Seismic Full Horizon Tracking Method Based on Multi-Attribute Knowledge Graph

Abstract

The present invention discloses a seismic full horizon tracking method based on a multi-attribute knowledge graph, which is applied to the field of oil and gas exploration. Aiming at the problem that the existing seismic full horizon tracking methods are difficult to effectively represent the stratigraphic relationships in seismic data, resulting in the phenomenon of time slippage easily occurring in areas with complex geological structures such as faults in the tracking results and being difficult to meet the actual application requirements. The present invention first uses the dynamic time warping algorithm to construct horizon segments, and calculates the fault attribute and dip attribute by using transfer learning and gradient structure tensor respectively. Subsequently, the multi-attribute knowledge graph is used to characterize the stratigraphic relationships in seismic data. Finally, under the constraints of the fault attribute, dip attribute and stratigraphic relationship, the nodes in the initial multi-attribute knowledge graph are continuously fused, and a complete horizon plane is obtained on the basis of conforming to seismic stratigraphy. Experiments show that the method of the present invention can effectively represent the stratigraphic relationships of seismic horizons in the underground three-dimensional space and accurately track the horizons located at complex geological structures such as faults.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas exploration, and specifically relates to a seismic full-layer tracking technology. Background Technology

[0002] Three-dimensional seismic full-layer tracing is of great significance for three-dimensional sequence stratigraphy analysis and three-dimensional reservoir modeling, and is one of the important tasks in oil and gas exploration. Accurate and efficient full-layer tracing helps to reveal the interrelationships between stratigraphic interfaces of different ages. The traced stratigraphic interfaces provide a stratigraphic framework for subsequent three-dimensional sequence stratigraphy analysis. The detailed stratigraphic distribution and comprehensive stratigraphic interface information obtained provide a basic stratigraphic framework for the establishment of three-dimensional reservoir models. As fundamental data, it helps to improve the accuracy and reliability of subsequent work such as elastic parameter inversion and reservoir prediction.

[0003] Compared to conventional single-layer and multi-layer seismic tracking, full-layer seismic tracking differs in the number and targets of stratigraphic interfaces it tracks. Single-layer and multi-layer seismic tracking primarily utilize seismic data processing techniques (such as seismic attributes like amplitude, phase, and frequency) and stratigraphic morphology features on seismic profiles to identify and track one or more specific stratigraphic interfaces. These methods often focus on one or more strata of exploration significance within the seismic data. In contrast, full-layer seismic tracking is a technique for identifying and tracking all stratigraphic interfaces in seismic data. This method offers significant advantages in constructing an overall stratigraphic framework, extracting macroscopic stratigraphic information, and describing sequence relationships between strata.

[0004] Researchers have developed numerous automated full-sequence tracking algorithms aimed at obtaining broader and more detailed sequence information. These methods fall into four categories. The first category is the slope-based seismic reflection surface normal vector method. While these slope-based methods are effective, they are prone to accumulating errors and fail to track the correct seismic reflections at faults. To address this, researchers preprocess faults before extracting the horizon, such as removing faults from seismic data, using manual control points (typically placed on both sides of the fault) as constraints, calculating multi-grid correlations, and using multiple seismic properties. However, these methods may lack a description of sequence relationships when tracking horizons. The second category involves calculating relative geological time (RGT) values ​​and extracting RGT contour lines to obtain horizons. While these methods are globally optimal, they tend to produce smooth horizon lines lacking detailed geological information and require significant manual intervention in complex geological structures. The third category is clustering-based methods that follow a process of horizon fragmentation and fragment grouping. However, these methods cannot maintain correct sequence relationships in geologically complex areas. The fourth category is methods that directly calculate geological models from seismic data using mathematical algorithms. These methods maintain global optimum while tracking stratigraphic levels, thus avoiding error accumulation. However, the resulting geological models may not match the actual geological structure, leading to time-lapses in tracking results in areas with complex geological structures.

[0005] It can be observed that whole-layer tracing (WLT) requires simultaneous consideration of the interactions and influences between all stratigraphic interfaces when performing stratigraphic segment matching. Therefore, a complete and accurate description of the sequence relationships between all strata is crucial for WLT; the lack of effective sequence constraints often leads to time-lapse phenomena in the tracing results. Existing WLT methods lack an effective computer representation method for describing the complex relationships between stratigraphic segments in three-dimensional space. This makes it difficult for computers to fully and effectively understand the sequence relationships between stratigraphic interfaces, resulting in time-lapse phenomena in the tracing results in areas with complex geological structures. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes a seismic full-layer tracing method based on multi-attribute knowledge graphs. This method enables computers to accurately describe the stratigraphic sequence relationships of all stratigraphic interfaces in three-dimensional space, effectively avoiding time-lapse phenomena caused by the lack of stratigraphic sequence constraints in complex geological structures.

[0007] The technical solution adopted in this invention is: a seismic full-layer tracing method based on multi-attribute knowledge graphs, comprising:

[0008] S1. Preprocess the seismic data to obtain layer segments, fault attributes, and dip attributes;

[0009] S2. Use the layer fragments as nodes and create edges between nodes within the distance threshold to construct an initial multi-attribute knowledge graph;

[0010] S3. For the current multi-attribute knowledge graph, based on whether the two layer segments represented by the nodes have a hierarchical relationship, the edges between the nodes are divided into two types: directed edges and undirected edges; a directed graph is constructed based on the directed edges, and an undirected graph is constructed based on the undirected edges.

[0011] S4. Calculate the properties of undirected edges in an undirected graph based on the tilt angle property;

[0012] S5. Nodes with attributes of undirected edges greater than a relevant threshold are classified as nodes to be merged; based on the fault attributes, the nodes to be merged are divided into nodes with connectivity and nodes with faults. All nodes connected by the same undirected edge with connectivity are directly merged to form a new node. All nodes connected by undirected edges with faults are merged by judging the connectivity of the nodes; the distance threshold is expanded, and edges are created between nodes within the distance threshold to obtain a new multi-attribute knowledge graph.

[0013] S6. If the new multi-attribute knowledge graph still has nodes that are not connected across faults, return to step S3; otherwise, obtain the final full-layer tracking result.

[0014] The beneficial effects of this invention are as follows: First, this invention uses the Dynamic Time Planning (DTW) algorithm to construct stratigraphic fragments, and then utilizes transfer learning and Gradient Structure Tensor (GST) to calculate fault attributes and dip attributes, respectively. Subsequently, a multi-attribute knowledge graph is used to represent the stratigraphic sequence relationships in seismic data. Here, the initial multi-attribute knowledge graph represents the overall distribution of stratigraphic fragments in the data space, where nodes represent stratigraphic fragments, node attributes encompass information such as the stratigraphic sequence relationships of the fragments and the geographical location of faults, edge attributes provide the connectivity between stratigraphic fragments, and the magnitude of the attribute values ​​represents the probability of node fusion. Finally, under the constraints of fault attributes, dip attributes, and stratigraphic sequence relationships, nodes in the initial multi-attribute knowledge graph are continuously fused, obtaining a complete stratigraphic plane based on seismic stratigraphy and avoiding time-lapse errors. Experiments show that the method of this invention can effectively represent the stratigraphic sequence relationships of seismic horizons in three-dimensional subsurface space and accurately track horizons located at complex geological structures such as faults. Attached Figure Description

[0015] Figure 1 The DTW algorithm flow;

[0016] Wherein, (a) is the original seismic data; (b) is the seismic data sampled to obtain seismic data blocks; (c) is the identification of extreme points; and (d) is the extension of the matching results of extreme points into the sampled seismic data volume to form layer fragments.

[0017] Figure 2 A flowchart for full-level tracing using multi-attribute knowledge graphs;

[0018] Figure 3 The process of creating edges;

[0019] Figure 4 There are three types of edges between two nodes;

[0020] Among them, (a) is an edge of type E1, (b) is an edge of type E2, and (c) is an edge of type E3;

[0021] Figure 5 Selecting extreme points;

[0022] Figure 6 This is a schematic diagram illustrating node correlation calculation.

[0023] Figure 7 This is a schematic diagram of point correlation calculation;

[0024] Figure 8 A four-layer, multi-attribute knowledge graph for F3 data;

[0025] Figure 9 Visualizing multi-attribute knowledge graph data;

[0026] Figure 10 The Paleoscan software is used to visualize the fault.

[0027] Figure 11 This demonstrates the effect of the method of the present invention at the fault.

[0028] Figure 12 A comparison of the effects at layer P2;

[0029] (a) shows the effect of Paleoscan software, and (b) shows the effect of the method of the present invention.

[0030] Figure 13 Comparison of effects at Q7 layer;

[0031] (a) shows the effect of Paleoscan software, and (b) shows the effect of the method of the present invention.

[0032] Figure 14 For comparison of effects across multiple layers;

[0033] In the figure, (a) shows the effect of Paleoscan software, and (b) shows the effect of the method of the present invention. Detailed Implementation

[0034] To facilitate understanding of the technical content of this invention by those skilled in the art, the following description, in conjunction with the accompanying drawings, further illustrates the invention.

[0035] This invention proposes a method for representing sequence relations between strata in three-dimensional space using a multi-attribute knowledge graph. This method enables computers to accurately describe the sequence relations of all stratigraphic interfaces in three-dimensional space, effectively avoiding the time-travel phenomenon caused by the lack of sequence constraints in tracking results in complex geological structures. First, this invention merges data extrema points into stratigraphic fragments and calculates fault and dip attributes. Then, it constructs an initial multi-attribute knowledge graph that can represent the three-dimensional spatial relationships of sequence strata. To achieve stratigraphic fragment fusion, the initial multi-attribute knowledge graph merges nodes under the constraints of fault attributes, dip attributes, and sequence relations, and obtains all non-time-travel stratigraphic surfaces at once through connectivity criteria. Simultaneously, since the stratigraphic fragments in the knowledge graph are independent, this method can effectively avoid error accumulation during node fusion. Experiments using data from two actual work areas show that the proposed method can effectively represent the sequence relations of seismic horizons in subsurface three-dimensional space and effectively avoid the time-travel phenomenon in three-dimensional full-stratigraphic tracking in geologically complex areas.

[0036] The implementation process of the method of the present invention includes the following steps:

[0037] 1. First, prepare the input data:

[0038] Seismic data preprocessing includes generating horizon fragments and calculating fault and dip attributes. The generated horizon fragments, fault attributes, and dip attributes will provide data support for the construction of the initial multi-attribute knowledge graph.

[0039] (1) Generate layer fragments

[0040] Stratified fragments are generated using the DTW (Dynamic Time Warping) algorithm. First, a smoothing filter is applied along obvious geological structures in the seismic data to enhance structural features while preserving important discontinuities such as faults or unconformities. Then, the DTW algorithm is used to generate the stratified fragments. The workflow is as follows: Figure 1 As shown, it mainly consists of three steps.

[0041] The first step is to create a regular grid with a side length of 3 to sample the seismic data, thus obtaining sampled seismic data blocks. The second step is to identify extreme points and match them within the seismic data blocks. The third step is to extend the matching results of the extreme points into the sampled seismic data volumes to form layer fragments. Typically, for larger datasets, this embodiment can set the grid side length to 5 or 7. Although the DTW algorithm accumulates errors, setting the grid side length to a smaller value can limit the errors to an acceptable range.

[0042] (2) Calculate fault properties

[0043] Fault attributes are calculated using transfer learning methods. One part of the deep transfer learning model learns fault-related features, while another part mines common features between real and synthetic seismic data, making the deep transfer learning model more suitable for real-world seismic data.

[0044] (3) Calculate the tilt angle property

[0045] Dip properties are calculated using a gradient structural tensor (GST). First, gradient values ​​in three directions are calculated, and the gradient structural tensor is constructed. Then, eigenvalue decomposition is performed on the gradient structural tensor; the resulting eigenvalues ​​and corresponding eigenvectors contain information about the local structure of the seismic data. Finally, the eigenvalues ​​and eigenvectors are analyzed to obtain the dip properties.

[0046] 2. Multi-attribute knowledge graph method

[0047] The innovation of this invention lies in using a multi-attribute knowledge graph for full-level tracing. The workflow mainly consists of two steps, such as... Figure 2 As shown: (1) Initial multi-attribute knowledge graph construction, (2) Multi-attribute knowledge graph node fusion. The first step is to construct an initial multi-attribute knowledge graph and convert earthquake data into graphic data, so that the computer can accurately represent the stratigraphic sequence relationship and the overall distribution of stratigraphic segments in the data space. The second step is to continuously fuse the nodes in the initial multi-attribute knowledge graph using stratigraphic sequence relationship, fault attributes and dip angle attributes to form a multi-layered multi-attribute knowledge graph. The specific details of these two steps are as follows:

[0048] (1) Initial multi-attribute knowledge graph construction

[0049] During the input data preparation phase, horizon segments, fault attributes, and dip attributes are obtained. To better represent the overall distribution of horizon segments and faults, this embodiment constructs an initial multi-attribute knowledge graph to manage the seismic data representing the graphical data. Before constructing this initial multi-attribute knowledge graph, this embodiment needs to first construct an initial graph to convert the seismic data into graphical data, defined as G. 0 (V, E). This initial graph contains nodes and edges, denoted by V and E respectively. Nodes are directly derived from seismic data. Each node represents a stratigraphic segment, and node attributes include the stratigraphic sequence relationship of the segment and the geographical location information of the fault. Edges are created based on the distances between nodes. Edges represent the relationships between nodes, and edge attributes represent the probability of merging these nodes. In this embodiment, a distance threshold d is preset, and Chebyshev distance is used to calculate the distances between nodes. If the calculated distance value is less than the preset distance threshold d, an edge is created between the nodes. For any two points P and Q in three-dimensional space, such that P = (px p y p z ) and Q = (q x q y q z The Chebyshev distance between P and Q is:

[0050] D chebyshev (P, Q) = max(|p x -q x |,|p y -q y |,|p z -q z |)

[0051] Where, p x Let p represent the x-coordinate of point P. y Let p represent the y-coordinate of point P. z q represents the z-axis coordinate of point P; x Let q represent the x-coordinate of point Q. y Let q represent the y-coordinate of point Q. z This represents the z-axis coordinate of point Q;

[0052] For two nodes N1 and N2 in the initial graph, such that N1 = {P1, P2, ..., P...} m} and N2={Q1,Q2,…,Q n}, where m and n represent the number of extreme points at each node. The distance between N1 and N2 is defined as follows:

[0053]

[0054] If Dist(N1, N2) < d, an edge can be created between these two nodes. Generally, this embodiment sets d to a small value to ensure that edges are created primarily between the closest nodes. In the experiments of this invention, d is set to 2, and its selection is determined by the quality of the seismic data and the edge length of the sampled grid. The process of creating the edge is as follows... Figure 3 As shown. If the seismic data quality is very low or the grid side length becomes large, d should be set to a larger value.

[0055] Having completed the edge creation, this embodiment categorizes edges into three types, represented as E1, E2, and E3. These three types of edges correspond to three types of relationships between nodes: hierarchical relationships, connectivity relationships, and discontinuity relationships, such as... Figure 4As shown. Stratigraphic sequence relationships indicate a strict hierarchical relationship between two stratigraphic segments, suggesting that the centers of these segments lie on the same seismic trace. Connectivity relationships indicate that two stratigraphic segments may connect. Fault relationships indicate that two stratigraphic segments are spatially close to a fault; the fault relationship between two stratigraphic segments is determined by fault properties. Thus, this embodiment completes the construction of the initial map.

[0056] Based on whether the two layer segments represented by the nodes have a hierarchical relationship, E1, E2, and E3 can be divided into two types: initially directed edges and initially undirected edges. Initially directed edges have a hierarchical relationship and include E1. Initially undirected edges lack a hierarchical relationship and include E2 and E3. Clearly, nodes connected by initially directed edges cannot be merged, while nodes connected by initially undirected edges have the potential to be merged. After determining the initial directed edges and initial undirected edges, this embodiment can construct an initially directed graph and an initially undirected graph, denoted as […]. and Together they constructed the initial multi-attribute knowledge graph, using G... 0 (V, E) represents the initial multi-attribute knowledge graph, which depicts the hierarchical relationships of layer fragments, further enhancing the computer's understanding of complex relationships in underground three-dimensional space. In subsequent steps, this embodiment will utilize the combined constraints of hierarchical relationships, fault attributes, and dip angle attributes to fuse nodes, and construct a multi-layered multi-attribute knowledge graph by creating edges that are further apart.

[0057] (2) Multi-attribute knowledge graph node fusion

[0058] like Figure 2 As shown, node fusion in a multi-attribute knowledge graph is a cyclical process comprising six steps: calculating undirected edge attributes, node fusion, edge creation, redetermining directed and undirected edges, constructing a higher-level multi-attribute knowledge graph, and determining the iteration stopping criterion. Undirected edge attributes represent the connection potential between pairs of nodes connected by the undirected edge, and their calculation forms the basis for subsequent node fusion. However, each node in the initial multi-attribute undirected graph still has undirected edges with multiple other nodes. Therefore, this embodiment requires combining tilt angle attributes and utilizing correlations to calculate each undirected edge attribute. For two nodes N1 and N2 with undirected edges along the tilt angle direction, their extreme point sets are N1 = {P1, ..., P...} m} and N2={Q 1， ..., Q n} where m and n represent the number of extreme points in N1 and N2, respectively. Next, this embodiment finds the two closest extreme points in N1 and N2 as...

[0059]

[0060] Given an integer parameter k, this embodiment finds the closest P in N1 and N2 respectively. a and Q b The k extreme points are denoted as S(P) a ) and S(Q b It is worth noting that S(P) a ) and S(Q b It also contains P. a and Q b This indicates that S(P) a ) and S(Q b Each of the following has k+1 points. In this embodiment of the invention, k is set to 5, such as... Figure 5 As shown. To prevent errors when there are too few extreme points in the nodes, k can be updated immediately after being given, using the minimum of k, m, and n. Now, this embodiment uses these extreme points to calculate the correlation, where two nodes with high correlation correspond to a high probability of connection. For S(P a ) and S(Q b For each extreme point in the data, this embodiment selects L seismic data points along the tline (time) direction, where L is set to 5, to form k+1 seismic microchannels. Furthermore, the center point of these seismic microchannels is parallel to S(P). a ) and S(Q b Each extreme point in the equation coincides with the other extreme point, such as... Figure 6 As shown. The correlation between computing nodes is defined as...

[0061]

[0062] Where P i ∈S(P a ) and Q j ∈S(Q b ). and Representing P respectively i and Q j The earthquake microchannel centered on the earthquake. and Represented by and The average value of the calculated seismic micro-traces is used. This calculation strategy improves the robustness of the results. At the same time, the independence of each node prevents the accumulation of errors during the calculation of correlations between nodes.

[0063] By determining the properties of the edges, this embodiment obtains the possibility of establishing connections between adjacent nodes. If two adjacent nodes N1 and N2 belong to the same level, they should exhibit the maximum correlation, reflected in the maximum value R. node In contrast, if N1 and N2 are not part of the same layer, then their R...node The value should not be the largest. However, the largest R is... node The value does not necessarily mean that the two nodes belong to the same layer. For two nodes with a fault relationship, setting the distance threshold too small may cause the calculation of undirected edge attributes of nodes in the same layer to be ignored. To prevent such erroneous connections, this embodiment proposes an undirected edge attribute judgment function R(N1, N2). Given the relevant threshold λ, the function R(N1, N2) is defined as follows:

[0064]

[0065] Here, the value of λ is determined based on the quality of the seismic data; in this embodiment, λ is 0.7. After evaluating all undirected edge attributes using the proposed function R(N1, N2), this embodiment can continue to merge nodes according to the principle of point set topology. All undirected edge attributes are sorted in descending order, and nodes with the largest attribute values ​​are merged first. However, merging nodes leads to an update of the directed graph, which guides the merging of subsequent nodes. If this guidance is ignored, incorrect node merging may lead to stratigraphic time travel. To address this issue, this embodiment guides the node merging process under the constraints of fault attributes, dip attributes, and stratigraphic sequences. Finally, this embodiment uses the DFS (Deep First Search) algorithm to determine the connectivity of the directed graph. Figure 7 The process of determining connectivity between nodes is shown. For nodes to be merged, this embodiment needs to determine whether they can remain connected in the directed graph. If they are connected, the nodes are merged; otherwise, they are skipped.

[0066] Guided by fault properties, dip properties, and stratigraphic relationships, this embodiment can merge all nodes connected by the same edge to form a new node. This embodiment requires creating edges between the new nodes, but with a larger distance threshold to gradually connect nodes belonging to fault relationships. In the experiments of this invention, the distance threshold is set to 4. This allows nodes to establish connections with more distant nodes. Subsequently, this embodiment can redetermine directed and undirected edges, using the same method described above. After redetermining the directed and undirected edges, this embodiment can construct a new directed graph and a new undirected graph, denoted as follows: and Together they construct a higher-level multi-attribute knowledge graph, using G 1 (V, E represent this.) This embodiment has now completed one iteration. Each iteration forms a new layer of the multi-attribute knowledge graph, ultimately creating a multi-layered multi-attribute knowledge graph. Specifically, each iteration is based on the results of the previous iteration.

[0067] As can be seen, this embodiment requires a pre-defined set of distance thresholds to construct a multi-layer, multi-attribute knowledge graph, where the number of distance thresholds equals the number of layers in the multi-attribute knowledge graph. Specifically, iteration stops when the number of iterations in constructing the multi-attribute knowledge graph equals the number of distance thresholds, thus completing the construction of the multi-layer, multi-attribute knowledge graph. The maximum distance threshold in the distance threshold set is related to the quality of the seismic data. For high-quality data with clear stratigraphic layers, its value is smaller; for low-quality data with discontinuous stratigraphic layers, its value is larger.

[0068] The technical effects of the method of the present invention will be explained below with reference to specific data:

[0069] To demonstrate the effectiveness of the method proposed in this embodiment, the method was used to perform full-layer tracing on the F3 public dataset in the Netherlands and in a work area in the western Bohai Bay to verify the effect.

[0070] For F3 data, this embodiment first performs initial smoothing on the seismic data, and then preprocesses it to obtain layer fragments, fault attributes, and dip attributes. Here, this embodiment pre-provides a set of distance thresholds {2, 4, 6, 8}, and then constructs a four-layer multi-attribute knowledge graph, as follows: Figure 8 As shown, this demonstrates the process of node fusion forming a single node at the highest level. The bottom row shows the initial multi-attribute knowledge graph and the full-layer tracking effect against seismic data. Gradually moving upwards, nodes in the left column are continuously fused, while the right column shows the continuous fusion of layer fragments. In this embodiment, it can be observed that at the lowest level, a layer is split into several nodes, while at the highest level, a single node represents the entire layer. From the lowest to the highest level, layer fragments are continuously merged.

[0071] The distance threshold value in the distance threshold set needs to take into account that the layer crosses the fault. For example, the minimum fault distance of F3 data is 3 points, so the minimum distance threshold value in the distance threshold set is 2, and it increases by 2 each time until the maximum distance threshold is slightly greater than the maximum fault distance in the data. The maximum fault distance of F3 data is 7 points, so the maximum distance threshold value in the distance threshold set is 8.

[0072] In the data from a certain work area in the western Bohai Bay, this embodiment first performs initial smoothing on the seismic data, and then preprocesses it to obtain layer segments, fault attributes and dip angle attributes. Figure 9 The visualization effect of the last layer of multi-attribute knowledge graph for full-level tracking in this work area using the method of the present invention is shown.

[0073] In this embodiment, the Paleoscan software was used to compare the effects at the fault location. Figure 10 For the effect of Paleoscan software, Figure 11 This describes the effect of the method of the present invention.

[0074] from Figure 10 and Figure 11 As can be seen, the Paleoscan software effect exhibits time-travel at the fault locations. At the left fault, the Paleoscan software effect shows time-travel at points P1, P2, P4, and P5 when crossing the fault. At the right fault, the Paleoscan software effect shows time-travel at points Q7 and Q8 when crossing the fault. However, the method of this invention does not exhibit time-travel, demonstrating the effectiveness and accuracy of the method proposed in this embodiment. To provide a more intuitive display from the 3D effect, this embodiment selects layers P2 and Q7 for demonstration. Here, the multiple layers of time-travel in the Paleoscan software effect of this embodiment are represented by different colors, such as... Figure 12 and 13 As shown.

[0075] The method proposed in this invention also demonstrates better performance in terms of the continuity of tracking results. Figure 13 and Figure 14 The previous example already demonstrated that the traced stratigraphic planes in this embodiment exhibit better continuity. To better illustrate the effect of continuity across multiple stratigraphic planes, this embodiment... Figure 10 and Figure 11 Six additional layer planes were selected for display, such as Figure 14 As shown, compared to the tracking results of Paleoscan software, the tracking results in this embodiment have better continuity.

[0076] Figure 8 In this context, "small" indicates small, "large" indicates large, "Number of contained horizon patches" indicates the number of contained horizon segments, and "Time (sample)" indicates the time (sampling point). Figure 9 In this context, Shallow means shallow, Average depth (m) means average depth, and Deep means deep. Figure 10-14 In this context, Inline(trace number) represents the line number, Crossline(trace number) represents the trace number, Depth(sample) represents the depth, and Amplitude represents the amplitude.

[0077] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A seismic full-layer seismic tracing method based on multi-attribute knowledge graphs, characterized in that, include: S1. Preprocess the seismic data to obtain layer segments, fault attributes, and dip attributes; S2. Use the layer fragments as nodes and create edges between nodes within the distance threshold to construct an initial multi-attribute knowledge graph; S3. For the current multi-attribute knowledge graph, based on whether the two layer segments represented by the nodes have a hierarchical relationship, the edges between the nodes are divided into two types: directed edges and undirected edges; a directed graph is constructed based on the directed edges, and an undirected graph is constructed based on the undirected edges. S4. Calculate the properties of undirected edges in an undirected graph based on the tilt angle property; S5. Classify nodes whose attributes of undirected edges are greater than or equal to the relevant threshold as nodes to be merged; divide the nodes to be merged into nodes with connection relationship and nodes with fault relationship according to the fault attribute. For all nodes connected by the same undirected edge with connection relationship, merge them directly to form a new node. For all nodes connected by undirected edges with fault relationship, merge them by judging the connectivity of the nodes. Expand the distance threshold and create edges between nodes within the distance threshold to obtain a new multi-attribute knowledge graph; S6. If the new multi-attribute knowledge graph still has nodes that are not connected across faults, return to step S3; otherwise, obtain the final full-layer tracking result.

2. The seismic full-layer seismic tracing method based on multi-attribute knowledge graphs according to claim 1, characterized in that, The distance threshold in step S2 is set to 2.

3. The seismic full-layer seismic tracing method based on multi-attribute knowledge graphs according to claim 1, characterized in that, In step S3, if the centers of the two layer segments represented by the nodes are located on the same seismic trace, then the relationship between the two nodes is a stratigraphic relationship.

4. The seismic full-layer seismic tracing method based on multi-attribute knowledge graphs according to claim 3, characterized in that, The specific process of step S4 is as follows: For two nodes N1 and N2 with undirected edges in the direction of the inclination, the extreme point sets they contain are N1 = {P1,..., Pm} and N2 = {Q1,..., Qn} respectively, where m and n represent the number of extreme points in N1 and N2 respectively. m} and N2 = {Q1,..., Q n} respectively, where m and n represent the number of extreme points in N1 and N2 respectively. Find the two closest extreme points between N1 and N2. Among them, D Chebyshev (P i Q j ) represents P i and Q j Chebyshev distance between them; Given an integer parameter k, find the closest P in N1 and N2 respectively. a and Q b The k extreme points are denoted as S(P) a ) and S(Q b );S(P a P is included in ) a S(Q) b Q is included in the text. b ; For S(P) a ) and S(Q b For each extreme point in S(P), L seismic data points are selected along the tline direction to form k+1 seismic traces; the center point of these seismic traces is connected to S(P). a ) and S(Q b Each extreme point in the graph coincides with another extreme point. The expression for calculating the correlation between nodes is: Among them, P i ∈S(P a ), Q j ∈S(Q b ), and Representing P respectively i and Q j Earthquake microchannels centered on the center and Represented by and The average value of the calculated seismic microchannels.

5. The seismic full-layer seismic tracing method based on multi-attribute knowledge graphs according to claim 4, characterized in that, In step S5, a depth-first search algorithm is used to determine the connectivity between nodes.

6. The seismic full-layer seismic tracing method based on multi-attribute knowledge graphs according to claim 1, characterized in that, In step S5, increasing the distance threshold specifically involves adding 2 to the current distance threshold each time.

Images