Graph neural network three-dimensional geologic structure modeling method

Abstract

The invention relates to the technical field of geologic structure modeling, and particularly discloses a graph neural network three-dimensional geologic structure modeling method, which takes a whole geologic map as input data, and can automatically identify and simulate the influence of a fault on a stratum scalar field by establishing a new fault coding method. Especially, the discontinuous jump phenomenon of the scalar field of the fault position is avoided, the problem that fault displacement parameters need to be set manually in a traditional method is avoided, accurate simulation of the fault effect can be achieved under the condition that manual intervention is not relied on, and the fault displacement accuracy is improved. The problem that a traditional graph neural network scalar field prediction model is difficult to express the phenomenon of discontinuous jump of a fault position scalar field due to the influence of a fault effect is solved.

Description

Technical Field

[0001] The present application relates to the technical field of geological structure modeling, and specifically discloses a graph neural network three-dimensional geological structure modeling method. Background Art

[0002] The three-dimensional geological structure model is an important model for understanding the geometric morphology of underground hidden geological bodies and numerical simulation of geological processes. The construction of three-dimensional geological structure models can be divided into two categories: explicit modeling and implicit modeling. In the prior art, the implicit modeling method is usually used to represent the geological surface using the isosurface of the three-dimensional scalar field.

[0003] With the development of big data, the neural network geological modeling method, especially the one using graph neural network, has developed rapidly. In the existing technology, the graph neural network modeling method consists of a scalar field GNN model and a lithology category GNN model.

[0004] In the scalar field GNN model, by inputting the graph node coordinate matrix and node adjacency matrix to be calculated, the model parameters are trained under the loss function constraints of the occurrence data and interface property data, and the scalar field values ​​of the unknown nodes are calculated to obtain the scalar field values ​​of the nodes.

[0005] In the lithology category GNN model, the lithology category sampling point data is extracted through the virtual drilling of the geological profile. The scalar field value predicted by the scalar field GNN can be spliced ​​with the graph node coordinate matrix to form a joint graph feature matrix. Under the constraint of the loss function of the lithology category data, the model parameters are trained, the lithology category of the unknown node is calculated, and the lithology category of the node is obtained.

[0006] The scalar field GNN model and the lithology category GNN model can obtain the scalar field representing the stratigraphic interface to be modeled and the lithology category representing the stratigraphic entity.

[0007] However, the current graph neural network modeling method generally directly uses virtual boreholes to take discrete data points in the exploration line profile for lithology sampling points, and the topological relationship hidden in the map is lost during the sampling process. In addition, in actual modeling applications, we often encounter scenes where faults cut through strata, and fault cutting will cause the same stratum surface to be discontinuous in space. The current graph neural network scalar field prediction model is difficult to express the problem of discontinuous jumps in the scalar field at the fault position caused by the fault effect.

[0008] The present invention provides a graph neural network three-dimensional geological structure modeling method to solve the above problems. Summary of the invention

[0009] The purpose of the present invention is to solve the problem that the traditional graph neural network scalar field prediction model is difficult to express the phenomenon of discontinuity jump of the scalar field at the fault position caused by the fault effect.

[0010] In order to achieve the above object, the basic scheme of the present invention provides a graph neural network three-dimensional geological structure modeling method, comprising the following steps: Step A1: obtaining a three-dimensional geological map data set based on three-dimensional geological map preprocessing, and extracting scalar field attribute sampling points and lithology unit sampling points based on the three-dimensional geological map data set; Step A2: tetrahedroning the point data of the obtained scalar field attribute sampling points and lithology unit sampling points, and generating graph structure data with the tetrahedroning results; Step A3: constructing a graph neural network structure based on a dual-task GNN model, wherein the dual-task GNN model is used to predict the scalar field value of the node and to classify the lithology category of the node; Step A4: inputting the preprocessed graph structure data into the graph neural network structure, and generating the scalar field prediction value and the node lithology category obtained by the division by the graph neural network structure; Step A5: Based on the predicted value of the scalar field, isosurfaces are extracted from it by the tetrahedron algorithm as the stratigraphic interface, the tetrahedron units are classified by the node lithology labels, the tetrahedron units of the same lithology category are merged to generate a closed three-dimensional entity, and different stratigraphic entity models are obtained based on different lithology categories.

[0011] Furthermore, the three-dimensional geological map includes a plan geological map, an exploration line profile map and a mid-section plan map.

[0012] Further, in step A2, the graph structure data is represented as: G = (V, E); In the formula, G represents graph structure data, G is an undirected graph, V is a node set, including all nodes, V={v1,v2,...,v n}, where each node v i Corresponding to a node in the graph, v i ∈V, i=1,2,…,n; E is a set of edges, which represents the connection relationship between nodes. E={(i,j), i,j∈V}, (i,j) represents the edge between node i and node j.

[0013] Furthermore, each node v i They all have input features and input attributes. The input features of nodes include node coordinate features and fault coding features. The input attributes of nodes include scalar field attributes, occurrence attributes and lithology labels.

[0014] Furthermore, for the i-th node in the graph structure data, i=1,2,…, all have input features and input attributes, and their original input features and input attributes are expressed as follows: Feature i ={C i ,F i ,S i ,A i ,L i}; In the formula, C i represents the coordinate information of the i-th node, and C i ∈R n , R n is an n-dimensional geometric space, F i represents the fault coding feature of the i-th node, and F i ∈{1,2,…,K f}, K f is the number of faults, S i represents the scalar field property of the ith node, and S i ∈R p , R p is a p-dimensional attribute space, A i represents the occurrence attribute of the i-th node, and A i ∈R q , R q is a q-dimensional attribute space, L i represents the lithology label of the i-th node, and L i ∈{1,2,…,K l}, K l The number of lithologies.

[0015] Furthermore, for node v i and fault F k , k=1,2,…, the expression of its fault coding characteristics is as follows: ; In the formula, when the node v i Located on fault F k When the upper wall of , when node v i Located on fault F k When the footwall is .

[0016] Furthermore, when node v i When affected by k cross faults, the final fault coding feature F i for: ; In the formula, For node vi The encoding value on the kth fault indicates the position and relative relationship of the node on different faults.

[0017] Furthermore, the graph structure data is divided into continuous data and discrete data. In step A4, the preprocessing operation performed on the graph structure data includes preprocessing the continuous data by range normalization and encoding the discrete data by One-Hot encoding.

[0018] Furthermore, the dual-task GNN model includes a scalar field GNN model for predicting the scalar field value of a node and a lithology category GNN model for classifying the lithology category of the node. Both the scalar field GNN model and the lithology category GNN model are equipped with a three-layer stacked graph attention network module.

[0019] Further, in step A4, the steps of generating scalar field prediction values ​​and node lithology categories obtained by the graph neural network structure are as follows: Step B1, taking the node coordinate features, fault coding features and edge connection relationships as feature inputs of the scalar field GNN model, and outputting the scalar field prediction value of each node through the scalar field GNN model equipped with a three-layer stacked graph attention network module.

[0020] In step B2, the scalar field prediction value is concatenated with the node coordinate feature to obtain a feature matrix, and the concatenated feature matrix and edge connection matrix are used as the input of the lithology category GNN model. The lithology category GNN model equipped with a three-layer stacked graph attention network module outputs the predicted lithology category of each node.

[0021] The principle and effect of this scheme are: 1. Compared with the prior art, the present invention uses the entire geological map as input data. This method significantly enriches the lithological unit sampling point set while maintaining the internal topological relationship of the stratum entity, thereby more comprehensively reflecting the geological characteristics.

[0022] 2. Compared with the prior art, the fault encoding method of the present invention can automatically identify and simulate the impact of faults on the scalar field of the formation, especially the discontinuity jump phenomenon of the scalar field at the fault location, avoiding the problem of manually setting the fault displacement parameters in the traditional method, and can achieve accurate simulation of the fault effect without relying on human intervention, solving the problem that the traditional graph neural network scalar field prediction model is difficult to express the phenomenon of discontinuity jump of the scalar field at the fault location caused by the influence of the fault effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. It is obvious that the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 A schematic diagram of a graph neural network three-dimensional geological structure modeling method proposed in an embodiment of the present application is shown; Figure 2 The sampling diagram proposed in the embodiment of the present application is shown, wherein (a) is a three-dimensional model, (b) is a cross-sectional view of the three-dimensional model in the A-A' direction, and (c) is a plan geological map of the three-dimensional model in the B-B' direction; Figure 3 The sampling difference graphs proposed in the embodiments of the present application are shown, wherein (a) is a sampling graph obtained based on the traditional GNN sampling method, and (b) is a sampling graph obtained based on the sampling method proposed in the present application; Figure 4 The schematic diagram of tetrahedron decomposition proposed in the embodiment of the present application is shown, wherein (a) is a schematic diagram of the process of deleting equidistant interpolation points, and (b) is a schematic diagram of the process of tetrahedron decomposition with boundary and node constraints; Figure 5 The plan geological map of the modeling area proposed in the embodiment of the present application is shown; Figure 6 A schematic diagram of the graph data proposed in the embodiment of the present application is shown, wherein (a) is a three-dimensional view of the modeling graph data, (b) is a three-dimensional view of the sampling data point set, and (c) is an east-west three-dimensional view of the sampling data point set; Figure 7 A schematic diagram of cross-fault coding proposed in an embodiment of the present application is shown, wherein (a) is a fault schematic diagram, (b) is a new fault feature, (c) is an old fault feature, and (d) is a final fault feature; Figure 8 A schematic diagram of a three-dimensional formation scalar field model with fault effect proposed in an embodiment of the present application is shown, wherein (a) is a three-dimensional formation scalar field model with fault effect, (b) is a section 1 of the three-dimensional formation scalar field, (c) is a section 2 of the three-dimensional formation scalar field, and (d) is a section 3 of the three-dimensional formation scalar field; Fig. 9 A schematic diagram of a three-dimensional lithology category model with fault effect proposed in an embodiment of the present application is shown, wherein (a) is a three-dimensional lithology category model with fault effect, (b) is a section 1 of the three-dimensional lithology category model, (c) is a section 2 of the three-dimensional lithology category model, and (d) is a section 3 of the three-dimensional lithology category model; Fig.10A schematic diagram of a three-dimensional stratigraphic model with a fault effect proposed in an embodiment of the present application is shown, wherein (a) is an overlay display of the fault plane and the stratigraphic model, and (b) is an overlay display of the stratigraphic isosurface and the stratigraphic model. DETAILED DESCRIPTION

[0025] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0026] A graph neural network three-dimensional geological structure modeling method, implementing Figure 1 As shown, the following steps are included: Step A1: obtain a 3D geological map data set based on 3D geological map preprocessing, and extract sampling points based on the 3D geological map data set. The sampling points include scalar field attribute sampling points and lithology unit sampling points.

[0027] Three-dimensional geological maps include plan geological maps, exploration line profiles and mid-section plan maps.

[0028] The processing of three-dimensional geological maps includes vectorization of two-dimensional maps: plane geological maps, exploration line profiles and mid-section plane maps, and mapping from two-dimensional to three-dimensional based on spatial control points to generate three-dimensional geological map data sets.

[0029] Specifically, Figure 2 The sampling map shown extracts the stratigraphic boundaries, fault boundaries, and stratigraphic planes based on the 3D geological map data set. For the stratigraphic boundaries and fault boundaries, scalar field attribute sampling points are extracted at equal intervals on the lines.

[0030] Perform Delaunay triangulation on the stratigraphic surface and extract the nodes of the triangulation: the triangulated nodes are used as the sampling points of the lithological units. Figure 3 The sampling difference plot shown, Figure 3 (a) in the figure is the sampling graph obtained based on the traditional GNN sampling method. Figure 3 (b) is a sampling diagram obtained based on the sampling method proposed in this application, such as Figure 3 In (a), the topological information is lost due to the use of virtual boreholes for lithologic sampling. Compared with the traditional lithologic sampling method of first extracting a number of lithologic unit sampling points in the exploration profile through virtual boreholes and then performing triangulation, this embodiment performs uniform sampling in the entire exploration line profile, such as Figure 3 As shown in (b), the original topological relationship of the exploration line profile is retained, while the sampling data volume of the lithological unit is increased.

[0031] Step A2: Tetrahedroning is performed on the point data of the obtained scalar field attribute sampling points and lithology unit sampling points, and graph structure data is generated based on the tetrahedroning results.

[0032] Specifically, in the present invention, Figure 4 The tetrahedron division schematic diagram shown in the figure shows that in the process of tetrahedron division of the point data of the obtained scalar field attribute sampling points and lithological unit sampling points, the modeling space range and DEM surface are used as the boundary constraint surface of the tetrahedron division, and the cleaned equidistant interpolation encryption points and sampling points are used as the nodes of the tetrahedron division to ensure that the graph structure data is consistent with the spatial range of the modeling area, contains all known sampling points, and additionally adds encrypted control division points.

[0033] Specifically, the input modeling space range is used as the bottom boundary and side boundary of the tetrahedron; The DEM surface is used as the constraint boundary of the top boundary of the tetrahedron; All sampling points are used as nodes of tetrahedron division, and since the sampling points of lithological units are directly obtained from the triangulated network of geological maps, the topological relationship of the sampling points is preserved during tetrahedron division; Generate equidistant interpolation encryption points according to the specified modeling resolution. When the generated equidistant interpolation encryption points coincide with sampling points and DEM points, delete the overlapping encryption points.

[0034] In this embodiment, the graph structure is defined by a node set V and an edge set E, and the graph structure data generated based on the tetrahedron decomposition result is represented as: G = (V, E); In the formula, G represents graph structure data, G is an undirected graph, V is a node set, including all nodes, V={v1,v2,...,v n}, where each node v i Corresponding to a node in the graph, v i ∈V, i=1,2,…,n; E is a set of edges, which represents the connection relationship between nodes. E={(i,j), i,j∈V}, (i,j) represents the edge between node i and node j. And each node v i ∈V have input features and input attributes, and edge (i,j)∈E represents the connection relationship between node i and node j.

[0035] The input features of the node include node coordinate features and fault coding features, and the input attributes of the node include scalar field attributes, occurrence attributes, and lithology labels. The node coordinate features are obtained when the node is generated. The fault coding feature value is determined by the relative position of the node and the fault. The relative position of the node and the fault includes being located on the upper plate of the fault and being located on the lower plate of the fault. The scalar field attribute of the node is based on the top surface of the modeled stratum as the reference plane, and scalar field values ​​are assigned to different stratum interfaces according to the accumulated stratum thickness. The occurrence sampling points are extracted near the strata and fault lines, and the unit vector is calculated according to the dip and inclination of the occurrence points. The occurrence point data of the node is represented by the unit normal vector as the occurrence attribute. The lithology label of the node is represented by different integers.

[0036] For node v i , its original input features and input attributes are expressed as follows: Feature i ={C i ,F i ,S i ,A i ,L i}; In the formula, C i represents the coordinate information of the i-th node, and C i ∈R n , R n is an n-dimensional geometric space, F i represents the fault coding feature of the i-th node, and F i ∈{1,2,…,K f}, K f is the number of faults, S i represents the scalar field property of the ith node, and S i ∈R p , R p is a p-dimensional attribute space, A i represents the occurrence attribute of the i-th node, and A i ∈R q , R q is a q-dimensional attribute space, L i represents the lithology label of the i-th node, and L i ∈{1,2,…,K l}, K l The number of lithologies.

[0037] In this embodiment, the following fault coding method is established based on the relative relationship between nodes and faults: i Represents the spatial relative relationship between the node and different faults. i and fault F k, i=1,2,…, k=1,2,…, the expression of its fault coding characteristics is as follows: ; In the formula, when the node v i Located on fault F k When the upper wall of , when node v i Located on fault F k When the footwall is .

[0038] When there are cross faults, the final fault code of the node can be expressed as a vector that integrates the effects of multiple faults. i When affected by k cross faults, the final fault coding feature F i for: ; In the formula, For node v i The encoding value on the kth fault indicates the position and relative relationship of the node on different faults.

[0039] Specifically, the node located at the foot of the fault is assigned a value of 0, and the node located at the hanging wall of the fault is assigned a value of 1. For cross-faults, the partition fault code values ​​of different faults are superimposed as the final fault code characteristic value of a certain node.

[0040] In this implementation, the constructed graph structure data is divided into continuous data and discrete data. Continuous data includes coordinate features, scalar field attributes and occurrence attributes, and discrete data includes fault codes and lithology codes (lithology labels). Before inputting graph structure data as training data, the graph structure data needs to be preprocessed, including preprocessing continuous data by range normalization to accelerate the training convergence of the subsequent graph neural network structure, and encoding discrete data by One-Hot encoding. Specifically: For coordinate features, scalar field attributes, and occurrence attributes belonging to continuous data, the input features of each graph node are located in the three-dimensional space The normalized spatial coordinates of the nodes in and the fault encoding features.

[0041] Normalized spatial coordinates: Cartesian coordinates are calculated by rescaling the range of each node feature based on the maximum coordinate range, centering it around 0. Normalized spatial coordinates The coordinates are calculated as follows: ; ; ; ; In the formula, x v ,y v ,z v are the unnormalized x, y, and z coordinates of the node, respectively. , , They are the center values ​​of the x, y, and z coordinates, respectively. max ,y max ,z max They are the maximum coordinate values ​​in the x, y, and z directions respectively. min ,y min ,z min are the minimum coordinate values ​​in the x, y, and z directions respectively. x, y, and z are the scaling factors in the x, y, and z directions. In the present invention, the coordinate range is scaled between [-1, 1], and the scalar field attributes are scaled to between [-1, 1].

[0042] One-Hot coding is used to process the fault coding and lithology coding of discrete data. The principle is to convert the categorical variable into a binary vector, where each category corresponds to a unique position, and only the corresponding position of the category is assigned a value of 1, and the values ​​of other positions are 0. In this embodiment, Var is a categorical variable with n possible values. For the category value c i ∈c, its One-Hot encoding vector v i ∈{0,1} n The i-th component of is 1, and the other components are 0, that is: v i =[0,0,…,1,…,0]; In the formula, 1 appears in the i-th position, indicating the existence of the category.

[0043] Step A3: Construct a graph neural network structure based on a dual-task GNN model, where the dual-task GNN model is used to predict the scalar field value of the node and to classify the node lithology category.

[0044] Specifically, the dual-task GNN model includes a scalar field GNN model for predicting the scalar field value of a node and a lithology category GNN model for classifying the lithology category of the node.

[0045] The scalar field GNN model is based on the scalar field sampling point data (stratum boundary scalar field value point S i , Stratigraphic occurrence point A i , Fault feature code F i etc.) to generate scalar field prediction values ​​for each input graph node for subsequent stratum interface extraction.

[0046] The lithology category GNN model uses the prediction results from the scalar field GNN model: the scalar field prediction value as input to classify the lithology category of the node.

[0047] Both the scalar field GNN model and the lithology category GNN model include a three-layer stacked graph attention network module.

[0048] The scalar field GNN model includes an input layer, a three-layer stacked graph attention network module, a PReLU activation function, a linear layer, and an output layer, which outputs the scalar field prediction value.

[0049] The lithology category GNN model includes an input layer, a three-layer stacked graph attention network module, a PReLU activation function, a linear layer, and a Softmax classification layer.

[0050] Step A4: Input the preprocessed graph structure data into the graph neural network structure, and generate the scalar field prediction value and the node lithology category obtained by the division by the graph neural network structure.

[0051] Input graph structure data into the graph neural network structure, including node attributes (coordinates, fault codes, occurrence vectors, lithology categories, stratigraphic scalar values) and edge connection relationships.

[0052] After receiving the preprocessed graph structure data, the processing flow of the scalar field GNN model and lithology category GNN model carried inside the graph neural network structure includes the following steps: Step B1, take the node coordinate features, fault coding features and edge connection relationships as the feature input of the scalar field GNN model, embed the input features into high-dimensional vectors through a three-layer stacked graph attention network module, perform nonlinear transformation through the PReLU activation function, and use a linear layer to reduce the dimension of the features. A joint loss function is constructed through the scalar field loss function and the state vector loss function. Combined with the GradNorm loss weight adjustment strategy, the network parameters are iteratively updated in the back-propagation process to output the scalar field prediction value of each node.

[0053] Step B2, concatenate the scalar field prediction value with the node coordinate feature to obtain the feature matrix, use the concatenated feature matrix and edge connection matrix as the input of the lithology category GNN model, embed the input features into a high-dimensional feature vector through a three-layer stacked graph attention network module, use a linear layer to reduce the dimension to the number of lithology categories, use the Softmax function to use the lithology category with the highest prediction probability as the lithology formation category predicted by the current node, and use the formation category label loss function based on the cross entropy loss function to iteratively update the network parameters, and output the lithology category predicted for each node after the lithology category GNN model converges.

[0054] Specifically, after inputting the preprocessed graph structure data, each layer of the neural network i The convolution operation of its neighborhood is calculated on it to generate a new node embedding. i Embed high-dimensional features h i Initialized to node feature x i , and used as the input of the first layer of the network. In this way, the graph neural network can gradually update the node embedding to better capture the spatial relationship between nodes and the graph structure information.

[0055] like Figure 5 As shown, in view of the particularity of geological modeling, the present invention selects GAT, which belongs to spatial domain convolution, as a method for node embedding and representation. Spatial domain convolution directly performs convolution between nodes in the graph, and uses the connection relationship between nodes to perform weighted fusion of the features of each node's neighbors. In the present invention, a neighbor aggregation strategy based on the self-attention mechanism is used.

[0056] Specifically, for node v i and its neighbor node v j , calculate the attention coefficient α ij , which reflects the node v j For node v i The importance of information transmission is calculated as follows: ; In the formula, a is the learned attention parameter, || represents the concatenation operation of the feature vector, W is the weight matrix, and h i and h j They are node v i and node v j high-dimensional feature representation.

[0057] For complex graph learning tasks, a multi-head attention mechanism is introduced. The embedding of each node is calculated through multiple attention heads, each of which independently aggregates the features of its neighbors and learns through different weight matrices. The final node representation is the concatenation or average of the outputs of all heads. When there are K attention heads, the output of the multi-head attention is: ; In the formula, The output of the kth self-attention head is given by the formula: ; Each self-attention head independently learns a set of attention coefficients and the weight matrix W k .

[0058] In this embodiment, different loss functions are designed according to the type of network prediction (e.g., continuous variables, discrete classes) and learning objectives, and based on the different characteristics of the scalar field GNN model and the lithology category GNN model regression, classification tasks, and data input conditions.

[0059] For the scalar field GNN model, a scalar field loss function of the formation interface sampling point, a scalar field loss function of the lithology unit sampling point, and an occurrence vector loss function are established, as follows: Scalar field loss function of the formation interface sampling point: Scalar field prediction value It is generated by each graph node v of the scalar field GNN module and is constrained by the interface data points of the stratum and structural direction (inclination, dip). It is represented in three-dimensional space as a scalar function f(p), where f represents the buried depth of the stratum at point p in three-dimensional space relative to a certain stratum surface (the latest modeled geological surface is selected as the reference surface in this invention). For a series of H stratum interfaces in space, it can be represented as: ; In the formula, f1 and f H are the oldest and newest stratigraphic scalar field values, respectively.

[0060] In the process of geological structure modeling, different stratigraphic models can be obtained by extracting the scalar field isosurfaces of two adjacent stratigraphic layers and using the adjacent isosurfaces as constraints.

[0061] Given a scalar field f v The subset of graph nodes v sampled from a restricted level I can be expressed as follows: I = {v∈V|s v =f v} Scalar field value s at the formation interface sampling point v and the scalar field value predicted by the network The difference between them can be expressed using the loss function L interface To measure: ; Scalar field loss function of lithology unit sampling point: The lithology unit sampling point is sampled in the stratum between the two stratum interfaces, and the scalar field value of the lithology unit sampling point should also be between the scalar field values ​​of the two stratum interfaces. v ) is a lithology-dependent label L v Function: R(L v )=[a v ,b v ] In the formula, a v Is the lithology label L v The corresponding minimum scalar field value, a v =R(L v ) min , b v Is the lithology label L v The corresponding maximum scalar field value, b v =R(L v ) max The scalar field loss function at a certain lithological unit sampling point can be defined as: ; In the formula, is the scalar value of the node with the true lithology label predicted by the current network. v The maximum value of its scalar value is R(L v ) max , the minimum value is R(L v ) min , the correct scalar value should be between the two ranges. >R(L v ) max or <R(L v ) min , indicating that there is a deviation between the scalar field value predicted by the current network and the true field value range, then the absolute value of the deviation is taken as the loss value L of the current lithological unit sampling point litho .

[0062] For all lithology unit sampling point sets in the entire graph, the final total loss function L sum_litho It can be expressed as: ; Output vector loss function: Nodes containing occurrence data can be obtained from geological maps, representing the actual changes in the surface near the point. In order to measure the angular error between the sampling point and the predicted node, the gradient of the predicted scalar field on these graph nodes must be estimated. In the present invention, the scalar field value is approximated by a first-order Taylor series within the first-order neighborhood of the node to be gradiented, and the directional derivative of the predicted scalar field is obtained to perform least squares fitting, as shown in the following formula: ; In the formula, is the estimated gradient of the scalar field at node v, the matrix P v and S v Given by: ; ; In the formula, , , is the normalized spatial coordinate of a given node, the predicted scalar field gradient at node v is equal to the known direction vector α v The estimated angle between is: ; This angle is used to measure the constraints between the predicted angle and the observed angle (true angle) The angle error. For a given angle constraint The loss function is: ; But for normal data , and its loss function is: ; For tangent data , and its loss function is: ; During the scalar field GNN model training process, L interface and L orientation As a loss function, different weights are given to the loss function according to the learning objectives. Different loss function units and numerical ranges may be inconsistent, resulting in unbalanced gradient updates. The present invention introduces the GradNorm (Gradient Normalization) idea to optimize the training process. It adaptively adjusts the weight of the loss function of each task to ensure that the gradient of each task has a similar scale during the training process, thereby improving the stability and efficiency of the training.

[0063] For the lithology category GNN model, a stratum category label loss function is established, as follows: In the process of modeling practice, the data of geological boundary sampling points are relatively scarce. In order to make full use of the data of geological maps, the lithological unit sampling points are included in the modeling data in the form of stratigraphic category loss function.

[0064] Create a measured data set R of lithologic units, which contains N related to a subset of graph nodes. class possible lithology categories, and each lithology unit measured data point is assigned a one-hot encoding (One-Hot) vector label v , indicating the class it belongs to. A one-hot encoded vector is a vector whose components are all 0 except for a given component with a value of 1.

[0065] For the inclusion category N class =3 data set, the corresponding one-hot encoding vectors of categories a, b, and c are: ; ; ; The loss function used by one-hot encoding is the cross entropy loss function L rockunit : ; In the formula, is the probability value of predicting a certain category obtained from the following formula. The calculation formula is as follows: ; In the formula, For vector The Softmax function is commonly used in the field of machine learning to normalize the network output to the value of the lithology category corresponding to the i-th element of N. class The probability distribution among the categories.

[0066] Step A5: Based on the scalar field prediction value, isosurfaces are extracted from it through the tetrahedron algorithm as the stratigraphic interface, tetrahedral units are classified by node lithology labels, tetrahedral units of the same lithology category are merged to generate closed three-dimensional entities, and different lithology categories are used as different stratigraphic entity models.

[0067] Modeling example Figure 5 As shown, Figure 5 The exposed strata in the area are, from old to new, the Neoproterozoic Qingbaikou System Lengjiaxi Group Xiaomuping Formation (Pt x ) and Banxi Group Madiyi Group (Pt m )、Tongtawan Formation(Pt t ), Cretaceous (K) and Quaternary (Q). According to the thickness range of each stratum in the comprehensive stratigraphic column, the burial depth of each geological interface relative to the top of the Quaternary was determined, and the relative burial depth was used as the field value used by the stratigraphic boundary node attribute.

[0068] The planar geological map and the cross-sectional geological map are triangulated and combined with the determined spatial range to generate tetrahedral map data, such as Figure 6 shown.

[0069] like Figure 7 As shown in (a), the new fault F1 runs through the entire area and cuts off the old fault F2, dividing the area into three parts. Figure 7 In (b), the region is divided into two regions of 1 and 0 according to the rise and fall of the fault. P1, P2, and P3 are points at three different locations in the region, and F is the coded fault feature, such as The fault characteristic code representing the F1 fault at point P2 is 1. Figure 7 In (c), since the old fault is cut by the new fault, the right side of the F2 fault is area 1, and the other areas are area 0. The fault code of point P2 is , Figure 7 In (d), the cross-fault coding features are fused, and F2=[1,0] means that the fault feature coding of the final point P2 is [1,0].

[0070] After preparing the graph data for modeling, the model is trained according to the following model parameters. The network depth of the model is set to 3 layers. A shallower network depth helps avoid overfitting and reduce the problem of gradient disappearance.

[0071] The initial learning rate is set to 0.01. During the training process, the learning rate adaptive decay strategy is used. If the loss function decreases by less than 0.01 every 10 cycles, the learning rate is decayed by 50%. The PReLU (Parametric ReLU) activation function is selected. The embedding dimension is selected as 128, and the number of attention heads is set to 2. Then, the scalar field value and lithology category of the node are trained and predicted through the scalar field GNN model and the lithology category GNN model.

[0072] After the training, the formation interface and formation entity model are extracted based on the scalar field GNN model and lithology category GNN model generated above. For the scalar field, the marching tetrahedron algorithm is used to extract different isosurfaces as the formation interface, and different lithology categories are directly extracted as different formation entity models. The method of the present invention constructs the following Figure 8 The three-dimensional stratigraphic scalar field model with fault effect shown in Fig. 9 The three-dimensional lithology model with fault effects shown in Fig.10 The three-dimensional stratigraphic model with fault effects is shown.

[0073] Compared with the traditional graph neural network that only uses virtual borehole data as input, the present invention innovatively uses the entire geological map as input data. This method significantly enriches the lithological unit sampling point set while maintaining the internal topological relationship of the stratigraphic entity, thereby more comprehensively reflecting the geological characteristics. In the tetrahedron partitioning process, the present invention introduces the digital elevation model (DEM) surface as the upper boundary surface constraint of the graph structure, and considers the impact of the DEM topological relationship on the model, so that the partitioning process not only retains the stratigraphic surface characteristics, but also reflects the local geological topological information, thereby enhancing the model's ability to capture local changes and details.

[0074] In addition, the present invention proposes a fault coding method, through which the influence of faults on the scalar field of the formation can be automatically identified and simulated, especially the discontinuity jump phenomenon of the scalar field at the fault position, avoiding the problem of manually setting the fault displacement parameters in the traditional method, and can achieve accurate simulation of the fault effect without relying on human intervention.

[0075] The present invention also constructs a scalar field GNN model and a lithology category GNN model based on a three-layer stacked graph attention network module (GAT) on the basis of the graph neural network, and introduces a self-attention mechanism and a multi-head attention mechanism respectively, which can dynamically adjust the weight difference of each node according to the importance of each node, thereby enhancing the expression ability of the model in a complex geological environment. At the same time, the appropriate GNN model can be selected according to different data types, such as continuous or discrete, to expand the applicability of modeling data. In terms of multi-task learning, the present invention constructs a multi-task loss function including scalar fields, occurrence vectors, stratum categories, etc., and introduces the GradNorm algorithm to automatically adjust the weights of different loss functions during the training process to achieve adaptive loss balance. This mechanism avoids the tediousness of manually setting loss weights in traditional methods, and significantly improves the learning performance and stability of the model.

[0076] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modifications to equivalent embodiments of equivalent changes by using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A graph neural network three-dimensional geological structure modeling method, characterized in that: The steps include: Step A1: obtaining a three-dimensional geological map data set based on three-dimensional geological map preprocessing, and extracting scalar field attribute sampling points and lithology unit sampling points based on the three-dimensional geological map data set; Step A2: tetrahedroning the point data of the obtained scalar field attribute sampling points and lithology unit sampling points, and generating graph structure data with the tetrahedroning results; Step A3: constructing a graph neural network structure based on a dual-task GNN model, wherein the dual-task GNN model is used to predict the scalar field value of the node and to classify the lithology category of the node; Step A4: inputting the preprocessed graph structure data into the graph neural network structure, and generating the scalar field prediction value and the node lithology category obtained by the division by the graph neural network structure; Step A5: Based on the predicted value of the scalar field, isosurfaces are extracted from it by the tetrahedron algorithm as the stratigraphic interface, the tetrahedron units are classified by the node lithology labels, the tetrahedron units of the same lithology category are merged to generate a closed three-dimensional entity, and different stratigraphic entity models are obtained based on different lithology categories.

2. A graph neural network three-dimensional geological structure modeling method according to claim 1, characterized in that: The three-dimensional geological map includes a plan geological map, an exploration line profile map and a mid-section plan map.

3. A graph neural network three-dimensional geological structure modeling method according to claim 1, characterized in that: In step A2, the graph structure data is represented as: G = (V, E); In the formula, G represents graph structure data, G is an undirected graph, V is a node set, including all nodes, V={v1,v2,...,v n }, where each node v i Corresponding to a node in the graph, v i ∈V, i=1,2,…,n; E is a set of edges, which represents the connection relationship between nodes. E={(i,j), i,j∈V}, (i,j) represents the edge between node i and node j.

4. A graph neural network three-dimensional geological structure modeling method according to claim 3, characterized in that: Each node v i They all have input features and input attributes. The input features of nodes include node coordinate features and fault coding features. The input attributes of nodes include scalar field attributes, occurrence attributes and lithology labels.

5. A graph neural network three-dimensional geological structure modeling method according to claim 3, characterized in that: For the i-th node in the graph structure data, i=1,2,…, all have input features and input attributes, and their original input features and input attributes are expressed as follows: Feature i ={C i ,F i ,S i ,A i ,L i }; In the formula, C i represents the coordinate information of the i-th node, and C i ∈R n , R n is an n-dimensional geometric space, F i represents the fault coding feature of the i-th node, and F i ∈{1,2,…,K f }, K f is the number of faults, S i represents the scalar field property of the ith node, and S i ∈R p , R p is a p-dimensional attribute space, A i represents the occurrence attribute of the i-th node, and A i ∈R q , R q is a q-dimensional attribute space, L i represents the lithology label of the i-th node, and L i ∈{1,2,…,K l }, K l The number of lithologies.

6. A graph neural network three-dimensional geological structure modeling method according to claim 5, characterized in that: For node v i and fault F k , k=1,2,…, the expression of its fault coding characteristics is as follows: ； In the formula, when the node v i Located on fault F k When the upper wall of , when node v i Located on fault F k When the footwall is .

7. A graph neural network three-dimensional geological structure modeling method according to claim 6, characterized in that: When node v i When affected by k cross faults, the final fault coding feature F i for: ； In the formula, For node v i The encoding value on the kth fault indicates the position and relative relationship of the node on different faults.

8. A graph neural network three-dimensional geological structure modeling method according to claim 4, characterized in that: The graph structure data is divided into continuous data and discrete data. In step A4, the preprocessing operation performed on the graph structure data includes preprocessing the continuous data by range normalization and encoding the discrete data by One-Hot encoding.

9. A graph neural network three-dimensional geological structure modeling method according to claim 4, characterized in that: The dual-task GNN model includes a scalar field GNN model for predicting the scalar field value of a node and a lithology category GNN model for classifying the lithology category of the node. Both the scalar field GNN model and the lithology category GNN model are equipped with a three-layer stacked graph attention network module.

10. A graph neural network three-dimensional geological structure modeling method according to claim 9, characterized in that: In step A4, the steps of generating scalar field prediction values ​​and node lithology categories obtained by the graph neural network structure are as follows: Step B1, taking the node coordinate features, fault coding features and edge connection relationships as feature inputs of the scalar field GNN model, and outputting the scalar field prediction value of each node through the scalar field GNN model equipped with a three-layer stacked graph attention network module; In step B2, the scalar field prediction value is concatenated with the node coordinate feature to obtain a feature matrix, and the concatenated feature matrix and edge connection matrix are used as the input of the lithology category GNN model. The lithology category GNN model equipped with a three-layer stacked graph attention network module outputs the predicted lithology category of each node.

Images