Multi-component digital core reconstruction method, device, equipment and medium

Abstract

The invention discloses a multi-component digital rock core reconstruction method, device, equipment and medium, and relates to the technical field of reservoir logging identification, the Wasserstein distance is introduced, the Wasserstein distance between a generated sample and real sample distribution is calculated, and the Wasserstein distance is minimized layer by layer in training, so that the problem of non-convergence in the training process is solved, and the accuracy of the multi-component digital rock core reconstruction is improved. The training stability and the quality of generated samples are improved; a priori knowledge decision is embedded into a discriminator to perform regularization constraint on mineral distribution, porosity and crack characteristics, so that the judgment capability of the discriminator on the authenticity of a generated sample and coincidence domain characteristics is improved, and the reconstruction accuracy of a multi-component structure is improved; meanwhile, a gradient penalty mechanism is introduced, the stability of model training and the distribution consistency of generated samples are improved, finally, the trained network model is used for reconstructing a multi-component digital core, and the complex pore structure and the multi-component structure of the reservoir are clearly displayed.

Description

Technical Field

[0001] The present invention relates to the technical field of reservoir logging identification, and particularly relates to a multi-component digital core reconstruction method, device, equipment and medium. Background Art

[0002] Carbonate reservoirs play an important role in the global energy field, with rich reserves of oil and natural gas. However, due to the complex pore structure and multi-component composition of carbonate reservoirs, traditional reservoir characterization methods face great challenges, such as logging technology and microscopic imaging technology, and it is difficult for such methods to realize the microscopic multi-component pore structure of carbonate rocks. In addition, technologies such as Qemscan technology (Quantitative Evaluation of Minerals by Scanning Electron Microscopy) that can identify the multi-component pore structure of microscopic carbonate cores have the disadvantage of high cost.

[0003] In recent years, artificial intelligence algorithms such as generative adversarial networks (GANs) have shown great potential in processing logging and physical property data and have advantages in the structural characterization of carbonate porous media. At the current stage, when dealing with the multi-component structure reconstruction of carbonate reservoirs, since the training process of the generative adversarial network depends on the discriminator's evaluation of the authenticity of the generated images, a large amount of training data is required to support it. However, the carbonate reservoir structure is complex and variable, and it is difficult to obtain more core images showing the carbonate reservoir structure, resulting in inaccurate reconstruction ratios when reconstructing the multi-component of carbonate reservoir cores, and making it difficult for the reconstructed multi-component digital cores to characterize the complex pore structure and multi-component structure of carbonate reservoirs. Summary of the Invention

[0004] Embodiments of the present invention provide a multi-component digital core reconstruction method, device, equipment and medium, which can solve the problem in the prior art that the reconstructed multi-component digital cores are difficult to characterize the complex pore structure and multi-component structure of carbonate reservoirs.

[0005] Embodiments of the present invention provide a multi-component digital core reconstruction method, including the following steps: Obtain multiple core images of small samples of carbonate rocks and the core multi-component structures corresponding to each core image; Introduce the Wasserstein distance as the loss function of the generator in the generator of the generative adversarial network model GAN to form an improved generator; embed prior knowledge decision-making and gradient penalty in the discriminator of the generative adversarial network model GAN to form an improved discriminator; and form an improved generative adversarial network model WGAN-GP including the improved generator and the improved discriminator; Improve the generative adversarial network model WGAN-GP by training with the multi-component structure of cores corresponding to multiple core images; during training, the generator inputs a random noise vector and a corresponding domain knowledge vector, layer by layer optimizing the network model parameters. At the same time, the generator uses the Wasserstein distance as the loss function, adjusting the distribution of the generated samples through the feedback information of the discriminator for adversarial training; the discriminator makes decisions through the embedded prior knowledge, imposing constraints on the mineral distribution, porosity, and fracture connectivity, and evaluating the matching degree between the generated samples and the real samples; and during the adversarial training of the generator and the discriminator, the generator adopts a gradient penalty mechanism to perform gradient constraints on the interpolation points between the real samples and the generated samples to obtain the trained generative adversarial network model WGAN-GP; Input the multi-component structure of the core corresponding to the core image to be reconstructed into the generator of the trained generative adversarial network model WGAN-GP, and output the reconstructed multi-component digital core.

[0006] Preferably, the obtaining of multiple core images of carbonate rocks and the multi-component structure of the core corresponding to the core images includes: Select core samples from carbonate rock core columns, use a slicing machine to cut the core samples into thin slices with a thickness of 5-10 mm, ensuring that the core sample slices contain the main pore and fracture characteristics; After placing the core sample slices in a vacuum drying oven and drying for a preset time, gradually scan the selected area to obtain high-resolution backscattered electron images BSE and multi-spectral energy spectrum diagrams EDS of the samples; use the QEMSCAN method to quantitatively evaluate the sample mineral components in the high-resolution backscattered electron images BSE and multi-spectral energy spectrum diagrams EDS, and obtain small-sample core images of carbonate rocks and the multi-component structure of the core corresponding to the core images.

[0007] Preferably, the improvement of the generative adversarial network model GAN includes: The log-likelihood loss of the objective function in the generative adversarial network model GAN is: ; Replace the log-likelihood loss with the Wasserstein distance, and the equation of the Wasserstein distance is: ; Where: and are two possible distributions; is the set of joint distributions of the two distributions and , and γ is a possible joint distribution of ( , ); x represents the real sample,y Denote the generated sample, which is a sample drawn from γ; Denote two samples x and y the distance between; f represents a 1-Lipschitz function; Embed prior knowledge decision in the discriminator of the generative adversarial network model GAN to obtain the loss error of the multi-component ratio, and the loss function is expressed as: ; Where: Denote the predicted percentage of the sample of the i-th component ratio; Denote the true percentage of the sample of the i-th component ratio; Denote the number of rock component types included in the label; Denote the importance ratio of the original loss function in training the neural network; Denote the importance ratio of the prior knowledge in training the neural network; To maintain the condition of the 1-Lipschitz function, replace the weight clipping strategy in WGAN with gradient penalty, and penalize the gradient of the discriminator output. The gradient penalty equation is: ; Where: is a random variable sampled from the uniform distribution ; is a hyperparameter.

[0008] Preferably, the training of the improved generative adversarial network model WGAN-GP includes: During training, the generator inputs a random noise vector z and the corresponding domain knowledge vector k. The random noise vector z and the corresponding domain knowledge vector k are the mineral composition and pore distribution obtained from experiments, and optimize the parameters of the network model layer by layer according to the mineral composition and pore distribution information obtained from experiments to generate a realistic digital core; The discriminator receives real samples and generated samples. The discriminator imposes constraints on the mineral distribution, porosity, and fracture connectivity through the embedded prior knowledge decision and outputs a scalar as an approximation of the Wasserstein distance to evaluate the difference between the distribution of the generated samples and the real samples; the generator uses the Wasserstein distance as the loss function and updates according to the feedback of the discriminator to minimize the Wasserstein distance and adjust the distribution of the generated samples; At the same time, the discriminator, according to the gradient penalty mechanism, regularizes and constrains the gradient of the interpolation points between the real samples and the generated samples to ensure that the discriminator satisfies the continuity condition of the 1-Lipschitz function.

[0009] Preferably, the output of the trained generative adversarial network model WGAN-GP includes: After the generative adversarial network model WGAN-GP is trained, the input of the generator of the model is a random noise vector z and a domain knowledge vector k; the random noise vector z is a high-dimensional vector sampled from a standard normal distribution, and the knowledge vector k is the mineral composition ratio extracted from the actually collected data. After the random noise vector z and the domain knowledge vector k are concatenated, the generator maps them to a high-dimensional feature space through a fully connected layer and expands them into an initial low-resolution feature map; the generator gradually upsamples the resolution of the feature map through multi-layer transposed convolutions with three-dimensional convolutional kernels, and each layer of transposed convolution refines the features, gradually generating a multi-component digital core with mineral distribution, pore structure, and fracture connectivity.

[0010] An embodiment of the present invention further provides a multi-component digital core reconstruction device, including: A data module for obtaining multiple core images of small samples of carbonate rocks and the core multi-component structures corresponding to each core image; A model building module for introducing the Wasserstein distance as the loss function of the generator in the generator of the generative adversarial network model GAN to form an improved generator; embedding prior knowledge decision-making and gradient penalty in the discriminator of the generative adversarial network model GAN to form an improved discriminator; and forming an improved generative adversarial network model WGAN-GP including the improved generator and the improved discriminator. A model training module for training the improved generative adversarial network model WGAN-GP by using the core multi-component structures corresponding to multiple core images; during training, the generator inputs a random noise vector and a corresponding domain knowledge vector, layer by layer optimizing the network model parameters, and at the same time the generator uses the Wasserstein distance as the loss function, adjusting the distribution of the generated samples through the feedback information of the discriminator for adversarial training; the discriminator applies constraints to the mineral distribution, porosity, and fracture connectivity through the embedded prior knowledge decision-making, evaluating the matching degree between the generated samples and the real samples; and in the adversarial training of the generator and the discriminator, the generator adopts a gradient penalty mechanism to perform gradient constraint on the interpolation points between the real samples and the generated samples to obtain the trained generative adversarial network model WGAN-GP. A reconstruction module for inputting the core multi-component structure corresponding to the core image to be reconstructed into the generator of the trained generative adversarial network model WGAN-GP and outputting the reconstructed multi-component digital core.

[0011] An embodiment of the present invention further provides an electronic device, including a memory and a processor; The memory is used to store a computer program; The processor is configured to implement the steps of the multi-component digital core reconstruction method as described above when executing the computer program stored in the memory.

[0012] An embodiment of the present invention further provides a computer-readable storage medium for storing a computer program, and when the computer program is executed by a processor, the steps of the multi-component digital core reconstruction method as described above are implemented.

[0013] An embodiment of the present invention provides a multi-component digital core reconstruction method, device, equipment and medium. Compared with the prior art, the beneficial effects are as follows: In the present invention, prior knowledge decision-making and gradient penalty are embedded in the discriminator of the generative adversarial network model GAN to form an improved network model WGAN-GP of the generative adversarial network model GAN. By embedding prior knowledge decision-making in the discriminator, the model can consider geological constraints during the training process, that is, regularize and constrain domain knowledge such as mineral distribution, porosity, and fracture characteristics, improving the discriminator's ability to judge the authenticity of generated samples and their compliance with domain characteristics. This process only requires a small number of core images to train the model WGAN-GP, thereby improving the reconstruction accuracy of the multi-component structure of carbonate reservoir cores.

[0014] Moreover, in the present invention, the Wasserstein distance is introduced as the loss function of the generator in the generative adversarial network model GAN. Taking the Wasserstein distance as the optimization target, by calculating the Wasserstein distance between the distribution of generated samples and real samples, and minimizing the Wasserstein distance layer by layer during training, the problem of non-convergence during model training is solved, thereby improving training stability and the quality of generated samples. At the same time, the present invention also introduces a gradient penalty mechanism to constrain the gradient of the interpolation points between real samples and generated samples, improving the stability of model training and the distribution consistency of generated samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the overall process of a multi-component digital core reconstruction method provided by an embodiment of the present invention; Figure 2 It is a schematic diagram of the neural network structure of the WGAN-GP model established for a multi-component digital core reconstruction method provided by an embodiment of the present invention; Figure 3 It is a core image and a multi-component structure diagram of a small part of carbonate rock collected by the Qemscan method for Sample 1 of a multi-component digital core reconstruction method provided by an embodiment of the present invention; Figure 4Sample 2 of a multi-component digital core reconstruction method provided by an embodiment of the present invention uses the Qemscan method to collect core images and multi-component structure schematic diagrams of a small part of carbonate rock; Figure 5 Sample 3 of a multi-component digital core reconstruction method provided by an embodiment of the present invention uses the Qemscan method to collect core images and multi-component structure schematic diagrams of a small part of carbonate rock; Figure 6 Schematic diagram of a multi-component digital core reconstructed by a multi-component digital core reconstruction method provided by an embodiment of the present invention. Detailed implementation manners

[0016] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention in conjunction with the accompanying drawings. Many specific details are set forth in the following description to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0017] See Figure 1 , an embodiment of the present invention provides a multi-component digital core reconstruction method, specifically a multi-component digital core reconstruction method based on knowledge embedding to optimize WGAN-GP. This method improves the training stability and the quality of generated samples by introducing the Wasserstein distance to improve the generative adversarial network (GAN), and combines the knowledge embedding technology to achieve the embedding of prior knowledge of multi-component ratios and improve the reconstruction accuracy of multi-component structures; at the same time, considering the conditions of inaccurate multi-component ratios and high experimental costs in the reconstruction of carbonate rock multi-component structures by traditional methods, the method of the present invention has high timeliness and low cost, can batch reconstruct carbonate rock multi-component digital cores, and is suitable for on-site measurement and analysis in oil fields.

[0018] During the training of the model, the generator inputs a random noise vector z and the corresponding domain knowledge vector k, and optimizes its network parameters layer by layer to generate realistic digital core samples. The generator uses the Wasserstein distance as the loss function and adjusts the distribution of the generated samples through the feedback information of the discriminator, so that it gradually approximates the multi-component characteristics of the real core samples. During the training of the discriminator, the embedded prior knowledge decision module imposes constraints on domain features such as mineral distribution, porosity, and fracture connectivity, ensuring that the discriminator can accurately evaluate the authenticity of the generated samples and their matching degree with the real samples. The entire training process adopts an alternating optimization method. First, the generator is fixed to update the discriminator, enabling the discriminator to distinguish between real samples and generated samples, and maintaining its 1-Lipschitz continuity through the gradient penalty mechanism. Subsequently, the discriminator is fixed to update the generator, and the generator minimizes the output of the discriminator, making the generated digital core samples gradually approach the real data in terms of multi-component distribution and geometric characteristics.

[0019] After training, the trained generator model is used to generate new digital core samples. By inputting a new random noise vector z and domain knowledge k, the generator can efficiently generate high-resolution digital cores that meet the specified mineral distribution and pore structure characteristics. These generated digital core samples are highly consistent with real cores in terms of mineral components and can be directly used for reservoir analysis, seepage simulation, and pore network modeling, significantly improving the efficiency and accuracy of reservoir research.

[0020] Specifically, it includes the following steps: Step 1: Select representative samples from carbonate core columns. Use a slicing machine to cut the core into thin slices with a thickness of 5-10 mm, ensuring that the slices contain the main pore and fracture characteristics. Subsequently, place the slices in a vacuum drying oven for 24 hours, and then gradually scan the selected areas to obtain high-resolution backscattered electron images (BSE) and multi-spectral energy-dispersive spectroscopy maps (EDS) of the samples. Use the automatic energy spectrum analysis function of the QEMSCAN method to quantitatively evaluate the mineral components of the samples, and collect core images and multi-component structures of a small part of carbonate rocks through the QEMSCAN method.

[0021] Step 2: One-hot encode the labels of each component in each image sample, so that the corresponding labels are stored in vector form. Specifically:

[0022] First, extract the distribution information of all components from the image, such as calcite, dolomite, quartz, and pore and fracture regions. In the mineral distribution map obtained by the QEMSCAN method, each pixel corresponds to a label of a mineral type or pore property. To store and process these discrete labels in a computer-friendly way, the one-hot encoding method is used to create a unique vector representation for each component. Specifically, assume that the sample contains N different components, and each label is represented by an N-dimensional vector, where the position corresponding to this component is 1 and the other positions are 0. For example, [1, 0, …, 0] represents dolomite, [0, 1, …, 0] represents quartz, etc. In this way, the two-dimensional image data of the component labels is converted into a three-dimensional array structure, where each pixel is represented by a one-hot encoding vector, enabling efficient use in subsequent data integration, analysis, and simulation processes. Through this method, it can be further combined with pore feature data to construct a multi-scale digital core including mineral distribution and pore properties, providing standardized and easily processed input data for reservoir research.

[0023] Step 3: Divide the data into a training set and a test set. The training set is used to train the data, and the test set is used to evaluate the performance of the model. Using the random division method, the data is split according to a certain ratio (such as 80% training set and 20% test set) to ensure that each subset is consistent in component distribution and feature space.

[0024] Step 4: Initialize the WGAN-GP model, whose model structure is as Figure 2 shown; the network structure of WGAN-GP (as Figure 2 shown) is divided into two parts: a generator and a discriminator. The core of the process of generating a multi-component digital core through the WGAN-GP network model architecture lies in leveraging the powerful learning ability of the generative adversarial network to accurately reconstruct and generate the multi-component characteristics (such as mineral distribution, pore structure, fracture connectivity, etc.) of carbonate reservoirs in the form of a digital core.

[0025] The entire model consists of two parts: a generator and a discriminator, each undertaking different tasks. Through mutual adversarial optimization, high-quality digital core generation is achieved. The main role of the generator is to generate realistic digital cores from random noise z and knowledge embedding vectors k (such as prior information like mineral composition and pore distribution obtained from experiments). The generator gradually maps the input into high-dimensional image data through multi-layer fully connected networks and convolutional layers. Each layer refines the mineral distribution and pore characteristics, thus generating multi-component digital cores with the expected resolution in the output layer. The introduction of knowledge embedding enables the generator to combine domain-specific features and generate digital cores that better conform to the actual reservoir characteristics. The task of the discriminator is to classify the generated samples and real samples and output a Wasserstein distance value to evaluate the difference between the distribution of the generated samples and the real samples. By introducing gradient penalty, the discriminator ensures the consistency of the distribution of the generated samples in terms of mineral composition and pore characteristics, while avoiding the mode collapse problem in traditional GANs. In addition, a knowledge correction module is embedded in the discriminator to verify whether the generated samples meet the constraints of domain knowledge, thereby improving the quality and authenticity of the generated samples. The generation process first extracts real sample features and knowledge vectors from the preprocessed dataset and initializes the generator and discriminator. Then, by inputting random noise z and knowledge vector k into the generator, digital cores with preliminary mineral distribution and pore structure characteristics are generated. The discriminator receives the generated samples and real samples and evaluates the authenticity and distribution difference of the samples through the Wasserstein distance and the knowledge correction module. The generator continuously optimizes its parameters according to the feedback of the discriminator, making the generated samples gradually approach the real sample distribution.

[0026] Among them, the Wasserstein distance is used to measure the difference between the distribution of the generated samples and the real samples, overcoming the training instability problem caused by the JS divergence in traditional GANs. Specifically, it refers to the minimum energy consumption along the optimal path. In WGAN-GP (Wasserstein GAN with Gradient Penalty), the Wasserstein distance is incorporated into the generative adversarial network in the following way. The objective function of the traditional GAN is based on the log-likelihood loss as:

[0027] 。

[0028] In WGAN-GP, this is replaced by the Wasserstein distance, and its formula is: 。

[0029] Where: and are two possible distributions, and the Wasserstein distance is defined by optimizing all possible joint distributions, whose marginal distributions are respectively and , and is used to quantify the difference between the two distributions. is the set of joint distributions of two distributions and , and γ is a possible joint distribution of ( , ); x represents the real samples, y represents the generated samples, is the sample drawn from γ; represents the distance between two samples x and y ; f is a 1-Lipschitz function (implemented by weight constraint or gradient penalty).

[0030] To maintain the 1-Lipschitz condition, WGAN-GP replaces the weight clipping strategy in WGAN with gradient penalty, and defines it by penalizing the gradient of the discriminator output as: .

[0031] where: is a random variable sampled from the uniform distribution . This interpolation method ensures that is the "mixed" point between real samples and generated samples, and is used to calculate the gradient of the discriminator output. is a hyperparameter used to control the weight of gradient penalty.

[0032] The generator receives the random noise z and the knowledge vector k, maps the input to the high-dimensional feature space using the fully connected layer, and then gradually upsamples the features through multiple transposed convolutional layers to generate a digital core with multi-scale mineral distribution and pore structure. Activation functions such as LeakyReLU further enhance the non-linear expression ability of the generator, making the output image closer to the real core samples. The discriminator receives the generated samples and real samples as inputs, extracts the local and global features of the samples using the convolutional layer, and outputs a scalar through the fully connected layer, representing the Wasserstein distance between the distribution of the generated samples and the real samples.

[0033] Step Five: Calculate the normal loss function according to the trained reconstructed image and the label image.

[0034] Step Six: Calculate the loss error of the multi-component ratio according to the knowledge embedding technology, and the specific form of the loss function is shown as follows: .

[0035] Wherein: is the predicted percentage of the sample of the i-th component ratio, is the true percentage of the sample of the i-th component ratio, represents the number of types of rock components included in the label, represents the importance ratio of the original loss function in training the neural network, while represents the importance ratio of the prior knowledge in training the neural network.

[0036] Step Seven: Perform error propagation and feedback regulation on the constructed WGAN-GP neural network, and finally update the weights according to the error.

[0037] During the training process of generating multi-component digital cores by WGAN-GP, high-quality samples are generated through adversarial optimization between the generator and the discriminator. The training starts from initializing the model parameters. The generator and the discriminator initialize their weights respectively, and configure optimizers (usually Adam) to ensure the efficient convergence of the model. In each iteration step of the training, first, the discriminator is updated multiple times to ensure that it can accurately evaluate the Wasserstein distance between the generated samples and the real samples. Specifically, the discriminator receives real samples and generated samples, extracts the local and global features of the samples through convolutional layers, and outputs a scalar as an approximation of the Wasserstein distance. At the same time, Gradient Penalty is introduced to regularize the gradients of the interpolation points between real samples and generated samples, ensuring that the discriminator satisfies the 1-Lipschitz continuity condition, thereby improving the training stability and the accuracy of sample distribution matching. After the discriminator training is completed, the generator is updated according to the feedback of the discriminator, with the goal of minimizing the Wasserstein distance, making the generated sample distribution gradually approach the real sample distribution. The input of the generator consists of a random noise vector z and a knowledge vector k. The noise is used to generate the diversity of samples, while the knowledge vector provides domain information to guide the generator to learn the key features of the samples. The generator gradually constructs a digital core with multi-component features through multi-layer fully connected and transposed convolutional layers, and the mineral distribution and pore structure of the output image gradually approach the real data. The entire training process alternates between optimizing the generator and the discriminator. The generator challenges the discriminator by continuously improving the quality of the generated samples, while the discriminator forces the generator to improve the generation level by continuously enhancing the ability to distinguish true and false samples. During the training, to ensure the convergence and generation quality of the model, the number of optimization steps between the generator and the discriminator is usually set asymmetrically (for example, before each update of the generator, the discriminator is updated multiple times first). When the model converges, the generator can generate digital cores that are highly consistent with real core samples in terms of multi-component features and spatial structure.

[0038] Step 8: Continuously loop Steps 6 and 7 until the iteration times requirement is met, and then output the model.

[0039] After training, the WGAN-GP model generates digital cores with multi-component characteristics through the generator G based on the input random noise z and knowledge vector k. The specific generation process is as follows: First, the input of the generator includes two parts: the random noise vector z and the domain knowledge vector k. The random noise z is a high-dimensional vector sampled from the standard normal distribution, which is used to introduce the diversity of the generated samples; the knowledge vector k is the mineral composition ratio extracted from the actually collected data, which is used to guide the generator to generate digital core samples that meet specific reservoir conditions. The input vector [z, k] is concatenated and then processed layer by layer through the neural network of the generator. Inside the generator, the input is first mapped to a high-dimensional feature space through a fully connected layer and expanded into an initial low-resolution feature map. Subsequently, the generator gradually upsamples the resolution of the feature map through multiple transposed convolutions with 3D convolutional kernels. Each layer of transposed convolution refines the features and gradually generates a multi-component digital core with mineral distribution, pore structure, and fracture connectivity. During the generation process, activation functions (such as LeakyReLU and Tanh) are used to increase the non-linear expression ability, making the generated samples closer to the real samples in terms of mineral distribution and pore geometry. The output of the generator is a high-resolution 3D image, and each pixel (voxel) in the image represents a local point component of the digital core. To reflect the multi-component characteristics, the generated core samples contain multiple channels, and each channel corresponds to a mineral distribution or pore feature. For example, the voxel value [1, 0, …, 0] at the coordinates (X, Y, Z) represents that the generated component at this position is dolomite.

[0040] The present invention introduces the Wasserstein distance as an optimization objective in the generator of the generative adversarial network model GAN, and realizes the precise generation of the multi-component characteristics of the generator by calculating the Wasserstein distance between the generated samples and the real sample distribution, and combining the input of the random noise and the prior knowledge vector.

[0041] The present invention embeds a prior knowledge decision module in the discriminator of the generative adversarial network model GAN, and improves the discriminator's ability to judge the authenticity of the generated samples and their compliance with domain characteristics through regularization constraints on domain knowledge such as mineral distribution, porosity, and fracture characteristics.

[0042] The present invention adopts a gradient penalty mechanism during the training process of the generative adversarial network model WGAN-GP. By constraining the gradients of the interpolation points between the real samples and the generated samples, it ensures that the discriminator meets the 1-Lipschitz continuity condition, thereby improving the stability of model training and the distribution consistency of the generated samples.

[0043] The generator of the Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) of the present invention performs progressive upsampling through multiple fully connected layers and transposed convolutional layers, and combines the Tanh activation function to limit the output range, thereby realizing the generation of digital cores with high resolution and multi-component features.

[0044] The present invention introduces the concatenated vector of the domain knowledge vector k and the random noise z as the input of the generator, so that the digital cores generated by the generator are more in line with the actual reservoir conditions in terms of features such as mineral distribution, pore structure, and fracture connectivity, meeting the reservoir analysis requirements.

[0045] The present invention only requires a small amount of sample data collected by the Qemscan technology to quickly and accurately reconstruct multi-component digital cores, with high timeliness, low cost, and good generalization ability. It realizes the characterization of multi-component pore structures in carbonate reservoirs and can provide guiding suggestions for the exploration and development of carbonate oil reservoirs.

[0046] The present invention improves the training stability and the quality of generated samples by introducing the Wasserstein distance to improve the generative adversarial neural network (GAN); aiming at the problem of inaccurate multi-component reconstruction ratio, and combining the knowledge embedding technology, thereby realizing the embedding of prior knowledge of multi-component ratio and improving the reconstruction accuracy of multi-component structures; through the improved WGAN-GP framework, it can realize the intelligent reconstruction of the complex pore structure and multi-component structure of carbonate reservoirs. This method reduces the time cost, improves the processing efficiency and reconstruction accuracy, and is suitable for measurement and analysis in the oilfield site, providing strong technical support for the development of conglomerate oil and gas reservoirs.

[0047] The Wasserstein distance is combined in the generative adversarial neural network model of the present invention, thereby realizing the improvement of the model training stability and the quality of generated samples. It can be applied to the generation of pore structures of carbonate reservoir core samples at various depths and has the characteristics of strong generalization; the entire evaluation process has no complex calculations, does not require manual processing, and does not require huge computing resources, making it suitable for popularization to the oilfield site.

[0048] Specific experiments: (1) Taking a certain carbonate oil reservoir as an example, a small part of the core images and multi-component structures of carbonate rocks were collected by the QEMSCAN method, such as Figure 3As shown. Before the experiment, the core samples were cut into thin slices with a thickness of about 10 mm and vacuum dried to remove moisture and impurities in the pores. Subsequently, the QEMSCAN method was used to scan each sample layer by layer to collect high-resolution two-dimensional slice images and mineral component distribution data. The left figure shows a typical carbonate core image, in which pores, fractures and matrix structures are clearly visible; the right figure is the corresponding multi-component structure acquisition result, and different colors respectively mark mineral components such as calcite, dolomite, quartz and the pore area. These data were preprocessed to generate a standardized sample matrix, which contains both the spatial characteristics of mineral distribution and the geometric parameters of the pore network, providing comprehensive input information for subsequent deep learning model training. To ensure the scientific nature of model training and verification, the collected core images and multi-component structure data were divided into a training set and a test set. The random sampling method was adopted during the division process to ensure the distribution consistency of the data in terms of mineral composition and pore structure. Finally, 80% of the samples were selected as the training set to optimize the model parameters of the generator and discriminator; the remaining 20% of the samples were used as the test set to verify the effect of the model in generating digital cores.

[0049] (2) Construct a WGAN-GP neural network model and initialize the parameters. The structure of the generator consists of multiple fully connected layers and transposed convolutional layers. Its input is random noise z and knowledge vector k, and high-resolution digital cores are generated through upsampling layer by layer. The structure of the discriminator consists of convolutional layers and fully connected layers, which are mainly used to evaluate the distribution difference between the generated samples and the real samples and output the Wasserstein distance value. When the model is initialized, the Xavier initialization method is adopted to optimize the distribution of the weights of the generator and discriminator to ensure the stability of the parameters in the initial stage. In addition, to accelerate the training process, the Adam optimizer is selected, and the learning rate is set to 10 −4 , and a gradient penalty term is introduced into the discriminator to meet the 1-Lipschitz continuity condition. Among them, the set K vector is [0.02, 0.01, 0.26, 0.01, 0.26, 0.06, 0.06, 0.21, 0.25, 0.01, 0.01]. Each digit of this vector represents the component ratio of albite, quartz, calcite, illite, chlorite, pyrite, rutile, dolomite, anhydrite, other minerals, and pores respectively. This ratio is the average value obtained from a large number of core samples produced from adjacent wells through core analysis experiments.

[0050] (4) During the process of training the WGAN-GP neural network model, an alternating optimization method of the generator and the discriminator is adopted. In each iteration, the discriminator is first updated multiple times to enable it to more accurately calculate the Wasserstein distance between real samples and generated samples. The generator then generates realistic digital cores by minimizing the feedback loss of the discriminator. In each training iteration, the generator inputs random noise z and domain knowledge k, and the discriminator receives generated samples and real samples, and adjusts its parameters through the gradient penalty term to make the calculation of the Wasserstein distance more stable. During the training process, the loss curve of the model gradually converges, indicating that the output quality of the generator is continuously improving, and the matching degree between the generated samples and the real samples in terms of mineral composition and pore structure is significantly improved.

[0051] (5) Save the parameters of the WGAN-GP neural network model after iteration. After training is completed, save the final parameters of the generator and the discriminator as model files (HDF5 or PT format), and record the key hyperparameters (such as learning rate, batch size, and number of training epochs) to ensure the reproducibility of the model. In addition, the saved model contains the weight configuration and network structure of the generator, which can be directly used for subsequent digital core generation.

[0052] (6) Read the saved parameters of the WGAN-GP neural network model, input the test set to verify the model effect, and the generated digital cores are as Figure 4 shown. During the verification process, use the test set data to input the saved generator model to generate digital core samples with multi-component characteristics. The generated digital cores are not only highly consistent with the real samples in terms of mineral distribution, but also the connectivity of pores and fractures conforms to the physical characteristics of the actual reservoir. Figure 4 and Figure 5 show the generated digital core samples. The left figure shows the overall geometric characteristics of the core, and the right figure further shows the distribution characteristics of different mineral components and pore regions through multi-component annotation. The digital cores generated by the model can be directly applied to reservoir seepage simulation and pore network modeling, providing high-precision digital model support for the development of carbonate reservoirs; the finally reconstructed multi-component digital core is as Figure 6 shown, which includes albite, quartz, calcite, illite, chlorite, pyrite, rutile, dolomite, anhydrite, other minerals, pores.

[0053] The above-described embodiments merely represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A multi-component digital core reconstruction method, characterized in that: The following steps are involved: Obtain multiple core images of small samples of carbonate rocks and the core multi-component structure corresponding to each core image; The Wasserstein distance is introduced as the loss function of the generator in the generative adversarial network model GAN ​​to form an improved generator; the prior knowledge decision and gradient penalty are embedded in the discriminator of the generative adversarial network model GAN ​​to form an improved discriminator; and forming an improved generative adversarial network model WGAN-GP including an improved generator and an improved discriminator; The improved generative adversarial network model WGAN-GP is trained using the core multi-component structure corresponding to multiple core images; During training, the generator optimizes the network model parameters layer by layer by inputting random noise vectors and corresponding domain knowledge vectors. At the same time, the generator uses Wasserstein distance as the loss function and adjusts the distribution of generated samples through the feedback information of the discriminator for adversarial training. The discriminator imposes constraints on mineral distribution, porosity and fracture connectivity through embedded prior knowledge decisions to evaluate the matching degree between generated samples and real samples. And in the adversarial training of the generator and the discriminator, the generator uses a gradient penalty mechanism to impose gradient constraints on the interpolation points between the real samples and the generated samples to obtain the trained generative adversarial network model WGAN-GP; The core multi-component structure corresponding to the core image to be reconstructed is input into the generator of the trained generative adversarial network model WGAN-GP, and the reconstructed multi-component digital core is output.

2. A multi-component digital core reconstruction method according to claim 1, characterized in that: The method of obtaining multiple core images of carbonate rocks and core multi-component structures corresponding to the core images includes: Select core samples from carbonate rock cores and cut them into slices with a thickness of 5-10 mm using a slicer to ensure that the core sample slices contain the main pore and fracture features; After the core sample slices are placed in a vacuum drying oven and dried for a preset time, the selected area is scanned step by step to obtain the high-resolution backscattered electron image BSE and multi-spectral energy spectrum EDS of the sample; the QEMSCAN method is used to quantitatively evaluate the sample mineral components in the high-resolution backscattered electron image BSE and multi-spectral energy spectrum EDS to obtain a small sample core image of carbonate rock and the multi-component structure of the core corresponding to the core image.

3. A multi-component digital core reconstruction method according to claim 1, characterized in that: The improvements of the generative adversarial network model GAN ​​include: The log-likelihood loss of the objective function in the generative adversarial network model GAN ​​is: ； Replace the log-likelihood loss with the Wasserstein distance, the equation for the Wasserstein distance is: ； in: and are two possible distributions; There are two distributions and The set of joint distributions of , γ is ( , ) is a possible joint distribution of x represents the real sample, y Indicates the generated sample, is a sample drawn from γ; Represents two samples x and y The distance between them; f represents the 1-Lipschitz function; Prior knowledge decisions are embedded in the discriminator of the generative adversarial network model GAN ​​to obtain the loss error of multi-component proportions, where the loss function is expressed as: ； in: represents the sample prediction percentage of the proportion of the i-th component; Indicates the true percentage of the sample of the i-th component ratio; Indicates the number of rock component types included in the label; Represents the importance ratio of the original loss function in training the neural network; Represents the importance ratio of prior knowledge in training neural network; In order to maintain the condition of the 1-Lipschitz function, the weight clipping strategy in WGAN is replaced by gradient penalty to penalize the gradient of the discriminator output. The gradient penalty equation is: ； in: From a uniform distribution A random variable sampled from ; is a hyperparameter.

4. A multi-component digital core reconstruction method according to claim 1, characterized in that: The training of the improved generative adversarial network model WGAN-GP includes: During training, the generator inputs a random noise vector z and a corresponding domain knowledge vector k, which are the mineral composition and pore distribution obtained experimentally, and optimizes the parameters of the network model layer by layer according to the mineral composition and pore distribution information obtained experimentally to generate a realistic digital core; The discriminator receives real samples and generated samples. The discriminator imposes constraints on mineral distribution, porosity, and fracture connectivity through embedded prior knowledge decisions, and outputs a scalar as an approximation of the Wasserstein distance to evaluate the difference between the generated samples and the real sample distributions. The generator uses the Wasserstein distance as a loss function and is updated according to the feedback from the discriminator to minimize the Wasserstein distance and adjust the distribution of the generated samples. At the same time, the discriminator uses the gradient penalty mechanism to regularize the gradient of the interpolation points between the real samples and the generated samples to ensure that the discriminator meets the continuity condition of the 1-Lipschitz function.

5. A multi-component digital core reconstruction method according to claim 1, characterized in that: The output of the trained generative adversarial network model WGAN-GP includes: After the training of the generative adversarial network model WGAN-GP is completed, the input of the model generator is a random noise vector z and a domain knowledge vector k; the random noise vector z is a high-dimensional vector sampled from a standard normal distribution, and the knowledge vector k is a mineral composition ratio extracted from the actual collected data; After the random noise vector z and the domain knowledge vector k are concatenated, the generator is mapped to the high-dimensional feature space through a fully connected layer and expanded to the initial low-resolution feature map; the generator gradually upsamples the resolution of the feature map through multi-layer transposed convolutions using a three-dimensional convolution kernel. Each layer of transposed convolution refines the features and gradually generates a multi-component digital core with mineral distribution, pore structure, and fracture connectivity.

6. A multi-component digital core reconstruction device, characterized in that: include: A data module, used to obtain multiple core images of small samples of carbonate rocks and the core multi-component structure corresponding to each core image; A model building module, for introducing Wasserstein distance as the loss function of the generator in the generative adversarial network model GAN ​​to form an improved generator; embedding prior knowledge decision and gradient penalty in the discriminator of the generative adversarial network model GAN ​​to form an improved discriminator; and forming an improved generative adversarial network model WGAN-GP including an improved generator and an improved discriminator; Model training module, used to train the improved generative adversarial network model WGAN-GP using the core multi-component structure corresponding to multiple core images; During training, the generator optimizes the network model parameters layer by layer by inputting random noise vectors and corresponding domain knowledge vectors. At the same time, the generator uses Wasserstein distance as the loss function and adjusts the distribution of generated samples through the feedback information of the discriminator for adversarial training. The discriminator imposes constraints on mineral distribution, porosity and fracture connectivity through embedded prior knowledge decisions to evaluate the matching degree between generated samples and real samples. And in the adversarial training of the generator and the discriminator, the generator uses a gradient penalty mechanism to impose gradient constraints on the interpolation points between the real samples and the generated samples to obtain the trained generative adversarial network model WGAN-GP; The reconstruction module is used to input the core multi-component structure corresponding to the core image to be reconstructed into the generator of the trained generative adversarial network model WGAN-GP, and output the reconstructed multi-component digital core.

7. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer programs; The processor is used to implement the steps of a multi-component digital core reconstruction method as described in any one of claims 1 to 5 when executing the computer program stored in the memory.

8. A computer-readable storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, implements the steps of a multi-component digital core reconstruction method as described in any one of claims 1 to 5.

Images