Automatic 3D construction method of multi-constrained active faults based on machine learning

Abstract

The present invention discloses a method for automatic three-dimensional construction of multi-constrained active faults based on machine learning, which relates to the technical field of three-dimensional modeling of active faults, including quantitative analysis of earthquake catalogs in three-dimensional space to determine whether they have cluster characteristics of fault structures; identifying clusters of small earthquakes on each fault through adaptive threshold hierarchical clustering, combined with earthquake source mechanism parameters and seismic geological information such as surface rupture; using high-density automated three-dimensional slicing technology to robustly fit fault lines, and combining surface rupture data to construct a three-dimensional model of the fault with multi-constraints; calculating the distance from the relevant earthquakes of the fitted fault to the fault plane and its kernel density function to evaluate the rationality of the constructed model. Therefore, by adopting the above-mentioned automatic three-dimensional construction method of multi-constrained active faults based on machine learning and combining multi-source data, a data-driven modeling method is established, which can effectively reduce manual intervention, improve modeling efficiency and reduce errors, and is suitable for the study of various earthquake scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional modeling of active faults, and in particular to a method for automatic three-dimensional construction of multi-constrained active faults based on machine learning. Background Art

[0002] When the main shock location is not precise enough, relocating earthquake sequences can better constrain the geometry of the seismogenic fault. Specifically, based on the distribution of clusters of small earthquakes on and near the fault plane of a major earthquake, precise earthquake location data can be used to construct a three-dimensional fault model. The key to constructing a three-dimensional model of an active fault lies in identifying the relevant small earthquake clusters and surface rupture data for the modeled fault, and in properly fitting the fine-grained fault geometry.

[0003] Currently, traditional methods rely heavily on qualitative constraints from small earthquake catalogs, which are highly subjective. Furthermore, existing modeling methods often require manual identification of various relevant data in the study area (such as earthquake catalogs, surface rupture data, and focal mechanism solutions) to determine the nature of active faults. This makes the 3D modeling of active faults largely dependent on manual experience, resulting in a low level of automation and difficulty meeting the needs of large-scale earthquake scenario research.

[0004] Based on this, the present invention uses machine learning techniques, combined with multi-source data such as earthquake catalogs, surface ruptures, and focal mechanism solutions, to establish a data-driven, multi-constrained modeling approach. This approach effectively reduces manual intervention, improves modeling efficiency, and reduces errors, making it suitable for studying a variety of earthquake scenarios. Furthermore, the present invention provides an automated workflow that is applicable not only to single faults but also to multiple intersecting faults, providing a highly efficient solution for the detailed three-dimensional modeling of active faults. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for automatic three-dimensional construction of multi-constrained active faults based on machine learning, which can improve modeling efficiency and reduce errors while reducing manual intervention through the automatic modeling method of multi-constrained active faults.

[0006] To achieve the above objectives, the present invention provides a method for automatic three-dimensional construction of multi-constrained active faults based on machine learning, comprising the following steps:

[0007] S1. Collect relevant data on the epicenter area, analyze the minimum complete magnitude of the earthquake catalog in the relevant data, extract the minimum complete subcatalog, and quantitatively analyze whether the seismic activity has the characteristics of small earthquake clusters with fault structure using the spatial nearest neighbor index method;

[0008] S2. Identify all earthquake clusters based on the minimum complete sub-catalog. Preliminary determination of fault distribution and total number of faults is performed through spatial distance, surface rupture characteristics, and consistency analysis of focal mechanism solutions. Small earthquake clusters along each fault are also identified.

[0009] S3. Based on the identified small earthquake clusters on each fault, the earthquake cluster trend line is fitted through local weighted regression, and then the position and orientation of the 3D slices are automatically determined. The fault interpretation line on each slice is fitted using the least squares method.

[0010] S4. Based on the fault interpretation lines obtained by fitting and combined with surface rupture data constraints, a three-dimensional initial fault model is constructed using spatial interpolation. The distance distribution of small earthquake clusters is then analyzed using kernel density estimation to test and improve the three-dimensional fault model and construct a three-dimensional fine fault model.

[0011] S5. Determine the activity properties of each fault based on the seismic geological data of the study area, including surface ruptures, earthquake catalogs, and focal mechanism solutions. For fault zones containing multiple branch faults, repeat steps S2 to S4, analyze the spatial relationships of the branch faults, and calculate the properties of each fault.

[0012] In one possible implementation, in step S2, earthquake clusters are identified by adaptive threshold hierarchical clustering.

[0013] In a possible implementation, in step S2, earthquake clusters that are close in location and have the same strike and dip are grouped as the same fault.

[0014] In one possible implementation, in step S3, the earthquake cluster trend line is determined based on local weighted regression, and then the step size of the three-dimensional slice is customized, the overlap of each slice profile for querying earthquakes is set, and the coverage range of the three-dimensional slice is optimized based on the overlap, so as to automatically determine the position and direction of each slice so that each slice is perpendicular to the direction of the small earthquake cluster.

[0015] In one possible implementation, step S4 includes screening out modeling points for constructing a three-dimensional fault model based on the consistency between the surface rupture data and the identified fault earthquake clusters, and directly interpolating the modeling points with the fitted fault interpretation lines to construct an initial three-dimensional fault model.

[0016] In one possible implementation, step S4 includes calculating the distance from the small earthquake cluster to the three-dimensional structure of the fault, analyzing the spatial distribution characteristics of the distance using the kernel density estimation method, and making it present a nearly symmetrical distribution pattern through repeated modeling.

[0017] In a possible implementation, in step S4, a set of distance parameters is determined, and a kernel density estimation method is used to analyze the distribution of small earthquake clusters on both sides of the fault under the corresponding distance parameters.

[0018] In one possible implementation, in step S4, a distance parameter is set according to the occurrence of a cluster of small earthquakes along the fault.

[0019] Therefore, the present invention adopts the above-mentioned automatic three-dimensional construction method of multi-constrained active faults based on machine learning, which has the following technical effects:

[0020] (1) This paper combines multi-source data such as earthquake catalogs, surface ruptures, and focal mechanism parameters to establish a data-driven modeling approach, which effectively reduces manual intervention, improves modeling efficiency, and reduces errors. It is suitable for the study of various earthquake scenarios.

[0021] (2) The present invention uses automated high-density three-dimensional slicing technology to robustly fit fault lines, overcome the subjectivity of manual judgment, generate automated fault interpretation lines, and provide an efficient solution for fine modeling.

[0022] (3) The present invention combines surface rupture data to constrain the construction of a three-dimensional fault model, providing a new fault plane fitting method that can improve the accuracy of model construction. At the same time, by calculating the distance from the earthquake to the fault plane and its kernel density function, the rationality of the constructed model is evaluated, and automatic optimization of the refined model is achieved.

[0023] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a flowchart of the automatic three-dimensional construction method of multi-constrained active faults based on machine learning;

[0025] Figure 2 This is an embodiment of an automatic three-dimensional construction method for multi-constrained active faults based on machine learning, in which an adaptive threshold hierarchical clustering analysis of earthquake clusters is performed, wherein (a) is a dendrogram of the hierarchical clustering results, and (b) is a clustering result diagram of earthquake activity;

[0026] Figure 3 Schematic diagram of high-density three-dimensional slicing and robust fault line fitting in an embodiment of a method for automatic three-dimensional construction of multivariate constrained active faults based on machine learning, wherein (a) is automatic slicing based on local weighted regression, (a1) is slicing without using local weighted regression, (a2) is slicing using local weighted regression, (a3) ​​is a schematic diagram of a local comparison between the two, and (b) is a schematic diagram of a fault fitting line based on least squares.

[0027] Figure 4 2. Schematic diagram of a three-dimensional model of a multi-constrained active fault in an embodiment of a method for automatic three-dimensional construction of a multi-constrained active fault based on machine learning, wherein (a) is a three-dimensional fine model of the active fault, and (b) is a distance statistic and kernel density curve of small earthquake clusters on both sides of the fault. DETAILED DESCRIPTION

[0028] The present invention can be explained in more detail by the following examples. The purpose of disclosing the present invention is to protect all changes and improvements within the scope of the present invention. The present invention is not limited to the following examples.

[0029] See Figure 1 The present invention provides a method for automatically constructing three-dimensional active faults with multi-constraints based on machine learning, comprising the following steps:

[0030] S1. Collect surface rupture data, earthquake catalogs, and focal mechanism solutions in the epicenter area, analyze the minimum complete magnitude of the earthquake catalog, and extract the minimum complete subcatalog. At the same time, use the spatial nearest neighbor index method to quantitatively analyze in three-dimensional space whether the earthquake catalog has cluster characteristics of fault structure.

[0031] S2, see Figure 2 Based on the minimum complete subdirectory extracted in step S1, all earthquake clusters are first identified through adaptive threshold hierarchical clustering; then, through spatial distance, surface rupture characteristics and consistency analysis of focal mechanism solutions, earthquakes with similar locations and consistent strikes and dips are clustered into the same fault, and the fault distribution is preliminarily determined; then, combined with earthquake geological information such as focal mechanism parameters and surface rupture, the total number of faults is determined, and the small earthquake clusters on each fault are identified.

[0032] S3, see Figure 3 This embodiment uses high-density automated 3D slicing technology to robustly fit fault lines. It can automatically determine the position and direction of 3D slices based on small earthquake cluster data to improve modeling efficiency and accuracy. The details are as follows:

[0033] First, based on the small earthquake clusters on each fault identified in step S1, the earthquake cluster trend line is fitted using Locally Weighted Scatterplot Smoothing.

[0034] Secondly, customize the 3D slice step size (such as 2km, 4km, 6km) and set the overlap of earthquake queries in each slice profile. The overlap can be flexibly adjusted to 10%, 20%, etc. according to the step size to optimize the slice coverage.

[0035] Based on the obtained earthquake cluster trend line, slice step size and overlap, the three-dimensional slice position and direction are automatically determined so that each slice is nearly perpendicular to the direction of the small earthquake cluster.

[0036] Then, the aftershocks were projected onto each slice, and the least squares method was used to fit the fault interpretation line on each slice.

[0037] This embodiment uses the above-mentioned automated high-density three-dimensional slicing technology to robustly fit fault lines, overcome the subjectivity of manual judgment, and automatically generate fault interpretation lines, providing an efficient solution for fine modeling.

[0038] S4. Based on the fitted fault interpretation lines and combined with surface rupture data constraints, a 3D initial fault model is directly constructed through spatial interpolation. By directly interpolating surface ruptures and fault interpretation lines, surface rupture data is directly involved in the calculation, enabling the construction of a 3D fault model with quantitative constraints from multivariate data. This method is more reliable than traditional auxiliary judgment methods.

[0039] In addition, small earthquake clusters occur along both sides of the fault, but due to the complexity of natural earthquakes, in practice they are usually not symmetrically distributed, but rather approximately symmetrically distributed. Therefore, this example uses the kernel density estimation method to analyze the spatial distribution characteristics of the distance from the small earthquake cluster to the fault plane to test and improve the three-dimensional fault model. By repeatedly modeling it to present a nearly symmetrical distribution pattern, a detailed three-dimensional fault model is constructed. Figure 4 . Figure 4 The red line range shown at the top of (a) represents the area with surface rupture constraint, and the purple line range represents the area without surface rupture constraint.

[0040] During this process, this embodiment proposes a novel fault plane fitting method that combines surface rupture data with fault lines fitted from earthquake catalogs to implement multi-factorial constraints on the 3D fault model. Specifically, the method involves performing a consistency analysis between the surface rupture data obtained from the survey and the identified earthquake cluster data. The filtered surface rupture data is then used as modeling points. The selected modeling points are then directly interpolated with the fault interpretation lines fitted in step S3 to construct an initial 3D fault model. This method, in which points and lines are directly interpolated to calculate the fitted fault plane, enables the construction of a multi-factorially constrained 3D fault model, offering significant advantages over existing methods that rely on auxiliary judgment.

[0041] This example also proposes an automatic model evaluation and optimization method. This method analyzes the distribution characteristics of earthquakes on both sides of the fault by statistically analyzing the distribution of small earthquake clusters at 0.5 km, 1 km, 1.5 km, 2 km, and 2.5 km on either side of the fault. Kernel density estimation is then used to generate a kernel density curve, which quantitatively analyzes whether it exhibits a nearly symmetrical distribution pattern, ensuring that the small earthquake clusters occur along the fault, thereby improving modeling accuracy. If so, a corresponding 3D fault fine-scale model is obtained. If not, modeling is restarted (repeating steps S2-S4) until the distance from the small earthquake cluster to the fault plane exhibits a concentrated and nearly symmetrical distribution pattern, at which point the earthquakes are considered to be distributed along the fault.

[0042] This embodiment combines multi-source data such as earthquake catalogs, surface ruptures, and focal mechanism solutions to establish the above-mentioned data-driven modeling method, which can effectively reduce manual intervention, improve modeling efficiency, and reduce errors, and is suitable for the study of various earthquake scenarios.

[0043] S5. Constructing a detailed three-dimensional model of the active fault, including:

[0044] Based on the surface rupture conditions and focal mechanism solutions in the study area, the activity nature of each fault (such as exposed surface fault or hidden active fault) is determined.

[0045] For a fault zone containing multiple branch faults, substeps S2 to S4 need to be performed on each branch fault to construct its three-dimensional fine model. Combined with seismic data and seismic geological information such as surface ruptures, the spatial relationship (intersection, conjugation, parallelism, etc.) of each branch fault is analyzed to determine the three-dimensional fault structure.

[0046] In addition, the properties of each fault need to be calculated, including fault scale (such as length, cutting depth, area) and attitude information (strike, dip, inclination).

[0047] Therefore, the present invention adopts the above-mentioned automatic three-dimensional construction method of multi-constrained active faults based on machine learning, combined with surface rupture data to constrain fault modeling, which can effectively reduce manual intervention and improve modeling efficiency; at the same time, the kernel density estimation method is used to evaluate and optimize the three-dimensional fault model, which can reduce modeling errors and realize an automated model optimization process.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for automatic 3D construction of multi-constrained active faults based on machine learning, characterized by: The following steps are involved: S1. Collect relevant data on the epicenter area, analyze the minimum complete magnitude of the earthquake catalog in the relevant data, extract the minimum complete subcatalog, and quantitatively analyze whether the seismic activity has the characteristics of small earthquake clusters with fault structure using the spatial nearest neighbor index method; S2. Identify all earthquake clusters based on the minimum complete sub-catalog. Preliminary determination of fault distribution and total number of faults is performed through spatial distance, surface rupture characteristics, and consistency analysis of focal mechanism solutions. Small earthquake clusters along each fault are also identified. S3. Based on the identified small earthquake clusters on each fault, the earthquake cluster trend line is fitted through local weighted regression, and then the position and orientation of the 3D slices are automatically determined. The fault interpretation line on each slice is fitted using the least squares method. S4. Based on the fault interpretation lines obtained by fitting and combined with surface rupture data constraints, a three-dimensional initial fault model is constructed using spatial interpolation. The distance distribution of small earthquake clusters is then analyzed using kernel density estimation to test and improve the three-dimensional fault model and construct a three-dimensional fine fault model. S5. Determine the activity properties of each fault based on the seismic geological data of the study area, including surface ruptures, earthquake catalogs, and focal mechanism solutions. For fault zones containing multiple branch faults, repeat steps S2 to S4, analyze the spatial relationships of the branch faults, and calculate the properties of each fault.

2. The automatic three-dimensional construction method of multi-constrained active faults based on machine learning according to claim 1 is characterized in that: In step S2, earthquake clusters are identified by adaptive threshold hierarchical clustering.

3. The automatic three-dimensional construction method of multi-constrained active faults based on machine learning according to claim 1 is characterized in that: In step S2, earthquake clusters with similar locations and consistent strikes and dips are grouped as the same fault.

4. The automatic three-dimensional construction method of multi-constrained active faults based on machine learning according to claim 1 is characterized in that: In step S3, the earthquake cluster trend line is determined based on local weighted regression, the step size of the three-dimensional slice is customized, the overlap of the earthquake query profiles of each slice is set, and based on the earthquake trend line, the step size of the three-dimensional slice and the set overlap, the position and direction of each slice are automatically determined so that each slice is perpendicular to the direction of the small earthquake cluster.

5. The automatic three-dimensional construction method of multi-constrained active faults based on machine learning according to claim 1 is characterized in that: In step S4, the modeling points for constructing a three-dimensional fault model are selected based on the consistency between the surface rupture data and the identified fault earthquake clusters, and the modeling points are directly interpolated with the fitted fault interpretation lines to construct an initial three-dimensional fault model.

6. The automatic three-dimensional construction method of multi-constrained active faults based on machine learning according to claim 1 is characterized in that: Step S4 includes calculating the distance from the small earthquake cluster to the three-dimensional structure of the fault, analyzing the spatial distribution characteristics of the distance using the kernel density estimation method, and making it present a nearly symmetrical distribution pattern through repeated modeling.

7. The automatic three-dimensional construction method of multi-constrained active faults based on machine learning according to claim 6 is characterized in that: In step S4, a set of distance parameters is determined and the kernel density estimation method is used to analyze the distribution of small earthquake clusters on both sides of the fault under the corresponding distance parameters.

8. The method for automatic three-dimensional construction of multi-constrained active faults based on machine learning according to claim 7, characterized in that: In step S4, a distance parameter is set according to the occurrence of small earthquake clusters along the fault.

Images