components, logging porosity, and logging pore distribution into the constructed second neural network model, and search for the most similar first digital core;
[0118] An index between the first digital core and the core features is created within the second neural network model. The core features are obtained by extracting features from the first digital core, and the core features include mineral composition, porosity size, and pore distribution.
[0119] 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 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. A person of ordinary skill in the art can understand and implement it without creative efforts.
[0120] On the other hand, an embodiment of the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a machine-readable storage medium. When the computer program is executed by a processor, the computer can execute the above-mentioned continuous deep formation parameter simulation method, which includes: constructing an initial three-dimensional digital core of the cored section in the study area; generating corresponding first digital cores by randomly changing the mineral composition, porosity size, and pore distribution to construct a digital core library; constructing a formation parameter simulation chart library; and performing continuous deep formation parameter simulation.
[0121] On another aspect, an embodiment of the present invention also provides a device, including 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 continuous deep formation parameter simulation method as described in the method embodiment.
[0122] On 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 continuous deep formation parameter simulation method as provided in the method embodiment.
[0123] 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 this understanding, the above technical solutions, in essence, 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.
[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended 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 cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.
Claims
1. A continuous deep formation parameter simulation method, characterized in that, Comprising: Reconstructing an initial three-dimensional digital core of the cored section in the study area using rock scan data, and establishing a correspondence between the gray scale of the rock sample image and the mineral composition in the initial three-dimensional digital core; Inputting the initial three-dimensional digital core with different condition information superimposed thereon into a constructed continuous-depth digital core generation model to generate different first digital cores one by one, wherein the continuous-depth digital core generation model is a first neural network model, and the first neural network model converts the received condition information into a condition constraint for its own generation result, and the condition information includes mineral composition, porosity size, and pore distribution; Performing formation parameter simulation on each first digital core to obtain a formation parameter simulation result of each first digital core; According to the well logging mineral composition, well logging porosity size, and well logging pore distribution at each depth point of the continuous formation, searching for the most similar first digital core at each depth point through machine learning, and correspondingly using the formation parameter simulation results of the found first digital cores as the formation parameters at each depth point.
2. The continuous deep formation parameter simulation method according to claim 1, characterized in that The first neural network model is CDCGAN.
3. A continuous deep formation parameter simulation method according to claim 1, characterized in that, The reconstructing the initial three-dimensional digital core of the cored section in the study area using rock scan data, and establishing a correspondence between the gray scale of the rock sample image and the mineral composition in the initial three-dimensional digital core, specifically: Performing multi-scale CT scanning and SEM scanning analysis on the rock samples in the cored section of the study area to obtain core images at different scales; Performing multi-scale QEMSCAN scanning analysis on the rock samples in the studied cored section to obtain QEMSCAN data, and determining the mineral composition of the rock samples according to the QEMSCAN data; Registering the core images at different scales; Segmenting the high-resolution core image, and establishing a correspondence between the gray scale and the mineral composition of the high-resolution core image; According to the image registration relationship, mapping the segmentation result to the low-resolution core image, and applying the correspondence to the low-resolution core image; After the mapping and application are completed, reconstructing to obtain the initial three-dimensional digital core of the rock sample.
4. A continuous deep formation parameter simulation method according to claim 1, characterized in that The formation parameters include rock electrical parameters and elastic parameters. When performing rock electrical parameter and elastic parameter simulation on the first digital core, the finite element method is used.
5. A continuous deep formation parameter simulation method according to claim 1, characterized in that, The searching for the most similar first digital core at each depth point through machine learning according to the well logging mineral composition, well logging porosity size, and well logging pore distribution at each depth point of the continuous formation, specifically: Selecting each depth point of the continuous formation respectively, and inputting the corresponding well logging mineral composition, well logging porosity size, and well logging pore distribution into a constructed second neural network model to search for the most similar first digital core thereto; An index between the first digital core and the core characteristics is created in the second neural network model, and the core characteristics are obtained by performing feature extraction on the first digital core, and the core characteristics include mineral composition, porosity size, and pore distribution.
6. A continuous deep formation parameter simulation device, characterized in that, Comprising: An initial three-dimensional digital core reconstruction module for reconstructing an initial three-dimensional digital core of the cored section in the study area using rock scan data, and establishing a correspondence between the gray scale of the rock sample image and the mineral composition in the initial three-dimensional digital core; The digital core library construction module is used to input the initial three-dimensional digital cores after superimposing different condition information into the constructed continuous-depth digital core generation model one by one to generate different first digital cores correspondingly. The continuous-depth digital core generation model is a first neural network model, and the first neural network model converts the received condition information into the condition constraints of its own generation result. The condition information includes mineral composition, porosity size, and pore distribution. The formation parameter simulation chart library generation module is used to perform formation parameter simulation on each first digital core to obtain the formation parameter simulation results of each first digital core. The continuous-depth formation parameter simulation module is used to search for the most similar first digital core for each depth point through machine learning according to the well logging mineral composition, well logging porosity size, and well logging pore distribution of each depth point in the continuous formation, and use the formation parameter simulation results of the found first digital cores as the formation parameters of each depth point one by one.
7. A continuous deep formation parameter simulation device according to claim 6, characterized in that, The first neural network model is CDCGAN.
8. A continuous deep formation parameter simulation device according to claim 6, characterized in that, In the initial three-dimensional digital core reconstruction module, the initial three-dimensional digital core of the cored section in the study area is reconstructed using rock scan data. The initial three-dimensional digital core establishes the correspondence between the gray level of the rock sample image and the mineral composition. The specific process includes: Performing multi-scale CT scanning and SEM scanning analysis on the rock samples in the cored section of the study area to obtain core images at different scales. Performing multi-scale QEMSCAN scanning analysis on the rock samples in the cored section of the study area to obtain QEMSCAN data, and determining the mineral composition of the rock samples according to the QEMSCAN data. Registering the core images at different scales. Segmenting the high-resolution core images and establishing the correspondence between the gray level and the mineral composition of the high-resolution core images. According to the image registration relationship, mapping the segmentation results to the low-resolution core images and applying the correspondence to the low-resolution core images. After the mapping and application are completed, the initial three-dimensional digital core of the rock sample is reconstructed.
9. A continuous deep formation parameter simulation device according to claim 6, characterized in that, The formation parameters include rock electrical parameters and elastic parameters. When performing rock electrical parameter and elastic parameter simulation on the first digital core, the finite element method is used.
10. A continuous deep formation parameter simulation device according to claim 6, characterized in that, In the continuous-depth formation parameter simulation module, according to the well logging mineral composition, well logging porosity size, and well logging pore distribution of each depth point in the continuous formation, the process of searching for the most similar first digital core for each depth point through machine learning is as follows: Selecting each depth point in the continuous formation respectively, inputting the corresponding well logging mineral composition, well logging porosity size, and well logging pore distribution into the constructed second neural network model, and searching for the most similar first digital core. An index between the first digital core and the core features is created in the second neural network model. The core features are obtained by extracting features from the first digital core. The core features include mineral composition, porosity size, and pore distribution.
11. A 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 the continuous-depth formation parameter simulation method according to any one of claims 1 to 5.
12. A storage medium, on which a computer program is stored, characterized in that, When the computer program is executed by a processor, it implements the continuous deep formation parameter simulation method according to any one of claims 1 to 5.