Rock mechanical parameter prediction method, device, equipment and medium

Abstract

The invention relates to a rock mechanical parameter prediction method, device and equipment and a medium, and belongs to the field of oil field stratum analysis, and the method comprises the steps: obtaining to-be-predicted logging data; performing feature extraction on the to-be-predicted logging data through a graph convolution layer of a completely trained GNN prediction model to obtain stratum feature information; and performing calculation according to the stratum characteristic information to obtain rock mechanical parameters of the stratum. According to the scheme, the graph structure of the stratum logging data can be modeled by training the complete GNN prediction model, the nonlinear interaction between stratums is extracted, the complex relation between nodes and edges is automatically learned, the rock mechanics parameter prediction problem of the complex fault zone of the northern oil field is solved, and the prediction efficiency is improved. And even under the condition of data missing or non-uniformity, relatively high prediction precision is kept, and the adaptability and prediction capability to complex stratums of oil fields along the north are greatly improved.

Description

Technical Field

[0001] The invention relates to the field of oilfield stratum analysis, and in particular to a rock mechanics parameter prediction method, device, equipment and medium. Background Art

[0002] Shunbei Oilfield is located in the Tarim Basin, a typical deep complex fault zone in China. The area is mainly carbonate rock, with deep formations, developed faults, complex structures, significant stress differences, and geological conditions with strong heterogeneity and developed faults. This geological structure has brought huge technical challenges to the drilling operations and reservoir development of the oilfield. Common problems include wellbore instability, well leakage, well collapse, etc., which are closely related to the rock mechanics properties of the formation.

[0003] The current geological complexity of Shunbei Oilfield has led to the limitations of traditional rock mechanics parameter prediction methods. Commonly used formation mechanical property analysis mainly relies on laboratory rock physics experiments and simple elastic mechanics models, which are difficult to cope with the complex and changeable geological environment in the fault zone. In addition, the existing prediction methods are highly sensitive to formation lithology changes and data loss, have low prediction accuracy when dealing with nonlinear complex formations, and lack effective tools to capture the physical interaction of faults and fractures.

[0004] The deep carbonate formations in Shunbei Oilfield are very prone to wellbore instability during drilling due to their brittle characteristics, so high-precision prediction of rock mechanical parameters (such as Young's modulus, Poisson's ratio, shear modulus, etc.) is crucial. Traditional rock experiments and regression models are difficult to fully capture the complexity of the formation and often require a large amount of measured data support, which is not always easy to obtain in deep fault zones. Summary of the invention

[0005] In view of this, it is necessary to provide a rock mechanics parameter prediction method, device, equipment and medium to solve the problem that the existing traditional rock mechanics parameter prediction methods are mostly based on linear regression models or simple regression models with limited experimental data, which cannot effectively deal with the complex nonlinear interactions between strata, resulting in large errors in the prediction of rock parameters.

[0006] In order to solve the problem that the existing traditional rock mechanics parameter prediction methods are mostly based on linear regression models or simple regression models with limited experimental data, which cannot effectively handle the complex nonlinear interactions between strata, resulting in large errors in the prediction of rock parameters, the present invention provides a rock mechanics parameter prediction method comprising: Acquire well logging data to be predicted; Extract features of the well logging data to be predicted by training the graph convolution layer of the complete GNN prediction model to obtain formation feature information; Calculation is performed based on the formation characteristic information to obtain rock mechanics parameters of the formation.

[0007] In a possible implementation, the method further includes: Acquire formation characteristic information of multiple nodes of the formation according to historical well logging data; Obtaining connection relationships between multiple nodes according to multiple nodes; The GNN prediction model is constructed based on the stratigraphic feature information of multiple nodes and the connection relationship between the multiple nodes in combination with the graph convolution layer.

[0008] In a possible implementation, the logging characteristics include: P-wave time difference, density, gamma ray value and porosity.

[0009] In a possible implementation manner, the method further includes: constructing a stress distribution between multiple nodes according to a finite element method, and obtaining a connection relationship between the multiple nodes according to the stress distribution.

[0010] In a possible implementation, the method further includes: Inputting training logging data into the GNN prediction model; The node information of the GNN prediction model is updated by combining the graph convolution layer of the GNN prediction model with the training logging data, and the model weight of the GNN prediction model is optimized by gradient descent and back propagation methods. After repeated iterations, a fully trained GNN prediction model is obtained.

[0011] In a possible implementation, the method further includes: optimizing the historical logging data by seismic inversion to obtain optimized historical logging data; The step of obtaining formation characteristic information of multiple nodes of a formation according to historical well logging data specifically includes: The stratigraphic characteristic information of multiple nodes of the stratigraphic formation is obtained based on the optimized historical logging data.

[0012] In a possible implementation, the rock mechanics parameters include: Young's modulus, Poisson's ratio and shear modulus.

[0013] The present invention also provides a rock mechanics parameter prediction device, comprising: a data processing module and a prediction module; The data processing module is used to obtain the well logging data to be predicted; The prediction module is used to extract features of the well logging data to be predicted by training the graph convolution layer of the complete GNN prediction model to obtain formation characteristic information, and perform calculations based on the formation characteristic information to obtain rock mechanical parameters of the formation.

[0014] Another technical solution of the present invention to solve the above technical problems is as follows: An electronic device includes a memory and a processor, wherein: The memory is used to store programs; The processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in a rock mechanics parameter prediction method of any of the above solutions.

[0015] Another technical solution of the present invention to solve the above technical problems is as follows: A computer-readable storage medium is used to store computer-readable programs or instructions, and when the program or instructions are executed by a processor, the steps in a rock mechanical parameter prediction method of any of the above schemes can be implemented.

[0016] The beneficial effects of the present invention are as follows: the scheme of the present invention can model the graph structure of formation logging data by training a complete GNN prediction model, extract nonlinear interactions between formations, and automatically learn the complex connections between nodes and edges, thereby solving the problem of rock mechanical parameter prediction in the complex fault zone of Shunbei Oilfield. Even in the absence of data or under non-uniform conditions, a high prediction accuracy is maintained, which greatly improves the adaptability and prediction ability of the complex formations in Shunbei Oilfield. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of an embodiment of a rock mechanics parameter prediction method provided by the present invention; Figure 2 A schematic diagram of a process for constructing an embodiment of the GNN prediction model provided by the present invention; Figure 3 A schematic diagram of a process flow of an embodiment of the GNN prediction model optimization provided by the present invention; Figure 4 A structural framework diagram of an embodiment of a rock mechanics parameter prediction device provided by the present invention; Figure 5 A schematic diagram of the structure of an electronic device according to an embodiment of the present invention; Figure 6 A schematic diagram of an embodiment of a technical route for predicting rock mechanics parameters in Shunbei Oilfield provided by the present invention; Figure 7 Schematic diagram of the visualization of the prediction results provided by the present invention Figure 1 Schematic diagram provided by the embodiment; Figure 8 A schematic diagram of an embodiment of a three-dimensional finite element stress distribution of a geostress field provided by the present invention; Fig. 9A schematic diagram of an embodiment of a three-dimensional visualization result of GNN, seismic inversion and finite element analysis (FEA) provided by the present invention; Fig.10 A schematic diagram provided for an embodiment of the dynamic visualization of the geostress field changing with time provided by the present invention; Fig.11 A schematic diagram of an embodiment of the dynamic effect of pressure changes in oil and gas reservoirs on well wall stress provided by the present invention; Fig.12 A schematic diagram of an embodiment of the three-dimensional simulation and analysis of the formation fracture propagation process provided by the present invention. DETAILED DESCRIPTION

[0018] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention but not to limit the scope of the present invention.

[0019] like Figure 1 As shown, a rock mechanics parameter prediction method provided by an embodiment of the present invention includes: S101, obtaining the well logging data to be predicted; it should be noted that, in a certain embodiment, these data come from the physical measurement of the wellbore, which is the core input of the model, including: P-wave time difference (Delta-T): reflects the delay time of the stratum to the propagation velocity of the P-wave. This feature is closely related to the elastic properties of the rock and is an important basis for predicting mechanical parameters such as Young's modulus.

[0020] Density: This is the direct measurement of the density of the formation rock, which is used to characterize the mass properties of the rock and is crucial for predicting shear modulus and other parameters.

[0021] Gamma Ray (GR): reflects the radioactive characteristics of the formation and is usually used to distinguish sandstone from shale. This is very useful for determining the lithology changes of the formation and identifying aquifers.

[0022] S102, extracting features of the well logging data to be predicted by training the graph convolution layer of the complete GNN prediction model to obtain formation feature information; S103: Calculate according to the formation characteristic information to obtain rock mechanics parameters of the formation.

[0023] It should be noted that, in a certain embodiment, the prediction model includes: input layer: node feature initialization of stratigraphic data; graph structure: building connection relationships between nodes and edges; graph convolution layers (GraphConvolutionLayers); feature aggregation and activation functions; output layer: rock mechanics parameter prediction; loss function and model optimization.

[0024] Input layer (node ​​feature initialization): It provides the basic physical characteristics of the formation through logging data and is the starting point for the model to predict rock mechanical parameters.

[0025] Graph structure (adjacency matrix): defines the physical connections between strata and helps the model understand the relationship between stress transfer and lithology changes.

[0026] Graph Convolutional Layer (GCN): Extract more abstract geological features layer by layer to capture the physical interactions under complex geological conditions.

[0027] Activation function and feature aggregation: Increase the nonlinear expression ability of the model so that it can adapt to the complex stress changes in the fault zone.

[0028] Output layer: Convert the extracted features into target rock mechanics parameters and provide results that can be directly used for geological engineering analysis.

[0029] Loss function and optimization: The model is continuously improved by minimizing the error to ensure the accuracy of the predicted rock mechanical parameters.

[0030] Through the close cooperation of the above parts, the GNN model can effectively extract key features from the logging data of complex fault zones, thereby accurately predicting rock mechanical parameters and providing important theoretical basis and data support for the exploration and development of oil and gas fields.

[0031] Optimizing the GNN prediction model specifically includes: Data preparation and preprocessing: Constructing node feature matrix and adjacency matrix, node features come from logging data. Model initialization: Define the structure of the graph neural network and initialize the model parameters. Forward propagation: Extract features through multi-layer graph convolution. Loss function calculation: Calculate the error between the predicted value and the true value. Back propagation and parameter update: Use the error to optimize the model weights. Iteration and convergence: Repeat the training steps until the model converges to obtain a fully trained GNN prediction model.

[0032] Through the GNN algorithm, we can perform deep rock mechanical parameter inversion and prediction based on the full use of existing logging data (such as density, P-wave time difference, gamma ray value, etc.). GNN can not only process large-scale complex data, but also maintain high prediction accuracy under data missing or non-uniform conditions, greatly improving the adaptability and prediction ability of complex formations in Shunbei Oilfield.

[0033] Preferably, in any of the above embodiments, Figure 2 As shown, S201, obtaining formation characteristic information of multiple nodes of the formation according to historical well logging data; S202. Obtain the connection relationship between multiple nodes based on multiple nodes; It should be noted that the Graph Neural Network (GNN) models the strata as nodes and models the interactions between nodes as edges: Each node represents a stratigraphic unit; The connection relationship between multiple nodes is also the edge, which reflects the stress transfer and physical connection between stratigraphic units. For example, how the stress between adjacent strata is transferred, these physical connections are modeled through edge features and updated during the training process.

[0034] S203. Construct the GNN prediction model based on the stratigraphic feature information of the multiple nodes and the connection relationship between the multiple nodes in combination with the graph convolution layer.

[0035] Preferably, in one of the above embodiments, the formation characteristic information includes: formation P-wave time difference information, formation density information and formation gamma-ray value information. It should be noted that P-wave time difference (Delta-T): is used to describe the P-wave velocity of the new well formation. Density (Density): provides quality information of the formation rock. Gamma-ray value (GammaRay, GR): is used to characterize the lithology of the new formation. These logging data are input into the trained model to initialize the features of each node, and then feature extraction is performed through the trained graph convolution layer.

[0036] Preferably, in one of the above embodiments, it further includes: constructing stress distribution between multiple nodes according to the finite element method, and obtaining the connection relationship between the multiple nodes according to the stress distribution.

[0037] It should be noted that, in a certain embodiment, the core of the finite element method is to discretize the mechanical behavior of the continuum into finite units and solve the stress-strain relationship of the material in these units. For each finite element unit, its displacement field can be expressed by the displacement vector of the node To interpolate: , in, is the shape function matrix, is the node displacement vector.

[0038] According to the generalized Hooke's law, the stress in the element With strain The relationship is: , in, is the elastic modulus matrix, which is in the form: , Through the finite element method, the stress and strain fields in complex formations can be discretized and solved to obtain the distribution of ground stress.

[0039] In another embodiment, the core of the finite element method is to discretize the mechanical behavior of the continuum into finite units and solve the stress-strain relationship of the material in these units. For each finite element unit, its displacement field can be expressed as the displacement vector of the node To interpolate: , in, is the shape function matrix, is the node displacement vector.

[0040] According to the generalized Hooke's law, the stress in the element With strain The relationship is: , in, is the elastic modulus matrix, which is in the form: , Through the finite element method, the stress and strain fields in complex formations can be discretized and solved to obtain the distribution of ground stress.

[0041] The finite element discretization of the Shunbei Oilfield formation includes: Due to the complex geometry and strong heterogeneity of the fault zone in the Shunbei Oilfield, the formation and the area near the wellbore need to be divided into many small finite element units. Common unit types include triangular units (suitable for two-dimensional models) and tetrahedral units (suitable for three-dimensional models).

[0042] The stress concentration phenomenon around the wellbore includes: In Shunbei Oilfield, the stress field around the wellbore will change dramatically due to formation damage and fluid injection during wellbore drilling, especially in the fault zone area, where stress concentration is prone to occur. According to the elastic mechanics model of the formation, the stress distribution near the wellbore can be expressed by the following equation: , in: is the hoop stress; and are the minimum and maximum horizontal stresses, respectively; is the wellbore radius, is the radial distance from the center of the wellbore.

[0043] This formula shows that when When the wellbore wall is at the maximum, the hoop stress reaches the maximum, which easily leads to wellbore instability. Therefore, the stress concentration phenomenon around the wellbore is the key to drilling stability analysis in the complex fault zone of Shunbei Oilfield.

[0044] Physical background and basic equations of elastic mechanics include: In three-dimensional elastic mechanics, the stress field prediction of the formation is based on the equilibrium equation of Newton's second law, which describes how an elastic body (formation) maintains static equilibrium through stress transfer under the action of external forces. This equation is applicable to complex formations such as Shunbei Oilfield.

[0045] The three-dimensional equilibrium equation includes: In three-dimensional space, for an elastic body, its equilibrium equation is: , in: is the stress component; is the body force component, usually expressed as: , is the density, is the gravitational acceleration component.

[0046] Stress-strain relationship: Generalized Hooke's law, including: For the complex rock mass in Shunbei Oilfield, assuming that the material is an isotropic linear elastic body, its stress-strain relationship is described by the generalized Hooke's law: , in: is the Lamé constant, defined as: , is the shear modulus, defined as: , is the strain component, expressed by the derivative of the displacement component u: , when hour, ;when hour, .

[0047] Total strain: The superposition of volumetric strain and shear strain includes: In order to analyze the deformation of rock materials more complexly, we decompose the total strain into volumetric strain and shear strain. Volumetric strain Indicates the volume change of the material in each direction, shear strain represents the deformation of the material. The volume strain and shear strain are given by the following formulas: Volumetric strain: , Shear strain: , Multi-equation coupling derivation: The interaction between stress field and displacement field includes: In order to more complexly describe the stress-strain relationship in the Shunbei Oilfield formation, multiple equations such as stress field, displacement field and elastic properties of materials must be coupled and derived.

[0048] The derivation of the coupling of stress tensor and strain tensor includes: First, coupling stress tensor and strain tensor can obtain more complex stress-strain relationship. According to the generalized Hooke's law, the three-dimensional stress tensor and the strain tensor The complete expression is: , in: represents the trace of the strain tensor, i.e., the volumetric strain , is the unit tensor; is the shear modulus.

[0049] After expanding the above formula, the three-dimensional stress components and strain components The relationship can be expressed as: , in, and are the elastic constants of the material.

[0050] Interaction between displacement field and stress field: displacement derivative, which converts displacement field Combined with the stress field, their interaction is derived. The gradient of the displacement field can be expressed in the form of a strain tensor: , At this point, combined with the generalized Hooke's law, the complete equations of the displacement field and stress field are obtained: , This is a set of coupled equations that represent the interaction between displacements, stresses, and material parameters.

[0051] Derivation of the Lame equation: from displacement to stress, including: through the relationship between strain and displacement, combined with the elastic parameters of the material, the Lame equation in the formation can be further derived to solve complex displacement and stress distribution.

[0052] For a three-dimensional elastic body, the basic form of the Lame equation is: , in: represents the divergence of the displacement field, i.e., the volume change; The Laplace operator representing the displacement field reflects the rate of change of deformation; is the external body force.

[0053] This is a set of second-order partial differential equations that describe how the displacement field is affected by external forces, ground stresses, and material properties.

[0054] The derivation of the finite element analysis (FEA) formula includes: In order to perform numerical solutions, we need to discretize the above complex displacement-stress coupling equations into finite element equations. This process includes the stiffness matrix Derivation of external force vector Calculation of displacement vector Numerical solution of .

[0055] The stiffness matrix derivation includes: first, for each finite element, the stiffness matrix It can be expressed by the following formula: , in: is the strain-displacement matrix; is the elastic modulus matrix of the material; is the unit volume.

[0056] Specific to three-dimensional problems, the elastic modulus matrix of the material As shown below: , The derivation of the external force vector F includes: It is composed of body force (such as gravity) and surface force (such as wellbore pressure). For each finite element unit, the external force vector can be expressed as: , in: is the shape function matrix; is the volume force; It is the surface force, usually caused by the wellbore pressure Given.

[0057] The solution of the total displacement equation includes: Through the above derivation, the total finite element equation of the ground stress field of the entire Shunbei Oilfield can be obtained: , in, is the stiffness matrix, is the node displacement vector, is the external load vector.

[0058] Iteratively solve the displacement vector using numerical methods (such as the conjugate gradient method) , the displacement component of each node in the formation can be obtained, and then the strain-displacement matrix Calculate the distribution of the stress field. In one embodiment, Fig.11As shown, it is a schematic diagram of the dynamic impact of pressure changes in oil and gas reservoirs on wellbore stress. The pressure field changes over time, the color indicates the degree of pressure concentration, and the stress around the wellbore also changes accordingly.

[0059] Preferably, in one of the above embodiments, it further comprises: Figure 3 As shown, S301. Input training logging data into the GNN prediction model. It should be noted that, in a certain embodiment, the training logging data may include: These data, derived from physical measurements of the wellbore, are the core inputs to the model and include: P-wave time difference (Delta-T): reflects the delay time of the stratum to the propagation velocity of the P-wave. This feature is closely related to the elastic properties of the rock and is an important basis for predicting mechanical parameters such as Young's modulus.

[0060] Density: This is the direct measurement of the density of the formation rock, which is used to characterize the mass properties of the rock and is crucial for predicting shear modulus and other parameters.

[0061] Gamma Ray (GR): reflects the radioactive characteristics of the formation and is usually used to distinguish sandstone from shale. This is very useful for determining the lithology changes of the formation and identifying aquifers.

[0062] S302, updating the node information of the GNN prediction model through the graph convolution layer of the GNN prediction model in combination with the training logging data, and optimizing the model weight of the GNN prediction model through gradient descent and back propagation methods, and obtaining a fully trained GNN prediction model through repeated iterations.

[0063] It should be noted that, in a certain embodiment, the optimization process of the GNN prediction model may specifically include: The key process of model training includes: Multi-layer graph convolution operation (GCN): GNN updates the features of nodes through multi-layer graph convolution. In each layer of convolution, the node features are updated according to the information of its neighboring nodes, and more complex features are extracted layer by layer.

[0064] Gradient Descent and Backpropagation: In each iteration, the model updates the weights through gradient descent to minimize the error between the prediction and the true measurement value, thereby improving the accuracy of the model.

[0065] Preferably, in one of the above embodiments, it also includes: optimizing historical logging data through seismic inversion to obtain optimized historical logging data; it should be noted that, in one embodiment, seismic inversion data is combined with logging data to provide a more comprehensive stratigraphic description for GNN. For example, the input node features of GNN may include Poisson's ratio or other macroscopic properties derived from seismic inversion data. By combining the results of seismic inversion with GNN model predictions, it is possible to supplement the deficiencies of GNN in certain areas due to data sparsity, so that the prediction results have better spatial continuity.

[0066] The step of obtaining formation characteristic information of multiple nodes of a formation according to historical well logging data specifically includes: The stratigraphic characteristic information of multiple nodes of the stratigraphic formation is obtained based on the optimized historical logging data.

[0067] Preferably, in one of the above embodiments, the rock mechanics parameters include: Young's modulus, Poisson's ratio and shear modulus.

[0068] It should be noted that, in a certain embodiment, Figure 6 As shown, the technical route for predicting rock mechanics parameters in Shunbei Oilfield, calculating Young's modulus, Poisson's ratio and shear modulus, can include: Based on the rock mechanics parameter model of well logging data, for the complex fault zone of Shunbei Oilfield, key physical properties (such as P-wave time difference) are first extracted through well logging data. ,density and gamma ray value GR), combined with the elastic mechanics model to calculate rock mechanics parameters: In view of the complex fault zone in Shunbei Oilfield, key physical characteristics such as P-wave time difference are first extracted through logging data. ,density And gamma ray value GR, combined with elastic mechanics model to calculate rock mechanics parameters: The relationship between Young's modulus ElasticModulus, E and Poisson's ratio Poisson'sRatio, v: , Among them, the longitudinal wave velocity ,density Poisson's ratio obtained from well logging data It can be obtained through seismic inversion data or experimental measurement.

[0069] The relationship between shear modulus (G) and Young's modulus: , These relationships can be used to invert the lithological characteristics of the Shunbei oilfield and thus determine the elastic parameters of the carbonate formations.

[0070] The derivation of elastic parameters driven by logging data includes: logging data (such as P-wave time difference ,density ) is usually used to invert Young's modulus and Poisson's ratio of rocks. P-wave velocity The relationship with Young's modulus is as follows: , According to the wave velocity and density in the well logging data, Young's modulus It can be calculated by the following formula: , in, is the density, is the longitudinal wave velocity, Poisson's ratio Usually obtained through experiments or seismic inversion.

[0071] The GNN prediction model calculates rock mechanical parameters based on the logging data and the model of rock mechanical parameters.

[0072] In one embodiment, the calculation of the geostress field includes: Predicting geostress is a key issue in oil and gas exploration and production, especially in deep formations where vertical stress . Usually determined by the density of the rock and depth Decide: , Maximum horizontal stress and minimum stress According to the relationship between Poisson's ratio and vertical stress in the elastic mechanics model, it is estimated that: , These stress models can provide a basis for evaluating wellbore stability and fault slip, but when faced with complex fault zones and areas of high stress concentration, the accuracy of elastic mechanics models is limited and it is difficult to deal with the heterogeneity and nonlinear behavior of rocks.

[0073] In one embodiment, the training process of the GNN prediction model includes: In the training phase, the goal is to learn the relationship between logging data and rock mechanical parameters through multiple iterations. The input data in this phase is very diverse, and the purpose is to provide the model with enough information to understand the relationship between formation properties and rock mechanical parameters.

[0074] The input data in the training phase specifically include: Well logging data (basic input features), which are the core input of the model, including: P-wave time difference: reflects the delay time of the formation to the propagation velocity of the P-wave. This feature is closely related to the elastic properties of the rock and is an important basis for predicting mechanical parameters such as Young's modulus. Density: This is the directly measured density of the formation rock, which is used to characterize the mass characteristics of the rock and is crucial for predicting the shear modulus and other parameters. Gamma ray value: reflects the radioactive characteristics of the formation and is usually used to distinguish between sandstone and shale. This is very useful for judging the lithology changes of the formation and identifying aquifers.

[0075] In order to better capture the physical relationship between strata, Graph Neural Network (GNN) models strata as nodes and the interactions between nodes as edges; Each node represents a stratigraphic unit, and the initialized feature vector consists of logging data (such as P-wave time difference, density, gamma ray value, etc.).

[0076] These edges reflect the stress transfer and physical connections between stratigraphic units. For example, how stress is transferred between adjacent stratigraphic units. These physical connections are modeled by edge features and updated during the training process.

[0077] True rock mechanical parameters: The goal of the model is to predict the rock mechanical parameters of the formation, such as Young's modulus, Poisson's ratio, shear modulus, etc. During training, these target parameters come from experimental measurements or seismic inversion data as true labels.

[0078] The GNN model updates its parameters by calculating the error between the predicted value and the true value. The loss function is used to measure the gap between the rock mechanical parameters predicted by the model and the true measured value, and the model learns by minimizing this loss function.

[0079] GNN updates node features through multiple layers of graph convolution. In each layer of convolution, node features are updated based on the information of its neighboring nodes, extracting more complex features layer by layer. In each iteration, the model updates weights through gradient descent to minimize the error between the prediction and the true measurement value, thereby improving the accuracy of the model. After training, the GNN model can be used to predict rock mechanical parameters for new formation data. The input data in the prediction phase has some similarities with the training phase, but there are also significant differences.

[0080] Preferably, in one of the above embodiments, during the prediction process, in the prediction stage, the model still relies on logging data as input. These data usually come from new drilling operations and include: P-wave time difference (Delta-T): used to describe the P-wave velocity of the new well formation. Density (Density): provides quality information of the formation rock. Gamma ray value (GammaRay, GR): used to characterize the lithology of the new formation. These logging data are input into the trained model to initialize the features of each node, and then feature extraction is performed through the trained graph convolution layer. In the prediction stage, the model does not need to be retrained, so label data is not used. Instead, the optimal model weights learned in the training stage are used to predict new data. It should be noted that the weights in the training process contain experience about stress transfer and rock properties between formations, and this information will be directly used to infer the rock mechanical parameters of the new formation in the prediction stage.

[0081] Key steps in the prediction process: Propagation and convolution of node features: New data is processed by each layer of the graph convolutional network (GCN), and node features are propagated and updated through the learned weights. Output prediction of rock mechanical parameters: Finally, the node features after multiple layers of convolution will be used to predict rock mechanical parameters such as Young's modulus and Poisson's ratio of the formation. These results are used for engineering decisions of oil and gas wells, such as wellbore stability assessment and drilling design.

[0082] The same things in the training phase and the prediction phase are: Well logging data: Both the training and prediction phases use similar well logging data types (P-wave time difference, density, gamma ray values, etc.) as the initial feature input of the model nodes.

[0083] Node and edge information of the graph structure: The physical modeling method of the stratum nodes and the structure of the graph are consistent in the two stages.

[0084] Differences in input data: Use of labeled data: In the training phase, real labeled data (such as Young's modulus and Poisson's ratio) is required to calculate the loss function and optimize the model through back propagation. In the prediction phase, the model only needs logging data as input to generate prediction results of rock mechanical parameters, and no labeled data is required.

[0085] The difference between the training phase and the prediction phase is: Training phase: By optimizing the model, the error between the predicted value and the true value is minimized to obtain an accurate prediction model.

[0086] Prediction stage: Through the trained model, the data of the new well is directly inferred to obtain the rock mechanical parameters.

[0087] Differences in processes: The training phase requires continuous model iteration and weight updates to find the optimal parameter settings.

[0088] Prediction phase: New data only needs to be inferred once through the trained model, and there is no need to optimize weights.

[0089] In both phases, logging data (such as P-wave time difference, density, gamma ray values, etc.) are always key input information. In the training phase, these data are paired with true labels for model learning and optimization. In the prediction phase, logging data are input into the trained model to generate predictions of formation rock mechanical parameters.

[0090] Preferably, in one of the above embodiments, Figure 7 As shown in the figure, a schematic diagram of the visualization of the prediction results is used to evaluate the risk of wellbore stability based on the comprehensive prediction results. The figure shows the high-risk areas around the wellbore. The darker the color, the more severe the stress concentration and the greater the risk of instability. This visualization helps identify high-risk locations on the wellbore so that preventive measures can be taken to reduce the risk of wellbore instability.

[0091] Preferably, in one of the above embodiments, the construction and training process of the CNN prediction model specifically includes: The GNN-based prediction model mainly includes the following parts during construction and training: input layer: node feature initialization of formation data; graph structure: building the connection relationship between nodes and edges; graph convolution layer (GraphConvolutionLayers); feature aggregation and activation function; output layer: rock mechanics parameter prediction; loss function and model optimization; Input layer: node feature initialization; node feature matrix construction: node feature initialization is the first step of the model and the most critical basic part. Each node represents a stratigraphic unit, and the node feature matrix contains features that reflect the physical properties of the formation. The node feature matrix can be represented as X, where the number of rows represents the number of nodes (i.e., the number of stratigraphic units), and the number of columns represents the number of features of each node. For example: For the strata in the Shunbei Fault Zone, we obtain node features through logging data, and each node contains the following logging features: P-wave time difference (Delta-T): reflects the P-wave velocity of the rock. Density: the density of the formation rock. Gamma ray value (GammaRay, GR): used to characterize the lithology of the formation. Porosity (Porosity) (if any): used to describe the reservoir characteristics of the formation. For example, suppose we have N stratigraphic units, each stratigraphic unit contains 4 logging features, and the node feature matrix}X is N×4 in shape, where each row represents the feature vector of a stratigraphic unit.

[0092] The node feature matrix provides the initial information of each unit in the formation and is the basic data of the entire model. These features reflect the differences in lithology and physical properties of the formation, and are very important for the model to predict mechanical parameters such as Young's modulus and Poisson's ratio.

[0093] Graph structure: construct the connection relationship between nodes and edges; adjacency matrix construction: the core of the graph structure is the connection relationship between nodes, which is represented by the adjacency matrix A. Each node represents a stratigraphic unit, and the edges between nodes represent the interaction between stratigraphic units, such as the stress transfer relationship between adjacent strata. For example: the adjacency matrix can be represented as an N×N square matrix, representing the connection relationship between all nodes. If there is a direct physical connection between node i and node j (such as stress transfer between adjacent strata), the corresponding position Aij=1 in the adjacency matrix, otherwise it is 0.

[0094] The adjacency matrix defines the physical connection between stratigraphic units, helping GNN to aggregate the information of neighboring nodes to the central node during graph convolution. For complex fault zones, such connection relationships can effectively capture the stress transmission path and the changing characteristics of the strata.

[0095] The graph convolutional layer (GCN) is the core of GNN, which is used to extract higher-level features from the initial features of the ground-level nodes. The features of each node are updated by aggregating and combining the features of its neighboring nodes. The update formula of the graph convolution can be expressed as: , in, :No. The node feature matrix of the layer. :No. The weight matrix of the layer. : Bias vector. : Activation function (e.g. ReLU).

[0096] For example: Assume that layer 0 (input layer) , that is, well logging data. The first layer of graph convolution aggregates the features of neighboring nodes to the central node to obtain a new feature representation .

[0097] The graph convolution layer extracts more abstract and high-level features layer by layer by sharing and combining information with neighboring nodes. This is particularly important for capturing complex physical relationships between strata (such as stress transfer), thereby improving the accuracy of rock mechanical parameter prediction.

[0098] Feature aggregation and activation function. Feature aggregation includes: after each layer of convolution operation, the node features need to be aggregated and nonlinearly transformed, which is usually achieved through activation functions, such as ReLU function. Activation functions are used to increase the nonlinear expression ability of the model so that it can fit complex stratigraphic relationships. In each layer after graph convolution, the ReLU function is used to truncate negative values, making the output more sparse and introducing nonlinearity.

[0099] The activation function increases the model's adaptability to complex geological conditions, enabling GNN to better capture nonlinear phenomena such as stress concentration and stratum damage in fault zones.

[0100] Output layer: rock mechanical parameter prediction, the goal of the output layer: the final feature matrix H(L) after multi-layer graph convolution is used as the input of the output layer to predict the target rock mechanical parameters of each stratigraphic unit, such as Young's modulus and Poisson's ratio. The final node features are mapped to the target output dimension using a fully connected layer. Assuming that the prediction target is two parameters (Young's modulus and Poisson's ratio), the output shape is N×2.

[0101] The output layer converts the extracted high-level features into specific rock mechanical parameters for subsequent geological analysis and engineering decision-making. For example, Young's modulus is used to evaluate the deformation characteristics of the formation, while Poisson's ratio is used to evaluate the volume change of the formation.

[0102] Loss function and model optimization, definition of loss function: During the training process, it is necessary to calculate the error between the predicted value and the true value to optimize the model. The loss function can be defined as the mean square error (MSE), which is used to measure the prediction performance of the model. For example, assuming the true value is Y and the predicted value is Y, the loss function can be expressed as: , Where N is the number of formation nodes.

[0103] By minimizing this loss function, the model can continuously update its weights to improve prediction accuracy.

[0104] The loss function optimizes the model parameters through back propagation and gradient descent, so that the model can accurately predict the rock mechanical parameters of the formation. These parameters are of great significance for wellbore stability analysis, drilling design, etc.

[0105] The role of each part of the prediction model in predicting rock mechanical parameters is summarized as follows: Input layer (node ​​feature initialization): provides basic physical characteristics of the formation through logging data, which is the starting point for the model to predict rock mechanical parameters. Graph structure (adjacency matrix): defines the physical connection between formations to help the model understand the relationship between stress transmission and lithology changes. Graph convolution layer (GCN): extracts more abstract formation features layer by layer to capture physical interactions under complex formation conditions. Activation function and feature aggregation: increases the nonlinear expression ability of the model so that it can adapt to complex stress changes in fault zones. Output layer: converts the extracted features into target rock mechanical parameters and provides results that can be directly used for geological engineering analysis. Loss function and optimization: continuously improves the model by minimizing errors to ensure the accuracy of the predicted rock mechanical parameters. Through the close cooperation of the above parts, the GNN model can effectively extract key features from logging data in complex fault zones, thereby accurately predicting rock mechanical parameters and providing important theoretical basis and data support for the exploration and development of oil and gas fields. In a certain embodiment, such as Fig.12 The figure shows a schematic diagram of the three-dimensional simulation and analysis of the formation fracture propagation process. The dark bar in the figure represents the fracture path, and the surrounding scattered points represent the stress distribution in the fracture area. Through this three-dimensional visualization, the expansion process of the crack in the formation and its impact on the surrounding stress field can be intuitively observed. This is very critical for predicting the formation and expansion of cracks near the wellbore, and helps to formulate countermeasures to ensure the stability and safety of the wellbore wall.

[0106] GNN graph structure definition: For the complex fault zone of Shunbei Oilfield, the graph neural network (GNN) models the stratigraphic nodes and lithological characteristics as a graph structure Among them, the node Well logging data representing each stratigraphic unit, including density , longitudinal wave time difference , Gamma ray value etc. Represents the physical relationship between stratigraphic units, such as stress transfer between adjacent strata. Node feature initialization: The initial feature vector of each node Expressed as: , in, is the density, is the longitudinal wave time difference, is the gamma ray value.

[0107] These logging parameters are the basis for GNN to extract lithology and physical properties.

[0108] Mathematical definition of graph convolutional layer (GCN): GNN extracts the relationship between nodes through multi-layer graph convolution operations. Layer, Node The feature update formula is: , in, Representation Node Neighborhood of is the normalization coefficient, is the weight matrix, is the feature vector of node j at layer k, is the bias term, is the activation function (such as ReLU). This formula is used to update node features layer by layer to capture the physical connection between layers.

[0109] Update of edge features: There are significant physical interactions (such as stress transfer) between the stratigraphic nodes in the Shunbei Oilfield fault zone. These interactions can be effectively extracted through the edge feature update formula: , in, is the feature vector of edge (i, j) at the kth layer, is the weight matrix of edge features, is the bias vector.

[0110] The updating of edge features enables GNN to learn the lithology and stress changes in complex fault zones and thus predict the mechanical parameters of the formation.

[0111] Loss function design of GNN: During the training process of GNN, the model is optimized by minimizing the error between the predicted rock mechanical parameters and the true measured values. The loss function C is designed as: , Where N is the number of formation nodes, and are the measured Young's modulus and Poisson's ratio, and is the GNN prediction value.

[0112] Model back propagation and optimization: During the back propagation process, the weights of GNN Updated by gradient descent, the gradient update formula is: , in, is the loss function, is the weight matrix, N is the number of stratum nodes, is the feature vector of node i in the kth layer.

[0113] After multiple iterations, the optimal model parameters were finally obtained for the prediction of rock mechanical parameters in the complex fault zone of Shunbei Oilfield.

[0114] In GNN-based model training, the whole process is usually divided into the following main steps: Data preparation and preprocessing: Construct node feature matrix and adjacency matrix, and the node features come from logging data. Model initialization: Define the structure of the graph neural network and initialize the model parameters. Forward propagation: Extract features through multi-layer graph convolution. Loss function calculation: Calculate the error between the predicted value and the true value. Back propagation and parameter update: Use the error to optimize the model weights. Iteration and convergence: Repeat the training steps until the model converges. Before training begins, the formation logging data of Shunbei Oilfield must be collected and preprocessed. Construct node feature matrix , assuming that there is There are stratum nodes, each with 3 features, then for The matrix of .

[0115] Adjacency matrix construction: The adjacency matrix of Used to describe the interconnection between stratigraphic nodes. In Shunbei Oilfield, the stress transfer relationship between adjacent stratigraphic units is represented as edges in the graph structure. and nodes Adjacent, then Otherwise, it is 0. Label data: The real rock mechanical parameters (such as Young's modulus and Poisson's ratio) are used as label data for supervised learning. This data can be obtained through experimental measurement or seismic inversion.

[0116] Model initialization specifically includes: Model structure definition: based on The method uses the graph convolution layer ( ) to process the relationship between nodes. Assume that two layers of graph convolution are used, each with 16 hidden units, in order to extract the interaction features between complex strata.

[0117] Parameter initialization: Initialize the weight matrix of each layer and the bias vector , usually initialized with random numbers to avoid symmetry in the early stages of the model.

[0118] Forward propagation specifically includes: Forward propagation refers to the process of calculating the final prediction from the input features through the model.

[0119] The first layer of graph convolution: the input node feature matrix Through the adjacency matrix With the first layer weight matrix Multiply, add the bias vector, and then pass the activation function (such as ReLU) to get the first layer output : , in, is the activation function, is the adjacency matrix, X is the node feature matrix, is the first layer weight matrix, is the first layer bias vector.

[0120] The second layer of graph convolution: the output of the first layer Input into the second layer of graph convolution to get the second layer output : , in, is the activation function, is the adjacency matrix, is the first layer output, is the second layer weight matrix, is the second layer bias vector.

[0121] Output layer: Finally, As the output of the last layer, it will be mapped to predicted values, such as Young's modulus and Poisson's ratio.

[0122] The loss function is used to measure the difference between the model's predicted value and the true value. In this model, the mean square error (MSE) is used as the loss function: Assume that the predicted value is , the true value is y, then the loss function is: , in, is the number of stratum nodes, is the number of parameters to be predicted (e.g. 2 for Young's modulus and Poisson's ratio).

[0123] The purpose of backpropagation is to update the parameters of the model by minimizing the loss function. Gradient calculation: The gradient of each weight parameter and bias term is calculated through the backpropagation algorithm, which represents the partial derivative of the loss function with respect to each parameter. Gradient update: Use an optimization algorithm (such as gradient descent or Adam) to update the parameters: , , Where L is the loss function, is the weight matrix, is the bias vector, is the learning rate, which is used to control the amplitude of parameter updates.

[0124] The forward propagation, loss calculation, and backpropagation steps are repeated, usually over multiple training cycles (epochs), until the loss function reaches a sufficiently small value or a preset maximum number of training rounds is reached.

[0125] After each training cycle, a portion of the data (validation set) is used to evaluate the performance of the model to ensure that the model is not overfitting (i.e., performing well in the training set but poorly in the validation set).

[0126] When the loss function of the model no longer changes significantly, the model is considered to have converged and training can be stopped. At this point, the model should be able to accurately predict the mechanical parameters of rocks in Shunbei Oilfield.

[0127] The combination of finite element simulation (FEA), seismic inversion and GNN (graph neural network) prediction model aims to accurately predict the geostress field and rock mechanical parameters in complex fault zones, especially in ultra-deep carbonate oil and gas reservoirs such as Shunbei Oilfield. The combination of the three can more effectively capture the mechanical behavior under complex geological conditions. The following is a detailed description of how the three are combined and interact with each other to achieve the overall technical effect. In one embodiment, Figure 8 As shown in the figure, the three-dimensional finite element stress distribution of the geostress field simulates the stress changes inside the stratum through 3D scatter plots and color gradients. The redder the color, the higher the stress concentration.

[0128] Input logging data to predict rock mechanical parameters: The GNN model mainly uses logging data (such as P-wave time difference, density, gamma ray value, etc.) to predict rock mechanical parameters of the formation, such as Young's modulus, Poisson's ratio, etc. These parameters are key indicators for describing rock mechanical properties and directly affect the distribution of the ground stress field. In one embodiment, Fig.10 As shown in the figure, it is a dynamic visualization diagram of the change of the ground stress field over time. The figure shows how the stress field changes over time, and the color change represents the evolution of the stress concentration area.

[0129] Model capability under complex geological conditions: The fault zone of Shunbei Oilfield has a complex geological structure, and traditional linear methods are difficult to effectively model. GNN constructs stratigraphic unit nodes and their adjacent relationships, and uses graph convolution to capture stress transfer and lithology changes between strata, thereby improving the accuracy of rock mechanics parameter prediction.

[0130] Interface with other methods: The prediction results obtained by GNN (such as Young's modulus and Poisson's ratio) can be used as initial input data for finite element analysis and seismic inversion, helping subsequent steps to more accurately reflect the formation characteristics.

[0131] Seismic inversion provides large-scale stratigraphic information. Seismic inversion obtains stratigraphic properties: Seismic inversion uses seismic exploration data (such as seismic wave velocity, reflection wave characteristics, etc.) to infer physical properties such as lithology distribution and porosity of underground strata. These properties are large-scale and can provide macroscopic geological structural characteristics for the entire oil and gas reservoir.

[0132] Combining inversion with logging data: Combining seismic inversion data with logging data can provide GNN with a more comprehensive description of the formation. For example, the input node features of GNN may include Poisson's ratio or other macroscopic properties derived from seismic inversion data. By combining the results of seismic inversion with GNN model predictions, it is possible to supplement the deficiencies of GNN in certain areas due to data sparsity, making the prediction results more continuous in space. Providing boundary conditions for finite element simulation: Seismic inversion results can also be used to determine the boundary conditions and initial conditions of finite element simulation, especially in the modeling of the entire geostress field. The formation elastic modulus and density obtained by seismic inversion can provide reasonable material parameter settings for the finite element model.

[0133] Finite element analysis is used for stress field simulation and wellbore stability analysis. Finite element analysis is used to simulate the stress field distribution of the wellbore and surrounding strata, especially in complex fault structure areas such as the Shunbei Fault Zone. FEA discretizes the strata into finite units and uses rock mechanics parameters (such as Young's modulus and Poisson's ratio predicted by GNN) to solve the stress and strain in each unit.

[0134] FEA can simulate the stress concentration around the wellbore in detail, which is crucial for wellbore stability analysis. During drilling and injection operations, the area near the wellbore is prone to damage due to stress concentration. FEA can predict these high-risk areas and help formulate engineering measures to reduce the risk of wellbore instability.

[0135] The elastic modulus, Poisson's ratio and other parameters in the finite element analysis come from the results predicted by GNN, which provide an accurate description of the FEA calculation based on well logging data and machine learning. In addition, the macro boundary conditions provided by seismic inversion also further improve the accuracy of the finite element model on a large scale.

[0136] Through the cooperation of finite element simulation, seismic inversion and GNN prediction model, accurate modeling and analysis of complex fault zone oil and gas reservoirs such as Shunbei Oilfield can be achieved. Fig. 9As shown in the figure, the three-dimensional visualization results of GNN, seismic inversion and finite element analysis (FEA) are comprehensively displayed. In the figure: nodes and edges represent the interaction of strata (GNN model). The color and size of the nodes correspond to the elastic modulus and Poisson's ratio obtained by seismic inversion. The color gradient points scattered throughout the area represent the stress field distribution of FEA.

[0137] Specifically: GNN prediction uses logging data to predict Young's modulus, Poisson's ratio and other mechanical parameters of rocks through GNN. These parameters are directly used as input for finite element simulation.

[0138] GNN extracts important stratigraphic features from complex stratigraphic data through graph convolution operations, overcoming the limitations of traditional methods in dealing with complex fault zones. Seismic inversion provides macroscopic lithology and physical properties of the strata, which serves as an important supplement to GNN and finite element analysis. The inversion results are used to define the boundary conditions of the FEA model and to supplement the GNN node features to ensure that the prediction model has a more comprehensive understanding of the entire fault zone. Using the rock mechanics parameters predicted by GNN, FEA simulates the stress field in the area around the wellbore to help analyze the stability of the wellbore wall. FEA is particularly suitable for dealing with local stress concentration phenomena, especially stress changes caused by drilling and fluid injection. Through the combination of the three, the overall technical effect has been improved in the following aspects: Accurate rock mechanics parameter inversion: GNN provides high-precision rock mechanics parameters by learning well logging data and seismic inversion data, which can better reflect the actual characteristics of the formation.

[0139] Global and local stress field simulation: Seismic inversion provides macroscopic stratigraphic structure, GNN predicts rock mechanical parameters, and finite element analysis plays a role in local stress field simulation. By combining these three methods, accurate simulation of the stress field of the entire oil and gas reservoir can be achieved, covering both the macroscopic level and the microscopic wellbore area. Wellbore stability and risk analysis: FEA combines the mechanical parameters predicted by GNN to perform wellbore stress concentration analysis to help identify high-risk areas. This plays an important guiding role in formulating drilling plans and reducing the risk of wellbore instability. Application example: complex fault zones in Shunbei Oilfield.

[0140] Through seismic inversion, the large-scale lithology distribution of the entire fault zone was obtained, and reliable geological background information was established. Using the GNN model, a mechanical parameter model of the stratigraphic unit was constructed based on the logging data, so that the lithological characteristics and mechanical parameters of the formation were inverted more accurately. Using finite element analysis, the stress concentration phenomenon near the wellbore was simulated in detail, especially during drilling and water injection, and the high-risk areas of wellbore instability were identified, thus providing guidance for drilling safety. Through this combination, the rock mechanical properties and stress field changes of the formation can be more accurately predicted under complex geological conditions, providing a comprehensive and accurate decision-making basis for the exploration and development of oil and gas reservoirs, especially in high-risk areas such as Shunbei Oilfield, which has significant application value.

[0141] Preferably, in one of the above embodiments, Figure 4 , a rock mechanics parameter prediction device 400, comprising: a data processing module 401 and a prediction module 402; The data processing module 401 is used to obtain the well logging data to be predicted; The prediction module 402 is used to extract features of the well logging data to be predicted through the graph convolution layer of the trained complete GNN prediction model to obtain formation characteristic information; and perform calculations based on the formation characteristic information to obtain rock mechanical parameters of the formation.

[0142] The rock mechanics parameter prediction device provided in the above embodiment can implement the technical solution described in the above embodiment of a rock mechanics parameter prediction method. The specific implementation principles of the above modules or units can refer to the corresponding contents in the above embodiment of a rock mechanics parameter prediction method, which will not be repeated here.

[0143] like Figure 5 As shown, the present invention also provides an electronic device 500. The electronic device 500 includes a processor 501, a memory 502 and a display 503. Figure 5 Only some components of the electronic device 500 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0144] In some embodiments, the memory 502 may be an internal storage unit of the electronic device 500, such as a hard disk or memory of the electronic device 500. In other embodiments, the memory 502 may also be an external storage device of the electronic device 500, such as a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), etc. equipped on the electronic device 500.

[0145] Furthermore, the memory 502 may include both an internal storage unit of the electronic device 500 and an external storage device. The memory 502 is used to store application software installed in the electronic device 500 and various data.

[0146] In some embodiments, the processor 501 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 502, such as a rock mechanics parameter prediction method in the present invention.

[0147] In some embodiments, the display 503 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 503 is used to display information on the electronic device 500 and to display a visual user interface. The components 501-503 of the electronic device 500 communicate with each other via a system bus.

[0148] In some embodiments of the present invention, when the processor 501 executes a rock mechanics parameter prediction program in the memory 502, the following steps may be implemented: Acquire well logging data to be predicted; Extract features of the well logging data to be predicted by training the graph convolution layer of the complete GNN prediction model to obtain formation feature information; Calculation is performed based on the formation characteristic information to obtain rock mechanics parameters of the formation.

[0149] It should be understood that: when the processor 501 executes a rock mechanics parameter prediction program in the memory 502, in addition to the above functions, other functions can also be implemented. For details, please refer to the description of the corresponding method embodiment above.

[0150] Furthermore, the embodiment of the present invention does not specifically limit the type of the electronic device 500 mentioned, and the electronic device 500 may be a portable electronic device such as a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop computer, etc. Exemplary embodiments of portable electronic devices include but are not limited to portable electronic devices equipped with IOS, Android, Microsoft or other operating systems. The above-mentioned portable electronic devices may also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, the electronic device 500 may not be a portable electronic device, but a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0151] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute a rock mechanical parameter prediction method provided by the above methods, the method comprising: Acquire the well logging data to be predicted; extract features of the well logging data to be predicted by using the graph convolution layer of the well-trained GNN prediction model to obtain formation feature information; Calculation is performed based on the formation characteristic information to obtain rock mechanics parameters of the formation.

[0152] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0153] The rock mechanics parameter prediction method, device, equipment and medium provided by the present invention are introduced in detail above. The principles and implementation methods of the present invention are explained in this article using specific examples. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A rock mechanics parameter prediction method, characterized in that: include: Acquire well logging data to be predicted; Extract features of the well logging data to be predicted by training the graph convolution layer of the complete GNN prediction model to obtain formation feature information; Calculation is performed based on the formation characteristic information to obtain rock mechanics parameters of the formation.

2. A rock mechanics parameter prediction method according to claim 1, characterized in that: Also includes: Acquire formation characteristic information of multiple nodes of the formation according to historical well logging data; Obtaining connection relationships between multiple nodes according to multiple nodes; The GNN prediction model is constructed based on the stratigraphic feature information of multiple nodes and the connection relationship between the multiple nodes in combination with the graph convolution layer.

3. A rock mechanics parameter prediction method according to claim 2, characterized in that: The formation characteristic information includes: formation compressional wave time difference information, formation density information and formation gamma ray value information.

4. A rock mechanics parameter prediction method according to claim 2 or 3, characterized in that: Also includes: The stress distribution among the multiple nodes is constructed according to the finite element method, and the connection relationship among the multiple nodes is obtained according to the stress distribution.

5. A rock mechanics parameter prediction method according to claim 1 or 2, characterized in that: Also includes: Inputting training logging data into the GNN prediction model; The node information of the GNN prediction model is updated by combining the graph convolution layer of the GNN prediction model with the training logging data, and the model weight of the GNN prediction model is optimized by gradient descent and back propagation methods. After repeated iterations, a fully trained GNN prediction model is obtained.

6. A rock mechanics parameter prediction method according to claim 2, characterized in that: Also includes: Optimize historical logging data through seismic inversion to obtain optimized historical logging data; The step of obtaining formation characteristic information of multiple nodes of a formation according to historical well logging data specifically includes: The stratigraphic characteristic information of multiple nodes of the stratigraphic formation is obtained based on the optimized historical logging data.

7. A rock mechanics parameter prediction method according to claim 1, characterized in that: The rock mechanics parameters include: Young's modulus, Poisson's ratio and shear modulus.

8. A rock mechanics parameter prediction device, characterized in that: include: Data processing module and prediction module; The data processing module is used to obtain the well logging data to be predicted; The prediction module is used to extract features of the well logging data to be predicted by training the graph convolution layer of the complete GNN prediction model to obtain formation characteristic information, and perform calculations based on the formation characteristic information to obtain rock mechanical parameters of the formation.

9. An electronic device, characterized in that: comprising a memory and a processor, wherein: The memory is used to store programs; The processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in the rock mechanics parameter prediction method described in any one of claims 1 to 7 above.

10. A computer-readable storage medium, characterized in that: Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in a rock mechanics parameter prediction method as described in any one of claims 1 to 7.

Images