A method, device, equipment and storage medium for extracting time function of apparent earthquake source

Abstract

The present invention relates to the technical field of analyzing earthquake source rupture processes, and specifically to a method, device, equipment and storage medium for extracting apparent source time functions. The present invention first obtains the station waveform record of the target earthquake, locates each earthquake event associated with the target earthquake, obtains the historical Green's function of each earthquake event, and then applies a neural network model to the station waveform of the target event and each historical Green's function to obtain each apparent source time function of the target earthquake. Thereafter, the historical Green's function and the apparent source time function are convolved, and the optimal apparent source time function of the target earthquake is screened and determined based on the convolution results corresponding to each apparent source time function. The present invention calculates the target apparent source time function of the target earthquake from two deterministic parameters, namely the known station waveform and the historical Green's function. The entire process does not involve parameters that need to be manually adjusted or marked, so the present invention can ensure the stability of the accuracy of the target apparent source time function.

Description

Technical Field

[0001] The present invention relates to the technical field of analyzing earthquake source rupture processes, and in particular to a method, device, equipment and storage medium for extracting an apparent earthquake source time function. Background Art

[0002] The apparent focal time function of the earthquake source area is calculated. Through the apparent focal time function, the source rupture information of the earthquake source area (including the finite fault parameters such as the characteristic length, characteristic width, characteristic velocity, characteristic duration, and rupture directionality of the rupture) can be inverted. The existing technology uses the following method to calculate the apparent focal time function of the earthquake source area:

[0003] First, the empirical Green's function of small earthquake events (magnitude less than 2.5) that have occurred around the source area is collected, and the waveform of the source area is collected. Then, the starting time (recorded as the waveform starting time) and the ending time (recorded as the waveform ending time) of the waveform are marked, and the starting time of the empirical Green's function corresponding to the waveform starting time is marked (recorded as the function starting time), and the ending time of the empirical Green's function corresponding to the waveform ending time is marked (recorded as the function ending time). The PLD method is applied to the waveform within the time period between the waveform starting time and the waveform ending time, and the function within the time period between the function starting time and the function ending time, and the effective time of the apparent source time function is set (the effective time that the apparent source time function calculated at the end needs to reach, and the effective time is the time duration that the function value is non-zero) to obtain the apparent source time function of the source area by deconvolution in the frequency domain.

[0004] The PLD algorithm used in the existing technology requires continuous manual marking of five parameters: the waveform start time, the waveform end time, the Green function start time, the Green function end time, and the effective duration of the apparent source time function, so as to continuously adjust the waveform and the empirical Green function input to the PLD algorithm until the convolution of the apparent source time function obtained by iterative calculation and the waveform currently input to the PLD algorithm meets the set conditions, and the apparent source time function at this time is used as the final apparent source time function of the source area.

[0005] From the above records, it can be seen that the above PLD algorithm involves five parameters, each of which needs to be manually marked. Any marking error in any parameter will affect the accuracy of the final apparent source time function. Therefore, the accuracy of the apparent source time function calculated by the existing technology is unstable.

[0006] In summary, the accuracy of the apparent source time function calculated by the existing technology is unstable.

[0007] Therefore, the existing technology needs to be improved and enhanced. Summary of the Invention

[0008] In order to solve the above technical problems, the present invention provides a method for extracting the apparent source time function, which solves the problem that the apparent source time function calculated by the prior art has an unstable accuracy.

[0009] To achieve the above object, the present invention adopts the following technical solutions:

[0010] In a first aspect, the present invention provides a method for extracting an apparent source time function, comprising:

[0011] Locating each seismic event around the target earthquake and obtaining a historical Green's function of each of the seismic events;

[0012] Obtaining a station waveform of the target earthquake, and applying a trained neural network model to the station waveform and each of the historical Green's functions to obtain each apparent source time function;

[0013] The historical Green's function and each of the apparent source time functions are convolved to obtain a convolution result corresponding to each of the apparent source time functions, and based on each of the convolution results, the target apparent source time function of the target earthquake is screened out from each of the apparent source time functions, and the target apparent source time function is used to calculate the source information of the target earthquake.

[0014] In one implementation, locating each seismic event around the target earthquake and obtaining a historical Green's function of each seismic event includes:

[0015] Acquire a data set, the data set including a plurality of earthquake sample events and a sample Green's function of each earthquake sample event;

[0016] The distances between several of the earthquake sample events and the target earthquake are obtained, and based on the several distances, the various earthquake events of the target earthquake are located from the several earthquake sample events, and the sample Green's function of each earthquake event is used as the historical Green's function of each earthquake event.

[0017] In one implementation, the data set further includes target simulated sample earthquakes corresponding to the earthquake sample events, simulated sample waveforms of the target simulated sample earthquakes, and apparent source time sample functions of the target simulated sample earthquakes. The data set is constructed by:

[0018] Acquire the sample Green's function of the earthquake sample event through a station;

[0019] determining the fault plane of each of the earthquake sample events based on the sample Green's function;

[0020] Simulating an earthquake by adjusting the earthquake rupture parameters of the fault plane to obtain a target simulated sample earthquake, and calculating an apparent source time sample function of the target simulated sample earthquake;

[0021] The sample Green's function and the apparent source time sample function are convolved to obtain the simulated sample waveform of the target simulated sample earthquake, and the data set is constructed through the sample Green's function of the earthquake sample event, the apparent source time sample function of the target simulated sample earthquake, and the simulated sample waveform of the target simulated sample earthquake.

[0022] In one implementation, the seismic sample events are covered by the stations in an omnidirectional manner;

[0023] and / or, the seismic sample events are covered by the stations at an off-source angle of 60°-180°;

[0024] and / or, the magnitude of the earthquake sample event is less than a set magnitude;

[0025] and / or, the waveform signal-to-noise ratio of the seismic sample event is greater than a set value;

[0026] And / or, the focal mechanism quality of the earthquake sample event is greater than a threshold.

[0027] In one implementation, the data set is used to train the neural network model, and the training method of the neural network model includes:

[0028] Inputting the simulated sample waveform and the sample Green's function into the neural network model to obtain an apparent source time training function output by the neural network model;

[0029] The loss function of the neural network model is determined according to the apparent source time training function and the apparent source time sample function, and the neural network model is iteratively trained according to the loss function.

[0030] In one implementation, determining the loss function of the neural network model based on the apparent source time training function and the apparent source time sample function includes:

[0031] Determining a mean square error between the apparent source time training function and the apparent source time sample function;

[0032] Counting the characteristic duration of the apparent source time training function and the characteristic duration of the apparent source time sample function, and calculating the duration difference between the characteristic duration of the apparent source time training function and the characteristic duration of the apparent source time sample function;

[0033] Counting the non-zero point lengths of the apparent source time training function and the non-zero point lengths of the apparent source time sample function, and calculating the length difference between the non-zero point lengths of the apparent source time training function and the non-zero point lengths of the apparent source time sample function;

[0034] A loss function is constructed based on the mean square error, the duration difference, and the length difference.

[0035] In one implementation, convolving the historical Green's function with each of the apparent source time functions to obtain a convolution result corresponding to each of the apparent source time functions, and screening out a target apparent source time function of the target earthquake from each of the apparent source time functions based on each of the convolution results, includes:

[0036] Convolving the historical Green's function with each of the apparent source time functions to obtain a simulation waveform corresponding to each of the apparent source time functions, and using each of the simulation waveforms as a convolution result;

[0037] Determine the similarity between the station waveform and each of the simulated waveforms, and based on the similarity corresponding to each of the apparent source time functions, screen out the target apparent source time function from each of the apparent source time functions, wherein the similarity is calculated using the normalized mean square error or waveform cross-correlation.

[0038] In a second aspect, an embodiment of the present invention further provides a device for extracting a time function of an apparent seismic source, wherein the device comprises the following components:

[0039] A positioning module, configured to locate each seismic event around the target earthquake and obtain the historical Green's function of each seismic event;

[0040] an apparent source time function calculation module, configured to obtain the station waveform of the target earthquake and apply a trained neural network model to the station waveform and each of the historical Green's functions to obtain each apparent source time function;

[0041] The apparent source time function prediction module is used to convolve the historical Green's function and each apparent source time function to obtain a convolution result corresponding to each apparent source time function, and based on each convolution result, screen out the target apparent source time function of the target earthquake from each apparent source time function, and the target apparent source time function is used to calculate the source information of the target earthquake.

[0042] In a third aspect, an embodiment of the present invention further provides a terminal device, wherein the terminal device includes a memory, a processor, and a visual source time function extraction program stored in the memory and runnable on the processor, and when the processor executes the visual source time function extraction program, the steps of the visual source time function extraction method described above are implemented.

[0043] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a visual source time function extraction program is stored. When the visual source time function extraction program is executed by a processor, the steps of the visual source time function extraction method described above are implemented.

[0044] Beneficial effects: The present invention first locates each seismic event associated with the target earthquake, and obtains the historical Green's function of each seismic event, as well as the station waveform of the target earthquake, and then applies a neural network model to the station waveform and each historical Green's function to obtain each apparent source time function, and then convolves the historical Green's function with the apparent source time function, and screens the apparent source time function according to the convolution results corresponding to each apparent source time function to obtain the target apparent source time function of the target earthquake. From the above analysis, it can be seen that the present invention calculates the target apparent source time function of the target earthquake from two deterministic parameters, namely the known station waveform and the historical Green's function. The entire process does not involve parameters that need to be manually adjusted or marked, so the present invention can ensure the stability of the accuracy of the target apparent source time function. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is the overall flow chart of the present invention;

[0046] Figure 2 is a flow chart for calculating source parameters in an embodiment of the present invention;

[0047] Figure 3 is a schematic diagram of a fault plane in an embodiment of the present invention;

[0048] Figure 4 A schematic diagram of a speed structure in an embodiment of the present invention;

[0049] Figure 5 4 is a simulation diagram of the apparent source time function and the empirical Green's function in an embodiment of the present invention;

[0050] Figure 6 This is a simulation diagram of a synthetic station receiving waveform in an embodiment of the present invention;

[0051] Figure 7 Schematic diagram of input and output of a deep learning network in an embodiment of the present invention;

[0052] Figure 8Schematic diagram of test results of the entire event in an embodiment of the present invention;

[0053] Figure 9 Schematic diagram of test results of the mutual coefficient of a single-channel waveform in an embodiment of the present invention;

[0054] Figure 10 Schematic diagram of the corresponding relationship between the output of the network and the duration of the label feature rupture in an embodiment of the present invention;

[0055] Figure 11 A structural diagram of the device for extracting the time function of the apparent source provided by the present invention;

[0056] Figure 12 This is a block diagram of the internal structure of a terminal device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following is a clear and complete description of the technical solutions of the present invention in conjunction with the embodiments and the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0058] Research has found that by calculating the apparent focal time function of the earthquake source area, the source rupture information of the earthquake source area (including finite fault parameters such as the characteristic length, characteristic width, characteristic velocity, characteristic duration, and rupture directionality of the rupture) can be inverted through the apparent focal time function. The existing technology uses the following method to calculate the apparent focal time function of the earthquake source area:

[0059] First, the empirical Green's function of small earthquake events (magnitude less than 2.5) that have occurred around the source area is collected, and the waveform of the source area is collected. Then, the starting time (recorded as the waveform starting time) and the ending time (recorded as the waveform ending time) of the waveform are marked, and the starting time of the empirical Green's function corresponding to the waveform starting time is marked (recorded as the function starting time), and the ending time of the empirical Green's function corresponding to the waveform ending time is marked (recorded as the function ending time). The PLD method is applied to the waveform within the time period between the waveform starting time and the waveform ending time, and the function within the time period between the function starting time and the function ending time, and the effective time length of the apparent source time function is set (the effective time length that the apparent source time function needs to reach, which is the time length during which the function value is non-zero) to obtain the apparent source time function of the source area by deconvolution in the frequency domain.

[0060] The PLD algorithm used in the existing technology requires continuous manual marking of five parameters: the waveform start time, the waveform end time, the Green function start time, the Green function end time, and the effective duration of the apparent source time function, so as to continuously adjust the waveform and the empirical Green function input to the PLD algorithm until the convolution of the apparent source time function obtained by iterative calculation and the waveform currently input to the PLD algorithm meets the set conditions, and the apparent source time function at this time is used as the final apparent source time function of the source area.

[0061] From the above records, it can be seen that the above PLD algorithm involves five parameters, each of which needs to be manually marked. Any marking error in any parameter will affect the accuracy of the final apparent source time function. Therefore, the accuracy of the apparent source time function calculated by the existing technology is unstable.

[0062] In order to solve the above technical problems, the present invention provides a method for extracting the apparent source time function, which solves the problem of instability in the accuracy of the apparent source time function calculated by the prior art. In specific implementation, first, the various seismic events around the target earthquake are located, and the historical Green's function of each of the seismic events is obtained; then, the station waveform of the target earthquake is obtained, and the trained neural network model is applied to the station waveform and each of the historical Green's functions to obtain each apparent source time function; finally, the historical Green's function and each of the apparent source time functions are convolved to obtain the convolution results corresponding to each of the apparent source time functions, and based on each of the convolution results, the target apparent source time function of the target earthquake is screened out from each of the apparent source time functions, and the target apparent source time function is used to calculate the source information of the target earthquake.

[0063] For example, the Earthquake Administration has set up 64 stations for monitoring earthquakes. When multiple stations among the 64 stations detect an earthquake in place A, that is, multiple stations detect the station waveform of place A, the earthquake is the target earthquake. Earthquakes have occurred in places B and C around place A, and places B and C are within the coverage of these 64 stations. The earthquakes in places B and C are earthquake events. The historical Green's function when an earthquake occurred in place B and the historical Green's function when an earthquake occurred in place C are obtained. The historical Green's function and station waveform of place B are then input into the trained neural network model, and the neural network model outputs the apparent source time function (denoted as ASTF). B ), the historical Green's function and station waveform of site C are input into the trained neural network model, and the neural network model outputs the apparent source time function (denoted as ASTF C ). B Convolve with the historical Green's function at point B to get the simulated waveform W B , ASTF CConvolve with the historical Green's function at C to get the simulated waveform W C .Simulation waveform W B and the analog waveform W C If the two waveforms are simulated, the waveform W B If the waveform of the station at location A is consistent with that of the station at location A, then ASTF B As the target apparent source time function of site A; if the simulated waveform W C If the waveform of the station at location A is consistent with that of the station at location A, then ASTF C As the target apparent source time function of site A.

[0064] The method for extracting the time function of the apparent source of earthquake in this embodiment can be applied to a terminal device, which can be a terminal product with a data processing function, such as a seismic analyzer. Figure 1 As shown in , the method for extracting the apparent source time function specifically includes the following steps:

[0065] S100, locating each seismic event around the target earthquake, and obtaining a historical Green's function of each seismic event;

[0066] S200, obtaining a station waveform of the target earthquake, and applying a trained neural network model to the station waveform and each of the historical Green's functions to obtain each apparent source time function;

[0067] S300, convolve the historical Green's function and each of the apparent source time functions to obtain a convolution result corresponding to each of the apparent source time functions, and based on each of the convolution results, filter out the target apparent source time function of the target earthquake from each of the apparent source time functions, and the target apparent source time function is used to calculate the source information of the target earthquake.

[0068] In order to implement steps S100, S200, and S300, it is necessary to first construct Figure 2 The data set shown is used to train a neural network model, and the trained neural network model is used to implement step S200. In this embodiment, the following specific steps S001, S002, S003, and S004 are included:

[0069] S001, obtaining the sample Green's function of the earthquake sample event through a station.

[0070] The sample Green's function eGf of each earthquake sample event includes the Green's functions collected by several stations. Several stations and earthquake sample events meet at least one or all of the following six conditions:

[0071] Condition 1: Several stations are high signal-to-noise ratio waveform stations;

[0072] Condition 2: Each earthquake sample event (the earthquake sample event is the earthquake zone or earthquake site where the earthquake event occurred) is covered by these stations from 0 to 360 degrees (that is, covered in a way that there is no obvious azimuth gap);

[0073] Condition three: each earthquake sample event is covered by several of the above-mentioned stations with an off-source angle of 60-180 degrees, that is, there is no obvious off-source angle gap;

[0074] Condition 4: The waveform signal-to-noise ratio of the earthquake sample event is greater than the set value, which is set to 3 to ensure that there is no interference from large noise;

[0075] Condition 5: The focal mechanism quality of the earthquake sample event is greater than the threshold, which is level C, that is, the focal mechanism quality is equivalent to the A and B classifications of the HASH focal mechanism calculation software;

[0076] Condition 6: M of earthquake sample events L Magnitude (M L The magnitude (i.e., the magnitude of a near-earthquake) is less than 2.5 to ensure that the target event is of a relatively large magnitude, that is, the earthquake sample event in this embodiment is a small earthquake event that can be regarded as a unit pulse.

[0077] When the seismic sample event is an event with a horizontal multi-layer half-space velocity structure, this embodiment can also obtain the sample Green's function in the following manner:

[0078] First, the reflection and transmission coefficients of a single interface in a horizontal multi-layer half-space velocity structure are considered. Then, all interface interactions are comprehensively considered to calculate the generalized reflection and transmission coefficients. Based on the generalized reflection and transmission coefficients and the frequency-domain elastodynamic equations describing the propagation of seismic waves in multi-layer media, the unknown quantities in the wavefield coefficient vector are solved. The wavefield coefficients are substituted into the vector function basis expansion, and the time-domain synthetic seismogram, i.e., the Green's function, is obtained through an inverse Fourier transform.

[0079] When the seismic sample event is an event with a complex three-dimensional velocity structure, this embodiment can also obtain the sample Green's function in the following manner:

[0080] First, establish a subsurface velocity model and set boundary conditions (free boundaries, absorbing boundaries, etc.). Use the finite difference method (FDM) to discretize the continuous physical equations into difference equations, and then obtain the wave field through iterative solution. Alternatively, use the finite element method (FEM) to spatially discretize the calculation area into finite elements, establish local equations on each grid, and then assemble them into a global system of equations for solution. Alternatively, use the spectral element method (SEM) to solve the wave field using high-order polynomial interpolation on each element to obtain the Green's function.

[0081] This embodiment also adds noise or coda to the Green's function to enhance the noise resistance of the model for training and testing the noise resistance of the neural network. The noise sources include:

[0082] The real noise recorded by the detector is randomly selected in a period before the arrival of the P wave of the earthquake event, and the real noise in this period is Fourier transformed. The amplitude spectrum is retained and the phase spectrum is randomly disrupted. After the inverse Fourier transform, the amplitude is multiplied by different coefficients and superimposed on the P wave.

[0083] S002: Determine the fault plane of each earthquake sample event based on the sample Green's function.

[0084] Fault plane Figure 3 As shown, Figure 3 The ellipse in the figure is the fault plane that determines the strike and dip, the gray five-pointed star represents the center of the rupture, θ represents the rupture direction, α represents the rupture length, and b represents the rupture width.

[0085] S003, simulating an earthquake by adjusting the earthquake rupture parameters of the fault plane to obtain a target simulated sample earthquake, and calculating an apparent source time sample function of the target simulated sample earthquake.

[0086] Earthquake rupture parameters include slip distribution and rupture propagation velocity. By adjusting the earthquake rupture parameters to simulate different earthquake events (that is, target simulation sample earthquakes), the propagation velocity model of the seismic wave of the earthquake event in the underground medium is defined (the propagation velocity model is as follows: Figure 4 The seismic wave emission path is obtained by ray tracing, and the seismic moment release process of the simulated target earthquake event (that is, the target simulated sample earthquake) is calculated using the kinematic method. The process is projected onto the ray path to obtain the theoretical seismic moment release process of the simulated target earthquake event to different stations, that is, the apparent source time sample function ASTFs, where s represents multiple apparent source time sample functions corresponding to multiple stations.

[0087] This embodiment can also construct the apparent source time sample function of the target simulated sample earthquake in the following manner:

[0088] The fault plane is divided into several sub-faults, and the slip on each sub-fault is calculated using a dynamic model to obtain the slip distribution of the entire fault. Subsurface structure and rupture parameters are also set, and the Green's function on each sub-fault is calculated using generalized backscatter or wavefield simulation. The slip on the corresponding sub-fault is convolved with the Green's function to obtain the waveform of the sub-fault at the station. By superimposing the waveforms of each sub-fault at that location, the seismic waveform recorded at that station during the entire earthquake process is obtained. This seismic waveform is also known as the apparent source time sample function.

[0089] S004. Convolve the sample Green's function and the apparent source time sample function to obtain the simulated sample waveform of the target simulated sample earthquake, and construct the data set through the sample Green's function of the earthquake sample event, the apparent source time sample function of the target simulated sample earthquake, and the simulated sample waveform of the target simulated sample earthquake.

[0090] When the target simulated sample earthquake is a small earthquake event, this embodiment performs convolution on the sample Green's function and the apparent source time sample function to obtain a simulated sample waveform of the small earthquake event.

[0091] When the target simulated sample earthquake is a large earthquake event, this embodiment extracts the simulated sample waveform of the large earthquake event in the following manner:

[0092] The underground structure and rupture parameters of the large earthquake event are obtained. After calculating the Green's function using the generalized anti-transmission or wave field simulation scheme, the apparent source time sample function ASTFs is convolved with the calculated Green's function term to obtain the theoretical target waveform of the station (that is, the simulated sample waveform).

[0093] This embodiment also adds noise to the simulated sample waveforms in the dataset to train the neural network model trained using the dataset to be noise-resistant.

[0094] The Green's function in S002 of this embodiment is as follows Figure 5 As shown in b in Figure 1, the apparent source time sample function ASTFs in S003 is as follows: Figure 5 As shown in a in Figure 1, the analog sample waveform in S004 is as follows Figure 6 As shown, that is, Figure 5 Figure b, Figure a, and Figure 6 The Green's function, apparent source time function and waveform are shown in turn.

[0095] In Example 2, the dataset constructed in Example 1 is evaluated. The evaluation results can be used as a basis for selecting the required dataset for training the neural network model. The evaluation methods include:

[0096] Evaluation of eGfs in the dataset (where s represents the Green's function corresponding to multiple earthquake events): eGfs events are evaluated based on the uniformity of the distribution of the earthquake locations of the eGfs events and the size of the coverage area of the eGfs events.

[0097] Evaluating the ASTFs in the dataset: The ASTFs are compared with real ASTFs to assess their plausibility. This approach evaluates the reliability of the ASTFs in the dataset. This involves analyzing all simulated earthquake events in the dataset using the seismic second-order moment method to obtain finite fault parameters. The calculated finite fault parameters for the simulated events are compared with statistical results from the regional earthquake catalog to ensure that the characteristic rupture velocity is consistent with the regional geological conditions; the characteristic rupture duration and scale are similar to the statistical distribution of the real earthquake catalog; and the rupture directionality of the simulated events is consistent with the regional earthquake record.

[0098] Embodiment 3: This embodiment is based on embodiment 1 or embodiment 2. This embodiment provides a Figure 2 The input and output of the neural network model shown is:

[0099] Since the rupture duration of this type of earthquake is usually no longer than 0.5 seconds, for earthquake waveform data with a sampling rate of 500 Hz, the number of sampling points of ASTFs is usually within 256. In order to standardize the input and output formats of the neural network, the present invention designs a method such as Figure 7 The input and output method of the neural network model shown in the figure stacks the synthetic received waveforms of 64 stations and their corresponding eGf by azimuth to form two (64,512) matrices. These two matrices are then merged in the channel dimension to generate a (2,64,512) tensor as the fixed input of the neural network.

[0100] The neural network model of this embodiment can be a Dec-Net model structure, or a convolutional neural network (CNN), a recurrent neural network (RNN), or a Transformer. According to the complexity of the problem, an appropriate model is selected to improve the model performance.

[0101] It is also possible to use one-dimensional or two-dimensional convolution or a fully connected layer as the decoding part in the middle of the network within the framework of CNN encoding (Encoder) and decoding (Decoder).

[0102] The neural network model of this embodiment can also be a U-Net structure, in which an attention mechanism or multi-scale feature fusion is introduced to further improve the model performance.

[0103] This embodiment can also use Transformer to replace the encoding and decoding structures in the Dec-Net model, enhance the model's ability to process time series data, and improve the effect of ASTFs deconvolution.

[0104] This embodiment can also combine the Transformer and U-Net structures to obtain a new model, which combines the temporal processing advantages of the Transformer with the spatial feature extraction capabilities of the U-Net to construct a hybrid model for ASTFs deconvolution.

[0105] Example 4, based on Example 1, Example 2, or Example 3, this example trains a neural network model based on the data set of Example 1. Before training, the data set of Example 1 is first divided into a training set, a validation set, and a test set. The three sets cover the same eGfs events, and the ASTFs obtained by forward modeling have similar visual duration distributions. The training set is used to train the neural network model, the validation set is used to verify the trained neural network model, and the test set is used to test the verified neural network model.

[0106] This embodiment trains a neural network model, including the following specific steps: inputting the simulated sample waveform and the sample Green's function into the neural network model to obtain an apparent source time training function output by the neural network model; determining a loss function of the neural network model based on the apparent source time training function and the apparent source time sample function, and iteratively training the neural network model based on the loss function.

[0107] The sample Green's function includes the sample Green's function of the same earthquake sample event collected by 64 stations, that is, the number of sample Green's functions is 64.

[0108] The simulated sample waveforms include waveforms of the target simulated sample earthquakes collected by 64 stations, that is, 64 waveforms. These 64 waveforms and 64 sample Green's functions are stacked and input into the neural network model. The neural network model outputs 64 apparent source time training functions. A loss function is determined by combining these 64 apparent source time training functions with the 64 apparent source time sample functions. The neural network model is iteratively trained based on this loss function until the model converges or the number of iterations is reached, resulting in a trained neural network model. During this training process, the simulated ASTFs of the 64 stations are stacked and expanded into a (1, 64, 512) tensor, which serves as the training label and fixed output dimension of the neural network.

[0109] Among them, the specific steps of constructing the loss function include: determining the mean square error MSE between the apparent source time training function and the apparent source time sample function; counting the characteristic duration of the apparent source time training function and the characteristic duration of the apparent source time sample function, and calculating the duration difference between the characteristic duration of the apparent source time training function and the characteristic duration of the apparent source time sample function; counting the non-zero point length of the apparent source time training function and the non-zero point length of the apparent source time sample function, and calculating the length difference between the non-zero point length of the apparent source time training function and the non-zero point length of the apparent source time sample function; constructing the loss function based on the mean square error, the duration difference, and the length difference.

[0110] The mean square error, duration difference, and length difference are added together according to different proportional coefficients to obtain the loss function.

[0111] Example 5. Step S300 in this example includes the following specific steps: convolving the historical Green's function with each of the apparent seismic source time functions to obtain a simulation waveform corresponding to each of the apparent seismic source time functions, and using each of the simulation waveforms as a convolution result; determining the similarity between the station waveform and each of the simulation waveforms, and screening out the target apparent seismic source time function ASTFs from each of the apparent seismic source time functions based on the similarity corresponding to each of the apparent seismic source time functions.

[0112] After obtaining the target apparent source time function ASTFs of the target earthquake, this embodiment applies the seismic second-order moment method to ASTFs to obtain the characteristic rupture length, width, duration, rupture velocity and rupture directionality of the target earthquake, which are the source rupture parameters. This embodiment combines ASTFs with the seismic second-order moment to change the final output content, and can achieve fully automated acquisition from the original waveform recording of the earthquake to the source rupture parameters, and more quickly output the source parameters of large quantities of small and medium-sized earthquakes without manual intervention. Moreover, through the source rupture parameters, more detailed underground fault structural characteristics can be obtained, and a more detailed explanation of the earthquake rupture mode can be provided.

[0113] Example 6: This example verifies the advantages of the apparent source time function extraction method of the present invention over the prior art:

[0114] Figure 8 、 Figure 9 、 Figure 10 The results of testing the invention using 4556 simulated small and medium earthquake events corresponding to an empirical Green's function are shown. Figure 8 and Figure 9The bars filled with diagonal lines show the correlation coefficients between events and overall ASTFs, while the bars filled with dots show the correlation coefficients between individual ASTFs. Over 99.97% of the output and labeled overall ASTFs for earthquake events have correlation coefficients above 0.998; over 99.93% of the output and labeled single-channel ASTF waveforms have correlation coefficients above 0.998. Figure 10 The horizontal and vertical axes represent the feature duration of the label and the predicted ASTF, respectively. The black dots represent the corresponding relationship between the two. The dark black dotted line represents that the predicted and label feature durations are equal. The light gray dotted line frames the error percentage range between the feature duration of the predicted ASTF and the feature duration of the label ASTF. The number of waveforms with a feature duration error within 3% accounts for 99.80%, and the number of waveforms with a feature duration error within 5% accounts for 99.97%. The determination coefficient R is used to 2 It represents the relative error between the two, and the coefficient of determination is 0.9999.

[0115] The cross-correlation coefficient between each ASTF (i.e., apparent source time training function) output by the neural network and each ASTF (i.e., apparent source time sample function) of the label is above 0.998 for 99.93% of the waveforms. After obtaining the ASTFs of different stations, the characteristic rupture duration can be obtained by taking twice the standard deviation of the ASTFs. The number of waveforms with an error of less than 3% for the characteristic duration is 99.80%. The coefficient of determination R is used to determine the characteristic rupture duration. 2 The relative error between the two is expressed as 0.9999. After loading the model and data on a single NVIDIA GeForce RTX 4090 GPU, it takes only 4.1 seconds to output ASTFs for 4556 earthquake events. This invention not only significantly improves the deconvolution speed of ASTFs, but also enhances the ability to subsequently use the seismic second-order moment method to process a large number of small and medium earthquakes and obtain finite fault parameters.

[0116] In summary, the present invention realizes the accurate and rapid completion of ASTFs automated large-scale extraction, and then obtains the rupture parameters of large quantities of regional small and medium earthquakes through the earthquake second-order moment method, completing the full-process automated analysis of the earthquake rupture process. The eGfs data set adopted by the present invention covers most of the focal mechanisms in the specified area. After a target event of a slightly larger magnitude occurs, the network target waveform (Target Wave) of the event and the different eGfs in the neural network training data set are input into the trained neural network in pairs. At this time, the neural network will output the ASTFs extraction results under different eGf conditions. The station receiving waveform and the ASTFs output by the neural network under different eGf combinations are convolved with the corresponding eGf used. After the convolution result is cross-correlated with the target waveform, the unique eGf that best matches the target earthquake can be selected, and the optimal ASTFs for the target earthquake can be determined.

[0117] This embodiment also provides a device for extracting the time function of the apparent source, such as Figure 11 As shown, the device includes the following components:

[0118] Positioning module 01, used to locate each earthquake event around the target earthquake and obtain the historical Green's function of each earthquake event;

[0119] The apparent source time function calculation module 02 is used to obtain the station waveform of the target earthquake and apply the trained neural network model to the station waveform and each of the historical Green's functions to obtain each apparent source time function;

[0120] The apparent source time function prediction module 03 is used to convolve the historical Green's function and each apparent source time function to obtain a convolution result corresponding to each apparent source time function, and based on each convolution result, screen out the target apparent source time function of the target earthquake from each apparent source time function, and the target apparent source time function is used to calculate the source information of the target earthquake.

[0121] Based on the above embodiment, the present invention further provides a terminal device, whose principle block diagram can be shown as follows: Figure 12As shown. The terminal device includes a processor, memory, a network interface, and a display screen connected via a system bus. The processor of the terminal device is used to provide computing and control capabilities. The memory of the terminal device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the terminal device is used to communicate with an external terminal via a network connection. When executed by the processor, the computer program implements a method for extracting a time function of a visual seismic source. The display screen of the terminal device can be a liquid crystal display or an electronic ink display.

[0122] Those skilled in the art will understand that Figure 12 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the solution of the present invention, and does not constitute a limitation on the terminal device to which the solution of the present invention is applied. The specific terminal device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0123] In one embodiment, a terminal device is provided. The terminal device includes a memory, a processor, and a program for extracting the time function of the apparent seismic source stored in the memory and executable on the processor. When the processor executes the program for extracting the time function of the apparent seismic source, the following operating instructions are implemented:

[0124] Locating each seismic event around the target earthquake and obtaining a historical Green's function of each of the seismic events;

[0125] Obtaining a station waveform of the target earthquake, and applying a trained neural network model to the station waveform and each of the historical Green's functions to obtain each apparent source time function;

[0126] The historical Green's function and each of the apparent source time functions are convolved to obtain a convolution result corresponding to each of the apparent source time functions, and based on each of the convolution results, the target apparent source time function of the target earthquake is screened out from each of the apparent source time functions, and the target apparent source time function is used to calculate the source information of the target earthquake.

[0127] Those skilled in the art will appreciate that all or part of the processes in the above-described embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described embodiments. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAM bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for extracting the time function of the apparent source, characterized in that: include: Locating each seismic event around the target earthquake and obtaining a historical Green's function of each of the seismic events; Obtaining a station waveform of the target earthquake, and applying a trained neural network model to the station waveform and each of the historical Green's functions to obtain each apparent source time function; Convolving the historical Green's function with each of the apparent source time functions to obtain a convolution result corresponding to each of the apparent source time functions, and screening out a target apparent source time function of the target earthquake from each of the apparent source time functions based on each of the convolution results, wherein the target apparent source time function is used to calculate the source information of the target earthquake; The convolution of the historical Green's function and each of the apparent source time functions to obtain a convolution result corresponding to each of the apparent source time functions, and screening out a target apparent source time function of the target earthquake from each of the apparent source time functions based on each of the convolution results, comprises: Convolving the historical Green's function with each of the apparent source time functions to obtain a simulation waveform corresponding to each of the apparent source time functions, and using each of the simulation waveforms as a convolution result; Determine the similarity between the station waveform and each of the simulated waveforms, and based on the similarity corresponding to each of the apparent source time functions, screen out the target apparent source time function from each of the apparent source time functions, wherein the similarity is calculated using the normalized mean square error or waveform cross-correlation.

2. The method for extracting the apparent source time function according to claim 1, wherein: The locating of each seismic event around the target earthquake and obtaining the historical Green's function of each seismic event includes: Acquire a data set, the data set including a plurality of earthquake sample events and a sample Green's function of each earthquake sample event; The distances between several of the earthquake sample events and the target earthquake are obtained, and based on the several distances, the various earthquake events of the target earthquake are located from the several earthquake sample events, and the sample Green's function of each earthquake event is used as the historical Green's function of each earthquake event.

3. The method for extracting the apparent source time function according to claim 2, wherein: The data set also includes a target simulated sample earthquake corresponding to the earthquake sample event, a simulated sample waveform of the target simulated sample earthquake, and an apparent source time sample function of the target simulated sample earthquake; The method for constructing the dataset includes: Acquire the sample Green's function of the earthquake sample event through a station; determining the fault plane of each of the earthquake sample events based on the sample Green's function; Simulating an earthquake by adjusting the earthquake rupture parameters of the fault plane to obtain a target simulated sample earthquake, and calculating an apparent source time sample function of the target simulated sample earthquake; The sample Green's function and the apparent source time sample function are convolved to obtain the simulated sample waveform of the target simulated sample earthquake, and the data set is constructed through the sample Green's function of the earthquake sample event, the apparent source time sample function of the target simulated sample earthquake, and the simulated sample waveform of the target simulated sample earthquake.

4. The method for extracting the apparent source time function according to claim 3, wherein: The earthquake sample events are covered by the stations in an omnidirectional manner; And / or, the earthquake sample events are recorded by the station at 60 0 -180 0 Coverage in the form of angles away from the source; and / or, the magnitude of the earthquake sample event is less than a set magnitude; and / or, the waveform signal-to-noise ratio of the seismic sample event is greater than a set value; And / or, the focal mechanism quality of the earthquake sample event is greater than a threshold.

5. The method for extracting the apparent source time function according to claim 3, wherein: The data set is used to train the neural network model, and the training method of the neural network model includes: Inputting the simulated sample waveform and the sample Green's function into the neural network model to obtain an apparent source time training function output by the neural network model; The loss function of the neural network model is determined according to the apparent source time training function and the apparent source time sample function, and the neural network model is iteratively trained according to the loss function.

6. The method for extracting the apparent source time function according to claim 5, wherein: Determining the loss function of the neural network model based on the apparent source time training function and the apparent source time sample function includes: Determining a mean square error between the apparent source time training function and the apparent source time sample function; Counting the characteristic duration of the apparent source time training function and the characteristic duration of the apparent source time sample function, and calculating the duration difference between the characteristic duration of the apparent source time training function and the characteristic duration of the apparent source time sample function; Counting the non-zero point lengths of the apparent source time training function and the non-zero point lengths of the apparent source time sample function, and calculating the length difference between the non-zero point lengths of the apparent source time training function and the non-zero point lengths of the apparent source time sample function; A loss function is constructed based on the mean square error, the duration difference, and the length difference.

7. A device for extracting the time function of an apparent earthquake source, characterized in that: The device comprises the following components: A positioning module, configured to locate each seismic event around the target earthquake and obtain the historical Green's function of each seismic event; an apparent source time function calculation module, configured to obtain the station waveform of the target earthquake and apply a trained neural network model to the station waveform and each of the historical Green's functions to obtain each apparent source time function; an apparent source time function prediction module, configured to convolve the historical Green's function with each of the apparent source time functions to obtain a convolution result corresponding to each of the apparent source time functions, and to screen out a target apparent source time function of the target earthquake from each of the apparent source time functions based on each of the convolution results, wherein the target apparent source time function is used to calculate the source information of the target earthquake; The convolution of the historical Green's function and each of the apparent source time functions to obtain a convolution result corresponding to each of the apparent source time functions, and screening out a target apparent source time function of the target earthquake from each of the apparent source time functions based on each of the convolution results, comprises: Convolving the historical Green's function with each of the apparent source time functions to obtain a simulation waveform corresponding to each of the apparent source time functions, and using each of the simulation waveforms as a convolution result; Determine the similarity between the station waveform and each of the simulated waveforms, and based on the similarity corresponding to each of the apparent source time functions, screen out the target apparent source time function from each of the apparent source time functions, wherein the similarity is calculated using the normalized mean square error or waveform cross-correlation.

8. A terminal device, characterized in that: The terminal device includes a memory, a processor, and a visual source time function extraction program stored in the memory and runnable on the processor. When the processor executes the visual source time function extraction program, the steps of the visual source time function extraction method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program for extracting the time function of the apparent seismic source. When the program is executed by the processor, the steps of the method for extracting the time function of the apparent seismic source according to any one of claims 1 to 6 are implemented.

Images