Training method, simulation method, device and medium of simulation model of explosive fracturing fracture propagation

Abstract

The application discloses a training method, a simulation method, equipment and a medium of a simulation model of combustion fracturing crack propagation. The training method of the simulation model of combustion fracturing crack propagation based on a graph neural network comprises the following steps: obtaining combustion fracturing crack propagation simulation parameters; generating a triangular mesh geometric model according to the geometric model parameters; performing numerical simulation of combustion fracturing crack propagation according to the generated triangular mesh geometric model and the combustion fracturing engineering parameters to obtain displacement data and crack state data of the triangular mesh geometric model at multiple time steps; constructing a graph structure and determining features according to the displacement data and the crack state data of the triangular mesh geometric model at the multiple time steps; taking graph data of the graph structure as a training data set according to the constructed graph structure; and training a first graph neural network model constructed in advance according to the training data set.

Description

Technical Field

[0001] This invention relates to the field of fracturing technology, and in particular to a training method, simulation method, equipment, and medium for a simulation model of fission-explosive fracturing fracture propagation. Background Technology

[0002] Hydraulic fracturing is currently the most widely used and mature technology for shale reservoir stimulation. However, it also faces challenges such as difficulty in creating good seepage channels, water scarcity, water locking, clay expansion, and severe reservoir contamination. Therefore, the use of in-situ methane combustion-explosion has been proposed to reduce reservoir contamination and promote the propagation of multiple fractures that are not controlled by stress. As a novel method for shale gas reservoir stimulation, methane combustion-explosion fracturing still presents many difficulties and challenges in its development model and parameter optimization, requiring further investigation.

[0003] In shale oil and gas reservoirs, the purpose of traditional hydraulic fracturing for production enhancement is to maximize oil and gas production by creating an effective reservoir stimulation volume (SRV) around the well. In combustion-explosion fracturing operations, the fracturing effect is evaluated by calculating the range of the fracture zone generated near the combustion point. However, many factors affect the fracturing effect, including reservoir geological static parameters, fracturing operation parameters, and production dynamic parameters. The relationships between these parameters are complex, and the relationship between numerous parameters and the fracturing effect is not a simple function, making it difficult to express the relationship with a single expression.

[0004] Typically, specific numerical calculation methods (such as the finite element method and the phase field method) are used to establish simulation models for the propagation of fractures in combustion-explosion fracturing. For example, in related existing technologies, a numerical model for methane combustion-explosion fracturing has been established based on the continuous-discontinuous element method (CDEM), combined with the Landau blast source model and the linear elastic tension-shear composite fracture constitutive model. This model is used to simulate the process of methane combustion-explosion fracturing in vertical wells (including fracture initiation and propagation). However, although this CDEM-based numerical model for combustion-explosion fracturing can basically complete the simulation accurately and has a certain degree of correctness, it suffers from problems such as long simulation time per numerical value and low computational efficiency.

[0005] For example, existing technologies employ machine learning to optimize parameters for combustion-explosion fracturing engineering. First, the engineering parameters and evaluation indices (fracture degree and modified area) are obtained using the CDEM-based numerical simulation software GDEM. Then, data normalization is used to preprocess the raw data. Finally, machine learning methods are applied to construct a hybrid surrogate model for predicting the evaluation indices of combustion-explosion fracturing engineering. While this approach directly obtains the evaluation indices of combustion-explosion fracturing engineering through the surrogate model, thus determining the optimal parameters more quickly, it only predicts the evaluation indices and cannot simulate the evolution of the fractures, lacking a degree of intuitiveness.

[0006] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0007] The main objective of this invention is to provide a training method, simulation method, equipment, and medium for a simulation model of combustion-explosive fracturing crack propagation, aiming to solve the above-mentioned problems.

[0008] To achieve the above objectives, this invention provides a training method for a simulation model of combustion-explosion fracturing crack propagation based on graph neural networks, comprising:

[0009] Obtain simulation parameters for fracture propagation in combustion-explosion fracturing, which include geometric model parameters and combustion-explosion fracturing engineering parameters;

[0010] Generate a triangular mesh geometric model based on the geometric model parameters;

[0011] Based on the generated triangular mesh geometric model and the combustion-explosion fracturing engineering parameters, numerical simulation of the propagation of combustion-explosion fracturing fractures is performed to obtain displacement data and fracture state data of the triangular mesh geometric model at multiple time steps.

[0012] Based on the displacement data and crack state data of the triangular mesh geometric model at multiple time steps, a graph structure and feature determination are constructed, wherein the vertices on the triangular mesh model are used as nodes of the graph structure, and the edges on the triangular mesh model are used as edges of the graph structure.

[0013] Based on the constructed graph structure, the graph data of the graph structure is used as the training dataset;

[0014] The first graph neural network model is trained based on the training dataset.

[0015] Preferably, in the training method of the simulation model of combustion and explosion fracturing crack propagation based on graph neural network, in the step of constructing graph structure and determining features based on displacement data and crack state data of triangular mesh geometric model at multiple time steps, the features of the nodes of the graph structure include vertex coordinates, vertex displacement, distance between vertex and combustion center, vector cosine value of vertex and nearest crack edge, distance between vertex and nearest crack edge, vector cosine value of vertex and nearest crack edge, distance between vertex and nearest natural crack, and vertex displacement vector cosine value.

[0016] The features of the edges in the graph structure include the crack state data, the node index numbers of the two vertices of the edge, the length of the edge, the absolute value of the cosine of the edge's angle, and the natural crack state of the edge.

[0017] Preferably, in the training method of the simulation model of combustion and explosion fracturing crack propagation based on graph neural network, the distance between the vertex and the combustion and explosion center is the Euclidean distance between the vertex coordinates and the combustion and explosion center coordinates;

[0018] The method for calculating the cosine value of the vector between the vertex and the explosion center is to form a first vector with the explosion center as the starting point and the vertex as the ending point, and then calculate the cosine value of the angle between the first vector and the positive direction of the X-axis.

[0019] The method for calculating the vector cosine value between the vertex and the nearest crack edge is to take the vertex coordinates as the starting point and the midpoint coordinates of the nearest crack edge as the ending point, and finally form a second vector, and calculate the cosine value of the angle between the second vector and the positive direction of the X-axis.

[0020] The method for calculating the distance between the vertex and the nearest crack edge is as follows: the crack edge is replaced by its midpoint coordinates; the Euclidean distance between the vertex and all crack edges at the current time step is calculated, and only the vertex with the closest distance is retained;

[0021] The method for calculating the vector cosine value between the vertex and the nearest crack edge is to take the vertex coordinates as the starting point and the midpoint coordinates of the nearest crack edge as the ending point, and finally form a third vector, and calculate the cosine value of the angle between the third vector and the positive direction of the X-axis.

[0022] The method for calculating the distance between the vertex and the nearest natural crack is to calculate the Euclidean distance between the vertex and all line segments, and select the distance corresponding to the nearest natural crack.

[0023] The method for calculating the cosine value of the vertex displacement vector is to calculate the cosine value of the angle between the vertex displacement vector and the positive direction of the X-axis;

[0024] The method for determining the node index number of the two vertices of the edge is that when storing the graph structure, the node information of the graph structure is stored in a specific order, and the index number can determine the relationship between the nodes and edges of the graph structure.

[0025] The length of the edge is calculated by directly calculating the Euclidean distance between the two vertices of the edge;

[0026] The method for calculating the absolute value of the angle cosine of the side is to form a fourth vector naturally from the two vertices of the side, calculate the cosine of the angle between this vector and the positive direction of the X-axis, and then process the absolute value of this cosine.

[0027] The method for determining the natural crack state of the edge is to calculate the nearest distance between the midpoint of the edge and the natural crack line segment. When the nearest distance is less than a threshold, the edge is marked as a natural crack.

[0028] Preferably, in the training method of the simulation model of combustion-explosion fracturing crack propagation based on graph neural networks, before the step of training the pre-constructed first graph neural network model according to the training dataset, the training method further includes:

[0029] A first graph neural network model is established based on equivariant graph neural networks and message passing neural networks. The first graph neural network model includes a first linear layer, an aggregation layer, a connection layer, a second linear layer, and a first multilayer perceptron layer connected in sequence. The first graph neural network model also includes a third linear layer, an attention layer, a second multilayer perceptron layer, and a softmax layer connected in sequence, wherein the third linear layer is connected to the connection layer, and the second linear layer is connected to the attention layer.

[0030] Accordingly, training the pre-built first graph neural network model based on the training dataset includes:

[0031] The node features and edge features of the training dataset are input into the first linear layer and the second linear layer, respectively, to form their own independent node feature representations and edge feature representations;

[0032] The node feature representation is then aggregated through an aggregation layer, and the aggregated node feature representation is updated through the edge feature representation and connection layer. The updated node features are then obtained through the second linear layer.

[0033] The updated node features are used to output the predicted node displacement through the first multilayer perceptron layer; the updated node features and the edge feature representation are used to obtain the updated edge feature representation through the attention layer and the second multilayer perceptron layer; the updated edge feature representation is used to output the predicted crack state through the softmax layer.

[0034] Preferably, in the training method of the simulation model of combustion-explosion fracturing crack propagation based on graph neural networks, after the step of using the graph data of the constructed graph structure as the training dataset, and before the step of training the pre-constructed first graph neural network model based on the training dataset, the training method further includes:

[0035] The training dataset is preprocessed, wherein the preprocessing includes:

[0036] The node coordinates remain unchanged;

[0037] The formula for processing nodal displacement data is:

[0038] x′=ax

[0039] The length of the side and the cosine value of the angle are standardized.

[0040]

[0041] Where x represents the node displacement data before preprocessing;

[0042] x' represents the preprocessed node displacement data;

[0043] y represents the length of the edge or the cosine of the angle before preprocessing;

[0044] y' is the length of the preprocessed edge or the cosine value of the angle;

[0045] 'a' is the scaling factor;

[0046] μ is the mean of the corresponding feature;

[0047] σ represents the variance of the corresponding feature.

[0048] Preferably, in the training method of the simulation model of combustion-explosion fracturing crack propagation based on graph neural networks, the step of training the pre-constructed first graph neural network model according to the training dataset includes:

[0049] The first graph neural network model is trained by taking the node or edge of the current time step as input and the displacement data or crack state data of the next time step as output.

[0050] To achieve the above objectives, the present invention also provides a simulation method for combustion-explosion fracturing fracture propagation based on graph neural networks, comprising:

[0051] The training method of the above-mentioned simulation model of combustion-explosion fracturing crack propagation based on graph neural network is used to train the simulation model for simulation.

[0052] To achieve the above objectives, the present invention also provides a training device for a simulation model of combustion-explosion fracturing crack propagation based on a graph neural network, comprising:

[0053] The acquisition unit is used to acquire simulation parameters of fibrillation-explosion fracturing crack propagation, which include geometric model parameters and fibrillation-explosion fracturing engineering parameters.

[0054] A generation unit is used to generate a triangular mesh geometric model based on the geometric model parameters;

[0055] The simulation unit is used to perform numerical simulation of the propagation of the combustion-explosion fracturing crack based on the generated triangular mesh geometric model and the combustion-explosion fracturing engineering parameters, and to obtain displacement data and crack state data of the triangular mesh geometric model at multiple time steps.

[0056] A construction unit is used to construct a graph structure and determine features based on displacement data and crack state data of a triangular mesh geometric model at multiple time steps, wherein the vertices on the triangular mesh model are used as nodes of the graph structure, and the edges on the triangular mesh model are used as edges of the graph structure.

[0057] The selection unit is used to select the graph data of the constructed graph structure as the training dataset.

[0058] The training unit is used to train a pre-built first graph neural network model based on the training dataset.

[0059] To achieve the above objectives, the present invention also provides a computer device, comprising:

[0060] At least one processor; and,

[0061] A memory communicatively connected to the at least one processor; wherein,

[0062] The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the training method of the above-described simulation model of combustion-explosion fracturing crack propagation based on graph neural networks, and / or the above-described simulation method of combustion-explosion fracturing crack propagation based on graph neural networks.

[0063] To achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the training method for the simulation model of combustion-explosion fracturing crack propagation based on graph neural networks, and / or the simulation method for combustion-explosion fracturing crack propagation based on graph neural networks.

[0064] The present invention has the following beneficial effects:

[0065] This invention obtains simulation parameters for the propagation of fractures in combustion-explosion fracturing, including geometric model parameters and engineering parameters. Based on the geometric model parameters, a triangular mesh geometric model is generated. Numerical simulations of fracture propagation are performed using the generated triangular mesh geometric model and the engineering parameters to obtain displacement data and fracture state data for the triangular mesh geometric model at multiple time steps. Based on the displacement data and fracture state data of the triangular mesh geometric model at multiple time steps, a graph structure and feature determination are constructed, where vertices on the triangular mesh model are used as nodes of the graph structure, and edges on the triangular mesh model are used as edges of the graph structure. The graph data of the constructed graph structure is used as a training dataset. A pre-constructed first graph neural network model is trained using the training dataset. This improves simulation speed and computational efficiency.

[0066] Furthermore, by using graph neural network technology, the original data is constructed into a graph and trained into a graph neural network model, which can replace part of the numerical simulation of crack propagation. This allows the graph neural network model to meet the accuracy requirements of combustion and explosion fracturing crack propagation simulation while significantly improving the simulation speed.

[0067] Furthermore, this invention obtains displacement data and fracture state data of a triangular mesh geometric model with multiple time steps through numerical simulation of fission-induced fracturing fracture propagation. Then, it combines graph structure and a first graph neural network model for training to complete the final simulation. This results in more accurate simulation results and higher computational efficiency. In contrast, using numerical simulation alone requires tens of thousands of meshes and relatively small time steps, typically requiring at least several hours or even more than ten hours to simulate the results. However, this invention, through numerical simulation followed by training with graph structure and a first graph neural network model, obtains a model that can simulate results in about ten minutes.

[0068] Furthermore, this invention successfully avoids the high computational costs of numerical simulation, significantly reducing computational complexity and thus greatly improving the efficiency of crack morphology evolution simulation. This improvement not only saves computational resources but also accelerates the simulation process, making the prediction and analysis of crack morphology evolution more efficient and accurate. Attached Figure Description

[0069] Figure 1 A visual representation of the results generated from the raw data of the numerical simulation.

[0070] Figure 2 This is a schematic diagram illustrating the calculation of the distance between the vertex and the explosion center and the vector cosine value in this invention;

[0071] Figure 3 This is a schematic diagram illustrating the calculation of the distance between the vertex and the nearest crack and the vector cosine value in this invention;

[0072] Figure 4 This is a schematic diagram illustrating the calculation of the distance between the vertex and the nearest natural crack in this invention;

[0073] Figure 5 This is a schematic diagram of the vertex displacement vector of the present invention;

[0074] Figure 6 This is a schematic diagram of a local structure of the triangular mesh model of the present invention;

[0075] Figure 7 This is a schematic diagram of the neural network model of the first figure of the present invention;

[0076] Figure 8 This is a schematic diagram of the training device for the fault diagnosis model of the present invention in a first embodiment.

[0077] Figure 9 This is a schematic diagram of the computer device of the present invention in a first embodiment.

[0078] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0079] In this embodiment of the invention, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0080] It should be noted that the terms "first," "second," etc., in the specification, claims, and drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0081] In this embodiment of the invention, the term "multiple" refers to two or more, and other quantifiers are similar.

[0082] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details are presented in the embodiments of the present invention to facilitate a better understanding of the invention. However, the technical solutions claimed in the present invention can be implemented even without these technical details and various variations and modifications based on the following embodiments. The division of the following embodiments is for ease of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined with and referenced by each other without contradiction.

[0083] To address the aforementioned issues, this embodiment relates to a training method for a simulation model of combustion-explosion fracturing crack propagation based on graph neural networks. This method can be applied to computer devices, such as desktop computers, tablets, laptops, and other electronic devices with data processing capabilities. In other embodiments, it can also be other electronic devices with data processing capabilities, without any specific limitations.

[0084] The following describes the implementation details of the training method for the simulation model of combustion-explosion fracturing crack propagation based on graph neural networks according to the first embodiment of the present invention. The following implementation details are provided for ease of understanding only and are not necessary for implementing this solution.

[0085] The specific process of this implementation method is as follows: Figure 1 As shown, it specifically includes:

[0086] Step S100: Obtain simulation parameters for the propagation of fibrillation fractures, which include geometric model parameters and fibrillation engineering parameters.

[0087] It should be understood that geometric model parameters include the model's length and height, wellbore radius, natural fractures, etc. Explosive fracturing engineering parameters include: elastic modulus, Poisson's ratio, blast source density, detonation velocity, heat of explosion, etc.

[0088] Step S200: Generate a triangular mesh geometric model based on the geometric model parameters;

[0089] Step S300: Based on the generated triangular mesh geometric model and the combustion-explosion fracturing engineering parameters, perform numerical simulation of the propagation of combustion-explosion fracturing cracks to obtain displacement data and crack state data of the triangular mesh geometric model at multiple time steps.

[0090] It should be understood that numerical simulation of combustion-explosion fracturing fracture propagation can be performed, but is not limited to, by constructing a combustion-explosion fracturing fracture propagation model using CDEM (Discrete Element Method based on Continuum Mechanics) provided by GDEM software and completing the corresponding numerical simulation.

[0091] After completing the numerical simulation using CDEM, displacement data and crack status data of the triangular mesh geometric model at several time steps can be obtained. The displacement data is stored as the vertex coordinates and corresponding displacement data of each element in the triangular mesh geometric model (sample data is shown in Table 1); the crack status data is stored as the midpoint coordinates of each edge in the triangular mesh geometric model and the crack status of the edge (a crack status value of 0 indicates that the edge is not a crack, and a crack status value of 1 indicates that it is a crack) (sample data is shown in Table 2). Visualizing the numerically calculated displacement data and crack status data yields... Figure 1 .exist Figure 1 In the diagram, the central black circular area represents the combustion-fractured wellbore, with the center of the combustion being the center of the wellbore; the white line segments represent the fractures generated by the combustion-fracture or natural fractures, while the other colors represent the numerical distribution of displacement values ​​of each vertex in the X direction in the triangular mesh model.

[0092] Table 1. Examples of displacement data from numerical simulation of combustion-explosion fracturing.

[0093]

[0094] Table 2. Examples of fracture state data from numerical simulation of combustion-explosion fracturing.

[0095]

[0096] Step S400: Based on the displacement data and crack state data of the triangular mesh geometric model at multiple time steps, construct the graph structure and determine the features, wherein the vertices on the triangular mesh model are used as nodes of the graph structure, and the edges on the triangular mesh model are used as edges of the graph structure.

[0097] Specifically, in step S400, the features of the nodes of the graph structure include vertex coordinates, vertex displacement, distance between the vertex and the explosion center, vector cosine value of the vertex and the nearest crack edge, distance between the vertex and the nearest crack edge, vector cosine value of the vertex and the nearest crack edge, distance between the vertex and the nearest natural crack, and vertex displacement vector cosine value; the features of the edges of the graph structure include crack state data, node index numbers of the two vertices of the edge, edge length, absolute value of the angle cosine value of the edge, and natural crack state of the edge.

[0098] in,

[0099] The distance between the vertex and the explosion center: the Euclidean distance between the vertex coordinates and the explosion center coordinates (the explosion center coordinates are fixed during a continuous numerical simulation); such as Figure 2 The line segment is thickened.

[0100] The cosine of the vector between the vertex and the explosion center: A first vector is formed with the explosion center as the starting point and the vertex as the ending point. The cosine of the angle between this first vector and the positive X-axis is calculated. Figure 2 As shown.

[0101] The cosine of the vector between the vertex and the nearest crack edge: taking the vertex coordinates as the starting point and the midpoint coordinates of the nearest crack edge as the ending point, a second vector is formed, and the cosine of the angle between this second vector and the positive X-axis is calculated; for example... Figure 3 As shown.

[0102] The distance between the vertex and the nearest crack edge: the crack edge will be replaced with its midpoint coordinates (this also simplifies the calculation); calculate the Euclidean distance between the vertex and all crack edges at the current time step, and only retain the vertex with the closest distance; such as Figure 3 The line segment is thickened.

[0103] The cosine value of the vector between the vertex and the nearest crack edge: taking the vertex coordinates as the starting point and the midpoint coordinates of the nearest crack edge as the ending point, a third vector is finally formed, and the cosine value of the angle between the third vector and the positive direction of the X-axis is calculated.

[0104] The distance between the vertex and the nearest natural crack: The numerical simulation did not directly include the natural crack state of the edges, but only the coordinates of the starting and ending points of the natural cracks in the triangular mesh model (i.e., the coordinates of the two endpoints of the line segment). Therefore, the Euclidean distances between the vertex and all line segments are calculated, and the distance corresponding to the nearest natural crack is selected; for example... Figure 4 The bolded line segment shown represents the distance between the vertex and the nearest natural crack.

[0105] The cosine value of the vertex displacement vector: calculate the cosine of the angle between the vertex displacement vector and the positive X-axis; as shown in Table 1, the vertex displacement data is in vector form, therefore the displacement data is enhanced. For example... Figure 5 As shown, the arrowed lines represent vertex displacement vectors.

[0106] The node index numbers of the two vertices of the edge: When storing the graph structure, the node information of the graph structure is stored in a specific order, and the index number can determine the relationship between the nodes and edges of the graph structure.

[0107] The length of the edge is calculated by directly calculating the Euclidean distance between the two vertices of the edge.

[0108] The absolute value of the angle cosine of the side: The two vertices of the side naturally form a fourth vector. Calculate the cosine of the angle between this vector and the positive direction of the X-axis, and process the absolute value of this cosine.

[0109] The natural crack status of the edge is determined by calculating the nearest distance between the midpoint of the edge and the natural crack segment. When the nearest distance is less than a threshold, the edge is marked as a natural crack. In this embodiment, the threshold can be 1e-6.

[0110] Step S500: Based on the constructed graph structure, use the graph data of the graph structure as the training dataset;

[0111] It should be understood that after step S500 and before step S600, the training method further includes step S700.

[0112] Step S700: Preprocess the training dataset, wherein the preprocessing includes:

[0113] The node coordinates remain unchanged;

[0114] The formula for processing nodal displacement data is:

[0115] x′=ax

[0116] The length of the side and the cosine value of the angle are standardized.

[0117]

[0118] Where x represents the node displacement data before preprocessing;

[0119] x' represents the preprocessed node displacement data;

[0120] y represents the length of the edge or the cosine of the angle before preprocessing;

[0121] y' is the length of the preprocessed edge or the cosine value of the angle;

[0122] 'a' is the scaling factor;

[0123] μ is the mean of the corresponding feature;

[0124] σ represents the variance of the corresponding feature.

[0125] Step S600: Train the pre-built first graph neural network model based on the training dataset.

[0126] In practice, the nodes or edges at the current time step are used as input, and the displacement data or crack state data at the next time step are used as output to train the pre-built first graph neural network model. For the graph structure at each time step, both its nodes and edges include label data. The label data of a node is the displacement data of the node at the next time step, and the label data of an edge is the crack state data at the next time step.

[0127] It should be understood that prior to step S600, the training method further includes:

[0128] Based on the equivariant graph neural network and the message-passing neural network, a first graph neural network model is established; such as... Figure 7 As shown, the first graph neural network model includes a first linear layer, an aggregation layer, a connection layer, a second linear layer, and a first multilayer perceptron layer connected in sequence; the first graph neural network model also includes a third linear layer, an attention layer, a second multilayer perceptron layer, and a softmax layer connected in sequence, wherein the third linear layer is connected to the connection layer, and the second linear layer is connected to the attention layer;

[0129] Accordingly, step S600 includes:

[0130] Step S610: Input the node features and edge features of the training dataset into the first linear layer and the second linear layer respectively to form their own independent node feature representations and edge feature representations;

[0131] It should be understood that the node features and edge features (edge ​​features do not include node index information) of the training dataset are passed through the first linear layer and the second linear layer (multiple linear layers can be stacked) respectively, without considering the relationship between nodes and edges, to obtain their own independent node feature representations and edge feature representations.

[0132] Step S620: The node feature representation is aggregated through an aggregation layer, the aggregated node feature representation is updated through the edge feature representation and the connection layer, and the updated node features are obtained through the second linear layer;

[0133] It should be understood that the node feature representation is updated using node and edge feature representations through aggregation and concatenation layers. The update method can be any conventional method and is not specifically limited here. The updated node feature representation can be used to update the edge feature representation, or as input to the next graph neural network (GNN) layer. The node features updated by the last GNN layer are then used to predict node displacement data through the first multilayer perceptron (MLP) layer.

[0134] Step S630: The updated node features are output as predicted node displacements through the first multilayer perceptron layer; the updated node features and the edge feature representations are passed through the attention layer and the second multilayer perceptron layer to obtain the updated edge feature representations; the updated edge feature representations are output as predicted crack states through the softmax layer.

[0135] It should be understood that during the edge feature representation update stage, the attention layer is used to learn the attention value of each component in the edge feature representation vector to obtain the updated edge feature representation (the update method can be similar to the edge feature update method of the Graph Equivalent Neural Network (EGNN), which extends the node update method to more dimensions. The updated edge feature representation can also be used as the input of the next GNN layer. The updated edge feature representation after the last GNN layer also utilizes the second multilayer perceptron layer (MLP layer) and the Softmax layer to predict the crack state at the next time step).

[0136] Additionally, it should be noted that the Adam optimizer can be used for iteration during the iteration process. For node displacement prediction and edge crack state prediction, mean square error (MSE) and binary cross-entropy (BCE) are used for optimization, respectively. The above iteration process ends when the model converges or the loss change is less than a certain threshold, thus obtaining a graph neural network model suitable for combustion and explosion fracturing simulation of the training dataset.

[0137] To achieve the above objectives, the present invention also provides a simulation method for combustion-explosion fracturing fracture propagation based on graph neural networks, comprising:

[0138] The training method of the simulation model of combustion-explosion fracturing crack propagation based on graph neural network described above is used to train the simulation model for simulation.

[0139] To achieve the above objectives, the present invention also provides a training device for a simulation model of combustion-explosion fracturing fracture propagation based on a graph neural network, such as... Figure 8As shown, the training device for the simulation model of combustion-explosion fracturing crack propagation based on graph neural networks includes:

[0140] The acquisition unit 810 is used to acquire simulation parameters of fibrillation-explosion fracturing crack propagation, which include geometric model parameters and fibrillation-explosion fracturing engineering parameters.

[0141] The generation unit 820 is used to generate a triangular mesh geometric model based on the geometric model parameters;

[0142] The simulation unit 830 is used to perform numerical simulation of the propagation of the combustion-explosion fracturing crack based on the generated triangular mesh geometric model and the combustion-explosion fracturing engineering parameters, and to obtain displacement data and crack state data of the triangular mesh geometric model at multiple time steps.

[0143] The construction unit 840 is used to construct a graph structure and determine features based on the displacement data and crack state data of the triangular mesh geometric model at multiple time steps, wherein the vertices on the triangular mesh model are used as nodes of the graph structure, and the edges on the triangular mesh model are used as edges of the graph structure.

[0144] Selecting unit 850 is used to use the graph data of the constructed graph structure as a training dataset according to the constructed graph structure.

[0145] Training unit 860 is used to train a pre-built first graph neural network model based on the training dataset.

[0146] To achieve the above objectives, the present invention also provides a computer device, such as... Figure 9 As shown, the computer device includes at least one processor 901; and a memory 902 communicatively connected to the at least one processor 901; wherein the memory 902 stores instructions executable by the at least one processor 901, the instructions being executed by the at least one processor 901 to enable the at least one processor 901 to execute the training method of the above-described simulation model of combustion-explosion fracturing crack propagation based on graph neural networks, and / or the above-described simulation method of combustion-explosion fracturing crack propagation based on graph neural networks.

[0147] The memory 902 and processor 901 are connected via a bus, which can include any number of interconnecting buses and bridges. The bus connects various circuits of one or more processors 901 and memory 902. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 901 is transmitted over a wireless medium via an antenna, which further receives data and transmits it to processor 901.

[0148] Processor 901 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory 902 can be used to store data used by processor 901 during operation.

[0149] To achieve the above objectives, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the training method for the simulation model of combustion-explosion fracturing crack propagation based on graph neural networks, and / or the simulation method for combustion-explosion fracturing crack propagation based on graph neural networks.

[0150] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0151] Obviously, the embodiments described above are merely some, not all, embodiments of the present invention. Based on the embodiments of the present invention, those skilled in the art can make other variations or modifications without creative effort, and all such variations or modifications should fall within the scope of protection of the present invention.

Claims

1. A training method for a simulation model of combustion-explosion fracturing crack propagation based on graph neural networks, characterized in that, include: Obtain simulation parameters for fracture propagation in combustion-explosion fracturing, which include geometric model parameters and combustion-explosion fracturing engineering parameters; Generate a triangular mesh geometric model based on the geometric model parameters; Based on the generated triangular mesh geometric model and the combustion-explosion fracturing engineering parameters, a numerical simulation of the propagation of combustion-explosion fracturing fractures is performed to obtain displacement data and fracture state data of the triangular mesh geometric model at multiple time steps. The displacement data is stored as the vertex coordinates and corresponding displacement data of each element in the triangular mesh geometric model; the fracture state data is stored as the midpoint coordinates and fracture state of each edge in the triangular mesh geometric model. Based on displacement data and crack state data of a triangular mesh geometric model at multiple time steps, a graph structure is constructed, and the characteristics of the nodes and edges of the graph structure are determined. The vertices of the triangular mesh geometric model are used as nodes of the graph structure, and the edges of the triangular mesh geometric model are used as edges of the graph structure. The characteristics of the nodes of the graph structure include vertex coordinates, vertex displacement, distance between the vertex and the explosion center, vector cosine value of the vertex and the nearest crack edge, distance between the vertex and the nearest crack edge, vector cosine value of the vertex and the nearest natural crack, and vertex displacement vector cosine value. The characteristics of the edges of the graph structure include the crack state data, node index numbers of the two vertices of the edge, edge length, absolute value of the angle cosine value of the edge, and the natural crack state of the edge. Based on the constructed graph structure, the graph data of the graph structure is used as the training dataset; The first pre-built graph neural network model is trained based on the training dataset, where graph data at one time step is used as input and graph data at the next time step is used as output.

2. The training method for the simulation model of combustion-explosion fracturing fracture propagation based on graph neural networks as described in claim 1, characterized in that, The distance between the vertex and the explosion center is the Euclidean distance between the vertex coordinates and the explosion center coordinates; The method for calculating the cosine value of the vector between the vertex and the explosion center is to form a first vector with the explosion center as the starting point and the vertex as the ending point, and then calculate the cosine value of the angle between the first vector and the positive direction of the X-axis. The method for calculating the vector cosine value between the vertex and the nearest crack edge is to take the vertex coordinates as the starting point and the midpoint coordinates of the nearest crack edge as the ending point, and finally form a second vector, and calculate the cosine value of the angle between the second vector and the positive direction of the X-axis. The method for calculating the distance between the vertex and the nearest crack edge is as follows: the crack edge is replaced by its midpoint coordinates; the Euclidean distance between the vertex and all crack edges at the current time step is calculated, and only the vertex with the closest distance is retained; The method for calculating the vector cosine value between the vertex and the nearest crack edge is to take the vertex coordinates as the starting point and the midpoint coordinates of the nearest crack edge as the ending point, and finally form a third vector, and calculate the cosine value of the angle between the third vector and the positive direction of the X-axis. The method for calculating the distance between the vertex and the nearest natural crack is to calculate the Euclidean distance between the vertex and all line segments, and select the distance corresponding to the nearest natural crack. The method for calculating the cosine value of the vertex displacement vector is to calculate the cosine value of the angle between the vertex displacement vector and the positive direction of the X-axis; The method for determining the node index number of the two vertices of the edge is that when storing the graph structure, the node information of the graph structure is stored in a specific order, and the index number can determine the relationship between the nodes and edges of the graph structure. The length of the edge is calculated by directly calculating the Euclidean distance between the two vertices of the edge; The method for calculating the absolute value of the angle cosine of the side is to form a fourth vector naturally from the two vertices of the side, calculate the cosine of the angle between this vector and the positive direction of the X-axis, and then process the absolute value of this cosine. The method for determining the natural crack state of the edge is to calculate the nearest distance between the midpoint of the edge and the natural crack line segment. When the nearest distance is less than a threshold, the edge is marked as a natural crack.

3. The training method for the simulation model of combustion-explosion fracturing crack propagation based on graph neural networks as described in claim 2, characterized in that, Before the step of training the pre-constructed first graph neural network model based on the training dataset, the training method further includes: A first graph neural network model is established based on equivariant graph neural networks and message passing neural networks. The first graph neural network model includes a first linear layer, an aggregation layer, a connection layer, a second linear layer, and a first multilayer perceptron layer connected in sequence. The first graph neural network model also includes a third linear layer, an attention layer, a second multilayer perceptron layer, and a softmax layer connected in sequence, wherein the third linear layer is connected to the connection layer, and the second linear layer is connected to the attention layer. Accordingly, training the pre-built first graph neural network model based on the training dataset includes: The node features and edge features of the training dataset are input into the first linear layer and the second linear layer, respectively, to form their own independent node feature representations and edge feature representations; The node feature representation is aggregated through an aggregation layer, and the aggregated node feature representation is updated through the edge feature representation and the connection layer. The updated node features are obtained through the second linear layer. The updated node features are used to output the predicted node displacement through the first multilayer perceptron layer; the updated node features and the edge feature representation are used to obtain the updated edge feature representation through the attention layer and the second multilayer perceptron layer; the updated edge feature representation is used to output the predicted crack state through the softmax layer.

4. The training method for the simulation model of combustion-explosion fracturing crack propagation based on graph neural networks as described in claim 1, characterized in that, After the step of using the graph data of the constructed graph structure as a training dataset, and before the step of training the pre-constructed first graph neural network model based on the training dataset, the training method further includes: The training dataset is preprocessed, wherein the preprocessing includes: The node coordinates remain unchanged; The formula for processing nodal displacement data is: The length of the side and the cosine value of the angle are standardized. ； Where x represents the node displacement data before preprocessing; x' represents the preprocessed node displacement data; y represents the length of the edge or the cosine of the angle before preprocessing; y' is the length of the preprocessed edge or the cosine value of the angle; 'a' is the scaling factor; μ is the mean of the corresponding feature; σ represents the variance of the corresponding feature.

5. The training method for the simulation model of combustion-explosion fracturing crack propagation based on graph neural networks as described in claim 1, characterized in that, The step of training a pre-built first graph neural network model based on a training dataset includes: The first graph neural network model is trained by taking the node or edge of the current time step as input and the displacement data or crack state data of the next time step as output.

6. A simulation method for combustion-explosion fracturing fracture propagation based on graph neural networks, characterized in that, include: The simulation model for combustion-explosion fracturing crack propagation based on graph neural networks, as described in any one of claims 1 to 5, is trained and used for simulation.

7. A training device for a simulation model of combustion-explosion fracturing crack propagation based on graph neural networks, characterized in that, include: The acquisition unit is used to acquire simulation parameters of fibrillation-explosion fracturing crack propagation, which include geometric model parameters and fibrillation-explosion fracturing engineering parameters. A generation unit is used to generate a triangular mesh geometric model based on the geometric model parameters; The simulation unit is used to perform numerical simulation of the propagation of the combustion-explosion fracturing fracture based on the generated triangular mesh geometric model and the combustion-explosion fracturing engineering parameters, and to obtain displacement data and fracture state data of the triangular mesh geometric model at multiple time steps. The displacement data is stored as the vertex coordinates and corresponding displacement data of each cell in the triangular mesh geometric model; the fracture state data is stored as the midpoint coordinates and fracture state of each edge in the triangular mesh geometric model. A construction unit is used to construct a graph structure and determine the features of the nodes and edges of the graph structure based on displacement data and crack state data of a triangular mesh geometric model at multiple time steps. The vertices of the triangular mesh geometric model are used as nodes of the graph structure, and the edges of the triangular mesh geometric model are used as edges of the graph structure. The features of the nodes of the graph structure include vertex coordinates, vertex displacement, distance between the vertex and the explosion center, vector cosine value of the vertex and the nearest crack edge, distance between the vertex and the nearest crack edge, vector cosine value of the vertex and the nearest natural crack, and vertex displacement vector cosine value. The features of the edges of the graph structure include the crack state data, node index numbers of the two vertices of the edge, edge length, absolute value of the angle cosine value of the edge, and natural crack state of the edge. The selection unit is used to select the graph data of the constructed graph structure as the training dataset. The training unit is used to train a pre-built first graph neural network model based on the training dataset.

8. A computer device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the training method for a simulation model of combustion-explosion fracturing fracture propagation based on a graph neural network as described in any one of claims 1 to 5, and / or the simulation method for combustion-explosion fracturing fracture propagation based on a graph neural network as described in claim 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the training method for the simulation model of combustion-explosion fracturing fracture propagation based on graph neural networks as described in any one of claims 1 to 5, and / or the simulation method for combustion-explosion fracturing fracture propagation based on graph neural networks as described in claim 6.

Images