Ground surface displacement field construction method, device, equipment, medium and product

Abstract

The application discloses a method, device, equipment, medium and product for constructing a ground surface displacement field, and the method comprises the following steps: acquiring first measured ground time history data at a first position during an earthquake; constructing a first measured ground feature vector based on the first measured ground time history data; performing source feature prediction based on the first measured ground feature vector through a pre-trained first source prediction model to obtain a first predicted source feature vector; constructing first predicted source time history data based on the first predicted source feature vector; and calculating predicted ground time history data at any position in a target region through a time-domain boundary element method based on the first predicted source time history data to obtain a ground surface displacement field of the target region, wherein the first position is located in the target region. In this way, the ground surface displacement field of the target region can be quickly and accurately reconstructed after an earthquake, and the timeliness and accuracy requirements of post-earthquake emergency rescue can be met.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a method, apparatus, equipment, medium, and product for constructing a surface displacement field. Background Technology

[0002] As a sudden natural disaster, earthquakes generate a surface displacement field that is a key basis for assessing earthquake damage distribution and guiding emergency rescue.

[0003] Currently, surface displacement is usually measured directly at the location of the station. However, due to the low density of station deployment, only surface displacement at a small number of locations can be obtained, making it impossible to construct a regional surface displacement field.

[0004] Therefore, in order to improve the timeliness of earthquake rescue, a solution is needed that can quickly reconstruct the surface displacement field after an earthquake.

[0005] The above statements are for the purpose of providing background information in relation to this application only and do not necessarily constitute prior art. Summary of the Invention

[0006] In view of the above problems, this application provides a method, apparatus, equipment, medium and product for constructing a surface displacement field, which can quickly and accurately reconstruct the surface displacement field of a target area after an earthquake, meeting the timeliness and accuracy requirements of post-earthquake emergency rescue.

[0007] In a first aspect, this application provides a method for constructing a surface displacement field, comprising: Acquire the first measured surface time history data at the first location during the earthquake; Construct the first measured surface feature vector based on the first measured surface time history data; The first predicted earthquake source feature vector is obtained by predicting the source features based on the first measured surface feature vector using a pre-trained first source prediction model. Construct the time history data of the first predicted earthquake source based on the feature vector of the first predicted earthquake source; Based on the time history data of the first predicted seismic source, the predicted surface time history data at any location within the target area is calculated using the time-domain boundary element method to obtain the surface displacement field of the target area. The first location is located within the target area.

[0008] Secondly, this application provides an apparatus for constructing a surface displacement field, comprising: The acquisition module is used to acquire the first measured surface time history data at the first location during the earthquake. The first construction module is used to construct the first measured surface feature vector based on the first measured surface time history data; The prediction module is used to predict the source features based on the first measured surface feature vector using a pre-trained first source prediction model, and obtain the first predicted source feature vector. The second construction module is used to construct the time history data of the first predicted earthquake source based on the feature vector of the first predicted earthquake source. The calculation module is used to calculate the predicted surface time history data at any location within the target area based on the first predicted seismic source time history data and the time domain boundary element method, so as to obtain the surface displacement field of the target area. The first location is located within the target area.

[0009] Thirdly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in the first aspect above.

[0010] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the method described in the first aspect above.

[0011] Fifthly, this application provides a computer program product, including a computer program that is executed by a processor to implement the method described in the first aspect above.

[0012] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application.

[0013] Therefore, with only the first measured surface time history data at the first location, the first predicted source feature vector can be quickly predicted using a pre-trained first source prediction model. Then, based on the first predicted source time history data corresponding to the first predicted source feature vector, the predicted surface time history data at any location within the target area can be accurately calculated using the time-domain boundary element method, thus obtaining the surface displacement field of the target area. This allows for the rapid and accurate reconstruction of the surface displacement field of the target area after an earthquake, meeting the timeliness and accuracy requirements of post-earthquake emergency rescue. Attached Figure Description

[0014] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A schematic diagram of a seismic dislocation model provided for some embodiments of this application; Figure 2 A schematic diagram illustrating an earthquake rupture process provided for some embodiments of this application; Figure 3 One of the flowcharts for a method of constructing a surface displacement field provided in some embodiments of this application; Figure 4(a) is one of the schematic diagrams of obtaining the start-up time of oscillation using the long and short window method according to some embodiments of this application; Figure 4(b) is a second schematic diagram of a long and short window method for obtaining the start-up time according to some embodiments of this application; Figure 4(c) is a third schematic diagram of a long and short window method for obtaining the start-up time according to some embodiments of this application; Figure 4(d) is a schematic diagram of a long and short window method for obtaining the start-up time according to some embodiments of this application; Figure 5 A schematic diagram illustrating an iterative scheme provided for some embodiments of this application; Figure 6 A schematic diagram of a LeakyReLu function provided for some embodiments of this application; Figure 7(a) is one of the waveform diagrams of a verification sample provided in some embodiments of this application; Figure 7(b) is a second waveform diagram of a verification sample provided in some embodiments of this application; Figure 7(c) is a third waveform diagram of a verification sample provided in some embodiments of this application; Figure 7(d) is a fourth waveform diagram of a verification sample provided in some embodiments of this application; Figure 8 A waveform diagram of an equivalent source wave provided for some embodiments of this application; Figure 9 A schematic diagram comparing a reconstructed waveform with a station waveform provided for some embodiments of this application; Figure 10 A second flowchart illustrating a method for constructing a surface displacement field provided in some embodiments of this application; Figure 11 A schematic diagram of station distribution provided for some embodiments of this application; Figure 12 A topographical schematic diagram provided for some embodiments of this application; Figure 13 A schematic diagram of the vertical displacement of station A within 0-180s provided for some embodiments of this application; Figure 14 A schematic diagram of the vertical displacement of station A within 15-70 seconds provided for some embodiments of this application; Figure 15A schematic diagram of an equivalent source displacement-time function provided for some embodiments of this application; Figure 16 A schematic diagram illustrating the reconstruction results of different stations provided for some embodiments of this application; Figure 17 A schematic diagram illustrating station computational consistency for some embodiments of this application; Figure 18 A schematic diagram of a device for constructing a surface displacement field provided for some embodiments of this application; Figure 19 A schematic diagram of the hardware structure of an electronic device provided for some embodiments of this application; Figure 20 This is a schematic diagram of the structure of a storage medium provided for some embodiments of this application.

[0015] The accompanying drawings are not necessarily drawn to scale. Detailed Implementation

[0016] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0018] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0019] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0020] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0021] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0022] In the description of the embodiments of this application, the term "determine" may encompass a wide variety of actions. For example, "determine" may include calculation, operation, processing, deduction, investigation, search (e.g., searching in a table, database, or other data structure), assertion, etc. Furthermore, "determine" may include receiving (e.g., receiving information), accessing (e.g., accessing data in memory), etc. In addition, "determine" may include parsing, selecting, picking, building, etc.

[0023] It should be noted that the formulas listed in the specific embodiments of this application are numerous, covering multiple fields such as dynamics, seismology, boundary element method, signal processing, and machine learning. To conform to the conventional notation used in each field, there may be cases where the same letter represents different physical quantities or variables in different formulas. Those skilled in the art, by combining the context and physical or mathematical meaning of the specific formulas, can clearly distinguish and understand the technical content expressed by each formula, and will not cause confusion.

[0024] First, some basic theories and concepts involved in this application will be introduced.

[0025] 1. Problem Definition and Basic Assumptions When using the time-domain boundary element method to solve elastic dynamics problems, the dynamic differential governing equations are as follows:

[0026] The initial and boundary conditions of the problem can be described as follows:

[0027] in, Indicates the location ,time hour Displacement in direction, Let be the divergence of the displacement gradient. For the Laplace operator of displacement, For material density, and All are Lamé constants. As a physical weight, For acceleration, For the initial displacement field, For the initial velocity field, For displacement boundary, For a given displacement value, As the stress boundary, For the traction force component, Given a traction force value.

[0028] Then, the partial differential equation is transformed into a boundary integral equation. The solution at the points in the domain is expressed as an integral form on the boundary, and the final form is usually an integral equation (6) containing unknowns on the boundary:

[0029] in, These are the boundary geometric coefficients. As the source point, For the source point displacement, As the boundary, Within the domain, This is the fundamental solution for the displacement. The fundamental solution for traction force. Represents temporal convolution. For boundary traction force, For boundary displacement, For physical strength, The initial velocity, and These are all initial terms of the basic solution.

[0030] 2. Seismic dislocation model and rupture process Earthquakes are triggered by the sudden fracturing of underground rocks. Based on this concept, the epicenter of an earthquake can be represented by a displacement discontinuity located within the Earth; this discontinuity is called a dislocation surface. For example... Figure 1 As shown, in the elastomer Inside, there is a curved surface where both displacement and stress are discontinuous. curved surface The two sides are respectively denoted as and At a point on the curved surface Place, so as to follow point to The direction is Positive direction at the point ,use The difference in displacement across the discontinuity is represented by... This represents the difference in stress across the discontinuous surface. Among them, intersection Displacement on one side intersection Displacement on one side intersection Stress on one side intersection Stress on one side.

[0031] 2.1 Earthquake source described by seismic moment Seismic moment is an important physical quantity in seismology, and one of the fundamental parameters characterizing the magnitude of an earthquake and the mechanical state of its source. Consider two forces acting on... A couple is a set of concentrated forces of equal magnitude and opposite direction, whose points of application are located at... and , It is a small quantity that tends to zero. This couple can be expressed as:

[0032] in, It is a time history function. The direction vector of the force. The Dirac delta function represents the point where the first force is concentrated. , This indicates that the second force is concentrated at a point. , Let the direction vector of the lever arm be . Let be the partial derivative of the δ function with respect to the coordinates.

[0033] make ,and Bounded, Then the force couple Represented as:

[0034] Using Betti's theorem, choose the strength :

[0035] The corresponding displacement is used as an auxiliary state, and its auxiliary state is:

[0036] Let the force couple be:

[0037] The moment density tensor formed by all combinations of couples The resulting displacement is:

[0038] in, There are a total of nine components, six of which are independent variables. For at any time instantaneous force For the moment density tensor components, For couple strength, For the fundamental solution of the moment tensor, The derivative of the fundamental solution with respect to the coordinates of the source point. These are the moment density tensor components.

[0039] If a set of displacements is discontinuous The resulting displacement and moment density tensor If the resulting displacements are the same, then the two are said to be equivalent. If the discontinuity of body forces and stresses is not considered, then on a discontinuous plane... Discontinuous displacement on both sides The elastic displacement at any point within the domain caused by this is:

[0040] When the Green's function is taken as the global Kelvin fundamental solution,

[0041] Equation (13) can also be expanded and rewritten as:

[0042] Comparing equations (12) and (15), we get The equivalent moment density tensor is:

[0043] in, It is the elastic modulus of the fault medium. It is the cosine of the normal direction of the fault plane. For displacement components, For the normal component, Displacement discontinuity Directional components, For temporal convolution, For the Green function, Let Green's function tensor be the traction force. The fundamental solution for traction force. As the venue, As the source point, For the source point Dislocation quantity at the location. Let the elastic constant tensor be... Displacement discontinuity Directional components, For normal cosine, It is the spatial derivative of the fundamental solution.

[0044] definition:

[0045] It is the integral of the moment density tensor over the entire dislocation discontinuity, and is called the moment tensor.

[0046] in, For moment tensor components, For moment density tensor, For moment tensors, and These are unit basis vectors.

[0047] If the size of the earthquake source and its epicentral distance are much smaller than the wavelength of the frequency range of interest, the earthquake source can be considered as a point source in space.

[0048] If the duration of the earthquake source is very small compared to the period of the frequency range of interest, the time history of the source can be expressed using the Dirac delta method. Function description:

[0049] If the seismic moment is simultaneously a point moment tensor in both time and space, then it simplifies to:

[0050] in, It is the moment center of the point-distance tensor. This represents the location of the earthquake source (center of gravity). Let the moment tensor be a time-varying variable. Let be the Dirac delta function, representing the concentration at a single point. The time of the earthquake's origin. The moment tensor intensity is spatially distributed. It is a time-δ function. It is a constant moment tensor (source intensity).

[0051] 2.2 Earthquake Rupture Process The main fault parameters affecting seismic radiation include: the total or average displacement on the fault plane, the fault's geometric dimensions, the fault rupture velocity, and the slip velocity of particles on the fault plane. In the simplest case, the seismic source is a point source, and the displacement history on the fault plane is the motion history of the particles. Considering real-world physical scenarios, the displacement of particles on the fault plane requires a certain amount of time. This can be described using a slope function: The time period is the rising phase, in At any given moment, the particle stops moving. More generally, however, not just one particle moves along the fault plane, but multiple particles move sequentially, meaning multiple ruptures occur along the fault plane. Here, the Haskell model is introduced to describe the earthquake rupture process.

[0052] like Figure 2 As shown, The fracture propagation length at the fault plane. Let be the rupture velocity, h be the focal depth, and r be the equivalent sphere radius at the rupture point. Along the fault, the rupture propagates to one side, and the particles at each rupture point move sequentially. The displacement functions of all rupture points are linearly superimposed with delays, forming a complete time response function, i.e.,

[0053] in, It is the first The time delay between the rupture point and the initial rupture point. For surface observation points At the moment The displacement vector, This represents the total number of sub-sources (rupture points) after the fault plane is discretized. This is the number of the rupture point.

[0054] The fracture velocity can usually be considered constant, so equation (22) can be written as

[0055] in, It is the first The distance from the rupture point to the initial rupture point. Generally, the rupture velocity will not exceed the shear wave velocity.

[0056] If all particles on the fault plane have the same motion history except for their starting time... Furthermore, since the observation point is far enough away from the fault, the distance from each particle to the observation point can be considered constant.

[0057] The displacement of the observation point located on the boundary is:

[0058] For the case of continuous fracture of the fault layer, equation (25) can be rewritten as:

[0059] in, For the first The moment density tensor of each breakpoint (sub-source), Let be the amplitude of the moment density tensor. It is a displacement-time function. To delay time, These are the boundary geometric coefficients. For the boundary point displacement, It is a time function with a delay.

[0060] The Haskell model describes unilateral fracture. Other fracture models include bilateral fracture models where fractures propagate in two directions simultaneously, and disk-shaped fracture models where fractures propagate in all four directions simultaneously. Other fracture processes are similar to unilateral fracture in their implementation.

[0061] For a real earthquake, determining the rupture process at the epicenter requires combining observational data. However, due to the limited quantity and resolution of observational data, it is usually only possible to determine the rupture process of some large earthquakes. For small and medium-sized earthquakes, it is often difficult to distinguish the spatial complexity of the epicenter. Therefore, for earthquake-prone areas, conducting extensive numerical simulations to study the ground response triggered by different rupture processes in advance, and constructing a scenario calculation database, can enable rapid analysis of real earthquakes and provide a reference for rapid decision-making during sudden earthquakes.

[0062] According to equation (24), we only consider one break point here, which is equivalent to a break point with a radius of A sphere, with a focal depth of When an earthquake occurs, the hypocenter ruptures from its initial location, generating isotropic source waves that propagate outwards. After... After a certain period of time, the equivalent source stopped vibrating, and the seismic waves continued to propagate to the Earth's surface.

[0063] The technical background and technical issues are analyzed in detail below.

[0064] Earthquakes, as typical sudden geological disasters, are characterized by their highly transient, destructive, and unpredictable evolution. Strong earthquakes not only threaten the structural safety of buildings but also trigger chain reactions of secondary disasters. Earthquakes caused by geological structures transmit energy outward through seismic waves, causing surface deformation, damage to surface structures, and a series of secondary disasters. After an earthquake, to quickly launch a rescue response and accurately allocate rescue resources, it is necessary to rapidly obtain information on the damage in different areas. Therefore, it is essential to quickly reconstruct the surface displacement field after the earthquake to improve the timeliness of earthquake rescue. Accurately inverting the source dynamics and predicting the surface response field plays a decisive role in the rapid assessment of the disaster situation.

[0065] Existing seismic networks and observation methods can acquire focal location and moment magnitude in real time, but significant bottlenecks and shortcomings remain: relying solely on these two parameters makes it difficult to construct a complete mapping relationship between the focal location, propagation path, and site effects, hindering accurate calculation of surface displacement response after an earthquake. A dynamic approach is needed to obtain a full-process analysis of the earthquake process. However, conducting dynamic calculations requires precise geological structural information, but the heterogeneity of real geological structures makes obtaining accurate information difficult, leading to difficulties in forward modeling the wave equation. Furthermore, direct numerical calculations are limited by computational resources and the availability of real physical information; they are also slow when dealing with large-scale displacement response analysis, failing to provide rapid results. The lack of surface displacement field data significantly delays the critical window for emergency rescue.

[0066] There are two main approaches to studying the propagation of seismic waves within the Earth: dynamics and kinematics. Dynamics involves directly solving the wave equation. Kinematics simplifies the wave equation into a ray theory of wave propagation, studying the kinematic characteristics of seismic wave propagation within the Earth and thus gaining insight into the Earth's internal structure.

[0067] The main methods used in earthquake numerical simulation include: higher-order finite difference method, finite element method, pseudospectral method and spectral method, boundary element method, etc. Among them, the boundary element method (BEM) based on Green's formula has the advantages of low computational dimensionality and high computational accuracy. BEM only requires discretizing the mesh on the boundary, reducing the dimension of the problem domain, which significantly reduces the complexity and computational cost of solving the problem. Due to the existence of the fundamental solution, the boundary element method has the characteristics of combining analytical and numerical methods, and therefore usually has high accuracy. Moreover, the fundamental solution naturally satisfies the boundary conditions at infinity, so it is suitable for infinite and semi-infinite domain problems, and is particularly suitable for simulating seismic wave propagation and ground motion.

[0068] When using the boundary element method (BEM) to calculate dynamic problems, there are three main approaches: frequency domain BEM, double reciprocal BEM, and time domain BEM. Among these, the time domain BEM directly utilizes the time-dependent fundamental solution, allowing for the direct acquisition of the time-dependent dynamic response, making it the most direct and effective method. However, because the BEM requires prior knowledge of the fundamental solution, it faces certain difficulties in handling strongly nonlinear and large deformation problems.

[0069] The main methods for obtaining the surface displacement field after an earthquake are: Direct measurement by stations: This method uses pre-deployed high-frequency Global Positioning System (GPS) or strong-motion seismometers to acquire surface acceleration and displacement data after an earthquake, and then integrates these data to obtain the surface displacement field. However, the density of station deployment is low, making it impossible to obtain surface displacement data at arbitrary locations. In addition, large earthquakes often cause power system failures, rendering stations unable to function properly and resulting in data loss.

[0070] Direct numerical simulation methods: These methods, such as the finite element method or boundary element method, directly obtain the surface displacement field. However, the difficulty in obtaining geological information leads to poor calculation accuracy, and the calculation process is limited by computational resources and scale, making it difficult to handle large-scale problems.

[0071] Artificial intelligence (AI) methods: These methods acquire data from pre-positioned seismic stations and directly reconstruct the Earth's surface displacement field using AI. However, AI methods require manual selection of control parameters, which relies heavily on experience and lacks scientific theoretical basis, thus exhibiting a certain degree of "black box" nature.

[0072] This application provides a method, apparatus, device, medium, and product for constructing a surface displacement field. The method includes: acquiring first measured surface time history data at a first location during an earthquake; constructing a first measured surface feature vector based on the first measured surface time history data; performing source feature prediction based on the first measured surface feature vector using a pre-trained first source prediction model to obtain a first predicted source feature vector; constructing first predicted source time history data based on the first predicted source feature vector; and calculating predicted surface time history data at any location within a target area using the time-domain boundary element method based on the first predicted source time history data to obtain the surface displacement field of the target area, wherein the first location is located within the target area.

[0073] Therefore, with only the first measured surface time history data at the first location, the first predicted source feature vector can be quickly predicted using a pre-trained first source prediction model. Then, based on the first predicted source time history data corresponding to the first predicted source feature vector, the predicted surface time history data at any location within the target area can be accurately calculated using the time-domain boundary element method, thus obtaining the surface displacement field of the target area. This allows for the rapid and accurate reconstruction of the surface displacement field of the target area after an earthquake, meeting the timeliness and accuracy requirements of post-earthquake emergency rescue.

[0074] Below, in conjunction with Figure 3 The method for constructing the surface displacement field provided in the embodiments of this application will be described in detail.

[0075] Figure 3The diagram shows one of the flowcharts of a method for constructing a surface displacement field according to an embodiment of this application. It should be noted that the method for constructing a surface displacement field can be applied to a device for constructing a surface displacement field.

[0076] like Figure 3 As shown, the method for constructing this surface displacement field may include the following steps: S310, acquire the first measured surface time history data at the first location during the earthquake; S320, Construct the first measured surface feature vector based on the first measured surface time history data; S330, based on the first measured surface feature vector, the source feature is predicted by the pre-trained first source prediction model to obtain the first predicted source feature vector; S340, Construct the time history data of the first predicted earthquake source based on the feature vector of the first predicted earthquake source; S350, based on the time history data of the first predicted seismic source, calculates the predicted surface time history data at any location within the target area using the time-domain boundary element method, and obtains the surface displacement field of the target area.

[0077] The first location can be within the target area. The purpose here is to construct the surface displacement field of the target area, which can be based on measured surface time history data from any location within the target area. The first location can be any location within the target area. Since a seismic station is located at the first location within the target area, this seismic station can acquire the first measured surface time history data at its own location (i.e., the first location) during the earthquake. Therefore, the surface location field within the target area can be constructed based on the first measured surface time history data at the first location during the earthquake.

[0078] The first measured surface time history data can be the data on the displacement of the first position on the surface over time, which is actually measured by the seismic station during the earthquake.

[0079] By performing a Fourier transform on the first measured surface time history data, the first measured surface time history data can be converted into frequency domain data, namely the third measured surface frequency domain data. The third measured surface frequency domain data can be data on amplitude and phase variations with frequency. Then, based on one or more preset frequencies of the target area, the amplitude and phase corresponding to the one or more preset frequencies can be selected from the third measured surface frequency domain data as the first measured surface feature vector.

[0080] The preset frequency for the target area can be determined through statistical analysis of the spectrum of historical earthquakes in that area, specifically the frequency of measured surface frequency domain data of historical earthquakes. Different areas can correspond to different preset frequencies.

[0081] The first source prediction model can be a pre-trained neural network model. For example, the first source prediction model can be obtained by pre-training a Multi-Layer Perceptron (MLP) network.

[0082] Specifically, the first measured surface feature vector can be input into the first source prediction model. Based on the first measured surface feature vector, the first source prediction model performs source feature prediction and outputs the first predicted source feature vector. The first predicted source feature vector can be the amplitude and phase corresponding to one or more preset frequencies in the frequency domain data of the source predicted by the model.

[0083] Based on the one or more preset frequencies, and the amplitude and phase corresponding to the one or more preset frequencies in the first predicted seismic source feature vector, frequency domain data of the predicted seismic source can be obtained. Then, by superimposing sine waves onto the frequency domain data of the predicted seismic source, the time history data of the first predicted seismic source can be obtained. This first predicted seismic source time history data can be data showing the predicted source displacement changing over time.

[0084] Then, based on the time history data of the first predicted seismic source, the predicted surface time history data at any location within the target area can be calculated using the time-domain boundary element method. Based on this predicted surface time history data, the surface displacement field of the target area can be constructed. The predicted surface time history data can be the data on the change of surface displacement over time at any location within the target area predicted by the above method. The surface displacement field of the target area can include data on the change of displacement over time at multiple locations within the target area.

[0085] The first measured surface time history data can be collected starting from the moment the seismic waves of this earthquake arrive at the first seismic station. Different stations have different epicentral distances, therefore the arrival times of the seismic waves at each station differ. To uniformly represent the measured surface time history data recorded by the stations, it is necessary to calibrate the arrival times of the seismic waves at the stations. Specifically, the long-short window method can be used, identifying the starting point of the vibration event by comparing the average energy changes of the signal (i.e., the measured surface time history data) within short and long time windows. Two time windows are set: one shorter window to capture instantaneous changes in the signal; and another longer window to represent background noise or the long-term average level of the signal. The input signal is framed, and the average signal energy within each window is calculated. When vibration begins, the energy in the short time window suddenly increases, causing a significant increase in the energy ratio between the long and short windows. This energy ratio is the average energy of the signal in the short time window divided by the average energy of the signal in the long time window. The moment when this energy ratio first exceeds a preset threshold is the start of vibration, i.e., the moment the seismic waves arrive at the station. The data truncation positions obtained using the long and short window method can be shown in Figures 4(a), 4(b), 4(c), and 4(d).

[0086] In actual seismic excitation, source waves exhibit significant non-stationary characteristics and broadband dispersion. Here, we can focus on three key spectral parameters characterizing source dynamics: dominant frequency components, amplitude modulation characteristics, and phase evolution. Based on these, a spectral decomposition framework based on the Discrete Fourier Transform (DFT) can be constructed, expressed as follows:

[0087] in, It is a discretized source wave / surface displacement time history signal. For the corresponding frequency domain representation, The imaginary unit, The number of sampling points. For frequency indexing.

[0088] The Fourier transform used to convert the first measured surface time history data into the third measured surface frequency domain data can be the DFT.

[0089] By defining the spectral descriptor triplet ,in Characterizing frequency domain decomposition, For amplitude spectrum, Characterizing phase delay enables decoupled analysis of time-frequency domain signal features.

[0090] Suppose that the first measured surface time history data, after undergoing DFT transformation, has the following characteristics: Discrete frequency components:

[0091] Each frequency The corresponding amplitude and phase are respectively and .make

[0092] In this context, equation (28) represents multiple preset frequencies, equation (29) represents the amplitude corresponding to each of the multiple preset frequencies, and equation (30) represents the phase corresponding to each of the multiple preset frequencies.

[0093] The first measured surface feature vector may include equations (29) and (30).

[0094] The inversion theory system established in this application allows the equivalent source time function reconstruction process to be formalized into a four-stage calculation paradigm. First, source spectral characteristics are extracted, and then the time history data recorded by the stations are processed. Preprocessing and spectral decomposition are performed:

[0095] in, For discrete Fourier transform operators, The time history data recorded by the station. for The corresponding frequency domain data, express Restoration Space ( (This refers to the number of frequency points).

[0096] Then the dominant frequency components can be extracted. Constructing feature vectors Based on this, a numerical simulation database was constructed; finally, a neural network was built to learn the nonlinear mapping relationship between the earthquake source and the surface displacement.

[0097] Therefore, with only the first measured surface time history data at the first location, the first predicted source feature vector can be quickly predicted using a pre-trained first source prediction model. Then, based on the first predicted source time history data corresponding to the first predicted source feature vector, the predicted surface time history data at any location within the target area can be accurately calculated using the time-domain boundary element method, thus obtaining the surface displacement field of the target area. This allows for the rapid and accurate reconstruction of the surface displacement field of the target area after an earthquake, meeting the timeliness and accuracy requirements of post-earthquake emergency rescue.

[0098] In some embodiments, if the target region includes a region whose distance from the first location is greater than or equal to a first threshold, the method may further include, after S150: Based on the surface displacement field, determine the first predicted surface time history data of the second location in the target area, and obtain the second measured surface time history data at the second location; Based on the first predicted surface time history data and the second measured surface time history data at the second location, the correction coefficient is calculated; If the absolute value of the difference between the correction coefficient and the preset value is less than the second threshold, the surface displacement field is determined as the final surface displacement field of the target area.

[0099] The target area can include near-field and non-near-field areas. If the distance between any location in the target area and the first location is less than a first threshold, the target area can be considered to include only the non-near-field area; if the target area includes areas whose distance from the first location is greater than or equal to the first threshold, the target area can be considered to include both near-field and non-near-field areas. The first threshold can be the boundary threshold between the near-field and non-near-field areas, and can be set according to actual needs; for example, the first threshold can be 10 km.

[0100] For example, a point on the Earth's surface is recorded. The horizontal distance to the nearest station (first position) is The near-field region is .

[0101] For the near-field region, a surface displacement representation that meets the accuracy requirements can be obtained after one neural network fitting process. Therefore, if the distance between any position in the target region and the first position is less than the first threshold, the surface displacement field obtained by S350 above can be the final surface displacement field of the target region.

[0102] However, for non-near-field regions, a single nonlinear fitting is insufficient to accurately reconstruct the surface displacement field. To improve the overall accuracy of the displacement field reconstruction, an iterative calculation strategy is required to gradually correct the surface displacement field. Specifically, nonlinear fitting is embedded into the calculation process, and the equivalent source displacement-time function and neural network fitting parameters are updated synchronously in each iteration, thereby systematically improving the reconstruction accuracy of the displacement field.

[0103] In addition to the seismic station at the first location, a seismic station can also be located at a second location within the target area. The second location can be the location of any other seismic station in the target area besides the one at the first location. The seismic station located at the second location can acquire second measured surface time history data for its own location (i.e., the second location).

[0104] Furthermore, the first predicted surface time history data at the second location can be determined based on the surface displacement field obtained from S150. Then, correction coefficients can be calculated based on the predicted first surface time history data and the second measured surface time history data obtained from the seismic station at the second location.

[0105] For example, the correction factor can be the ratio of the second measured surface time history data to the first predicted surface time history data.

[0106] For example, the preset value can be 1. The second threshold can be set according to actual needs.

[0107] If the absolute value of the difference between the correction coefficient and 1 is less than the second threshold, the correction coefficient can be considered close to 1. The closer the correction coefficient is to 1, the closer the first predicted surface time history data is to the second measured surface time history data, meaning the higher the accuracy of the first predicted surface time history data. Therefore, if the absolute value of the difference between the correction coefficient and 1 is less than the second threshold, the current surface displacement field can be considered to have high accuracy and requires no correction. Thus, the currently determined surface displacement field can be identified as the final surface displacement field for the target area.

[0108] Thus, by using the correction coefficient determined based on the first predicted surface time history data at the second location in the predicted surface displacement field and the second measured surface time history data at the second location, the accuracy of the currently obtained surface displacement field can be evaluated, so as to obtain a surface displacement field with higher accuracy.

[0109] In some embodiments, the method may further include: If the absolute value of the difference between the correction coefficient and the preset value is greater than or equal to the second threshold, the simulated surface feature vectors in multiple first training samples are corrected based on the correction coefficient to obtain multiple second training samples. The first source prediction model is trained based on multiple second training samples to obtain the second source prediction model; The second source prediction model is updated to the first source prediction model. Based on the updated first source prediction model, the process returns to the step of predicting source features based on the first measured surface feature vector using the pre-trained first source prediction model, and obtaining the first predicted source feature vector. This process continues until the absolute value of the difference between the correction coefficient and the preset value is less than the second threshold. The last obtained surface displacement field is then determined as the final surface displacement field of the target area.

[0110] For example, the preset value can be 1. If the absolute value of the difference between the correction coefficient and 1 is greater than or equal to the second threshold, it can be considered that the difference between the correction coefficient and 1 is large. The larger the difference between the correction coefficient and 1, the greater the difference between the first predicted surface time history data and the second measured surface time history data, that is, the lower the accuracy of the first predicted surface time history data. Therefore, if the absolute value of the difference between the correction coefficient and 1 is greater than or equal to the second threshold, it can be considered that the accuracy of the current surface displacement field is low and needs to be corrected. Therefore, the surface displacement field of the target area can be re-determined.

[0111] The first source prediction model can be trained based on multiple first training samples.

[0112] Specifically, the multiple first training samples used to train the first source prediction model can be corrected based on the correction coefficient to obtain multiple corrected second training samples. Then, the first source prediction model is retrained based on the multiple second training samples to obtain a second source prediction model. The second source prediction model is updated to the first source prediction model. Based on the updated first source prediction model, the execution of steps S330 and subsequent steps is repeated until the absolute value of the difference between the correction coefficient and the preset value is less than a second threshold. The surface displacement field obtained at the last time is determined as the final surface displacement field of the target area.

[0113] For example, the iterative scheme can be as follows: Figure 5 As shown.

[0114] In this way, the accuracy of the current surface displacement field can be evaluated by the correction coefficient. If the accuracy is low, the training samples can be corrected based on the correction coefficient and the first source prediction model can be retrained. The surface displacement field can be re-determined based on the retrained first source prediction model until the accuracy of the surface displacement field meets the requirements, thus obtaining a final surface displacement field with higher accuracy.

[0115] In some embodiments, prior to S330, the method may further include: Obtain multiple first training samples; The preset earthquake source prediction model is trained in multiple rounds based on multiple first training samples; Determine the loss function value of the earthquake source prediction model after each round of training; The earthquake source prediction model corresponding to the round with the smallest loss function value is determined as the first earthquake source prediction model.

[0116] The first training sample may include a simulated surface feature vector at a first location and its corresponding simulated source feature vector. The simulated source feature vector may be the amplitude and phase of the frequency domain data of the simulated equivalent source wave, and the simulated surface feature vector may be the amplitude and phase of the surface frequency domain data corresponding to the simulated source feature vector.

[0117] Because obtaining the first training sample is difficult, and the geological information of underground layers is complex in real-world scenarios, the problem exhibits strong nonlinearity; the accuracy requirements for the reconstruction results are relatively relaxed at the kilometer-scale computational scale. Considering the characteristics of low feature vector dimensionality, limited training data, strong nonlinearity, and relatively low accuracy requirements, and taking into account the need for rapid response, deep neural networks can be used for training. For example, the preset source prediction model can be an MLP (Multi-Level Network), which has strong nonlinear mapping and generalization capabilities and can achieve fast convergence when the data dimensionality is low. In cases where extremely high accuracy is not required, reliable training results can be obtained without a large amount of data, and it has good interpretability, making it suitable for the problem to be solved in this application.

[0118] The first layer of an MLP is the input layer, and the last layer is the output layer; the layers in between are called hidden layers. Neurons in adjacent layers are typically fully connected. The MLP model passes data through the neurons in each layer. The linear combination is used for transmission, let Indicates the preceding A linear combination of neurons, i.e.

[0119] in, As weight, This represents the bias. After linear combination, it is passed through a nonlinear activation function. It can achieve nonlinear mapping capabilities and obtain the activation values ​​of neurons. :

[0120] Commonly used activation functions include the sigmoid function, the rectified linear unit (ReLU), and the leaky rectified linear unit (LeakyReLU). Different model structures can be obtained by adjusting the number of hidden layers and the types of activation functions.

[0121] Based on the activation process given in equation (32), in order to avoid overfitting to the data and enhance the generalization ability of the model, four hidden layers and the LeakyReLu activation function can be used, which yields better results. (33) gives the definition of the LeakyReLu function.

[0122]

[0123] in .when At that time, the graph of the LeakyReLu function can be as follows Figure 6 As shown.

[0124] The preset source prediction model can be trained in multiple rounds. Each round can be trained based on multiple first training samples. After each round of training, the loss function value is calculated once, and the source prediction model corresponding to the round with the smallest loss function value is determined as the first source prediction model.

[0125] Specifically, the loss function can be the mean squared error (MSE). For example, let the number of training epochs be... After each training round, the model's fitting ability after the current training round is evaluated using MSE as the evaluation metric. After the first round of learning, the model with the smallest MSE is selected as the first-source prediction model. The formula for calculating MSE is:

[0126] in, The number of the first training samples. For the first The first training sample includes simulated earthquake source feature vectors. To make the first The simulated surface feature vectors included in the first training sample are input into the first source prediction model, and the output value of the first source prediction model is obtained.

[0127] Thus, the above training process can enhance the model's fitting ability and obtain a first earthquake source prediction model with higher prediction accuracy.

[0128] In some embodiments, determining the source prediction model corresponding to the round with the smallest loss function value as the trained source prediction model may include: The source prediction model corresponding to the round with the smallest loss function value is determined as the third source prediction model; Using the third source prediction model, the source features are predicted based on the second measured surface feature vector included in the target verification sample, and the second predicted source feature vector is obtained. Based on the second predicted earthquake source time history data corresponding to the second predicted earthquake source feature vector, the predicted surface feature vector at the target verification location is calculated by the time domain boundary element method. Determine the degree of difference between the predicted surface feature vector and the third measured surface feature vector; Based on the fact that the average difference of at least one validation sample is less than the third threshold, the third source prediction model is determined as the first source prediction model.

[0129] Here, after evaluating the model's fitting ability through the loss function value and determining the third-source earthquake prediction model with stronger fitting ability, the model performance can be further evaluated to improve model accuracy. Specifically, the model performance can be further evaluated based on target validation samples. The number of target validation samples can be one or more. Target validation samples can include the second measured surface feature vector at the first location and the third measured surface feature vector at the target validation location. The target validation location can be located within the target area. The target validation sample can be any one of at least one validation sample. The second measured surface feature vector at the first location can be the measured surface feature vector of a historical earthquake that occurred at the first location, and the third measured surface feature vector at the target validation location can be the measured surface feature vector of a historical earthquake that occurred at the target validation location. The second and third measured surface feature vectors can be data from the same earthquake.

[0130] The methods for obtaining the second measured surface feature vector at the first location and the third measured surface feature vector at the target verification location can both refer to the method for obtaining the first measured surface feature vector at the first location, and will not be repeated here.

[0131] Using the third source prediction model, source features are predicted based on the second measured surface feature vector included in the target verification sample, resulting in the second predicted source feature vector. Second predicted source time history data is constructed based on the second predicted source feature vector. Predicted surface time history data at the target verification location is calculated using the time domain boundary element method based on the second predicted source time history data. Predicted surface feature vector is constructed based on the predicted surface time history data at the target verification location. The specific process can be found in S310-S350 and will not be elaborated here.

[0132] Because the numerical values ​​of each component of the equivalent source wave vary considerably, relative error can be used as an evaluation metric for the model's performance. Specifically, this can be measured by the degree of difference. The difference between each group of predicted surface feature vectors and the third measured surface feature vector is measured.

[0133]

[0134] in, To predict surface feature vectors, This is the third measured surface feature vector. The smaller the value, the higher the degree of overlap between the predicted surface feature vector and the third measured surface feature vector; conversely, the higher the value, the lower the degree of overlap.

[0135] After determining the degree of difference corresponding to each verification sample based on equation (36), if the number of verification samples is one, it can be directly determined whether the degree of difference corresponding to the verification sample is less than the third threshold. If it is, it indicates that the model effect of the third source prediction model is good and can effectively restore the time history data of the equivalent source wave. Therefore, the third source prediction model can be directly determined as the final first source prediction model. If not, it indicates that the model effect of the third source prediction model is poor and the third source prediction model can be retrained.

[0136] If there are multiple validation samples, it can be determined whether the average difference of the multiple validation samples is less than the third threshold. If so, it indicates that the third source prediction model has a good model effect and can effectively restore the time history data of the equivalent source wave. Therefore, the third source prediction model can be directly determined as the final first source prediction model. If not, it indicates that the third source prediction model has a poor model effect and the third source prediction model can be retrained.

[0137] The third threshold can be set according to actual needs.

[0138] For example, 140 validation samples can be set, and the average difference of the 140 validation samples is:

[0139] For example, the input waveform (i.e., the time history data waveform corresponding to the third measured surface feature vector) and the reconstruction result (i.e., the time history data waveform corresponding to the predicted surface feature vector) of four random verification samples out of 140 verification samples can be shown in Figures 7(a), 7(b), 7(c) and 7(d).

[0140] Therefore, by further validating the model's performance using validation samples, the model's performance and accuracy as a first-source earthquake prediction model can be further improved.

[0141] In some embodiments, obtaining a plurality of first training samples may include: Based on multiple preset frequencies, preset amplitude ranges, preset phase ranges, and preset durations of the target area, an equivalent source wave is constructed; Determine the eigenvectors of multiple simulated seismic sources for the equivalent source wave; Based on the target simulated seismic source feature vector and the preset frequency corresponding to the target simulated seismic source feature vector, the simulated surface feature vector corresponding to the target simulated seismic source feature vector is determined, and multiple first training samples are obtained.

[0142] The target simulated seismic source eigenvector can be any one of multiple simulated seismic source eigenvectors. The simulated seismic source eigenvector can include the amplitude and phase of the equivalent source wave.

[0143] Specifically, multiple amplitudes can be randomly determined within a preset amplitude range, and multiple phases can be randomly determined within each preset phase range, serving as the amplitudes and phases corresponding to multiple preset frequencies respectively, to obtain multiple simulated source feature vectors of the equivalent source wave, wherein each simulated source feature vector may include the amplitude and phase corresponding to a preset frequency.

[0144] For each simulated seismic source feature vector, a simulated surface feature vector corresponding to the simulated seismic source feature vector can be determined based on the simulated seismic source feature vector and the preset frequency corresponding to the simulated seismic source feature vector. Thus, multiple simulated surface feature vectors are obtained based on multiple simulated seismic source feature vectors, and then multiple first training samples are constructed. Each first training sample may include a simulated seismic source feature vector and its corresponding simulated surface feature vector.

[0145] For example, using equivalent source waves Simulates a real seismic source wave. The equivalent seismic source wave is generated by the vibration of an equivalent seismic source and can be used to simulate real seismic source waves. A superposition of sinusoidal waves. After calculation using the time-domain boundary element method, the surface response produced by the equivalent source wave is the same as that produced by the real source wave under real earthquake conditions. It is assumed that when the earthquake occurs, waves of all frequencies are emitted simultaneously from the equivalent source and continue... Time, set

[0146] The duration can be preset.

[0147] From (38), It is about The following functions:

[0148] Therefore, based on the time history data recorded by strong-motion stations, an equivalent source spectrum characterization model is constructed using DFT. From (28)-(30), retain... By identifying frequency components with a relatively high spectral energy percentage, i.e., multiple preset frequencies, energy loss can be reduced while decreasing data dimensionality. The resulting feature vector is:

[0149] The preset amplitude range can be:

[0150] in, A constant set for a specific earthquake.

[0151] Since all equivalent source waves of all frequencies are emitted simultaneously when the equivalent source begins to vibrate, and this application only focuses on the phase difference generated after the equivalent source waves propagate to the Earth's surface, the preset phase range can be:

[0152] set up:

[0153] Can Rewritten as:

[0154] For example, the four waveforms of the equivalent source wave can be represented as follows: Figure 8 As shown in Table 1, the amplitudes and phases corresponding to the five preset frequencies of the equivalent source wave are respectively shown in Table 1.

[0155] Table 1 - Equivalent source wave data

[0156] In this way, by constructing equivalent source waves based on preset frequencies of the target area, multiple simulated source feature vectors and their corresponding simulated surface feature vectors can be obtained, thereby constructing multiple first training samples and solving the problem of scarce real earthquake data.

[0157] In some embodiments, determining the simulated surface feature vector corresponding to the target simulated seismic source feature vector based on the target simulated seismic source feature vector and the preset frequency corresponding to the target simulated seismic source feature vector may include: Based on the target simulated seismic source feature vector and the preset frequency corresponding to the target simulated seismic source feature vector, determine the frequency domain data of the simulated seismic source; Sine waves are superimposed on the frequency domain data of the simulated seismic source to obtain the time history data of the simulated seismic source; Based on simulated earthquake source time history data, simulated surface time history data were calculated using the time-domain boundary element method. Fourier transform is performed on the simulated surface time history data to obtain the simulated surface frequency domain data; Based on the amplitude and phase of the simulated surface frequency domain data, the characteristic vector of the simulated surface is determined.

[0158] Specifically, based on the amplitude and phase of the target simulated earthquake source feature vector, and the preset frequency corresponding to the target simulated earthquake source feature vector, the simulated earthquake source frequency domain data can be determined. Then, since the input and output of the time-domain boundary element method are both time-domain data, it is necessary to first convert the simulated earthquake source frequency domain data into simulated earthquake source time history data through sine wave superposition, and then use the time-domain boundary element method to calculate the simulated surface time history data based on the simulated earthquake source time history data. Finally, the simulated surface time history data is converted into simulated surface frequency domain data through DFT, and the amplitude and phase of the simulated surface frequency domain data can be determined as the simulated surface feature vector.

[0159] Thus, by using the temporal boundary element method, the simulated surface feature vector corresponding to the target simulated seismic source feature vector can be determined, thereby constructing the first training sample.

[0160] In some embodiments, before performing source feature prediction based on the second measured surface feature vector included in the target verification sample using the third source prediction model to obtain the second predicted source feature vector, the method may further include: Acquire the third measured surface time history data at the first location and the fourth measured surface time history data at the target verification location during historical earthquakes; Fourier transform is performed on the third measured surface time history data to obtain the first measured surface frequency domain data, and Fourier transform is performed on the fourth measured surface time history data to obtain the second measured surface frequency domain data. Based on the amplitude and phase of the first measured surface frequency domain data, the second measured surface feature vector is determined, and based on the amplitude and phase of the second measured surface frequency domain data, the third measured surface feature vector is determined. Based on the second and third measured surface feature vectors, a target verification sample is constructed.

[0161] Here, before validating the third-source earthquake prediction model based on validation samples, at least one validation sample needs to be constructed. This validation sample can be constructed based on real historical earthquake data. The target validation sample can be any one of the at least one validation sample.

[0162] Specifically, measured surface time-history data from two locations during historical earthquakes can be acquired: the third measured surface time-history data at the first location and the fourth measured surface time-history data at the target verification location. Then, DFT is performed on the measured surface time-history data at both locations to convert them into frequency domain data, namely, the first measured surface frequency domain data and the second measured surface frequency domain data. The amplitude and phase of the first measured surface frequency domain data can then be used as the second measured surface feature vector, and the amplitude and phase of the second measured surface frequency domain data can be used as the third measured surface feature vector. Finally, the second and third measured surface feature vectors are used as target verification samples.

[0163] For example, after a historical earthquake, multiple seismic stations record measured time-history data of their respective locations. To ensure that the data recorded by the seismic stations can be adapted to the input of the neural network model, the data recorded by the seismic stations should be processed consistently. Specifically, refer to the construction method of the first training sample, which uses a vector representation of the station data containing frequency, amplitude, and phase to obtain the measured surface feature vector with the same form as equation (45). :

[0164] Appropriate dimensionality reduction can reduce the complexity of data representation while preserving key frequency components and waveform features and minimizing energy loss.

[0165] For example, station data and frequency domain reconstruction data can be as follows: Figure 9 As shown.

[0166] In some examples, for a single historical earthquake, a fixed [database / strategy] can be selected. The main frequency components constitute To randomly construct in a manner that satisfies equations (41) and (42) Group ,form :

[0167] From equation (46), we get:

[0168] Thus obtain A set of equivalent source waves that meet the requirements. Using the time-domain boundary element method, ... As an incentive, the corresponding calculation is obtained. .

[0169] From equations (27) and (40), ,have to:

[0170] That is, each group The components representing frequencies are all the same.

[0171] Similarly, for ,

[0172] Therefore, it is possible to use only Machine learning is applied to the portions representing amplitude and phase. (Discard) The former Each component and The former Each component yields:

[0173] In this way, validation samples can be constructed based on historical earthquake data, which facilitates the verification of the model effectiveness of the third-source earthquake prediction model.

[0174] For example, the input-output relationship of the first earthquake source prediction model can be expressed as:

[0175] in, The first training samples used for model training include simulated surface feature vectors. The measured surface feature vector is input for verification or practical application. This is the predicted source feature vector derived from the model inversion.

[0176] Vectors can be used Characteristic vectors representing the features of the Earth's surface, vectors Characterizing the source eigenvector, and

[0177] Here, the process of restoring the equivalent source parameters of the first source prediction model is from... arrive The inversion.

[0178] Furthermore, in typical machine learning processes, different evaluation metrics exist for various data points within a dataset, often resulting in different units or dimensions that influence the machine learning's data analysis results. To eliminate the influence of different units, data normalization is necessary. Therefore, data normalization can be performed during the construction of the first training or validation samples.

[0179] In the process of generating training data for equivalent seismic waves, since the vibrations in the low-frequency part of the equivalent seismic waves are more significant, directly using... When inputting data into a model for training, components with larger values ​​in the low-frequency range can significantly impact the training results, leading to unreasonable outcomes. Therefore, it is necessary to... Normalize each column so that all values ​​fall within the range of 0. Intervals eliminate differences in the magnitude of data.

[0180] During the generation of the equivalent source wave, due to the randomness of the amplitude value, Each component exhibits a uniform distribution. Therefore, linear normalization can be used to process the data.

[0181] right Perform the following normalization transformation : set up:

[0182] make:

[0183] Normalized column vectors:

[0184] After normalization operation All values ​​are in Within the interval. For Perform the same The transformation normalizes both the input and output of the sample. Since the original proportional relationships between the data are still preserved before and after normalization, the normalization operation can still retain the physical information contained in the data.

[0185] Note that, In the transformation methods of transformation, and The value of depends on the matrix to which the data belongs. Furthermore, after the machine learning process is complete, the same inverse normalization operation should be performed to obtain the true output data.

[0186] This application proposes a data representation method for earthquake physical processes. By introducing an equivalent source and equivalent source wave to replace the actual source, the generation and propagation characteristics of the source wave are analyzed. Using discrete Fourier transform, the source displacement-time function and surface displacement response are uniformly represented by the main amplitude and phase parameters, effectively simplifying data representation. This application also provides a data-model hybrid-driven method for restoring the equivalent source displacement-time function. This method, based on limited station data, digitizes the earthquake physical process and constructs a deep neural network model, achieving high-precision restoration of the equivalent source displacement-time function. Furthermore, this application combines the equivalent source time function restoration technique with the time-domain boundary element method to perform forward modeling of the seismic wave propagation process. By iterative solving and introducing a loss function constrained by physical information, the solution process is further optimized, significantly improving the accuracy and computational efficiency of displacement field reconstruction.

[0187] This application's embodiments characterize the strong nonlinearity of geological information using a neural network model and employ the linear boundary element method (BEM) for dynamic calculations, enabling accurate reconstruction of the surface displacement field even when geological information is unknown. By fusing the temporal BEM with machine learning techniques and pre-constructing a training set for a specific region to form a neural network, the algorithm can rapidly reconstruct the source displacement time function and the surface displacement field after an earthquake, offering faster speed and higher efficiency. Moreover, compared to the finite element method (FEM), the BEM only requires discrete boundaries, thus its computational scale is less limited by computational resources, allowing it to handle larger-scale problems within the same computation time. Furthermore, compared to general artificial intelligence methods, this application considers the physical constraints of the problem, and the proposed hybrid-driven algorithm retains the advantages of numerical computation, obtaining results that conform to the physical laws of seismic wave propagation and offering better interpretability.

[0188] Therefore, the embodiments of this application can obtain a solution that conforms to the real wave propagation law while rapidly obtaining the post-earthquake surface displacement field. Through the strategy of combining data-driven and physical models, it provides an effective solution for the rapid reconstruction of the surface displacement field after an earthquake. It performs well in terms of computational efficiency and accuracy, and has reference value for related engineering applications and disaster assessment.

[0189] In this embodiment, besides using Discrete Fourier Transform to obtain frequency domain data and form a unified data vector, data processing can also be performed directly on time domain data to form a standardized data representation. In addition to using one-dimensional vector representation, two-dimensional matrix representation can also be used to adapt to different data processing methods. Besides using fully connected neural networks for nonlinear fitting, convolutional neural networks (CNNs) or long short-term memory networks (LSTMs) can be used on two-dimensional data to process time or frequency domain data and obtain the relationship between site effects and source excitation. Besides using the aforementioned source prediction model, other nonlinear fitting methods, such as polynomial fitting, can also be used to obtain nonlinear fitting relationships. In addition to repeatedly performing machine learning and time-domain boundary element method calculations using the aforementioned iterative methods to improve accuracy, computational accuracy can also be improved by adjusting neural network parameters and increasing the dataset size.

[0190] In some examples, such as Figure 10 As shown, based on the surface displacement field construction method provided in this application embodiment, a large number of simulated seismic sources and simulated source waves can be pre-set for any surface area to conduct machine learning, obtaining the optimal model that maps the surface displacement response to the source displacement time function within the area. When a real earthquake occurs, the optimal model can be invoked to quickly reconstruct the equivalent source displacement time function and calculate the surface displacement response at any point within the area. After an earthquake, the surface displacement distribution can be quickly obtained, enabling a rapid understanding of the degree of damage at different locations on the ground. By pre-constructing a large number of earthquake scenarios to form a regional earthquake scenario library, rapid scenario matching can be achieved after an earthquake, quickly obtaining the surface displacement distribution at various locations within the region.

[0191] The following describes a test of the method for constructing the surface displacement field provided in the embodiments of this application.

[0192] Taking a magnitude 8.0 earthquake as an example, the epicenter's latitude and longitude were 31.000N 103.400E, and the focal depth was... The area is 14 km long. The example uses a square region with sides of 200 km, centered at the midpoint of the line connecting the epicenter and the stations, as the target area. Based on the vertical displacement data from station A, located 20 km west of the epicenter, the source displacement time function (i.e., time history data) is reconstructed, and the surface displacement field is rebuilt. The distribution of stations in this area is as follows: Figure 11 As shown, the terrain is as follows Figure 12 As shown.

[0193] Let the radius of the equivalent seismic source be... If the distance is 1km, then we have

[0194] The duration of the earthquake focal depth is set as follows:

[0195] Equivalent seismic source from It begins to vibrate, generating source waves of various frequencies that propagate in all directions.

[0196] Obtain the station recording data for this earthquake and analyze the vertical displacement data recorded by station A during this earthquake. Station A recorded vertical displacement data for 180 seconds, starting 20 seconds before the earthquake occurred, such as... Figure 13 As shown. The constructed training samples are trained through a neural network, and the validation samples are used to obtain the predicted output through the obtained model. Using the data of station A as input, the equivalent source displacement time function of this earthquake is calculated through this neural network. The time interval of the station data is 0.005s, and the total recording time of the station is 180s. In order to adapt to the input vector length of the MLP network, 64 data points need to be selected in the station data to construct the model given in (38). The sampling time interval is:

[0197] The total sampling time is:

[0198] Figure 13 This indicates that significant shaking began on the ground around the 15th second and ended with weak shaking around the 70th second. Based on the total sampling duration, data should be selected within the 15-70s interval. Data between 15-47s was selected for sampling, and the vertical displacement curve of the ground surface during this period was plotted. Figure 14 As shown.

[0199] Obtain the corresponding data from the station Vector. Input MLP, obtain output vector From equation (46), it is possible to obtain... Obtain the source wave waveform, such as Figure 15 As shown.

[0200] Other stations within the region were selected to reconstruct the surface displacement field at that location. Stations B and C were selected, and the surface response at each station was calculated and compared with the recorded data. Station B is located at 30.500N, 102.900E, and station C is located at 31.000N, 102.400E. The results of the first iteration are as follows: Figure 16 Table 2 shows the difference E of each station calculated by (57) under neural networks (ANN1, ANN2 and ANN3) with different iteration numbers.

[0201] Table 2 - Differences among stations under different iteration numbers of neural networks

[0202] As the number of iterations increases, the consistency of the reconstruction results gradually improves. Figure 17 The results show the consistency of the restoration of 17 stations within the calculation area after three rounds of iteration, indicating that the displacement response restoration results of most stations are good.

[0203] The surface displacement field construction method provided in this application is a seismic dynamics hybrid-driven calculation method based on the fusion boundary element method with limited observation data and machine learning technology. It involves setting simulated source waves (i.e., equivalent source waves) for the target area, and using the time-domain boundary element method to perform direct numerical calculations in advance to obtain a large amount of simulated seismic data within the target area. A unified representation method is used to form a seismic dataset (i.e., multiple first training samples) for subsequent calculations. Using the constructed seismic dataset, a neural network surrogate model containing physical constraints is designed. Inversion is performed using recorded data from seismic stations as input to obtain the displacement-time function of the source (i.e., the first predicted source time history data). Using the time-domain boundary element method, based on the displacement-time function of the source, seismic forward modeling is performed to obtain the surface displacement field within the target area, and the reconstruction accuracy is improved through iterative methods.

[0204] This application fully leverages the physical rigor of the boundary element method in solving wave equations and the adaptability of machine learning in high-dimensional nonlinear mappings, forming a complete technical route for rapidly inverting the source displacement time function and reconstructing the surface displacement field from limited observation data. This achieves rapid reconstruction of the surface displacement field without relying on geological information, improves the computational efficiency and inversion accuracy of earthquake dynamics processes, and provides data support for earthquake resistance and disaster reduction.

[0205] The embodiments of this application enable rapid and accurate large-scale earthquake dynamics calculations.

[0206] Based on the same inventive concept, this application also provides a device for constructing a surface displacement field. The following is in conjunction with... Figure 18 The apparatus for constructing the surface displacement field provided in the embodiments of this application will be described in detail.

[0207] Figure 18 A schematic diagram of a device for constructing a surface displacement field according to an embodiment of this application is shown.

[0208] like Figure 18 As shown, the apparatus for constructing the surface displacement field may include: The acquisition module 1801 is used to acquire the first measured surface time history data at the first location during the earthquake. The first construction module 1802 is used to construct the first measured surface feature vector based on the first measured surface time history data; Prediction module 1803 is used to predict the source features based on the first measured surface feature vector using a pre-trained first source prediction model, and obtain the first predicted source feature vector. The second construction module 1804 is used to construct the first predicted earthquake source time history data based on the first predicted earthquake source feature vector; The calculation module 1805 is used to calculate the predicted surface time history data at any location within the target area based on the first predicted seismic source time history data and the time domain boundary element method, so as to obtain the surface displacement field of the target area. The first location is located within the target area.

[0209] Therefore, with only the first measured surface time history data at the first location, the first predicted source feature vector can be quickly predicted using a pre-trained first source prediction model. Then, based on the first predicted source time history data corresponding to the first predicted source feature vector, the predicted surface time history data at any location within the target area can be accurately calculated using the time-domain boundary element method, thus obtaining the surface displacement field of the target area. This allows for the rapid and accurate reconstruction of the surface displacement field of the target area after an earthquake, meeting the timeliness and accuracy requirements of post-earthquake emergency rescue.

[0210] In some embodiments, the apparatus for constructing the surface displacement field may further include: The processing module is used to, in the case that the target area includes a region whose distance from the first location is greater than or equal to a first threshold, calculate the predicted surface time history data at any location in the target area based on the first predicted seismic source time history data and the time domain boundary element method to obtain the surface displacement field of the target area, determine the first predicted surface time history data of the second location in the target area based on the surface displacement field, and obtain the second measured surface time history data at the second location. The coefficient calculation module is used to calculate correction coefficients based on the first predicted land surface time history data and the second measured land surface time history data at the second location. The first determining module is used to determine the surface displacement field as the final surface displacement field of the target area when the absolute value of the difference between the correction coefficient and the preset value is less than the second threshold.

[0211] In some embodiments, the apparatus for constructing the surface displacement field may further include: The correction module is used to correct the simulated surface feature vectors in multiple first training samples based on the correction coefficient when the absolute value of the difference between the correction coefficient and the preset value is greater than or equal to the second threshold, so as to obtain multiple second training samples. The first source prediction model is trained based on the multiple first training samples. The first training module is used to train the first source prediction model based on multiple second training samples to obtain the second source prediction model. The iterative module is used to update the second source prediction model to the first source prediction model. Based on the updated first source prediction model, it returns to the step of performing source feature prediction based on the first measured surface feature vector using the pre-trained first source prediction model to obtain the first predicted source feature vector. This process continues until the absolute value of the difference between the correction coefficient and the preset value is less than the second threshold. The last obtained surface displacement field is then determined as the final surface displacement field of the target area.

[0212] In some embodiments, the apparatus for constructing the surface displacement field may further include: The sample acquisition module is used to acquire multiple first training samples before predicting the source features based on the first measured surface feature vector using the pre-trained first source prediction model to obtain the first predicted source feature vector. The first training samples include the simulated surface feature vector at the first location and its corresponding simulated source feature vector. The second training module is used to train the preset earthquake source prediction model in multiple rounds based on multiple first training samples. The second determination module is used to determine the loss function value of the earthquake source prediction model after each round of training; The third determination module is used to determine the source prediction model corresponding to the round with the smallest loss function value as the first source prediction model.

[0213] In some embodiments, the third determining module may include: The first determination submodule is used to determine the source prediction model corresponding to the round with the smallest loss function value as the third source prediction model; The prediction submodule is used to predict the source features based on the second measured surface feature vector included in the target verification sample using the third source prediction model, and obtain the second predicted source feature vector. The target verification sample includes the second measured surface feature vector at the first location and the third measured surface feature vector at the target verification location. The target verification location is located within the target area, and the target verification sample is any one of at least one verification sample. The calculation submodule is used to calculate the predicted surface feature vector at the target verification location based on the second predicted earthquake source time history data corresponding to the second predicted earthquake source feature vector and the time domain boundary element method. The second determination submodule is used to determine the degree of difference between the predicted surface feature vector and the third measured surface feature vector. The third determination submodule is used to determine the third source prediction model as the first source prediction model based on the fact that the average difference of at least one verification sample is less than a third threshold.

[0214] In some embodiments, the sample acquisition module may include: A submodule is constructed to construct an equivalent source wave based on multiple preset frequencies, preset amplitude ranges, preset phase ranges, and preset durations of the target region. The fourth determination submodule is used to determine multiple simulated source feature vectors of the equivalent source wave. The simulated source feature vectors include the amplitude and phase of the equivalent source wave. The fifth determination submodule is used to determine the simulated surface feature vector corresponding to the target simulated earthquake source feature vector based on the target simulated earthquake source feature vector and the preset frequency corresponding to the target simulated earthquake source feature vector, and to obtain multiple first training samples. The target simulated earthquake source feature vector is any one of the multiple simulated earthquake source feature vectors.

[0215] In some embodiments, the fifth determining submodule may include: The first determining unit is used to determine the frequency domain data of the simulated earthquake source based on the target simulated earthquake source feature vector and the preset frequency corresponding to the target simulated earthquake source feature vector; The processing unit is used to perform sinusoidal superposition on the frequency domain data of the simulated seismic source to obtain the time history data of the simulated seismic source. The computational unit is used to calculate simulated Earth surface time history data based on simulated seismic source time history data using the time-domain boundary element method. The transformation unit is used to perform Fourier transform on the simulated land surface time history data to obtain the simulated land surface frequency domain data; The second determining unit is used to determine the feature vector of the simulated land surface based on the amplitude and phase of the simulated land surface frequency domain data.

[0216] In some embodiments, the apparatus for constructing the surface displacement field may further include: The historical data acquisition module is used to acquire the third measured surface time history data at the first location and the fourth measured surface time history data at the target verification location during historical earthquakes, before predicting the source features based on the second measured surface feature vector included in the target verification sample using the third source prediction model. The transformation module is used to perform Fourier transform on the third measured surface time history data to obtain the first measured surface frequency domain data, and to perform Fourier transform on the fourth measured surface time history data to obtain the second measured surface frequency domain data. The fourth determining module is used to determine the second measured surface feature vector based on the amplitude and phase of the first measured surface frequency domain data, and to determine the third measured surface feature vector based on the amplitude and phase of the second measured surface frequency domain data. The sample construction module is used to construct target verification samples based on the second and third measured surface feature vectors.

[0217] The surface displacement field construction apparatus and the surface displacement field construction method provided in this application are based on the same inventive concept and have the same beneficial effects as the methods they employ, operate, or implement.

[0218] The specific implementation process of the functions and roles of each module in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0219] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components illustrated as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0220] Some embodiments of this application also provide an electronic device corresponding to the surface displacement field construction method provided in the foregoing embodiments, for executing the surface displacement field construction method described above.

[0221] Figure 19 The diagram illustrates a hardware structure of an electronic device according to an exemplary embodiment. The electronic device includes a communication interface 1901, a processor 1902, a memory 1903, and a bus 1904. The communication interface 1901, processor 1902, and memory 1903 communicate with each other via the bus 1904. The processor 1902 can execute the above-described method for constructing the surface displacement field by reading and executing machine-executable instructions in the memory 1903 corresponding to the control logic of the method for constructing the surface displacement field. The specific details of this method are described in the above embodiment and will not be repeated here.

[0222] The memory 1903 mentioned in this embodiment can be any electronic, magnetic, optical, or other physical storage device, and can contain stored information such as executable instructions, data, etc. Specifically, the memory 1903 can be RAM (Random Access Memory), flash memory, storage drive (such as hard disk drive), any type of storage disk (such as optical disc, DVD, etc.), or similar storage media, or combinations thereof. Communication between this system network element and at least one other network element is achieved through at least one communication interface 1901 (which can be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc., can be used.

[0223] Bus 1904 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. Memory 1903 is used to store programs, and processor 1902 executes the programs after receiving execution instructions.

[0224] Processor 1902 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed through integrated logic circuits in the hardware of processor 1902 or through software instructions. Processor 1902 can be a general-purpose processor, including a network processor (NP), digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware controls, etc. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor.

[0225] The electronic device provided in this application embodiment and the method for constructing the surface displacement field provided in this application embodiment are based on the same inventive concept and have the same beneficial effects as the methods they adopt, operate or implement.

[0226] This application also provides a computer-readable storage medium corresponding to the method for constructing the surface displacement field provided in the foregoing embodiments. Please refer to [link / reference]. Figure 20 As shown, the computer-readable storage medium is an optical disc 200, on which a computer program (i.e., a program product) is stored. When the computer program is run by a processor, it executes the method for constructing the surface displacement field provided in any of the foregoing embodiments.

[0227] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0228] The computer-readable storage medium provided in the above embodiments of this application and the method for constructing the surface displacement field provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0229] This application also provides a computer program product corresponding to the method for constructing the surface displacement field provided in the foregoing embodiments. The computer program product includes a computer program that is executed by a processor to implement the method for constructing the surface displacement field provided in the foregoing embodiments.

[0230] The computer program product provided in the above embodiments of this application and the method for constructing the surface displacement field provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods used, run or implemented by the application stored therein.

[0231] One or more of the components, steps, features, and / or functions disclosed in this application may be rearranged and / or combined into a single component, step, feature, or function, or distributed among several components, steps, or functions. Those skilled in the art may also add additional elements, components, steps, and / or functions without departing from the scope disclosed herein.

[0232] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0233] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0234] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and they should all be covered within the scope of the claims and specification of this application. In particular, as long as there is no structural conflict, the various technical features mentioned in the embodiments can be combined in any way. This application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for constructing a surface displacement field, characterized in that, include: Acquire the first measured surface time history data at the first location during the earthquake; A first measured surface feature vector is constructed based on the first measured surface time history data; The first predicted earthquake source feature vector is obtained by predicting the earthquake source features based on the first measured surface feature vector using a pre-trained first earthquake source prediction model. First predicted earthquake source time history data is constructed based on the first predicted earthquake source feature vector; Based on the first predicted earthquake source time history data, the predicted surface time history data at any location within the target area is calculated using the time-domain boundary element method to obtain the surface displacement field of the target area, wherein the first location is located within the target area.

2. The method according to claim 1, characterized in that, If the target region includes areas whose distance from the first location is greater than or equal to a first threshold, after calculating the predicted surface time history data at any location within the target region using the time-domain boundary element method based on the first predicted seismic source time history data to obtain the surface displacement field of the target region, the method further includes: Based on the surface displacement field, determine the first predicted surface time history data of the second location in the target area, and obtain the second measured surface time history data at the second location; Based on the first predicted surface time history data and the second measured surface time history data at the second location, a correction factor is calculated. If the absolute value of the difference between the correction coefficient and the preset value is less than the second threshold, the surface displacement field is determined as the final surface displacement field of the target area.

3. The method according to claim 2, characterized in that, The method further includes: If the absolute value of the difference between the correction coefficient and the preset value is greater than or equal to the second threshold, the simulated surface feature vectors in the multiple first training samples are corrected based on the correction coefficient to obtain multiple second training samples, and the first earthquake source prediction model is trained based on the multiple first training samples. The first source prediction model is trained based on the multiple second training samples to obtain the second source prediction model; The second source prediction model is updated to the first source prediction model. Based on the updated first source prediction model, the process of predicting source features using the pre-trained first source prediction model based on the first measured surface feature vector is returned to, and the first predicted source feature vector is obtained. The process continues until the absolute value of the difference between the correction coefficient and the preset value is less than the second threshold. The last obtained surface displacement field is determined as the final surface displacement field of the target area.

4. The method according to claim 1, characterized in that, Before the first predicted earthquake source feature vector is obtained by predicting the earthquake source features based on the first measured surface feature vector using a pre-trained first earthquake source prediction model, the method further includes: Multiple first training samples are obtained, and the first training samples include the simulated surface feature vector at the first location and its corresponding simulated seismic source feature vector. The preset earthquake source prediction model is trained multiple times based on the multiple first training samples; Determine the loss function value of the earthquake source prediction model after each round of training; The earthquake source prediction model corresponding to the round with the smallest loss function value is determined as the first earthquake source prediction model.

5. The method according to claim 4, characterized in that, The step of determining the source prediction model corresponding to the round with the smallest loss function value as the trained source prediction model includes: The source prediction model corresponding to the round with the smallest loss function value is determined as the third source prediction model; Using the third source prediction model, source features are predicted based on the second measured surface feature vector included in the target verification sample to obtain the second predicted source feature vector. The target verification sample includes the second measured surface feature vector at the first location and the third measured surface feature vector at the target verification location. The target verification location is located within the target area. The target verification sample is any one of at least one verification sample. Based on the second predicted earthquake source time history data corresponding to the second predicted earthquake source feature vector, the predicted surface feature vector at the target verification location is calculated by the time domain boundary element method. Determine the degree of difference between the predicted surface feature vector and the third measured surface feature vector; Based on the fact that the average difference of the at least one verification sample is less than the third threshold, the third source prediction model is determined as the first source prediction model.

6. The method according to claim 4, characterized in that, The acquisition of multiple first training samples includes: Based on multiple preset frequencies, preset amplitude ranges, preset phase ranges, and preset durations of the target region, an equivalent source wave is constructed; Determine multiple simulated source feature vectors of the equivalent source wave, wherein the simulated source feature vectors include the amplitude and phase of the equivalent source wave; Based on the target simulated earthquake source feature vector and the preset frequency corresponding to the target simulated earthquake source feature vector, the simulated surface feature vector corresponding to the target simulated earthquake source feature vector is determined, and the plurality of first training samples are obtained, wherein the target simulated earthquake source feature vector is any one of the plurality of simulated earthquake source feature vectors.

7. The method according to claim 6, characterized in that, The step of determining the simulated surface feature vector corresponding to the target simulated seismic source feature vector based on the target simulated seismic source feature vector and the preset frequency corresponding to the target simulated seismic source feature vector includes: Based on the target simulated earthquake source feature vector and the preset frequency corresponding to the target simulated earthquake source feature vector, the frequency domain data of the simulated earthquake source is determined; The simulated seismic source frequency domain data is superimposed with sine waves to obtain the simulated seismic source time history data; Based on the simulated earthquake source time history data, the simulated surface time history data is calculated using the time-domain boundary element method. Perform a Fourier transform on the simulated surface time history data to obtain the simulated surface frequency domain data; The simulated surface feature vector is determined based on the amplitude and phase of the simulated surface frequency domain data.

8. The method according to claim 5, characterized in that, Before obtaining the second predicted source feature vector by predicting the source features based on the second measured surface feature vector included in the target verification sample using the third source prediction model, the method further includes: During historical earthquakes, the third measured surface time history data at the first location and the fourth measured surface time history data at the target verification location were acquired. Fourier transform is performed on the third measured surface time history data to obtain the first measured surface frequency domain data, and Fourier transform is performed on the fourth measured surface time history data to obtain the second measured surface frequency domain data. Based on the amplitude and phase of the first measured surface frequency domain data, the second measured surface feature vector is determined, and based on the amplitude and phase of the second measured surface frequency domain data, the third measured surface feature vector is determined. The target verification sample is constructed based on the second measured surface feature vector and the third measured surface feature vector.

9. A device for constructing a surface displacement field, characterized in that, include: The acquisition module is used to acquire the first measured surface time history data at the first location during the earthquake. The first construction module is used to construct a first measured surface feature vector based on the first measured surface time history data; The prediction module is used to predict the source features based on the first measured surface feature vector using a pre-trained first source prediction model, and obtain the first predicted source feature vector. The second construction module is used to construct the first predicted earthquake source time history data based on the first predicted earthquake source feature vector; The calculation module is used to calculate the predicted surface time history data at any location within the target area based on the first predicted earthquake source time history data using the time domain boundary element method, so as to obtain the surface displacement field of the target area, wherein the first location is located within the target area.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the program to implement the method as described in any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by a processor to implement the method as described in any one of claims 1-8.

12. A computer program product, comprising a computer program, characterized in that, The computer program is executed by a processor to implement the method according to any one of claims 1-8.

Images