Machine learning-based multivariate constraint active fault automatic three-dimensional construction method

Abstract

The invention discloses a multivariate constraint active fault automatic three-dimensional construction method based on machine learning, and relates to the technical field of active fault three-dimensional modeling, and the method comprises the steps: carrying out the quantitative analysis of an earthquake directory in a three-dimensional space, and judging whether the earthquake directory has a cluster feature of a fault structure or not; identifying a small earthquake cluster of each fault through self-adaptive threshold hierarchical clustering in combination with seismic geological information such as seismic source mechanism parameters and surface fracture; adopting a high-density automatic three-dimensional slicing technology to stably fit fault lines, and constructing a multivariate constraint fault three-dimensional model in combination with surface fracture data; and calculating the distance from the related earthquake of the fitting fault to the fault plane and a kernel density function of the distance, and evaluating the rationality of the constructed model. Therefore, by adopting the multi-element constraint active fault automatic three-dimensional construction method based on machine learning and combining multi-source data, a data-driven modeling mode is established, manual intervention can be effectively reduced, the modeling efficiency can be improved, errors can be reduced, and the method is suitable for various earthquake scene researches.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional modeling of active faults, and more particularly to an automatic three-dimensional construction method for active faults with multiple constraints based on machine learning. Background Art

[0002] In the case where the main shock position is not accurate enough, relocating the earthquake sequence can better constrain the geometric structure of the seismogenic fault. Specifically, according to the law that small clustered earthquakes are distributed on and near the large earthquake fault plane, a three-dimensional fault model can be constructed through earthquake precise location data. The key to constructing a three-dimensional model of an active fault lies in identifying the relevant small earthquake clusters and surface rupture data of the modeled fault, and reasonably fitting the fine geometric structure of the fault.

[0003] Currently, traditional methods mostly rely on qualitative constraints of small earthquake catalogs, with strong subjectivity. At the same time, in existing modeling methods, it is usually necessary to manually discriminate various relevant data in the study area (such as earthquake catalogs, surface rupture data, focal mechanism solutions, etc.) to determine the nature of active faults. This makes the three-dimensional modeling process of active faults highly dependent on manual experience, with low automation levels and difficult to meet the needs of large-scale earthquake scenario studies.

[0004] Based on this, the present invention uses machine learning technology, combines multi-source data such as earthquake catalogs, surface ruptures, and focal mechanism solutions, and establishes a data-driven multi-constraint modeling method, effectively reducing manual intervention, improving modeling efficiency and reducing errors, and being applicable to various earthquake scenario studies. In addition, the present invention provides an automated workflow that is not only applicable to a single fault but also capable of handling multiple intersecting faults, providing an efficient solution for three-dimensional fine modeling of active faults. Summary of the Invention

[0005] The object of the present invention is to provide an automatic three-dimensional construction method for active faults with multiple constraints based on machine learning, which can improve modeling efficiency and reduce errors through an automated modeling method for active faults with multiple constraints, while reducing manual intervention.

[0006] To achieve the above object, the present invention provides an automatic three-dimensional construction method for active faults with multiple constraints based on machine learning, including the following steps: S1. Collect relevant information in the epicentral area, extract the minimum complete sub-catalog by analyzing the minimum completeness magnitude of the earthquake catalog in the relevant information, and quantitatively analyze whether the seismic activity has the small earthquake clustering characteristics of a fault structure through the spatial nearest neighbor index method; S2. Identify all earthquake clusters based on the minimum complete sub-catalog, preliminarily determine the fault distribution and the total number of faults through spatial distance, surface rupture characteristics, and focal mechanism solution consistency analysis, and identify the small earthquake clusters of each fault; S3. Based on the small earthquake clusters of each identified fault, use locally weighted regression to fit the earthquake cluster trend line, then automatically determine the position and direction of the 3D slices, and use the least squares method to fit the fault interpretation lines on each slice; S4. Based on the fault interpretation lines obtained by fitting, combined with the constraints of surface rupture data, use spatial interpolation to construct an initial 3D fault model, and then analyze the distance distribution of small earthquake clusters through kernel density estimation method, test and improve the 3D fault model, and construct a fine 3D fault model; S5. According to the seismic geological data of the study area, including surface rupture, earthquake catalog and focal mechanism solution, determine the activity nature of each fault. For a fault zone containing multiple branch faults, repeat steps S2 - S4, analyze the spatial relationship of each branch fault, and calculate the attributes of each fault.

[0007] In a possible implementation, in step S2, identify earthquake clusters through adaptive threshold hierarchical clustering.

[0008] In a possible implementation, in step S2, group earthquake clusters with close positions and the same strike and dip into the same fault.

[0009] In a possible implementation, in step S3, determine the earthquake cluster trend line according to locally weighted regression, then customize the step size of the 3D slices, set the overlap degree of earthquake queries for each slice profile, and optimize the coverage range of the 3D slices according to the overlap degree to automatically determine the position and direction of each slice, so that each slice is perpendicular to the strike of the small earthquake clusters.

[0010] In a possible implementation, in step S4, include screening out the modeling points for constructing the 3D fault model according to the consistency between the surface rupture data and the identified small earthquake clusters of the fault, and directly interpolate the modeling points with the fault interpretation lines obtained by fitting to construct the initial 3D fault model.

[0011] In a possible implementation, in step S4, include calculating the distance from the small earthquake clusters to the 3D fault structure, analyzing the spatial distribution characteristics of this distance through the kernel density estimation method, and making it present a nearly symmetric distribution pattern through repeated modeling.

[0012] In a possible implementation, in step S4, by determining a set of distance parameters and using the kernel density estimation method, analyze the distribution of small earthquake clusters on both sides of the fault under the corresponding distance parameters.

[0013] In a possible implementation, in step S4, set the distance parameters according to the occurrence of small earthquake clusters along the fault.

[0014] Therefore, the present invention adopts the above - mentioned automatic 3D construction method for active faults with multi - constraint based on machine learning, and has the following technical effects: (1) The present invention combines multi-source data such as earthquake catalogs, surface ruptures, and focal mechanism parameters to establish a data-driven modeling method, which effectively reduces manual intervention, improves modeling efficiency, and reduces errors, and is applicable to various earthquake scenario studies.

[0015] (2) The present invention uses an automated high-density three-dimensional slicing technique to robustly fit fault lines, overcomes the subjectivity of manual discrimination, generates automated fault interpretation lines, and provides an efficient solution for fine modeling.

[0016] (3) The present invention combines surface rupture data to constrain the construction of a three-dimensional fault model, provides a new way of fitting fault planes, and can improve the construction accuracy of the model; 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 the automatic optimization of the refined model is realized.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0018] Figure 1 is a flowchart of an automatic three-dimensional construction method for multi-constrained active faults based on machine learning; Figure 2 is an example of an automatic three-dimensional construction method for multi-constrained active faults based on machine learning, in which (a) is a dendrogram of hierarchical clustering results, and (b) is a clustering result map of seismic activities; Figure 3 is a schematic diagram of high-density three-dimensional slicing and robust fault line fitting in an example of an automatic three-dimensional construction method for multi-constrained active faults based on machine learning, in which (a) is an automatic slice based on locally weighted regression, (a1) is a slice without using locally weighted regression, (a2) is a slice using locally weighted regression, (a3) is a local comparison schematic diagram of the two, and (b) is a schematic diagram of a fault fitting line based on least squares; Figure 4 is a schematic diagram of a three-dimensional model of a multi-constrained active fault in an example of an automatic three-dimensional construction method for multi-constrained active faults based on machine learning, in which (a) is a three-dimensional fine model of the active fault, and (b) is a statistical graph of the distance and kernel density curve of small earthquake clusters on both sides of the fault. Detailed Embodiments

[0019] The present invention can be more detailedly explained through the following embodiments. 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 embodiments.

[0020] Please refer to Figure 1, the present invention provides a method for automatically constructing a three-dimensional multi-constraint active fault based on machine learning, including the following steps: S1. Collect data such as surface rupture data, earthquake catalog, and focal mechanism solutions in the epicentral area, analyze the minimum complete magnitude of the earthquake catalog, and extract the minimum complete sub-catalog; at the same time, use the spatial nearest neighbor index method to quantitatively analyze whether the earthquake catalog has the clustering characteristics of a fault structure in three-dimensional space.

[0021] S2. Please refer to Figure 2 , based on the minimum complete sub-catalog extracted in step S1, first identify all earthquake clusters through adaptive threshold hierarchical clustering; then, through the analysis of spatial distance, surface rupture characteristics, and focal mechanism solution consistency, classify earthquakes with close positions and consistent strikes and dips into the same fault, and preliminarily determine the fault distribution; then, combined with earthquake geological information such as focal mechanism parameters and surface rupture, determine the total number of faults and identify the small earthquake clusters of each fault.

[0022] S3. Please refer to Figure 3 , in this embodiment, a high-density automated three-dimensional slicing technique is used to robustly fit the fault line, which can automatically determine the position and direction of the three-dimensional slice based on the small earthquake cluster data to improve the modeling efficiency and accuracy, specifically as follows: First, according to the small earthquake clusters of each fault identified in step S1, use locally weighted scatterplot smoothing to fit the earthquake cluster trend line.

[0023] Secondly, customize the three-dimensional slice step size (such as 2 km, 4 km, 6 km), and set the overlap degree of the earthquakes queried in each slice profile. The overlap degree can be flexibly adjusted to 10%, 20%, etc. according to the step size to optimize the slice coverage range.

[0024] Based on the obtained earthquake cluster trend line, slice step size, and overlap degree, automatically determine the position and direction of the three-dimensional slice accordingly, so that each slice is nearly perpendicular to the strike of the small earthquake cluster.

[0025] Then, project the aftershocks onto each slice and use the least squares method to fit the fault interpretation line on each slice.

[0026] Through the above automated high-density three-dimensional slicing technique, this embodiment can robustly fit the fault line, overcome the subjectivity of manual discrimination, automatically generate the fault interpretation line, and provide an efficient solution for fine modeling.

[0027] S4. Based on the interpreted fault lines obtained by fitting, combined with the constraint of surface rupture data, a three-dimensional initial fault model is directly constructed by spatial interpolation method. Through the method of direct spatial interpolation between surface ruptures and interpreted fault lines, the surface rupture data directly participates in the calculation, realizing the construction of a three-dimensional fault model with quantitative constraints of multiple data. Compared with the traditional auxiliary judgment method, it has higher reliability.

[0028] In addition, small earthquake clusters occur along both sides of the fault. However, due to the complexity of natural earthquakes, they usually do not show a symmetric distribution in the actual process, but an approximately symmetric distribution. Therefore, in this embodiment, the kernel density estimation method is used to analyze the spatial distribution characteristics of the distance from small earthquake clusters to the fault plane, so as to test and improve the three-dimensional fault model. By repeating the modeling, it presents an approximately symmetric distribution pattern to construct a three-dimensional fine fault model. Please refer to Figure 4 . Figure 4 In (a) at the topmost part of [reference], the range of the red line represents the constraint of surface rupture, and the range of the purple line represents the absence of surface rupture constraint.

[0029] In this process, this embodiment proposes a brand-new method for fitting the fault plane. Combining the surface rupture data and the fault line fitted from the earthquake catalog, it can perform multiple constraints on the three-dimensional fault model. Specifically, it includes: conducting a consistency analysis on the surveyed surface ruptures and the identified earthquake cluster data. The screened surface rupture data is used as modeling points. The obtained modeling points and the interpreted fault lines fitted in step S3 are directly used to construct a three-dimensional initial fault model by spatial interpolation method. This way of directly participating in the interpolation calculation and fitting of the fault plane by points and lines can realize the construction of a three-dimensional fault model with multiple constraints, and has significant advantages compared with the existing method of auxiliary judgment.

[0030] This embodiment also proposes a method for automatic model evaluation and optimization. By statistically analyzing the distribution of small earthquake clusters participating in the fault fitting at 0.5 km, 1 km, 1.5 km, 2 km, and 2.5 km on both sides of the fault, the seismic distribution characteristics on both sides of the fault are analyzed; then the kernel density estimation method is used to generate a kernel density curve to quantitatively analyze whether it shows an approximately symmetric distribution pattern, ensuring that small earthquake clusters occur along the fault, thereby improving the modeling accuracy. If it meets the requirements, the corresponding three-dimensional fine fault model is obtained; if it does not meet the requirements, the modeling starts over (repeating steps S2 - S4) until the distance from small earthquake clusters to the fault plane presents a concentrated and approximately symmetric distribution pattern. At this time, it is considered that the earthquake is distributed along the fault.

[0031] 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 applicable to various earthquake scenario studies.

[0032] S5. Constructing a three-dimensional fine model of active faults further includes: Determine the activity nature of each fault (such as surface-exposed faults or buried active faults) according to the surface rupture situation and focal mechanism solutions in the study area.

[0033] For a fault zone containing multiple branch faults, it is necessary to perform sub-steps S2 - S4 for each branch fault to construct its three-dimensional fine model, and analyze the spatial relationships (intersecting, conjugate, parallel, etc.) of each branch fault by combining seismic data and seismic geological information such as surface ruptures, so as to determine the three-dimensional fault structure.

[0034] In addition, it is also necessary to calculate the attributes of each fault, including fault scale (such as length, cutting depth, area) and attitude information (strike, dip, dip angle).

[0035] Therefore, the present invention adopts the above-mentioned automatic three-dimensional construction method of active faults based on machine learning, combines surface rupture data to constrain fault modeling, can effectively reduce manual intervention and improve modeling efficiency; at the same time, adopts the kernel density estimation method to evaluate and optimize the three-dimensional fault model, can reduce modeling errors and realize an automated model optimization process.

[0036] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for automatically constructing a three-dimensional multi-constraint active fault based on machine learning, characterized in that It includes the following steps: S1. Collect relevant data in the epicenter area, extract the minimum complete sub-catalogue by analyzing the minimum completeness magnitude of the earthquake catalogue in the relevant data, and quantitatively analyze whether the seismic activity has the characteristics of small earthquake clustering of fault structure by the spatial nearest neighbor index method; S2. Identify all earthquake clusters based on the minimum complete sub-catalogue, preliminarily determine the fault distribution and the total number of faults through the analysis of spatial distance, surface rupture characteristics and the consistency of focal mechanism solutions, and identify the small earthquake clustering of each fault; S3. Based on the small earthquake clustering of each identified fault, fit the earthquake clustering trend line by local weighted regression, then automatically determine the position and direction of the three-dimensional slice, and use the least squares method to fit the fault interpretation line on each slice; S4. Based on the fitted fault interpretation line, combined with the constraint of surface rupture data, use the spatial interpolation method to construct an initial three-dimensional fault model, and then analyze the distance distribution of small earthquake clustering by the kernel density estimation method, test and improve the three-dimensional fault model, and construct a fine three-dimensional fault model; S5. According to the seismic geological data of the study area, including surface rupture, earthquake catalogue and focal mechanism solutions, determine the activity nature of each fault. For the fault zone containing multiple branch faults, repeat steps S2 - S4, analyze the spatial relationship of each branch fault, and calculate the attributes of each fault.

2. The method for automatically three-dimensionally constructing a multivariate constrained activity fault based on machine learning according to claim 1, wherein In step S2, identify earthquake clusters by adaptive threshold hierarchical clustering.

3. The method for automatically three-dimensionally constructing a multivariate constrained active fault based on machine learning according to claim 1, wherein In step S2, group earthquake clusters with close positions and the same strike and dip into the same fault.

4. The method for automatically three-dimensionally constructing a multivariate constrained active fault based on machine learning according to claim 1, wherein, In step S3, determine the earthquake clustering trend line according to local weighted regression, customize the step size of the three-dimensional slice, set the overlap degree of the earthquakes queried in each slice profile, and automatically determine the position and direction of each slice based on the earthquake trend line, the step size of the three-dimensional slice and the set overlap degree, so that each slice is perpendicular to the strike of the small earthquake clustering.

5. The method for automatically three-dimensionally constructing a multivariate constrained activity fault based on machine learning according to claim 1, characterized in that In step S4, it includes screening out the modeling points for constructing the three-dimensional fault model according to the consistency between the surface rupture data and the identified small earthquake clustering of the fault, and directly interpolating the modeling points with the fitted fault interpretation line to construct an initial three-dimensional fault model.

6. The method for automatically three-dimensionally constructing a multivariate constrained activity fault based on machine learning according to claim 1, wherein In step S4, it includes calculating the distance from the small earthquake clustering to the three-dimensional fault structure, analyzing the spatial distribution characteristics of this distance by the kernel density estimation method, and making it present a nearly symmetric distribution pattern through repeated modeling.

7. The method for automatically three-dimensionally constructing a multivariate constrained activity fault based on machine learning according to claim 6, wherein In step S4, by determining a set of distance parameters, use the kernel density estimation method to analyze the distribution of small earthquake clustering on both sides of the fault under the corresponding distance parameters.

8. The method for automatically three-dimensionally constructing a multi-constraint activity fault based on machine learning according to claim 7, characterized in that, In step S4, set the distance parameter according to the occurrence of small earthquake clustering along the fault.

Images