Method for generating multi-point seabed ground motion and related device

Abstract

Embodiments of the present application provide a kind of multi-point seabed seismic generation method and related equipment, can be synthesized multi-point seismic time-domain non-stationary and the spatial variability of complex topography when considering seismic.The method comprises: according to three direction seismic waveform data, random noise and target single-point random seismic model generation one-dimensional seismic waveform data;One-dimensional seismic waveform data is spliced with random data satisfying Dirichlet distribution, to obtain spliced multi-point seismic data;According to the size of semi-space homogeneous isotropic geologic body, the position of fault plane and the size of the fault plane, establish three-dimensional site model containing fault;According to the fault plane parameters corresponding to different moment magnitude and three-dimensional site model, the simulation analysis of seismic is carried out by spectral element method, to obtain the multi-point seismic data of each measuring point in measuring point set;Spliced multi-point seismic data and multi-point seismic data are input into multi-point seismic data generation model, to obtain multi-point seismic.

Description

Technical Field

[0001] The present invention relates to the field of earthquake motion, and in particular to a method for generating multi-point seabed earthquake motion and related equipment. Background Art

[0002] Reasonable ground motion synthesis methods are crucial for assessing the dynamic response of underground engineering structures to earthquakes. Real ground motion records exhibit significant time-frequency nonstationarity, so synthesized ground motion waveforms must consider numerous constraints, such as frequency, amplitude, duration, and response spectrum. Furthermore, offshore sites often contain unfavorable geological conditions, such as submarine basins, valleys, and faults, which can cause ground motion to exhibit significant spatial variability and inconsistency during propagation. Therefore, for long-span marine projects (such as ultra-long cross-sea tunnels and bridges), multi-point ground motion synthesis methods must consider both the temporal nonstationarity and spatial variability of ground motion.

[0003] Currently proposed methods for the time-frequency nonstationarity of ground motion mainly include the random finite fault method and the engineering synthesis method. The random finite fault method is complex in parameter determination and has limited applicability in practical engineering projects. The engineering synthesis method requires the pre-determination of a mathematical model based on a large number of seismic records. However, most offshore sites lack abundant seismic records, making it difficult to obtain a reliable mathematical statistical model for marine ground motion. Furthermore, neither of these two methods considers the duration and energy characteristics of seismic waves and cannot accommodate all time-frequency constraints of seismic waves. Therefore, currently proposed methods cannot accurately simulate the time-frequency nonstationarity of marine ground motion. Currently proposed methods for the spatial variability of ground motion mainly rely on power spectral density models and coherence models based on array data. Because the power spectral density model assumes a uniform local site, this method cannot account for complex offshore sites and is therefore unsuitable for simulating the spatial variability of ground motion. Summary of the Invention

[0004] The present invention provides a method and device for generating multi-point seabed seismic motions, which combines the spectral element numerical simulation method with a generative neural network, and takes into account the time domain non-stationarity of seismic motions and the spatial variability of complex terrain when synthesizing multi-point seismic motions.

[0005] A first aspect of the present invention provides a method for generating multi-point seabed seismic motions, the method comprising:

[0006] Obtain target three-directional seismic waveform data corresponding to the target area;

[0007] generating target one-dimensional seismic waveform data corresponding to the target three-directional seismic waveform data according to the target three-directional seismic waveform data, random noise, and a target single-point random seismic motion model;

[0008] splicing the target one-dimensional seismic waveform data with random data satisfying Dirichlet distribution to obtain target spliced ​​multi-point seismic motion data;

[0009] Establishing a three-dimensional site model containing the fault based on the size of the semi-space homogeneous isotropic geological body, the position of the fault plane and the size of the fault plane;

[0010] Based on the fault plane parameters corresponding to different moment magnitudes and the three-dimensional site model, a spectral element method is used to simulate and analyze the earthquake motion to obtain target multi-point earthquake motion data for each measuring point in the set of measuring points arranged in the target area;

[0011] The target spliced ​​multi-point seismic motion data and the target multi-point seismic motion data are input into a target multi-point seismic motion data generation model to obtain the multi-point seismic motion corresponding to the target area.

[0012] A second aspect of the present invention provides a device for generating multi-point seabed earthquake motions, comprising:

[0013] An acquisition module is used to acquire target three-directional seismic waveform data corresponding to the target area;

[0014] a generating module, configured to generate target one-dimensional seismic waveform data corresponding to the target three-directional seismic waveform data based on the target three-directional seismic waveform data, random noise, and a target single-point random seismic motion model;

[0015] a splicing module, configured to splice the target one-dimensional seismic waveform data with random data satisfying Dirichlet distribution to obtain target spliced ​​multi-point seismic motion data;

[0016] A construction module for establishing a three-dimensional site model including a fault according to the size of the semi-space uniform isotropic geological body, the position of the fault plane and the size of the fault plane;

[0017] a simulation analysis module for performing a simulation analysis of seismic motions using a spectral element method based on the fault plane parameters corresponding to different moment magnitudes and the three-dimensional site model, to obtain target multi-point seismic motion data for each measuring point in the set of measuring points arranged in the target area;

[0018] The synthesis module is used to input the target spliced ​​multi-point seismic motion data and the target multi-point seismic motion data into a target multi-point seismic motion data generation model to obtain the multi-point seismic motion corresponding to the target area.

[0019] A third aspect of an embodiment of the present invention provides an electronic device, including a memory and a processor, wherein the processor is configured to implement the steps of the method for generating multi-point seabed seismic motions as described in the first aspect above when executing a computer management program stored in the memory.

[0020] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium having a computer management program stored thereon. When the computer management program is executed by a processor, the steps of the method for generating multi-point seabed seismic motions as described in the first aspect above are implemented.

[0021] To sum up, it can be seen that in the embodiment provided by the present invention, the single-point random seismic motion model and the multi-point seismic motion data generation model are pre-trained by a generative adversarial neural network, and at the same time, the spectral element method is introduced when determining the multi-point seismic motion data of the measuring point. Therefore, when synthesizing the sea area seismic motion based on the single-point random seismic motion model and the multi-point seismic motion data generation model, the spectral element method numerical simulation method and the generative neural network are combined, and the time domain non-stationarity of the seismic motion and the spatial variability of the complex terrain are taken into account at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic flow chart of a method for generating multi-point seabed seismic motions provided in an embodiment of the present application;

[0023] Figure 2 A schematic diagram of the process of generating real data of earthquake motion images provided by an embodiment of the present invention;

[0024] Figure 3 A schematic diagram of the structure of the generator and discriminator in the generative neural network provided by the present invention;

[0025] Figure 4 A schematic diagram of a virtual structure of a device for generating multi-point seabed earthquake motions provided by an embodiment of the present invention;

[0026] Figure 5 A schematic diagram of the hardware structure of a device for generating multi-point seabed earthquake motions provided by an embodiment of the present invention;

[0027] Figure 6 A schematic diagram of an electronic device according to an embodiment of the present invention;

[0028] Figure 7 A schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0030] In the following description, the specific embodiments of the present invention will be described with reference to steps and symbols performed by one or more computers, unless otherwise stated. Therefore, these steps and operations will be mentioned several times as being performed by a computer, and the computer execution referred to herein includes the operation of a computer processing unit by electronic signals representing data in a structured form. This operation converts the data or maintains it at a location in the computer's memory system, which can be reconfigured or otherwise change the operation of the computer in a manner familiar to testers in the field. The data structure in which the data is maintained is a physical location in the memory, which has specific characteristics defined by the data format. However, the principles of the present invention are described in the above text, which does not represent a limitation, and testers in the field will understand that the various steps and operations described below can also be implemented in hardware.

[0031] The principles of the present invention may be implemented and operated using many other general-purpose or special-purpose computing and communication environments or configurations. Examples of well-known computing systems, environments, and configurations suitable for use with the present invention include, but are not limited to, handheld phones, personal computers, servers, multiprocessor systems, microcomputer-based systems, mainframe computers, and distributed computing environments, including any of the aforementioned systems or devices.

[0032] The terms "first", "second" and "third" in the present invention are used to distinguish different objects rather than to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions.

[0033] Numerical simulation has always been an important means for engineers to understand the propagation patterns of ground motion and earthquake response at a site. By establishing a specific marine site model, a wealth of spatial multi-point ground motion data can be obtained. Common numerical simulation methods include the spectral element method, the finite element method, and the finite difference method. Among them, the spectral element method has high accuracy and can be used for ground motion simulation of large-scale, complex geological structure models. It has good applicability in the simulation of multi-point ground motion in marine areas. At the same time, deep learning methods have also been widely applied to various engineering fields. They can help process and generate various data samples, obtain the characteristics of the data itself and the inherent laws between the data, and can flexibly and conveniently process marine ground motion data. They have considerable potential and advantages in the field of ground motion synthesis methods. The present invention combines the spectral element method with deep learning, wherein the spectral element method is used to generate a site ground motion database covering a wide range of geological conditions; deep learning is used to generate single-point random ground motions on the one hand, and on the other hand, the nonlinear mapping relationship between single-point ground motions and multi-point ground motions is learned based on the site ground motion database, thereby generating multi-point random ground motions that can reflect the distribution of ground motion in the real site.

[0034] The following describes a method for generating multi-point seabed seismic motions from the perspective of a device for generating multi-point seabed seismic motions. The device for generating multi-point seabed seismic motions may be a server or a service unit in the server, without limitation.

[0035] See also Figure 1 , Figure 1 A schematic flow chart of a method for generating multi-point seabed seismic motions provided in an embodiment of the present invention includes:

[0036] 101. Obtain target three-directional seismic waveform data corresponding to the target area.

[0037] In this embodiment, the device for generating multi-point seabed seismic motion can obtain the target three-directional seismic waveform data corresponding to the target area, wherein the target area is a sea area with an adjustable range, and the target three-directional seismic waveform data includes n data points. The method of obtaining the target three-directional seismic waveform data is not specifically limited here.

[0038] 102. Generate target one-dimensional seismic waveform data corresponding to the target three-directional seismic waveform data based on the target three-directional seismic waveform data, random noise, and a target single-point random seismic motion model.

[0039] In this embodiment, the device for generating multi-point seabed seismic motion first trains the training data set through an adversarial generative network to obtain a target single-point random seismic motion model, and then generates target one-dimensional seismic waveform data corresponding to the target three-directional seismic waveform data based on the target three-directional seismic waveform data, random noise, and the target single-point random seismic motion model. That is, the target three-directional seismic waveform data is first processed to obtain unidirectional seismic motion data, and seismic motion real image data X is generated based on the unidirectional seismic motion data, and X is input into the target single-point random seismic motion model to finally obtain the target one-dimensional seismic waveform data. The training method of the target single-point random seismic motion model is described below:

[0040] Step 1: acquiring m sets of three-directional seismic waveform data corresponding to each of the multiple regions according to the user's operation instruction;

[0041] In this step, the device for generating multi-point seabed seismic motion can manually select three-directional seismic waveform data for training and testing from the existing seismic wave data corresponding to each area in multiple areas, a total of m groups, where each seismic waveform data in the m groups of three-dimensional seismic waveform data includes n data points, where m and n are both integers greater than or equal to 1.

[0042] Step 2: constructing a seismic data matrix based on m sets of three-directional seismic waveform data;

[0043] In this step, the device for generating multi-point seabed seismic motions can construct a seismic data matrix X based on m sets of three-directional seismic waveform data, wherein the seismic data matrix is ​​represented by the following formula:

[0044]

[0045] Among them, x ij is the i-th data point in the j-th group of three-directional seismic waveform data, i is any one of the n data points, and j is any one of the m groups of data.

[0046] Step 3: performing principal component analysis on the seismic data matrix to obtain one-dimensional seismic waveform data corresponding to each seismic waveform data in the m groups of three-directional seismic waveform data;

[0047] In this step, the device for generating multi-point seabed seismic motion can first perform data standardization processing on each seismic waveform data in the target group of three-directional seismic waveform data, where the target group of three-directional seismic waveform data is any one of the m groups of three-directional seismic waveform data. Specifically, the data standardization processing can be performed on each seismic waveform data using the following formula:

[0048]

[0049] x′ ij is x ij After the data is standardized, var(x j ) is the mean square error, is the mean,

[0050] Afterwards, the correlation coefficient matrix corresponding to each seismic waveform data in the target group of three-directional seismic waveform data is calculated. Specifically, the correlation coefficient matrix can be expressed by the following formula:

[0051]

[0052] in,

[0053] Finally, the one-dimensional seismic waveform data corresponding to each seismic waveform data in the three-directional seismic waveform data of the target group is determined according to the singular value decomposition method (SVD) and the correlation coefficient matrix. Specifically,

[0054] R=UDV T

[0055] Among them, U corresponds to m m-dimensional standard orthogonal vectors, V T Corresponding to n n-dimensional standard orthogonal vectors, they correspond to RR T and RT The decomposed eigenvectors of R are used to compress the number of rows and columns, respectively. D corresponds to a diagonal matrix with m rows and n columns, where the diagonal elements represent the squares of the eigenvalues. Based on the eigenvectors and eigenvalues, the target one-dimensional seismic wave data can be obtained.

[0056] It should be noted that the target one-dimensional seismic wave data can be used as the input value of the parameterized generator combination, X m =X m×n V n , X m is the input data of the parameterized generator combination after dimensionality reduction, X m×n is the original sample data, V n V T The corresponding eigenvector in .

[0057] Step 4: Determine the frequency, amplitude, and wavelet coefficients of the one-dimensional seismic waveform data.

[0058] In this step, the device for generating multi-point seabed seismic motion can use wavelet transform to obtain three-dimensional data of frequency, time and wavelet coefficients based on one-dimensional seismic waveform data, wherein the wavelet basis function is a Morlet wave function. The wavelet basis function is represented by the following formula:

[0059]

[0060] Among them, f(t) is the original earthquake motion data, a is the wavelet transform scale, corresponding to the frequency (its inverse), controlling the expansion and contraction of the wavelet function, and b is the wavelet translation, corresponding to time, controlling the translation of the wavelet function. Figure 2 As shown, Figure 2 A schematic diagram of the process of generating real data of earthquake motion images provided by an embodiment of the present invention.

[0061] Step 5: Generate a seismic wave data image according to the frequency, amplitude and wavelet coefficient, and establish a ground motion image database according to the seismic wave data image;

[0062] In this step, after obtaining the frequency, time, and wavelet coefficients, the multi-point seafloor seismic motion generator can create a time-frequency diagram of the seismic waveform based on these frequencies, time, and wavelet coefficients. This diagram is called a single-point true seismic motion image, where the horizontal axis represents time and the vertical axis represents frequency. The color of each point represents the wavelet coefficient. Ultimately, a seismic motion image database is established based on this single-point true seismic motion image.

[0063] Step 6: Train the initial single-point random seismic motion model based on the seismic motion image database to obtain the target single-point random seismic motion model.

[0064] In this step, after the multi-point seabed seismic motion generating device has completed the construction of the seismic motion image database, it can train the initial single-point random seismic motion model based on the seismic motion image database to obtain the target single-point random seismic motion model. The following describes the model training process:

[0065] The multi-point seabed seismic motion generating device inputs random noise into the first generator of the initial single-point random seismic motion model to obtain seismic sample data. That is, the random noise is input into the first generator and processed through the convolution layer, deconvolution layer, and pooling layer to obtain a single-point synthetic seismic motion image generated by the first generator;

[0066] and inputting the single-point synthetic seismic image and the image sample data in the seismic image database into the first discriminator of the single-point random seismic model to obtain a predicted value;

[0067] Then, the loss values ​​of the first generator and the first discriminator are calculated based on the predicted values ​​corresponding to the single-point synthetic ground motion image and the single-point real ground motion image, and the weight parameters of the first generator and the first discriminator are updated using the Adam optimizer. The loss value is calculated according to the loss function, which can be, for example, the root mean square difference, which is expressed by the following formula:

[0068]

[0069] in, denote the response spectra of generated and real earthquake motions, respectively, ε = 0.01, z is the noise input to the generator, and S is the response spectrum of earthquake motion.

[0070] Finally, the above steps are iteratively executed until a preset iteration termination condition is met, resulting in the target single-point random seismic motion model. Specifically, the device generating multiple-point seafloor seismic motions can be configured with physical constraints. Through iterative cycles, the iteration terminates when the generated single-point seismic motion data satisfies both the loss function and the physical constraints, resulting in the target single-point random seismic motion model.

[0071] See also Figure 3 , Figure 3 This is a schematic diagram of the structure of the generator and discriminator in the generative adversarial neural network provided in an embodiment of the present invention. The target single-point random seismic motion model adopts a generative adversarial neural network, which includes a generator and a discriminator. The length of the random noise is set to 4096 sampling points, and the seismic motion data X generally takes a sampling value of 120 seconds. In the present invention, 4096 sampling points are set. At the same time, considering the large amplitude variation of the seismic waveform, the maximum absolute value normalization method is used to first place the seismic motion data in the [-1, 1] interval to eliminate the influence of amplitude differences, as shown in the following formula:

[0072]

[0073] in, is the normalized intercepted waveform, max(*) and abs(*) are the maximum value function and the absolute value function, respectively.

[0074] For the generator G, its input is random noise z. Subsequently, convolution and maximum pooling operations are used in sequence to extract and downsample the random noise features to obtain high-dimensional abstract features. Then, deconvolution and convolution operations are used in sequence to upsample and extract the high-dimensional abstract features to obtain the synthetic ground motion image. Specifically, the input random noise is first convolved to obtain the Gconv1-1 layer, followed by a maximum pooling operation to obtain the Gpool1 layer. This process is repeated to obtain the Gconv2-1, Gpool2, Gconv3-1, Gpool3, Gconv4-1, Gpool4, and Gconv5-1 layers respectively. Subsequently, deconvolution is used for upsampling and the Gconv4-1 layer is spliced ​​in the layer dimension to obtain the Gup4 layer. Subsequently, convolution is performed to obtain the Gconv4-2 layer. This process is repeated to obtain the Gup3, Gconv3-2, Gup2, Gconv2-2, Gup1, and Gconv1-2 layers respectively. The Gconv1-2 layer is the synthetic ground motion image. The convolution kernel size of each convolution operation in the generator is 5×5, the stride is 1×1, the pooling stride and receptive field of each pooling operation are 4×4, and the upsampling ratio of each deconvolution operation is 4×4. The number of layers of Gconv1-1, Gconv2-1, Gconv3-1, Gconv4-1, and Gconv5-1 are [4, 8, 16, 32, 64] respectively, and the number of layers of Gconv1-2, Gconv2-2, Gconv3-2, and Gconv4-2 are [1, 8, 16, 32] respectively.

[0075] For the discriminator D, the input of the discriminator D is the earthquake motion image; then the convolution layer and the pooling layer are used alternately to extract high-dimensional features; then, the multi-channel features of the fully connected layer are integrated and the sigmoid function is used to generate the true and false prediction values ​​S. The step size of the convolution layer Dconv1~Dconv5 is 1×1, the convolution kernel size is 5×5, and the number of output channels is [4,8,16,32,64] respectively; the pooling receptive field and step size of the pooling layer Dpool1~Dpool4 are both 4×4; the number of output nodes of the fully connected layer Dfull is 1.

[0076] For the loss function, the generator loss function is expressed as follows:

[0077] LG=-∑lns fake

[0078] wherein s fake is the predicted true or false probability of a single point synthetic ground motion image;

[0079] The loss function of the discriminator is shown in the following formula: LD=-∑lns real -∑ln(1-s fake )

[0080] wherein s real is the predicted true or false probability of a single point real ground motion image.

[0081] 103. The target one-dimensional seismic waveform data is spliced with random data satisfying the Dirichlet distribution to obtain target spliced multi-point ground motion data.

[0082] In the embodiment, after determining the target one-dimensional seismic waveform data, the multi-point seabed ground motion generation device can splice the target one-dimensional seismic waveform data with random data satisfying the Dirichlet distribution to obtain target spliced multi-point ground motion data.

[0083] 104. A three-dimensional site model containing a fault is established according to the size of the half-space homogeneous isotropic geological body, the position of the fault plane, and the size of the fault plane.

[0084] In the embodiment, the multi-point seabed ground motion generation device can determine the size of the half-space homogeneous isotropic geological body corresponding to the target region, the position of the fault plane, and the size of the fault plane, and establish a three-dimensional site model containing a fault.

[0085] 105. Seismic motion simulation analysis is performed by the spectral element method according to the fault plane parameters corresponding to different moment magnitudes and the three-dimensional site model to obtain target multi-point ground motion data of each measuring point in the measuring point set arranged in the target region.

[0086] In this embodiment, moment magnitude is a physical quantity used to describe the mechanical strength of an earthquake. It is based on the concept of seismic moment, which is obtained by multiplying the rupture area of ​​the seismic fault, the average displacement, and the shear modulus of the rock. Moment magnitude can reflect the amount of energy released by an earthquake and does not suffer from saturation issues like other magnitude scales. The device for generating multi-point submarine seismic motion can arrange measuring points at fixed intervals along the fault strikes parallel to and perpendicular to the fault strikes corresponding to the target area, thereby obtaining a set of measuring points arranged within the target area. Since a three-dimensional site model containing faults has been constructed in advance, the SPECFEM3D spectral element method program is used to simulate and analyze seismic motions based on the three-dimensional site model and set parameters to obtain seismic motion data for each measuring point. Therefore, after the set of measuring points is arranged within the target area, different moment magnitudes can be input into the three-dimensional site model to change the target area's stratum parameters, friction coefficient, initial stress, and other fault surface parameters. This allows the acquisition of multiple sets of marine multi-point seismic motion data for each measuring point parallel to and perpendicular to the fault strikes, i.e., the target multi-point seismic motion data.

[0087] It should be noted that the target spliced ​​multi-point seismic motion data can be determined through steps 102 to 103, and the target multi-point seismic motion data can be determined through steps 104 to 105. However, there is no restriction on the order of execution between steps 102 to 103 and steps 104 to 105. Steps 104 to 105 can be executed first, or steps 102 to 103 can be executed first, or they can be executed simultaneously. There is no specific limitation.

[0088] 106. Input the target spliced ​​multi-point seismic motion data and the target multi-point seismic motion data into a target multi-point seismic motion data generation model to obtain the multi-point seismic motion corresponding to the target area.

[0089] In this embodiment, the device for generating multi-point seabed seismic motions can pre-build a marine multi-point seismic motion database, and train an initial multi-point random seismic motion data generation model based on the marine multi-point seismic motion database to obtain a target multi-point seismic motion data generation model. The target spliced ​​multi-point seismic motion data and the target multi-point seismic motion data are then input into the target multi-point seismic motion data generation model to obtain the multi-point seismic motion corresponding to the target area. The training process of the target multi-point seismic motion data generation model is described below:

[0090] Step 1: determining sample multi-point seismic data of each measuring point in a set of sample measuring points arranged in each of the multiple areas;

[0091] In this step, the device for generating multi-point seafloor seismic motions can first determine the size of a semi-space uniform isotropic geological body and the location and size of the fault plane, establish a three-dimensional site model containing the fault, and arrange sample measuring points at fixed intervals along the fault strike and perpendicular to the fault strike to obtain a set of sample measuring points. Then, based on the three-dimensional site model and set parameters, the SPECFEM3D spectral element method program is used to simulate and analyze the seismic motions to obtain seismic motion data for each sample measuring point. Finally, different moment magnitudes are input, and formation parameters and fault plane parameters such as friction coefficient and initial stress are changed to obtain multi-point seismic motion data for each measuring point parallel to the fault strike and perpendicular to the fault strike. A marine multi-point seismic motion database is then established based on this multi-point seismic motion data.

[0092] Step 2: splicing the one-dimensional seismic waveform data with the random data to obtain spliced ​​multi-point seismic motion data;

[0093] In this step, the device for generating multi-point seabed seismic motions may splice the one-dimensional seismic waveform data with the random data to obtain spliced ​​multi-point seismic motion data, wherein the random data satisfies the Dirichlet distribution.

[0094] Step 3: inputting the spliced ​​multi-point earthquake motion data into the second generator of the initial multi-point earthquake motion data generation model to obtain generated data;

[0095] Step 4: inputting the sample multi-point seismic data and the generated data into the second discriminator of the initial multi-point seismic data generation model to obtain predicted data;

[0096] Step 5: Adjust the loss functions of the second generator and the second discriminator according to the predicted data and the generated data;

[0097] Step 6: Perform iterative training based on the second generator and the second discriminator after adjusting the loss function to obtain the target multi-point seismic data generation model.

[0098] That is, the device for generating multi-point seabed seismic motion splices single-point seismic motion data and random data (satisfying the Dirichlet distribution) as input data of the multi-point seismic motion data generation model generator; inputs the real seismic motion data into the discriminator of the multi-point seismic motion data generation model to obtain predicted data; finally, the loss function of the second generator and the second discriminator is adjusted according to the generated data and the predicted data, and the training is iteratively performed to finally obtain a maturely trained target multi-point seismic motion data generation model, which can obtain multi-point seismic motion.

[0099] It should be noted that the composition of the generative neural network, generator, discriminator, and loss function used in this target multi-point seismic data generation model is similar to that of the target single-point random seismic motion model. The only difference is that the output channel size of the generator Gconv1-2 is equal to the number of multi-point seismic measurement points. This has been explained in detail above and will not be repeated here. It is worth noting that the generator weight parameters and discriminator weight parameters in this target multi-point seismic data generation model are independent of the generator and discriminator parameters in the target single-point random seismic motion model and are not shared.

[0100] To sum up, it can be seen that in the embodiment provided by the present invention, the single-point random seismic motion model and the multi-point seismic motion data generation model are pre-trained by a generative adversarial neural network, and the spectral element method is introduced when determining the multi-point seismic motion data of the measuring point. Therefore, when synthesizing the sea area seismic motion according to the single-point random seismic motion model and the multi-point seismic motion data generation model, the spectral element method numerical simulation method and the generative neural network are combined, and the time domain non-stationarity of the seismic motion and the spatial variability of the complex terrain are taken into account.

[0101] The above describes an embodiment of the present invention from the perspective of a method for generating multi-point seabed seismic motions. The following describes an embodiment of the present invention from the perspective of an apparatus for generating multi-point seabed seismic motions.

[0102] See also Figure 4 , a virtual structural diagram of a device for generating multi-point seabed earthquake motion according to an embodiment of the present invention, wherein the device 400 for generating multi-point seabed earthquake motion comprises:

[0103] An acquisition module 401 is used to acquire target three-directional seismic waveform data corresponding to a target area;

[0104] A generating module 402 is configured to generate target one-dimensional seismic waveform data corresponding to the target three-directional seismic waveform data based on the target three-directional seismic waveform data, random noise, and a target single-point random seismic motion model;

[0105] A splicing module 403 is configured to splice the target one-dimensional seismic waveform data with random data satisfying Dirichlet distribution to obtain target spliced ​​multi-point seismic motion data;

[0106] A construction module 404 is used to establish a three-dimensional site model including faults according to the size of the semi-space uniform isotropic geological body, the position of the fault plane and the size of the fault plane;

[0107] A simulation analysis module 405 is configured to perform a simulation analysis of ground motion using a spectral element method based on the fault plane parameters corresponding to different moment magnitudes and the three-dimensional site model, thereby obtaining target multi-point ground motion data for each measuring point in the set of measuring points arranged in the target area.

[0108] The synthesis module 406 is configured to input the target spliced ​​multi-point seismic motion data and the target multi-point seismic motion data into a target multi-point seismic motion data generation model to obtain the multi-point seismic motion corresponding to the target area.

[0109] In one possible design, the generating module 402 is further configured to:

[0110] Acquire, according to a user's operating instruction, m sets of three-dimensional seismic waveform data corresponding to each of the plurality of regions, wherein each piece of seismic waveform data in the m sets of three-dimensional seismic waveform data includes n data points, where m and n are both integers greater than or equal to 1;

[0111] Constructing a seismic data matrix based on the m groups of three-directional seismic waveform data;

[0112] Performing principal component analysis on the seismic data matrix to obtain one-dimensional seismic waveform data corresponding to each piece of seismic waveform data in the m groups of three-directional seismic waveform data;

[0113] determining the frequency, amplitude, and wavelet coefficients of the one-dimensional seismic waveform data;

[0114] generating a seismic wave data image according to the frequency, the amplitude, and the wavelet coefficient, and establishing a seismic image database according to the seismic wave data image;

[0115] An initial single-point random seismic motion model is trained based on the seismic motion image database to obtain the target single-point random seismic motion model.

[0116] In one possible design, the generating module 402 performs principal component analysis on the seismic data matrix to obtain one-dimensional seismic waveform data including:

[0117] performing data standardization processing on each piece of seismic waveform data in a target group of three-directional seismic waveform data, wherein the target group of three-directional seismic waveform data is any one group of three-directional seismic waveform data among the m groups of three-directional seismic waveform data;

[0118] Calculating a correlation coefficient matrix corresponding to each piece of seismic waveform data in the target group of three-directional seismic waveform data;

[0119] The one-dimensional seismic waveform data corresponding to each piece of seismic waveform data in the target group of three-directional seismic waveform data is determined according to the singular value decomposition method and the correlation coefficient matrix.

[0120] In one possible design, the generating module 402 trains the initial single-point random seismic motion model based on the seismic image database to obtain the target single-point random seismic motion model, including:

[0121] Inputting random noise into the first generator of the initial single-point random seismic model to obtain a single-point synthetic seismic image;

[0122] Inputting the single-point synthetic seismic image and the image sample data in the seismic image database into the first discriminator of the single-point random seismic model to obtain a predicted value;

[0123] Calculating loss values ​​of the first generator and the first discriminator according to the predicted values ​​corresponding to the single-point synthetic ground motion image and the single-point real ground motion image, and updating weight parameters of the first generator and the first discriminator using an optimizer;

[0124] The above steps are iteratively performed until a preset iteration termination condition is reached, thereby obtaining the target single-point random seismic motion model.

[0125] In one possible design, the synthesis module 406 is further configured to:

[0126] Determining sample multi-point seismic data for each measuring point in a set of sample measuring points arranged in each of the plurality of areas;

[0127] splicing the one-dimensional seismic waveform data with the random data to obtain spliced ​​multi-point seismic motion data;

[0128] inputting the spliced ​​multi-point earthquake motion data into a second generator of an initial multi-point earthquake motion data generation model to obtain generated data;

[0129] Inputting the sample multi-point seismic data and the generated data into a second discriminator of the initial multi-point seismic data generation model to obtain predicted data;

[0130] Adjust the loss functions of the second generator and the second discriminator according to the predicted data and the generated data;

[0131] Iterative training is performed on the second generator and the second discriminator after the loss function is adjusted to obtain the target multi-point seismic data generation model.

[0132] In one possible design, the target single-point random seismic motion model and the target multi-point seismic motion data generation model both use a generative adversarial neural network, and the generative adversarial neural network includes a generator and a discriminator;

[0133] The dimension of the random noise is 12, the number of hidden units of the full connection layer of the generator is 32, the up-sampling rate of the de-convolution layer of the generator is 2, the number of output layers of the generator is all, and the length of the output layer is [64, 128, 256, 512, 1024, 2048, 4096] in turn;

[0134] The length of the convolution kernel of the convolution layer of the discriminator is 5, the compensation is 1, and the batch normalization is used in the convolution layer to adjust the distribution of the feature map; the compensation of the convolution layer of the discriminator is 1, the length of the convolution kernel is [3, 3, 3, 5, 5] respectively, the number of output channels is [4, 8, 16, 32, 64] respectively, the pooling receptive field and the step length of the pooling layer Dpool1 and Dpool2 of the discriminator are both 2, the pooling receptive field and the step length of the pooling layer Dpool3-Dpool5 of the discriminator are both 4, and the number of hidden units of the full connection layer Dfull1-Dfull3 of the discriminator is [128, 4, 1] respectively.

[0135] In a possible design, the loss function of the generator is represented by the following formula:

[0136] LG=-∑lns fake ;

[0137] The loss function of the discriminator is represented by the following formula:

[0138] LD=-∑lns real -∑ln(1-s fake );

[0139] Wherein, s fake is the predicted true or false probability of the single-point synthetic seismic image, and s real is the predicted true or false probability of the single-point real seismic image.

[0140] The above Figure 4 The multi-point seabed seismic motion generation device in the embodiment of the application is described from the perspective of modular functional entities, and the multi-point seabed seismic motion generation device in the embodiment of the application is described in detail from the perspective of hardware processing, please refer to Figure 5 , the embodiment schematic diagram of the multi-point seabed seismic motion generation device 500 in the embodiment of the application, the multi-point seabed seismic motion generation device 500 comprises:

[0141] An input device 501, an output device 502, a processor 503 and a memory 504 (wherein the number of the processor 503 can be one or more, Figure 5The processor 503 is taken as an example) In some embodiments of the present application, the input device 501, the output device 502, the processor 503 and the memory 504 can be connected through a communication bus or other means, wherein, Figure 5 The communication bus is taken as an example.

[0142] The processor 503 is configured to execute the following steps by calling the operation instructions stored in the memory 504:

[0143] Obtain target three-directional seismic waveform data corresponding to a target region;

[0144] Generate target one-dimensional seismic waveform data corresponding to the target three-directional seismic waveform data according to the target three-directional seismic waveform data, random noise and a target single-point random ground motion model;

[0145] Splice the target one-dimensional seismic waveform data and random data satisfying Dirichlet distribution to obtain target spliced multi-point ground motion data;

[0146] Establish a three-dimensional site model containing a fault according to the size of a semi-space homogeneous isotropic geological body, the position of a fault surface and the size of the fault surface;

[0147] Obtain target multi-point ground motion data of each measuring point in a measuring point set arranged in the target region by performing simulation analysis of ground motion through a spectral element method according to fault surface parameters corresponding to different moment magnitudes and the three-dimensional site model;

[0148] Input the target spliced multi-point ground motion data and the target multi-point ground motion data into a target multi-point ground motion data generation model to obtain multi-point ground motion corresponding to the target region.

[0149] The processor 503 is further configured to execute the following steps by calling the operation instructions stored in the memory 504: Figure 1 Any of the manners in the corresponding embodiments.

[0150] Please refer to Figure 6 , Figure 6 An embodiment schematic diagram of an electronic device provided by the present application.

[0151] As Figure 6 shown, the present application provides an electronic device, which comprises a memory 610, a processor 620 and a computer program 611 stored in the memory 610 and capable of running on the processor 620, and the processor 620 implements the following steps when executing the computer program 611:

[0152] Obtain target three-directional seismic waveform data corresponding to a target region;

[0153] generating target one-dimensional seismic waveform data corresponding to the target three-directional seismic waveform data according to the target three-directional seismic waveform data, random noise, and a target single-point random seismic motion model;

[0154] splicing the target one-dimensional seismic waveform data with random data satisfying Dirichlet distribution to obtain target spliced ​​multi-point seismic motion data;

[0155] Establishing a three-dimensional site model containing the fault based on the size of the semi-space homogeneous isotropic geological body, the position of the fault plane and the size of the fault plane;

[0156] Based on the fault plane parameters corresponding to different moment magnitudes and the three-dimensional site model, a spectral element method is used to simulate and analyze the earthquake motion to obtain target multi-point earthquake motion data for each measuring point in the set of measuring points arranged in the target area;

[0157] The target spliced ​​multi-point seismic motion data and the target multi-point seismic motion data are input into a target multi-point seismic motion data generation model to obtain the multi-point seismic motion corresponding to the target area.

[0158] In a specific implementation process, when the processor 620 executes the computer program 611, it can achieve Figure 1 Any implementation manner in the corresponding embodiments.

[0159] Since the electronic device introduced in this embodiment is a device used to implement a multi-point seabed seismic generation device in an embodiment of the present invention, based on the method introduced in the embodiment of the present invention, technical personnel in this field can understand the specific implementation of the electronic device of this embodiment and its various variations. Therefore, how the electronic device implements the method in the embodiment of the present invention will not be introduced in detail here. As long as the equipment used by technical personnel in this field to implement the method in the embodiment of the present invention falls within the scope of protection of the present invention.

[0160] See also Figure 7 , Figure 7 A schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present invention.

[0161] like Figure 7 As shown, an embodiment of the present invention further provides a computer-readable storage medium 700, on which a computer program 711 is stored. When the computer program 711 is executed by a processor, the following steps are implemented:

[0162] Obtain target three-directional seismic waveform data corresponding to the target area;

[0163] generating target one-dimensional seismic waveform data corresponding to the target three-directional seismic waveform data according to the target three-directional seismic waveform data, random noise, and a target single-point random seismic motion model;

[0164] splicing the target one-dimensional seismic waveform data with random data satisfying Dirichlet distribution to obtain target spliced ​​multi-point seismic motion data;

[0165] Establishing a three-dimensional site model containing the fault based on the size of the semi-space homogeneous isotropic geological body, the position of the fault plane and the size of the fault plane;

[0166] Based on the fault plane parameters corresponding to different moment magnitudes and the three-dimensional site model, a spectral element method is used to simulate and analyze the earthquake motion to obtain target multi-point earthquake motion data for each measuring point in the set of measuring points arranged in the target area;

[0167] The target spliced ​​multi-point seismic motion data and the target multi-point seismic motion data are input into a target multi-point seismic motion data generation model to obtain the multi-point seismic motion corresponding to the target area.

[0168] In a specific implementation process, the computer program 711 is executed by the processor to achieve Figure 1 Any implementation manner in the corresponding embodiments.

[0169] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0170] The embodiment of the present invention also provides a computer program product, which includes computer software instructions. When the computer software instructions are executed on a processing device, the processing device executes the following Figure 1 The process in the corresponding embodiment.

[0171] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part.

[0172] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0173] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. 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 embodiments of the present invention.

Claims

1. A method for generating multi-point seabed earthquake motions, characterized in that: include: Obtain target three-directional seismic waveform data corresponding to the target area; generating target one-dimensional seismic waveform data corresponding to the target three-directional seismic waveform data according to the target three-directional seismic waveform data, random noise, and a target single-point random seismic motion model; The target single-point random earthquake motion model adopts a trained generative adversarial neural network; splicing the target one-dimensional seismic waveform data with random data satisfying Dirichlet distribution to obtain target spliced ​​multi-point seismic motion data; Establishing a three-dimensional site model containing the fault based on the size of the semi-space homogeneous isotropic geological body, the position of the fault plane and the size of the fault plane; Based on the fault plane parameters corresponding to different moment magnitudes and the three-dimensional site model, a spectral element method is used to simulate and analyze the earthquake motion to obtain target multi-point earthquake motion data for each measuring point in the set of measuring points arranged in the target area; The target spliced ​​multi-point seismic motion data and the target multi-point seismic motion data are input into a target multi-point seismic motion data generation model to obtain the multi-point seismic motion corresponding to the target area. The target multi-point seismic motion data generation model adopts a trained generative adversarial neural network.

2. The method according to claim 1, characterized in that The method further comprises: Acquire, according to a user's operating instruction, m sets of three-dimensional seismic waveform data corresponding to each of the plurality of regions, wherein each piece of seismic waveform data in the m sets of three-dimensional seismic waveform data includes n data points, where m and n are both integers greater than or equal to 1; Constructing a seismic data matrix based on the m groups of three-directional seismic waveform data; Performing principal component analysis on the seismic data matrix to obtain target one-dimensional seismic waveform data corresponding to each piece of seismic waveform data in the m groups of three-directional seismic waveform data; Determining the frequency, amplitude, and wavelet coefficients of the target one-dimensional seismic waveform data; generating a seismic wave data image according to the frequency, the amplitude, and the wavelet coefficient, and establishing a seismic image database according to the seismic wave data image; An initial single-point random seismic motion model is trained based on the seismic motion image database to obtain the target single-point random seismic motion model.

3. The method according to claim 2, characterized in that The performing principal component analysis on the seismic data matrix to obtain target one-dimensional seismic waveform data includes: performing data standardization processing on each piece of seismic waveform data in the target three-directional seismic waveform data, wherein the target three-directional seismic waveform data is any one set of three-directional seismic waveform data among the m sets of three-directional seismic waveform data; Calculating a correlation coefficient matrix corresponding to each piece of seismic waveform data in the target three-directional seismic waveform data; The target one-dimensional seismic waveform data corresponding to each piece of seismic waveform data in the target three-directional seismic waveform data is determined according to the singular value decomposition method and the correlation coefficient matrix.

4. The method according to claim 2, characterized in that The training of the initial single-point random seismic motion model based on the seismic motion image database to obtain the target single-point random seismic motion model includes: Inputting random noise into the first generator of the initial single-point random seismic model to obtain a single-point synthetic seismic image; Inputting the single-point synthetic seismic image and the image sample data in the seismic image database into the first discriminator of the single-point random seismic model to obtain a predicted value; Calculating loss values ​​of the first generator and the first discriminator according to the predicted values ​​corresponding to the single-point synthetic ground motion image and the single-point real ground motion image, and updating weight parameters of the first generator and the first discriminator using an optimizer; The above steps are iteratively performed until a preset iteration termination condition is reached, thereby obtaining the target single-point random seismic motion model.

5. The method according to any one of claims 2 to 4, characterized in that The method further comprises: Determining sample multi-point seismic data for each measuring point in a set of sample measuring points arranged in each of the plurality of areas; splicing the target one-dimensional seismic waveform data with the random data to obtain spliced ​​multi-point seismic motion data; inputting the spliced ​​multi-point earthquake motion data into a second generator of an initial multi-point earthquake motion data generation model to obtain generated data; Inputting the sample multi-point seismic data and the generated data into a second discriminator of the initial multi-point seismic data generation model to obtain predicted data; Adjust the loss functions of the second generator and the second discriminator according to the predicted data and the generated data; Iterative training is performed on the second generator and the second discriminator after the loss function is adjusted to obtain the target multi-point seismic data generation model.

6. The method according to any one of claims 1 to 4, characterized in that The generative adversarial neural network includes a generator and a discriminator; The dimension of the random noise is set to 12, the number of hidden units in the fully connected layer of the generator is 32, the upsampling ratio of the deconvolution layer of the generator is 2, the number of layers of the output layer of the generator is , and the lengths of the output layers are [64, 128, 256, 512, 1024, 2048, 4096] in sequence; The convolution kernel length of the convolution layer of the discriminator is 5, the offset is 1, and batch normalization is used in the convolution layer to adjust the distribution of the feature map; the offset of the convolution layer of the discriminator is 1, the convolution kernel lengths are [3, 3, 3, 5, 5], and the number of output channels are [4, 8, 16, 32, 64]. The pooling receptive field and stride of the pooling layers Dpool1 and Dpool2 of the discriminator are both 2, the pooling receptive field and stride of the pooling layers Dpool3~Dpool5 of the discriminator are both 4, and the number of hidden units of the fully connected layers Dfull1~Dfull3 of the discriminator are [128, 4, 1] respectively.

7. The method according to claim 6, characterized in that The loss function of the generator is expressed as follows: ； The loss function of the discriminator is expressed by the following formula: ； in, is the predicted true or false probability of the single-point synthetic seismic image, is the predicted true or false probability of a single point real earthquake image.

8. A device for generating multi-point seabed earthquake motions, characterized in that: include: An acquisition module is used to acquire target three-directional seismic waveform data corresponding to the target area; a generating module, configured to generate target one-dimensional seismic waveform data corresponding to the target three-directional seismic waveform data based on the target three-directional seismic waveform data, random noise, and a target single-point random seismic motion model; The target single-point random earthquake motion model adopts a trained generative adversarial neural network; a splicing module, configured to splice the target one-dimensional seismic waveform data with random data satisfying Dirichlet distribution to obtain target spliced ​​multi-point seismic motion data; A construction module for establishing a three-dimensional site model including a fault according to the size of the semi-space uniform isotropic geological body, the position of the fault plane and the size of the fault plane; a simulation analysis module for performing a simulation analysis of seismic motions using a spectral element method based on the fault plane parameters corresponding to different moment magnitudes and the three-dimensional site model, to obtain target multi-point seismic motion data for each measuring point in the set of measuring points arranged in the target area; A synthesis module is used to input the target spliced ​​multi-point seismic motion data and the target multi-point seismic motion data into a target multi-point seismic motion data generation model to obtain the multi-point seismic motion corresponding to the target area. The target multi-point seismic motion data generation model adopts a trained generative adversarial neural network.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is configured to implement the steps of the method for generating multi-point seabed earthquake motions according to any one of claims 1 to 7 when executing a computer management program stored in the memory.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer management program, which, when executed by a processor, implements the steps of the method for generating multi-point seabed seismic motions according to any one of claims 1 to 7.

Images