A digital core permeability prediction method, device, equipment and storage medium
The hybrid modeling paradigm constructed by graph neural networks solves the problem of deep coupling between microscopic physical fields and macroscopic property derivation in digital core permeability prediction, and realizes rapid and interpretable permeability prediction, which is suitable for industrial applications in the oil and gas field.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA UNIV OF PETROLEUM (EAST CHINA)
- Filing Date
- 2026-04-23
- Publication Date
- 2026-07-14
AI Technical Summary
Existing digital core permeability prediction models cannot achieve deep coupling between microscopic physical field prediction and macroscopic property derivation, mainly because the existing PINN concept is still based on the continuous field assumption when applied to porous media.
A hybrid modeling paradigm of micro-prediction and macro-derivation is constructed using graph neural networks. By acquiring voxel data of target digital cores, graph structure data is extracted using a pore network extraction algorithm, and microscopic physical quantities are obtained using encoder and decoder networks. The graph neural network is then trained with multi-task physical constraints to achieve permeability prediction.
It achieves deep coupling between microscopic physical field prediction and macroscopic property derivation, shortens prediction time, improves model interpretability and credibility, satisfies microscopic flow laws and macroscopic mass conservation, and is suitable for large-scale industrial applications.
Smart Images
Figure CN122063030B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas, and in particular to a digital core permeability prediction method, apparatus, equipment, and storage medium. Background Technology
[0002] In recent years, Physics-Informed Neural Networks (PINNs) have provided a new paradigm for fusing data and physical knowledge, and have achieved success in continuum mechanics problems. However, most existing research applying PINN concepts to porous media is still based on the continuum field assumption (such as applying partial differential equation constraints on voxel grids), which makes it impossible for existing digital core permeability prediction models to achieve deep coupling between microscopic physical field prediction and macroscopic property derivation. Summary of the Invention
[0003] In view of this, the purpose of this invention is to provide a digital core permeability prediction method, apparatus, device, and storage medium, capable of achieving deep coupling between the prediction of the microscopic physical field and the derivation of macroscopic physical properties of digital core permeability. The specific solution is as follows:
[0004] In a first aspect, this application discloses a digital core permeability prediction method, including:
[0005] Voxel data of the target digital core is acquired, and the target pore network extraction algorithm is used to extract features from the voxel data to obtain corresponding graph structure data; the voxel data is a three-dimensional binary image obtained by CT scanning of the target digital core.
[0006] The target encoder network is used to obtain the target hidden features corresponding to the graph structure data, and the target graph neural network is used to update the target hidden features to obtain the corresponding updated hidden features. The target decoder network is then used to obtain the microscopic physical quantities of the voxel data at the pore scale based on the updated hidden features.
[0007] The target pressure field corresponding to the target digital core is determined based on the microscopic physical quantities, the total flow rate corresponding to the target digital core is determined based on the target pressure field and the microscopic physical quantities, and the target permeability corresponding to the target digital core is determined based on the total flow rate.
[0008] Optionally, the step of using a target pore network extraction algorithm to extract features from the voxel data to obtain corresponding graph structure data includes:
[0009] The target pore network extraction algorithm is used to extract features from the voxel data to determine the target pores and target throats in the pore space of the target digital core; the target pore network extraction algorithm includes the maximum sphere algorithm and the central axis algorithm;
[0010] Based on the target pores, the corresponding target node features and target node coordinates are determined, and the target edge features are determined based on the target throats, so as to determine the graph structure data of the target digital core based on the target node features, the target node coordinates and the target edge features.
[0011] Optionally, obtaining the target hidden features corresponding to the graph structure data using the target encoder network includes:
[0012] Based on the target node coordinates, corresponding target position features are generated, and the target position features and the target node features are concatenated to obtain a node feature vector. Based on the target edge features, corresponding edge feature vectors are determined.
[0013] The node feature vector and the edge feature vector are input into the target encoder network to obtain the corresponding node hidden features and edge hidden features; the target hidden features include the node hidden features and the edge hidden features; the target encoder network is an encoder network built based on a multilayer perceptron.
[0014] Optionally, the step of using the target decoder network to obtain the microscopic physical quantities corresponding to the voxel data at the pore scale based on the updated hidden features includes:
[0015] Using the first decoder in the target decoder network, the pressure value corresponding to each target pore is obtained based on the updated node hidden features and the voxel data. Using the second decoder in the target decoder network, the conductivity coefficient corresponding to each target throat is obtained based on the updated edge hidden features and the voxel data. The updated hidden features include updated node hidden features and updated edge hidden features. The microscopic physical quantities include the pressure value corresponding to the target pore and the conductivity coefficient corresponding to the target throat.
[0016] Optionally, determining the target pressure field corresponding to the target digital core based on the microscopic physical quantities includes:
[0017] Based on the target location characteristics, the linear pressure coefficient corresponding to the target digital core is determined, and the linear pressure coefficient and the microscopic physical quantity are fused based on a preset physical guidance mechanism to determine the fused physical quantity.
[0018] The target pressure field corresponding to the target digital core is determined based on the fused physical quantities.
[0019] Optionally, the digital core permeability prediction method further includes:
[0020] Based on the target pressure field, the spatial distribution law of the pressure value in the pore network of the target digital core is determined. Based on the spatial distribution law and the throat flow rate of the target digital core, the main permeation path of the target is determined, and the main permeation path of the target is highlighted on the core model corresponding to the target digital core.
[0021] Optionally, the training process of the target graph neural network further includes:
[0022] The mean square error between the target permeability and the actual permeability corresponding to the target digital core is determined, so as to determine the current data-driven loss of the target graph neural network based on the mean square error using a preset data-driven loss function;
[0023] The current physical loss of the target graph neural network is determined based on the microscopic physical quantities using the target physical loss function.
[0024] The data-driven loss and the physical loss are weighted based on preset weight coefficients to determine the current total target loss of the target graph neural network, and the target graph neural network is updated based on the total target loss.
[0025] Secondly, this application discloses a digital core permeability prediction device, comprising:
[0026] The structural data acquisition module is used to acquire voxel data of the target digital core and to extract features from the voxel data using a target pore network extraction algorithm to obtain corresponding graph structure data; the voxel data is a three-dimensional binary image obtained by CT scanning of the target digital core.
[0027] The data processing module is used to obtain the target hidden features corresponding to the graph structure data using the target encoder network, and to update the target hidden features using the target graph neural network to obtain the corresponding updated hidden features, so as to obtain the microscopic physical quantities of the voxel data at the pore scale using the target decoder network based on the updated hidden features.
[0028] The permeability prediction module is used to determine the target pressure field corresponding to the target digital core based on the microscopic physical quantity, to determine the total flow data corresponding to the target digital core based on the target pressure field and the microscopic physical quantity, and to determine the target permeability corresponding to the target digital core based on the total flow data.
[0029] Thirdly, this application discloses an electronic device, including:
[0030] Memory, used to store computer programs;
[0031] A processor is used to execute the computer program to implement the aforementioned digital core permeability prediction method.
[0032] Fourthly, this application discloses a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned digital core permeability prediction method.
[0033] In this application, when predicting digital core permeability, voxel data of the target digital core is acquired, and a target pore network extraction algorithm is used to extract features from the voxel data to obtain corresponding graph structure data. The voxel data is a three-dimensional binary image obtained by CT scanning of the target digital core. A target encoder network is used to obtain target hidden features corresponding to the graph structure data, and a target graph neural network is used to update the target hidden features to obtain corresponding updated hidden features. A target decoder network is then used to obtain the microscopic physical quantities corresponding to the voxel data at the pore scale based on the updated hidden features. The target pressure field corresponding to the target digital core is determined based on the microscopic physical quantities. The total flow rate data corresponding to the target digital core is determined based on the target pressure field and the microscopic physical quantities, and the target permeability corresponding to the target digital core is determined based on the total flow rate data. As can be seen, this application fully utilizes the advantage of graph neural networks being naturally adapted to the topology of pore networks, learning the connectivity and transport characteristics of pore space in non-Euclidean space. By constructing a hybrid modeling paradigm of "microscopic prediction-macroscopic derivation", the model is driven to simultaneously output physically meaningful microscopic physical quantities to determine the target pressure field corresponding to the target digital core while predicting macroscopic permeability. This ensures that the target permeability corresponding to the final target digital core can simultaneously satisfy the microscopic flow law and the macroscopic mass conservation, achieving deep coupling between microscopic physical field prediction and macroscopic property derivation. Attached Figure Description
[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0035] Figure 1 This is a flowchart of a digital core permeability prediction method disclosed in this application;
[0036] Figure 2 This is a schematic diagram of a specific digital core permeability prediction method disclosed in this application;
[0037] Figure 3 This is a schematic diagram of the structure of a digital core permeability prediction device disclosed in this application;
[0038] Figure 4 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation
[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0040] In recent years, Physical Information Neural Networks (PINNs) have provided a new paradigm for fusing data and physical knowledge, and have achieved success in continuum mechanics problems. However, most existing research applying PINN concepts to porous media still relies on the continuum field assumption (such as applying partial differential equation constraints to voxel grids), which prevents existing digital core permeability prediction models from achieving deep coupling between microscopic physical field prediction and macroscopic property derivation. To address these technical issues, this application discloses a digital core permeability prediction method that enables deep coupling between microscopic physical field prediction and macroscopic property derivation in digital core permeability.
[0041] See Figure 1 As shown in the figure, an embodiment of the present invention discloses a digital core permeability prediction method, including:
[0042] Step S11: Obtain voxel data of the target digital core and use the target pore network extraction algorithm to extract features from the voxel data to obtain corresponding graph structure data; the voxel data is a three-dimensional binary image obtained by CT scanning of the target digital core.
[0043] In this embodiment, voxel data of the target digital core for model training or prediction is first acquired. This voxel data is a three-dimensional binary image obtained by performing micron-level CT scanning on the target digital core, where a pixel value of 1 represents a pore and 0 represents the rock skeleton. A target pore network extraction algorithm is used to extract features from the voxel data to obtain corresponding graph structure data. This includes: extracting features from the voxel data using the target pore network extraction algorithm to determine target pores and target throats in the pore space of the target digital core; the target pore network extraction algorithm includes the maximum sphere algorithm and the central axis algorithm; determining the corresponding target node features and target node coordinates based on the target pores, and determining target edge features based on the target throats, to determine the graph structure data of the target digital core based on the target node features, target node coordinates, and target edge features. Specifically, as shown... Figure 2 As shown, pores and throats in the pore space are identified using the maximum sphere algorithm or the central axis algorithm. Each pore is abstracted as a node in the graph, and the throats connecting the pores are abstracted as edges, thus constructing a graph structure data representing the topological structure of the pore space. Normalization is performed on the extracted node and edge features, mapping the feature values to a standard normal distribution with zero mean and unit variance, and the normalized statistics (mean and standard deviation) are saved for subsequent inverse normalization operations. This graph structure data contains the following components:
[0044] Node features: These include at least geometric features such as the equivalent diameter, inscribed diameter, pore volume, and pore surface area of the pore, used to describe the physical properties of the node.
[0045] Edge features: These include at least the equivalent diameter, inscribed diameter, length, and cross-sectional area of the throat, which are used to describe the transport capacity of the edge.
[0046] Node coordinates: The three-dimensional spatial coordinates of the center point of each pore, used for subsequent boundary condition identification and physical constraint application.
[0047] Step S12: Use the target encoder network to obtain the target hidden features corresponding to the graph structure data, and use the target graph neural network to update the target hidden features to obtain the corresponding updated hidden features, so as to use the target decoder network to obtain the microscopic physical quantities corresponding to the voxel data at the pore scale based on the updated hidden features.
[0048] In this embodiment, the target encoder network is used to obtain the target hidden features corresponding to the graph structure data. Specifically, this may include: generating corresponding target position features based on the target node coordinates, concatenating the target position features and target node features to obtain node feature vectors, and determining corresponding edge feature vectors based on target edge features; inputting the node feature vectors and edge feature vectors into the target encoder network to obtain corresponding node hidden features and edge hidden features; the target hidden features include node hidden features and edge hidden features; the target encoder network is an encoder network built based on a multilayer perceptron. In other words, this embodiment inputs the graph structure data obtained in the above steps into the encoder network, which maps the original node features and edge features to a high-dimensional hidden space, providing a basic representation for subsequent message passing. The position features include at least the normalized X-coordinate, the normalized distance to the inlet boundary, and the normalized distance to the outlet boundary. These features provide the model with direct physical priors regarding the spatial distribution of the pressure field.
[0049] In this embodiment, the encoded graph structure data (target hidden features) is input into a processor (i.e., the target neural network). This processor consists of multiple stacked message passing modules used for multi-round information exchange in the graph, enabling nodes and edges to fully perceive the global topology and local geometric features. Each message passing module includes an edge update unit and a node update unit. The edge update unit aggregates the source node features and target node features, combines them with the current edge features, and updates the edge features using a multilayer perceptron. The calculation process is as follows: the updated edge features are obtained by a nonlinear transformation of the source node features, target node features, and current edge features. The node update unit aggregates the features of all edges connected to the node, combines them with the current node features, and updates the node features using a multilayer perceptron. The calculation process is as follows: the updated node features are obtained by a nonlinear transformation of the current node features and the aggregated features of adjacent edges. Each message passing module uses residual connections to add the input features to the output features, mitigating the gradient vanishing problem in deep networks.
[0050] In this embodiment, the target decoder network is used to obtain the microscopic physical quantities corresponding to voxel data at the pore scale based on the updated hidden features. This includes: using a first decoder in the target decoder network to obtain the pressure value corresponding to each target pore based on the updated node hidden features and voxel data; and using a second decoder in the target decoder network to obtain the conductivity coefficient corresponding to each target throat based on the updated edge hidden features and voxel data. The updated hidden features include updated node hidden features and updated edge hidden features, and the microscopic physical quantities include the pressure value corresponding to the target pore and the conductivity coefficient corresponding to the target throat. In other words, this embodiment can input the updated hidden features into two parallel decoder networks to predict the microscopic physical quantities at the pore scale. Specifically, the pressure decoder (i.e., the first decoder) takes the updated node hidden features as input and predicts the pressure value of each pore node using a multilayer perceptron. The predicted pressure value is represented as a nonlinear function of the node hidden features. The conductivity coefficient decoder (i.e., the second decoder) takes the updated edge hidden features as input and predicts the conductivity coefficient of each throat using a multilayer perceptron. The predicted conductivity coefficient is represented as a nonlinear function of the edge hidden features.
[0051] Step S13: Determine the target pressure field corresponding to the target digital core based on the microscopic physical quantities, determine the total flow data corresponding to the target digital core based on the target pressure field and the microscopic physical quantities, and determine the target permeability corresponding to the target digital core based on the total flow data.
[0052] In this embodiment, after obtaining the microscopic physical quantities, the physical calculation module aggregates these quantities into macroscopic permeability, achieving cross-scale connectivity between the micro and macro levels. This calculation process follows Darcy's law and specifically includes the following sub-steps:
[0053] (1) Boundary node identification: Identify the inlet and outlet boundaries based on node coordinates. Specifically, nodes with X coordinates less than a preset threshold are marked as inlet nodes, nodes with X coordinates greater than (domain size minus X-direction boundary threshold) are marked as outlet nodes, and the remaining nodes are internal nodes.
[0054] (2) Pressure field construction: The pressure of the boundary node is fixed to the physical experimental conditions, the inlet pressure is set to the preset pressure difference value, the outlet pressure is set to zero, and the pressure of the internal node is predicted, thereby constructing a complete pressure field.
[0055] (3) Throat flow calculation: Calculate the fluid flow rate in each throat based on the pressure field and conductivity coefficient. For a throat connecting two nodes, the flow rate is equal to the conductivity coefficient multiplied by the pressure difference between the two nodes, and the flow direction is determined by the pressure gradient.
[0056] (4) Total flow calculation: The total flow through the entire core is obtained by summing the flow through all throats at the inlet boundary.
[0057] (5) Permeability Calculation: Substitute the total flow rate, fluid viscosity, core length, cross-sectional area, and pressure difference into Darcy's law to calculate the macroscopic permeability. The calculation formula is: permeability equals total flow rate multiplied by fluid viscosity multiplied by core length, divided by cross-sectional area, and divided by pressure difference. The calculation result is in square meters in the International System of Units (SI), which is further converted to millidarcy, commonly used in the oil and gas industry.
[0058] In one specific implementation, determining the target pressure field corresponding to the target digital core based on microscopic physical quantities includes: determining the linear pressure coefficient corresponding to the target digital core based on the target location characteristics, and fusing the linear pressure coefficient and the microscopic physical quantities based on a preset physical guidance mechanism to determine the fused physical quantity; and determining the target pressure field corresponding to the target digital core based on the fused physical quantity. In other words, to ensure the physical rationality of the prediction results, this embodiment introduces a physical guidance mechanism. For internal nodes, the normalized pressure coefficient output by the decoder is weighted and fused with the linear pressure coefficient generated based on the location, enabling the predicted pressure field to quickly converge to a reasonable physical distribution in the early stages of training.
[0059] In one specific implementation, the training process of the target graph neural network further includes: determining the mean square error between the target permeability and the actual permeability corresponding to the target digital core, so as to determine the current data-driven loss of the target graph neural network based on the mean square error using a preset data-driven loss function; determining the current physical loss of the target graph neural network based on microscopic physical quantities using a target physical loss function; weighting the data-driven loss and physical loss based on preset weight coefficients to determine the current total target loss of the target graph neural network, and updating the target graph neural network based on the total target loss. In other words, this embodiment compares the obtained predicted permeability with the actual permeability label, uses the mean square error as a supervision signal to drive the model to learn the mapping relationship between input and output; compares the predicted microscopic physical quantities with known physical laws, forcing the model's predictions to conform to microscopic flow laws. These microscopic flow laws include at least the Hagen-Poiseuille equation and the law of conservation of mass. Finally, the above loss components are combined according to preset weights to form the final total loss function. Each weight coefficient can be dynamically adjusted according to the loss value during the training process to balance the contribution of different supervision signals, and deeply integrates data-driven and physical laws using a multi-task physical information loss function. This embodiment constructs differentiable loss terms based on physical laws such as the Hagen-Poiseuille equation and the law of conservation of mass. These terms, together with the data fitting loss, guide the training of the graph neural network, forcing the model to learn representations that conform to physical laws. This significantly improves the physical consistency of the prediction results. Even when faced with pore structure types that have not appeared in the training data, the model can still give physically credible predictions.
[0060] In this embodiment, the spatial distribution of pressure values in the pore network of the target digital core can be determined based on the target pressure field. The main permeability paths of the target are then determined based on the spatial distribution and the throat flow rate of the target digital core, and these main permeability paths are highlighted on the core model corresponding to the target digital core. Specifically, the interpretability analysis process in this embodiment includes:
[0061] Pressure field analysis: Output the pressure value of each pore node and analyze the spatial distribution of pressure in the pore network.
[0062] Flow distribution analysis: Outputs the flow value for each throat, identifying the main flow channels.
[0063] Seepage path visualization: The edges are sorted according to the flow rate of the throats, and the throats with the largest flow rates are selected to form the main seepage paths. These paths are highlighted on the original digital core or pore network model to intuitively show the paths that the fluid preferentially passes through.
[0064] The visualization analysis results provide an intuitive tool for understanding the impact of pore structure on macroscopic transport properties, significantly enhancing the interpretability and credibility of the model. Through the above process, this embodiment constructs a hybrid modeling paradigm of "microscopic prediction-macroscopic derivation," driving the model to simultaneously output physically meaningful microscopic flow fields such as pore pressure distribution and throat flow distribution while predicting macroscopic permeability. Based on this, the model can automatically identify and visualize the main seepage paths, achieving interpretability from microscopic mechanisms to macroscopic properties, significantly enhancing the model's credibility and traceability. By collaboratively introducing pore network graph structure modeling and physical information neural networks within a unified framework, this embodiment fully leverages the advantage of graph neural networks' natural adaptation to pore network topology, learning the connectivity and transport characteristics of pore space in non-Euclidean space. Simultaneously, through multi-task physical constraints, the model can simultaneously satisfy microscopic flow laws and macroscopic mass conservation. Compared to traditional physical experimental methods, the prediction time is reduced from hours to days to seconds, and it does not require destroying the core sample. Compared to pore-scale numerical simulation methods, it also has a speed advantage of several orders of magnitude, which can meet the needs of large-scale industrial applications for rapid evaluation of massive amounts of core samples.
[0065] As can be seen, this application fully utilizes the advantage of graph neural networks being naturally adapted to the topology of pore networks, learning the connectivity and transport characteristics of pore space in non-Euclidean space. By constructing a hybrid modeling paradigm of "microscopic prediction-macroscopic derivation", the model is driven to simultaneously output physically meaningful microscopic physical quantities to determine the target pressure field corresponding to the target digital core while predicting macroscopic permeability. This ensures that the target permeability corresponding to the final target digital core can simultaneously satisfy the microscopic flow law and the macroscopic mass conservation, achieving deep coupling between microscopic physical field prediction and macroscopic property derivation.
[0066] See Figure 3 As shown, this application discloses a digital core permeability prediction device, comprising:
[0067] The structural data acquisition module 11 is used to acquire voxel data of the target digital core and to extract features from the voxel data using a target pore network extraction algorithm to obtain corresponding graph structure data; the voxel data is a three-dimensional binary image obtained by CT scanning of the target digital core.
[0068] Data processing module 12 is used to obtain target hidden features corresponding to the graph structure data using a target encoder network, and to update the target hidden features using a target graph neural network to obtain corresponding updated hidden features, so as to obtain the microscopic physical quantities of the voxel data at the pore scale using a target decoder network based on the updated hidden features.
[0069] The permeability prediction module 13 is used to determine the target pressure field corresponding to the target digital core based on the microscopic physical quantity, to determine the total flow data corresponding to the target digital core based on the target pressure field and the microscopic physical quantity, and to determine the target permeability corresponding to the target digital core based on the total flow data.
[0070] As can be seen, this application fully utilizes the advantage of graph neural networks being naturally adapted to the topology of pore networks, learning the connectivity and transport characteristics of pore space in non-Euclidean space. By constructing a hybrid modeling paradigm of "microscopic prediction-macroscopic derivation", the model is driven to simultaneously output physically meaningful microscopic physical quantities to determine the target pressure field corresponding to the target digital core while predicting macroscopic permeability. This ensures that the target permeability corresponding to the final target digital core can simultaneously satisfy the microscopic flow law and the macroscopic mass conservation, achieving deep coupling between microscopic physical field prediction and macroscopic property derivation.
[0071] In one specific embodiment, the structural data acquisition module 11 may include:
[0072] The feature extraction unit is used to extract features from the voxel data using a target pore network extraction algorithm to determine the target pores and target throats in the pore space of the target digital core; the target pore network extraction algorithm includes the maximum sphere algorithm and the central axis algorithm;
[0073] The structural data acquisition unit is used to determine the corresponding target node features and target node coordinates based on the target pores, and to determine the target edge features based on the target throat, so as to determine the graph structure data of the target digital core based on the target node features, the target node coordinates and the target edge features.
[0074] In one specific embodiment, the data processing module 12 may include:
[0075] The feature vector determination unit is used to generate corresponding target position features based on the target node coordinates, to concatenate the target position features and the target node features to obtain a node feature vector, and to determine the corresponding edge feature vector based on the target edge features;
[0076] The feature encoding unit is used to input the node feature vector and the edge feature vector into the target encoder network to obtain the corresponding node hidden features and edge hidden features; the target hidden features include the node hidden features and the edge hidden features; the target encoder network is an encoder network built based on a multilayer perceptron.
[0077] In one specific embodiment, the data processing module 12 may include:
[0078] The data processing unit is used to obtain the pressure value corresponding to each target pore based on the updated node hidden features and the voxel data using the first decoder in the target decoder network, and to obtain the conductivity coefficient corresponding to each target throat based on the updated edge hidden features and the voxel data using the second decoder in the target decoder network; the updated hidden features include updated node hidden features and updated edge hidden features, and the microscopic physical quantities include the pressure value corresponding to the target pore and the conductivity coefficient corresponding to the target throat.
[0079] In one specific embodiment, the penetration rate prediction module 13 may include:
[0080] The physical quantity fusion unit is used to determine the linear pressure coefficient corresponding to the target digital core based on the target location characteristics, and to fuse the linear pressure coefficient and the microscopic physical quantity based on a preset physical guidance mechanism to determine the fused physical quantity.
[0081] The pressure field determination unit is used to determine the target pressure field corresponding to the target digital core based on the fused physical quantities.
[0082] In one specific embodiment, the device may further include:
[0083] The permeation path display module is used to determine the spatial distribution law of pressure values in the pore network of the target digital core based on the target pressure field, to determine the main permeation path of the target based on the spatial distribution law and the throat flow rate of the target digital core, and to highlight the main permeation path of the target on the core model corresponding to the target digital core.
[0084] In one specific embodiment, the device may further include:
[0085] The first loss determination module is used to determine the mean square error between the target permeability and the actual permeability corresponding to the target digital core, so as to determine the current data-driven loss of the target graph neural network based on the mean square error using a preset data-driven loss function.
[0086] The second loss determination module is used to determine the current physical loss of the target graph neural network based on the microscopic physical quantities using the target physical loss function.
[0087] The graph neural network update module is used to weight the data-driven loss and the physical loss based on preset weight coefficients to determine the current total target loss of the target graph neural network, and to update the target graph neural network based on the total target loss.
[0088] Furthermore, embodiments of this application also disclose an electronic device, Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.
[0089] Figure 4 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the digital core permeability prediction method disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.
[0090] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.
[0091] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored thereon can include an operating system 221, computer programs 222, etc., and the storage method can be temporary storage or permanent storage.
[0092] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the digital core permeability prediction method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.
[0093] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned digital core permeability prediction method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.
[0094] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.
[0095] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0096] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
[0097] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0098] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A digital core permeability prediction method, characterized in that, include: Voxel data of the target digital core is acquired, and the target pore network extraction algorithm is used to extract features from the voxel data to obtain corresponding graph structure data; the voxel data is a three-dimensional binary image obtained by CT scanning of the target digital core. The target encoder network is used to obtain the target hidden features corresponding to the graph structure data, and the target graph neural network is used to update the target hidden features to obtain the corresponding updated hidden features. The target decoder network is then used to obtain the microscopic physical quantities of the voxel data at the pore scale based on the updated hidden features. The target pressure field corresponding to the target digital core is determined based on the microscopic physical quantities, the total flow rate corresponding to the target digital core is determined based on the target pressure field and the microscopic physical quantities, and the target permeability corresponding to the target digital core is determined based on the total flow rate. The step of extracting features from the voxel data using a target pore network extraction algorithm to obtain corresponding graph structure data includes: The target pore network extraction algorithm is used to extract features from the voxel data to determine the target pores and target throats in the pore space of the target digital core; the target pore network extraction algorithm includes the maximum sphere algorithm and the central axis algorithm; Based on the target pores, the corresponding target node features and target node coordinates are determined, and based on the target throat, the target edge features are determined, so as to determine the graph structure data of the target digital core based on the target node features, the target node coordinates, and the target edge features; The step of obtaining the target hidden features corresponding to the graph structure data using a target encoder network includes: Based on the target node coordinates, corresponding target position features are generated, and the target position features and the target node features are concatenated to obtain a node feature vector. Based on the target edge features, corresponding edge feature vectors are determined. The node feature vector and the edge feature vector are input into the target encoder network to obtain the corresponding node hidden features and edge hidden features; the target hidden features include the node hidden features and the edge hidden features; the target encoder network is an encoder network built based on a multilayer perceptron; The step of using a target decoder network to obtain the microscopic physical quantities corresponding to the voxel data at the pore scale based on the updated hidden features includes: Using the first decoder in the target decoder network, the pressure value corresponding to each target pore is obtained based on the updated node hidden features and the voxel data. Using the second decoder in the target decoder network, the conductivity coefficient corresponding to each target throat is obtained based on the updated edge hidden features and the voxel data. The updated hidden features include updated node hidden features and updated edge hidden features. The microscopic physical quantities include the pressure value corresponding to the target pore and the conductivity coefficient corresponding to the target throat.
2. The digital core permeability prediction method according to claim 1, characterized in that, Determining the target pressure field corresponding to the target digital core based on the microscopic physical quantities includes: Based on the target location characteristics, the linear pressure coefficient corresponding to the target digital core is determined, and the linear pressure coefficient and the microscopic physical quantity are fused based on a preset physical guidance mechanism to determine the fused physical quantity. The target pressure field corresponding to the target digital core is determined based on the fused physical quantities.
3. The digital core permeability prediction method according to claim 1, characterized in that, Also includes: Based on the target pressure field, the spatial distribution law of the pressure value in the pore network of the target digital core is determined. Based on the spatial distribution law and the throat flow rate of the target digital core, the main permeation path of the target is determined, and the main permeation path of the target is highlighted on the core model corresponding to the target digital core.
4. The digital core permeability prediction method according to any one of claims 1 to 3, characterized in that, The training process of the target graph neural network also includes: The mean square error between the target permeability and the actual permeability corresponding to the target digital core is determined, so as to determine the current data-driven loss of the target graph neural network based on the mean square error using a preset data-driven loss function; The current physical loss of the target graph neural network is determined based on the microscopic physical quantities using the target physical loss function. The data-driven loss and the physical loss are weighted based on preset weight coefficients to determine the current total target loss of the target graph neural network, and the target graph neural network is updated based on the total target loss.
5. A digital core permeability prediction device, characterized in that, include: The structural data acquisition module is used to acquire voxel data of the target digital core and to extract features from the voxel data using a target pore network extraction algorithm to obtain corresponding graph structure data; the voxel data is a three-dimensional binary image obtained by CT scanning of the target digital core. The data processing module is used to obtain the target hidden features corresponding to the graph structure data using the target encoder network, and to update the target hidden features using the target graph neural network to obtain the corresponding updated hidden features, so as to obtain the microscopic physical quantities of the voxel data at the pore scale using the target decoder network based on the updated hidden features. The permeability prediction module is used to determine the target pressure field corresponding to the target digital core based on the microscopic physical quantity, to determine the total flow data corresponding to the target digital core based on the target pressure field and the microscopic physical quantity, and to determine the target permeability corresponding to the target digital core based on the total flow data. The structure data acquisition module specifically includes: The feature extraction unit is used to extract features from the voxel data using a target pore network extraction algorithm to determine the target pores and target throats in the pore space of the target digital core; the target pore network extraction algorithm includes the maximum sphere algorithm and the central axis algorithm; The structural data acquisition unit is used to determine the corresponding target node features and target node coordinates based on the target pores, and to determine the target edge features based on the target throat, so as to determine the graph structure data of the target digital core based on the target node features, the target node coordinates and the target edge features; The data processing module specifically includes: The feature vector determination unit is used to generate corresponding target position features based on the target node coordinates, to concatenate the target position features and the target node features to obtain a node feature vector, and to determine the corresponding edge feature vector based on the target edge features; A feature encoding unit is used to input the node feature vector and the edge feature vector into the target encoder network to obtain the corresponding node hidden features and edge hidden features; the target hidden features include the node hidden features and the edge hidden features; the target encoder network is an encoder network built based on a multilayer perceptron; The data processing module specifically includes: The data processing unit is used to obtain the pressure value corresponding to each target pore based on the updated node hidden features and the voxel data using the first decoder in the target decoder network, and to obtain the conductivity coefficient corresponding to each target throat based on the updated edge hidden features and the voxel data using the second decoder in the target decoder network; the updated hidden features include updated node hidden features and updated edge hidden features, and the microscopic physical quantities include the pressure value corresponding to the target pore and the conductivity coefficient corresponding to the target throat.
6. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the digital core permeability prediction method as described in any one of claims 1 to 4.
7. A computer-readable storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the digital core permeability prediction method as described in any one of claims 1 to 4.