Reef limestone crack connectivity prediction method and device in combination with graph neural network

Abstract

The invention provides a reef limestone crack connectivity prediction method and device in combination with a graph neural network, and the method comprises the following steps: S1, taking the intersection points of reef limestone crack line segments as graph nodes, and recording the three-dimensional space coordinates of each node; taking actual crack line segments between adjacent nodes as edges of the graph, and endowing each edge with an edge feature vector; the edge feature vector at least comprises a crack trend azimuth angle, a crack section length, a crack section curvature and a minimum included angle formed by the crack section and all adjacent crack sections; s2, constructing an adjacent matrix and a node feature matrix, and constructing structured graph data for describing the geometrical morphology and topological connection of the crack network in combination with the edge feature vectors; and S3, inputting the structured graph data into the graph neural network to obtain a connected probability graph. A reef limestone fracture network is converted into a graph structure and a graph neural network, and collaborative breakthrough of precision and efficiency of seepage capacity evaluation and decision reliability is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas reservoir fracture characterization and seepage prediction, and specifically to a method and device for predicting reef limestone fracture connectivity combined with a graph neural network. Background Art

[0002] Reef limestone reservoirs are important reservoirs in oil and gas exploration and development. The connectivity of their internal fracture networks directly influences fluid flow and development outcomes. A well-connected fracture network provides low-resistance flow pathways for oil and gas, increasing productivity; whereas poor connectivity restricts flow and reduces development efficiency. Therefore, accurate prediction and evaluation of reef limestone fracture connectivity is crucial for detailed reservoir characterization and optimized development strategies.

[0003] Traditional methods for assessing fracture connectivity rely primarily on empirical statistical models based on static geometric parameters or computationally expensive numerical simulations. These methods have significant limitations: empirical models struggle to capture the nonlinear effects of the complex topological structure and spatial configuration of fracture networks on seepage paths; while numerical simulations offer clear physical meaning, they are computationally time-consuming, making them difficult to efficiently evaluate, and often oversimplify the representation of fracture geometry. Furthermore, existing methods often lack effective quantification of the dynamic interactions between fractures and struggle to achieve end-to-end connectivity probability prediction, limiting their precise application in engineering decision-making.

[0004] Therefore, there is an urgent need for an intelligent method that can integrate fracture geometry and topological features, balance prediction accuracy and computational efficiency, and output probabilistic connectivity evaluation results, so as to improve the accuracy and engineering practicality of reef limestone reservoir connectivity prediction. Summary of the Invention

[0005] The present invention proposes a method and device for predicting the connectivity of reef limestone fractures combined with a graph neural network to solve the technical problem that the existing technology is difficult to output probabilistic connectivity evaluation results while taking into account both prediction accuracy and computational efficiency.

[0006] To solve the above technical problems, the present invention provides a method for predicting the connectivity of reef limestone fractures combined with a graph neural network, comprising the following steps:

[0007] Step S1: The intersection points of the reef limestone fracture segments are used as graph nodes, and the three-dimensional spatial coordinates of each node are recorded; the actual fracture segments between adjacent nodes are used as graph edges, and an edge feature vector is assigned to each edge;

[0008] The edge feature vector at least includes: the crack strike azimuth, the crack segment length, the crack segment curvature, and the minimum angle formed by the crack segment and all adjacent crack segments;

[0009] Step S2: constructing an adjacency matrix and a node feature matrix, and combining the edge feature vectors to construct structured graph data describing the geometry and topological connections of the fracture network;

[0010] Step S3: Input the structured graph data into a graph neural network to obtain a connectivity probability graph.

[0011] Preferably, the calculation expression of the minimum angle in step S1 is:

[0012] ;

[0013] Where, Representation node and nodes The minimum angle between the sides; Representation node The set of neighbors of Indicates from arrive direction vector; Represents the inverse cosine function.

[0014] Preferably, the loss function used in the training of the graph neural network is The expression is:

[0015] ;

[0016] ;

[0017] Where, represents the total number of edges in the graph; Represents the weight coefficient of the connected edge; Represents an edge The true connectivity label of Represents an edge The predicted connectivity probability of represents the number of disconnected edges; Indicates the number of connected edges.

[0018] Preferably, when the graph neural network is being trained, data enhancement is performed on the training data set, and the data enhancement includes:

[0019] 1) Random rotation: Increase the crack strike angle by a random amount uniformly distributed between 0° and 360°;

[0020] 2) Random translation: uniformly translate the node coordinates within the range of ±10% of the area size;

[0021] 3) Random scale jitter: Multiply the crack segment length by a random factor uniformly distributed between 0.8 and 1.2.

[0022] Preferably, when the graph neural network is trained, DropEdge regularization is used to prevent overfitting: in each training iteration, the elements in the adjacency matrix are randomly set to zero and discarded with a set probability.

[0023] Preferably, the graph neural network in step S3 includes: a feature initialization layer, a message passing layer and an edge prediction layer;

[0024] The feature initialization layer: normalizes the edge feature vector to obtain an initial feature vector;

[0025] Message passing layer: Based on the adjacency matrix and node feature matrix, a multi-head graph attention mechanism is used to iteratively update node features and generate node embedding vectors;

[0026] Edge prediction layer: After concatenating the embedding vectors of the nodes at both ends of the target edge with the initial edge feature vector of the target edge, the connectivity probability of the target edge is obtained by passing it through two layers of multi-layer perceptrons (MLPs) and sigmoid functions.

[0027] Preferably, the calculation expression of the message transmission layer is:

[0028] ;

[0029] ;

[0030] Where, node In the The embedding vector output after layer message passing; σ() represents the activation function; Representation node The set of neighbors of Representation node To neighbors Normalized attention weights of ; Indicates the The weight matrix of the attention head; LeakyReLU represents the activation function; represents the attention mechanism parameter vector; Representation node The feature embedding vector of Representation node The feature embedding vector of .

[0031] Preferably, the calculation expression of the edge prediction layer is:

[0032] ;

[0033] ;

[0034] ;

[0035] ;

[0036] Where, Represents an edge The fusion feature vector of Represents an edge The connectivity probability of represents the normalized edge eigenvector; is the activation function; Represents the weight matrix of the first layer of the multilayer perceptron; Represents the bias vector of the first layer of the multilayer perceptron; Represents the second layer weight matrix of the multilayer perceptron; Represents the bias vector of the second layer of the multilayer perceptron.

[0037] Preferably, the method further comprises the following steps:

[0038] Step S4: binarizing the connectivity probability graph, setting a threshold to segment the connectivity domain, and extracting the largest connected subgraph;

[0039] Step S5: converting the edge connectivity probability into a percolation path weight and assigning it to the maximum connected subgraph, and then using the Dijkstra algorithm to calculate the shortest path between any nodes on the weighted maximum connected subgraph to obtain the main percolation path;

[0040] Step S6: Generate a fracture network heat map based on the fracture connectivity probability and mark the main seepage path for visualization; calculate the global connectivity ratio to quantify the reservoir seepage capacity.

[0041] The present invention also provides an electronic device, comprising: a memory, a processor and a computer program, wherein the computer program is stored in the memory and configured to be executed by the processor to implement the above method.

[0042] The beneficial effects of the present invention include at least: by converting the reef limestone fracture network into a graph structure and combining it with a graph neural network, the present invention achieves a synergistic breakthrough in the accuracy, efficiency and decision-making reliability of seepage capacity assessment: in terms of accuracy, due to the explicit modeling of geometric topological features such as fracture curvature and dynamic angle and the use of graph neural networks to capture non-local dependencies, the accuracy of main seepage channel identification is significantly improved compared with traditional numerical simulation; in terms of efficiency, end-to-end graph reasoning replaces physical simulation, which greatly reduces the time consumption of evaluating networks with thousands of nodes; in engineering decision-making, probabilistic connectivity output combined with a threshold segmentation mechanism significantly quantifies uncertainty, and the misjudgment rate of drilling targets is reduced; at the same time, automated seepage path extraction and thermal map rendering greatly reduce manual analysis hours, and full-process intelligence shortens the solution deployment cycle from days to hours. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following is a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts are within the scope of protection of the present invention.

[0045] Example 1

[0046] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting reef limestone fracture connectivity in combination with a graph neural network, comprising the following steps:

[0047] Step S1: The intersection points of the reef limestone fracture segments are used as graph nodes, and the three-dimensional spatial coordinates of each node are recorded; the actual fracture segments between adjacent nodes are used as graph edges, and an edge feature vector is assigned to each edge.

[0048] Specifically, the core of this step is to abstract the complex reef limestone fracture network into a mathematical graph structure as the input basis of the graph neural network. First, the intersection points of the fracture segments are identified as graph nodes, and each node accurately records its three-dimensional spatial coordinates. Next, the actual fracture segments between adjacent nodes are defined as the edges of the graph. Each edge not only represents a connection relationship, but is also endowed with a multi-dimensional geometric attribute feature vector, including the strike azimuth, actual length, curvature of the fracture segment, and a key topological indicator - the minimum angle formed by the fracture segment and all adjacent fracture segments. Finally, the adjacency matrix is ​​constructed by defining the Euclidean distance between nodes to be less than the connection judgment radius, and the node feature matrix is ​​formed by combining the characteristics of the nodes themselves, thereby completely constructing structured graph data that describes the geometric morphology and topological connection of the fracture network.

[0049] Node coordinates:

[0050] ;

[0051] Represents the node space coordinate vector; Indicates the node's coordinate in the X direction; Indicates the Y coordinate of the node; Indicates the coordinate of the node in the Z direction; Represents a collection of nodes.

[0052] Edge eigenvectors:

[0053] ;

[0054] Representation node and nodes The edge eigenvectors that constitute the edge; Indicates the crack strike azimuth; represents the length of the crack segment; represents the crack curvature; Indicates the minimum angle with adjacent cracks.

[0055] Minimum angle calculation formula

[0056] ;

[0057] Where, Representation node and nodes The minimum angle between the sides; Representation node The set of neighbors of Indicates from arrive direction vector; Represents the inverse cosine function.

[0058] Step S2: Construct an adjacency matrix and a node feature matrix, and combine them with edge feature vectors to construct structured graph data that describes the geometry and topological connections of the fracture network.

[0059] Specifically, the constructed adjacency matrix and feature matrix are shown below.

[0060] Adjacency Matrix:

[0061] ;

[0062] represents the adjacency matrix element; Representation node and nodes The Euclidean distance between Indicates the connection determination radius.

[0063] Feature matrix:

[0064] ;

[0065] represents the node feature matrix; Representation node The eigenvector of Represents the node degree, i.e. the number of connected edges.

[0066] Step S3: Input the structured graph data into the graph neural network to obtain a connectivity probability graph.

[0067] Example 2

[0068] This embodiment further provides a training and optimization method for a graph neural network based on Example 1. Label preparation is based on high-precision CT scans or OpenPNM numerical simulation results, and the edges in the graph are marked as truly connected or non-connected. The loss function adopts weighted binary cross entropy, and by introducing weight coefficients, it significantly increases the penalty for prediction errors of sparse connected edges, forcing the model to pay more attention to the learning of connected samples. The data enhancement strategy includes random rotation, random translation and random scale jittering of the original crack network to enhance the robustness and generalization ability of the model to spatial transformation. Regularization adopts DropEdge technology to randomly discard some edges in the adjacency matrix in each training iteration, and prevents the model from overfitting to a specific graph structure by destroying some connection relationships, thereby improving the generalization performance of the model.

[0069] 1) Label preparation: Based on CT scanning or numerical simulation such as OpenPNM percolation simulation, the real connected paths are marked, and the proportion of non-connected edges is usually over 90%;

[0070] 2) The loss function is as follows:

[0071] ;

[0072] ;

[0073] Where, represents the total number of edges in the graph; Indicates the sum of all edges; Represents the weight coefficient of the connected edge; Represents an edge The true connectivity label of Represents an edge The predicted connectivity probability of represents the number of disconnected edges; represents the number of connected edges; log represents the natural logarithm function.

[0074] 3) Data augmentation: Random rotation (0°-360°), translation (±10% of the region size), and scale jitter (0.8-1.2 times) are applied to the original fracture network to improve the model's robustness to spatial transformations. The expression is:

[0075] ;

[0076] Indicates the crack strike angle after rotation; Indicates that the random rotation amount conforms to the uniform distribution; Represents the coordinates after translation; and Indicates that the random translation amount conforms to the uniform distribution; represents the size of the fracture network area; represents the scaled crack length; Indicates that the random scaling factor follows a uniform distribution;

[0077] 4) Regularization: Use DropEdge technology to randomly drop 15% of edges to prevent overfitting:

[0078] ;

[0079] ;

[0080] represents the adjacency matrix after random dropping; represents the original adjacency matrix; represents element-wise multiplication; represents a random mask matrix; Indicates the reserved flag of the edge (0 / 1); Indicates the edge drop probability. In this embodiment, the value is 0.15.

[0081] Example 3

[0082] This embodiment provides an improved graph neural network based on Example 1, replacing the graph neural network in Example 1. It includes three core processing layers. The first is the feature initialization layer, which normalizes the multidimensional geometric properties of the input edges and maps them into initial feature vectors, providing a standardized starting point for subsequent learning. The second is the core message passing layer, which employs a graph attention mechanism. This layer dynamically calculates the attention weights of a node on its neighbors using a multi-head attention mechanism and updates the embedding representation of its own node by weightedly aggregating the feature information of neighboring nodes. This process uses the LeakyReLU activation function and allows the model to focus on important relationships across different subspaces, effectively capturing the non-local dependencies and complex interactions in the fracture network. Finally, the edge prediction layer concatenates the final embedding vectors of the nodes at both ends of each target edge in the graph with the normalized initial geometric feature vector of the edge itself. The concatenated fused feature vector is input into a two-layer multilayer perceptron, which ultimately outputs the probability of the edge being connected using a sigmoid activation function, thereby generating a probabilistic connectivity graph for the entire fracture network.

[0083] Specifically, the message passing layer adopts the graph attention mechanism (GAT) to dynamically aggregate neighborhood information through multi-head attention weights. The calculation process is:

[0084] ;

[0085] ;

[0086] Where, node In the The embedding vector output after layer message passing; σ() represents the activation function; Representation node The set of neighbors of Representation node To neighbors Normalized attention weights of ; Indicates the The weight matrix of the attention head; LeakyReLU represents the activation function; represents the attention mechanism parameter vector; Representation node The feature embedding vector of Representation node The feature embedding vector of .

[0087] The edge prediction layer predicts each edge Calculating connectivity probability: initial features of the splicing target edge and the two end node embedding vectors and , after inputting two layers of MLP and sigmoid activation, the output probability value ∈[0,1], forming the full-graph connectivity probability matrix, which is expressed as:

[0088] ;

[0089] ;

[0090] ;

[0091] ;

[0092] Where, Represents an edge The fusion feature vector of represents the normalized edge eigenvector; is the activation function; Represents the weight matrix of the first layer of the multilayer perceptron; Represents the bias vector of the first layer of the multilayer perceptron; Represents the second layer weight matrix of the multilayer perceptron; Represents the bias vector of the second layer of the multilayer perceptron.

[0093] Example 4

[0094] This embodiment 1 provides a probabilistic graph post-processing and engineering application method based on embodiment 1. The probabilistic connectivity graph obtained by model prediction is converted into information and tools that can directly serve engineering decision-making.

[0095] Specifically, the probability graph is binarized by setting a probability threshold, and the connected domain is segmented to identify the connected subgraph components, and the largest connected subgraph is extracted as the key main seepage channel. The seepage path analysis converts the connectivity probability of each edge into a path weight. The lower the weight, the more likely the edge is to be connected and the lower the resistance. The Dijkstra algorithm is used to calculate the shortest path between any nodes on the weighted graph. Its physical meaning is the seepage path with the highest probability. The visual output generates an intuitive heat map, in which the color of the edge is rendered by linear interpolation from red to blue, clearly showing the connectivity possibility of various parts of the network; at the same time, the main seepage path arrows identified by the Dijkstra algorithm are superimposed and marked. Finally, the global connectivity rate, that is, the proportion of connected edges to the total number of edges, is calculated as the core quantitative indicator for evaluating the seepage capacity of the entire fracture network layer, providing a direct basis for engineering decision-making.

[0096] 1) Connected domain segmentation: Set the threshold τ, the default value is 0.7, identify connected components after binarizing the probability map, and extract the largest connected subgraph;

[0097] ;

[0098] Represents the binary connectivity flag, 1 indicates connected, 0 indicates disconnected; Represents an edge The predicted connectivity probability of Indicates the probability segmentation threshold, the default value is 0.7.

[0099] 2) Percolation Path Analysis: Converting Edge Probabilities into Weights

[0100] ;

[0101] Represents an edge The percolation path weight of the edge is set for the maximum connected subgraph, and then the Dijkstra algorithm is used to calculate the shortest path between any nodes on the weighted maximum connected subgraph, whose physical meaning is the main percolation path.

[0102] 3) Visualization output: Generate a fracture network heat map, with red high-probability edges and blue low-probability edges, superimpose arrows on the main seepage paths, and calculate the global connectivity C to quantify the reservoir seepage capacity:

[0103] ;

[0104] represents the global connectivity rate of the fracture network; represents the sum of the number of all connected edges, that is, = the number of 1s; represents the total number of edges in the graph; Represents the binary connectivity flag.

[0105] When expressing color, it is calculated using the following formula:

[0106] ;

[0107] Represents an edge The rendering color; Indicates red-blue color interpolation; the red end indicates high connectivity probability ; The blue end indicates low connectivity probability .

[0108] Example 5

[0109] This embodiment provides an electronic device, including: a memory, a processor, and a computer program. The computer program is stored in the memory and is configured to enable the processor to execute the method of any of the above embodiments or any combination thereof.

[0110] The technical features of the above embodiments may be combined in any manner. To simplify the description, not all possible combinations of the technical features in the above embodiments are described. Only preferred embodiments of the present invention are presented. While the description is relatively specific and detailed, it should not be construed as limiting the scope of the present invention. As long as there are no conflicts in the combination of these technical features, they should be considered to be within the scope of this specification.

[0111] It should be noted that, for those skilled in the art, various modifications and improvements can be made without departing from the scope of the present invention, and these modifications and improvements fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A method for predicting reef limestone fracture connectivity using a graph neural network, characterized by: The following steps are involved: Step S1: The intersection points of the reef limestone fracture segments are used as graph nodes, and the three-dimensional spatial coordinates of each node are recorded; the actual fracture segments between adjacent nodes are used as graph edges, and an edge feature vector is assigned to each edge; The edge feature vector at least includes: the crack strike azimuth, the crack segment length, the crack segment curvature, and the minimum angle formed by the crack segment and all adjacent crack segments; Step S2: constructing an adjacency matrix and a node feature matrix, and combining the edge feature vectors to construct structured graph data describing the geometry and topological connections of the fracture network; Step S3: Input the structured graph data into a graph neural network to obtain a connectivity probability graph.

2. The method for predicting reef limestone fracture connectivity using a graph neural network according to claim 1, characterized in that: The calculation expression of the minimum angle in step S1 is: ； Where, Representation node and nodes The minimum angle between the sides; Representation node The set of neighbors of Indicates from arrive The direction vector of Represents the inverse cosine function.

3. The method for predicting reef limestone fracture connectivity using a graph neural network according to claim 1, characterized in that: The loss function used by the graph neural network during training The expression is: ； ； Where, represents the total number of edges in the graph; Represents the weight coefficient of the connected edge; Represents an edge The true connectivity label of Represents an edge The predicted connectivity probability of represents the number of disconnected edges; Indicates the number of connected edges.

4. The method for predicting reef limestone fracture connectivity using a graph neural network according to claim 1, wherein: When the graph neural network is being trained, data enhancement is performed on the training data set, and the data enhancement includes: 1) Random rotation: Increase the crack strike angle by a random amount uniformly distributed between 0° and 360°; 2) Random translation: uniformly translate the node coordinates within the range of ±10% of the area size; 3) Random scale jitter: Multiply the crack segment length by a random factor uniformly distributed between 0.8 and 1.

2.

5. The method for predicting reef limestone fracture connectivity using a graph neural network according to claim 1, wherein: During training, the graph neural network uses DropEdge regularization to prevent overfitting: in each training iteration, the elements in the adjacency matrix are randomly set to zero and discarded with a set probability.

6. The method for predicting reef limestone fracture connectivity using a graph neural network according to claim 1, characterized in that: The graph neural network in step S3 includes: a feature initialization layer, a message passing layer, and an edge prediction layer; The feature initialization layer: normalizes the edge feature vector to obtain an initial feature vector; Message passing layer: Based on the adjacency matrix and node feature matrix, a multi-head graph attention mechanism is used to iteratively update node features and generate node embedding vectors; Edge prediction layer: After concatenating the embedding vectors of the nodes at both ends of the target edge with the initial edge feature vector of the target edge, the connectivity probability of the target edge is obtained by passing it through two layers of multi-layer perceptrons (MLPs) and sigmoid functions.

7. The method for predicting reef limestone fracture connectivity using a graph neural network according to claim 6, characterized in that: The calculation expression of the message transmission layer is: ； ； Where, node In the The embedding vector output after layer message passing; σ() represents the activation function; Representation node The set of neighbors of Representation node To neighbors Normalized attention weights of ; Indicates the The weight matrix of the attention head; LeakyReLU represents the activation function; represents the attention mechanism parameter vector; Representation node The feature embedding vector of Representation node The feature embedding vector of .

8. The method for predicting reef limestone fracture connectivity using a graph neural network according to claim 7, characterized in that: The calculation expression of the edge prediction layer is: ； ； ； ； Where, Represents an edge The fusion feature vector of Represents an edge The connectivity probability of represents the normalized edge eigenvector; is the activation function; Represents the weight matrix of the first layer of the multilayer perceptron; Represents the bias vector of the first layer of the multilayer perceptron; Represents the second layer weight matrix of the multilayer perceptron; Represents the bias vector of the second layer of the multilayer perceptron.

9. The method for predicting reef limestone fracture connectivity using a graph neural network according to claim 1, characterized in that: The method further comprises the following steps: Step S4: binarizing the connectivity probability graph, setting a threshold to segment the connectivity domain, and extracting the largest connected subgraph; Step S5: converting the edge connectivity probability into a percolation path weight and assigning it to the maximum connected subgraph, and then using the Dijkstra algorithm to calculate the shortest path between any nodes on the weighted maximum connected subgraph to obtain the main percolation path; Step S6: Generate a fracture network heat map based on the fracture connectivity probability and mark the main seepage path for visualization; calculate the global connectivity ratio to quantify the reservoir seepage capacity.

10. An electronic device comprising: A memory, a processor and a computer program, characterized in that: the computer program is stored in the memory and is configured to be executed by the processor to implement the method according to any one of claims 1 to 9.

Images