Physical-data driven sequential seismic oscillation generation method

Abstract

The invention relates to a physical-data-driven sequence type seismic oscillation generation method. The method comprises the following steps: acquiring a historical seismic record; constructing a main earthquake ground motion random field model based on the physical mechanism of the ground motion field; the physical mechanism comprises a seismic source, a propagation path and a local site model; obtaining main earthquake parameters based on historical earthquake records according to the main earthquake ground motion random field model; the main earthquake parameter is used for predicting an unknown main earthquake seismic record according to a main earthquake seismic oscillation random field model to obtain a main earthquake seismic oscillation time history; and constructing an aftershock seismic oscillation generation model based on a conditional generative adversarial network model according to the main seismic oscillation time history to realize prediction of a seismic oscillation field. According to the method, the parameters of the main earthquake are generated based on the earthquake motion field of the physical mechanism, the time history of the main earthquake is obtained, the aftershock prediction model is constructed according to the conditional generative adversarial network, the physical mechanism model is combined with artificial intelligence, the model prediction fitness and accuracy are improved, the aftershock is predicted, the prediction effect is improved, complex calculation is avoided, and the model construction efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of dynamic analysis, and particularly to a physical-data-driven sequential ground motion generation method. Background Art

[0002] As a highly destructive natural disaster, an earthquake will cause casualties once it occurs, and will cause major damage to the ecological environment, regional economy and social stability. Therefore, predicting the ground motion time history is crucial for reducing earthquake disaster losses.

[0003] Existing ground motion field models can be divided into empirical models for statistical regression of station data and semi-empirical and semi-theoretical models that obtain expressions based on ground motion propagation mechanisms and wave theory and then regress model parameters according to station data. The empirical model proposes a series of delay coherence models based on the ground motion field data recorded by the array. Successively, many scholars have proposed empirical coherence function models under different conditions, and the empirical model performs well in sites similar to the array site conditions. However, when applied to different earthquake sources and different types of local earthquakes, different empirical coherence function models show large differences. Compared with the empirical model, the semi-empirical and semi-theoretical model is a more scientific and reasonable method for estimating the ground motion field. The semi-empirical and semi-theoretical model often starts from physical mechanisms and considers the propagation of shear waves, traveling wave effects, site effects, etc. Models such as those based on random vibration theory and models established according to Fourier spectra have been proposed. When constructing the semi-empirical and semi-theoretical model, although traveling wave effects, site effects, dispersion effects, etc. are considered, its construction process is usually based on the analysis and parameter determination of the ground motion record data and related physical mechanisms of a single ground motion, and the factor of aftershocks is not included in the core consideration range. In actual earthquakes, aftershocks will cause non-negligible additional damage to building structures. Therefore, the influence of aftershocks in earthquake action cannot be ignored.

[0004] In recent years, scholars at home and abroad have proposed various ground motion field models. By separately establishing large- and small-scale site earthquake analysis models, selecting various earthquake source slip models, calculating the local site ground motion field in combination with boundary coordination relations, and then establishing an average time-frequency domain correlation relationship, the rapid construction of the local site ground motion field under any earthquake source slip model is realized; using deep learning algorithms, the spectral characteristics and spatial distribution of ground motion have been deeply studied, and through the learning and analysis of a large amount of historical ground motion data, an intelligent ground motion field simulation method has been proposed, which can quickly and accurately predict the ground motion field distribution under different site conditions.

[0005] In summary, some achievements have been made in the prediction of ground motion time history, but there are still deficiencies, mainly manifested in the following aspects: (1) Under different types of sites, the function models vary greatly and are not accurate enough; (2) The impact of aftershocks in earthquakes is ignored, and aftershocks are not included in the core consideration scope of the model; (3) There are too many parameters of the existing sequential ground motion field, and the parameter identification process is too complex.

[0006] Therefore, there is an urgent need to provide a physics-data-driven sequential ground motion generation method to solve the above-mentioned technical problems. Summary of the Invention

[0007] This application provides a physics-data-driven sequential ground motion generation method. By combining the physical mechanism model with artificial intelligence innovation, the prediction fitness and accuracy of the model are improved. Under different types of sites, the prediction accuracy is high. Aftershocks are included in the core consideration scope of the model, further improving the prediction effect, and avoiding overly complex formula calculations, thus enhancing the model construction efficiency.

[0008] In the first aspect, this application provides a physics-data-driven sequential ground motion generation method, including the following steps: obtaining historical earthquake records; constructing a main shock ground motion random field model based on the physical mechanism of the ground motion field; wherein, the physical mechanism includes the earthquake source, propagation path, and local site model; obtaining main shock parameters according to the main shock ground motion random field model based on the historical earthquake records; wherein, the main shock parameters are used to predict unknown main shock earthquake records according to the main shock ground motion random field model to obtain the main shock ground motion time history; or, obtaining measured earthquake records to obtain the main shock ground motion time history; constructing an aftershock ground motion generation model based on the conditional adversarial generation network model according to the main shock ground motion time history to realize the prediction of the ground motion field.

[0009] Optionally, constructing the main shock ground motion random field model based on the physical mechanism includes: based on the one-dimensional wave equation, representing the generation and propagation process of the main shock ground motion by the surface acceleration time history at a distance of R from the earthquake source and r from the center of the local site, which is described by formula (1): Based on the surface acceleration time history, decomposing the generation and propagation process of the ground motion according to the earthquake source, propagation path, and local site, and respectively constructing the amplitude spectrum and phase spectrum of the earthquake source, propagation path, and local site, which are described by formula (2) and (3): In the formula, is the amplitude spectrum of the earthquake source, , and They are the amplitude spectrum transfer functions of the propagation path, the local site soil layer, and the local site plane wave field, respectively. 、 、 and are the random vectors in the models of the seismic source, propagation path, local site soil layer, and plane wave field, respectively. is the phase spectrum of the seismic source. 、 and are the phase spectrum transfer functions of the propagation path, the local site soil layer, and the local site plane wave field, respectively. Based on the amplitude spectra and phase spectra of the seismic source, propagation path, and local site, the seismic source model, propagation path model, and local site model are constructed respectively to obtain the model parameters that describe the characteristics of the seismic source, propagation path, and local site, so as to construct the main shock ground motion random field model.

[0010] Optionally, based on the amplitude spectra and phase spectra of the seismic source, propagation path, and local site, the seismic source model, propagation path model, and local site model are constructed respectively to obtain the model parameters that describe the characteristics of the seismic source, propagation path, and local site, and the main shock ground motion random field model is constructed, including: obtaining the random variables of the seismic source, propagation path, and local site based on the amplitude spectra and phase spectra of the seismic source, propagation path, and local site, which are used to describe the characteristics of the seismic source, propagation path, and local site to construct the seismic source model, propagation path model, and local site model; where the random variables, that is, the model parameters, are described by formula (13): In the formula, represents the seismic source amplitude coefficient. represents the seismic source coefficient, and the random vector of the seismic source model is , 、 、 and are the random variables describing the variability of the propagation path. The random vector of the propagation path model is , is the equivalent damping ratio of the site. is the equivalent dominant circular frequency of the site. The random vector of the local soil layer is , is the site seismic wave amplitude change coefficient. is the average apparent wave velocity of the site. The local wave field can be represented by the random vector ; Based on the seismic source model, propagation path model, and local site model, the main shock ground motion random field model is constructed.

[0011] Optionally, obtaining the main shock parameters based on the historical earthquake records according to the main shock seismic random field model includes: selecting earthquake records for identification according to preset requirements based on the historical earthquake records; wherein the preset requirements include selecting earthquake records for identification whose earthquake matrix level is not less than a specific order of magnitude and whose peak acceleration is not less than a specific acceleration according to needs; obtaining the amplitude spectrum model and phase spectrum model of the main shock seismic random field model according to Fourier transform based on the main shock seismic random field model; and identifying and obtaining the model parameters based on the amplitude spectrum model and the phase spectrum model respectively based on the identification earthquake records. , , , , , , , as well as , Based on the identified model parameters, the corresponding optimal probability model is obtained according to the Bayesian information criterion, and the marginal probability density function of each model parameter is determined to obtain the model parameter probability distribution to generate the main shock parameters.

[0012] Optionally, the model parameters are identified based on the seismic record for identification according to the amplitude spectrum model and the phase spectrum model. , , , , , , , as well as , , specifically including: using the least squares criterion to fit the amplitude spectrum model, identifying the model parameters in the amplitude spectrum model , , and ; Based on the seismic records used for identification, the real ground motion phase spectrum is obtained; based on the model parameter source coefficient According to the phase spectrum model, the phase difference spectrum of the phase spectrum model is fitted with the real phase difference spectrum distribution by using the least squares criterion, and the parameters in the phase spectrum model are obtained by using the genetic algorithm. , , and ; Based on the model parameters obtained by identification , , and , , , and identify the model parameters in the plane wave field transfer function by the least squares method ; identify the apparent wave velocity in the model parameters based on the time alignment and lag determination method wherein, the obtained apparent wave velocity is the mean value in frequency and is described by formula (16): In the formula is the total number of groups of stations in pairs, is the distance of the i-th group of stations along the wave propagation direction, is the time lag between the i-th group of samples

[0013] Optionally, the apparent wave velocity identified based on the time alignment and lag determination method further includes: based on the apparent wave velocity which is the mean value in frequency obtained use a band-pass filter to filter the target ground motion group at different frequencies, and find the cross-correlation function of the filtered narrowband signal and then obtain the time lag corresponding to the maximum value of the cross-correlation function which is the one corresponding to ; based on the station distance and the obtained time lag obtain the corresponding apparent wave velocity ; repeat the above process to obtain the apparent wave velocities corresponding to different frequencies . .

[0014] Optionally, based on the main shock ground motion time history, construct an aftershock ground motion generation model based on the conditional adversarial generation network model to realize the prediction of the ground motion field, including: obtaining the main shock ground motion time history based on the historical earthquake records and selecting a training set and a test set; using the training set as input data to train the data according to the conditional adversarial generation network model to construct an aftershock ground motion generation model; using the test set as input data to predict the prediction accuracy of the model on unseen new data according to the aftershock ground motion generation model; comparing the predicted value and the actual value of the aftershock ground motion generation model to test the prediction accuracy of the aftershock ground motion generation model

[0015] Optionally, by comparing the predicted values with the actual values of the aftershock ground motion generation model, the prediction accuracy of the aftershock ground motion generation model is verified, and it further includes: after amplitude modulation of the predicted values according to PGA, calculating and comparing the intensity parameters of the predicted values with the intensity parameters of the actual values; as needed, randomly selecting a specific number of predicted aftershock time histories and comparing their response spectra with the true recorded response spectra; comparing the mean response spectra of the predicted time histories obtained by using the training set and the test set as input data respectively with the mean response spectra of the true records.

[0016] In a second aspect, the present application provides a physics-data-driven sequential ground motion generation system, which includes: an acquisition module for acquiring historical earthquake records; a main shock model module for constructing a main shock ground motion random field model based on the physical mechanism of the ground motion field; wherein, the physical mechanism includes the earthquake source, the propagation path, and the local site model; a main shock time history generation module for obtaining main shock parameters based on the historical earthquake records according to the main shock ground motion random field model; wherein, the main shock parameters are used to predict unknown main shock earthquake records according to the main shock ground motion random field model to obtain main shock ground motion time histories; or, acquiring measured earthquake records to obtain main shock ground motion time histories; an aftershock model prediction module for constructing an aftershock ground motion generation model based on the conditional adversarial generation network model according to the main shock ground motion time history to realize the prediction of the ground motion field.

[0017] In a third aspect, the present application further provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the above-mentioned method are implemented.

[0018] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned method are implemented.

[0019] The present application has at least the following advantages: The above steps mainly construct a main shock ground motion random field model based on the physical mechanism of the ground motion field to generate various parameters of the main shock, so as to comprehensively reflect various factors affecting the earthquake and provide an accurate data basis for subsequent aftershock prediction. Then, the main shock time history is obtained according to the obtained main shock parameters, and the aftershock time history is predicted by constructing a ground motion generation model based on the conditional adversarial generation network with the main shock time history as input data. By combining the physical mechanism model with artificial intelligence innovation, the model can perform high-precision prediction under different types of sites, improve the prediction fitness of the model, and include aftershocks in the core consideration range of the model. The ground motion generation model is constructed by using the conditional adversarial generation network, avoiding overly complex formula calculations, improving the model construction efficiency, and further improving the prediction effect. Description of the Drawings

[0020] Figure 1 It is an application environment diagram showing the physical-data-driven sequential ground motion generation method in an embodiment; Figure 2 It is a schematic flow chart showing the steps of the physical-data-driven sequential ground motion generation method in an embodiment; Figure 3 It is a schematic structural diagram showing the physical-data-driven sequential ground motion generation method in an embodiment; Figure 4 It is a schematic structural diagram showing the physical mechanism of ground motion in an embodiment; Figure 5 It is a schematic flow chart showing the process of identifying main shock parameters in an embodiment; Figure 6 It is a schematic structural diagram showing the identification of main shock parameters in an embodiment; Figure 7 It is a Hilbert curve diagram taking the side length of 8 as an example in an embodiment; Figure 8 It is a schematic structural diagram of the CGAN network showing two-dimensional convolution in an embodiment; Figure 9 It is a comparison diagram showing the predicted aftershock time history and the real aftershock when predicting according to the training set in an embodiment; Figure 10 It is a comparison diagram showing the predicted aftershock time history and the real aftershock when predicting according to the test set in an embodiment; Figure 11 It is a comparison diagram showing the calculation of the intensity parameters of the predicted aftershock time history in an embodiment; Figure 12 It is a comparison diagram showing the response spectrum of the predicted aftershock time history in an embodiment; Figure 13 It is a comparison diagram showing the predicted time history mean response spectrum and the real one according to the training set and the test set in an embodiment; Figure 14 It is a schematic diagram of the ground motion field at 13.5 s generated in an embodiment; Figure 15 It is a schematic diagram of five ground motion samples with a spacing of 1000 m in the generated ground motion field in an embodiment; Figure 16 It is a comparison diagram showing the response of the building group in an embodiment; Figure 17 It is a structural block diagram showing the physical-data-driven sequential ground motion generation system in an embodiment; Figure 18 It is a schematic structural diagram of a computer device in an embodiment. Specific implementation manners

[0021] The following further describes the present application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0022] For the convenience of understanding, the system applicable to the present application is first described. A ground motion field prediction method provided by the present application can be applied to a system architecture as Figure 1 shown. The system includes: a user space file server 103 and a terminal device 101. The terminal device 101 communicates with the user space file server 103 through a network. Among them, the user space file server 103 can be a file server based on the NFSv3\v4 protocol and runs in a Linux environment. NFS (Network File System) is a network abstraction above the file system, which allows remote clients running on the terminal device 101 to access through the network in a manner similar to the local file system. The terminal device 101 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, etc. The user space file server 103 can be implemented by an independent server or a server cluster composed of multiple servers.

[0023] Figure 2 The figure is a schematic flow chart of a physical-data-driven sequential ground motion generation method provided by an embodiment of the present application. The method may include the following steps: S201. Obtain historical earthquake records; S202. Construct a main shock ground motion random field model based on the physical mechanism of the ground motion field; wherein, the physical mechanism includes a seismic source, a propagation path, and a local site model; S203. Obtain main shock parameters according to the main shock ground motion random field model based on the historical earthquake records; the main shock parameters are used to predict unknown main shock earthquake records according to the main shock ground motion random field model to obtain the main shock ground motion time history; or, obtain measured earthquake records to obtain the main shock ground motion time history; S204. Construct an aftershock ground motion generation model based on the main shock ground motion time history based on a conditional adversarial generation network model to realize the prediction of the ground motion field.

[0024] In the embodiments of the present application, a main shock ground motion random field model is constructed based on the physical mechanism of the ground motion field to generate various parameters of the main shock, so as to comprehensively reflect various factors affecting the earthquake and provide an accurate data basis for subsequent aftershock prediction. Then, the main shock time history is obtained according to the obtained main shock parameters, and the main shock time history is used as input data to construct a ground motion generation model based on the conditional adversarial generation network to predict the aftershock time history. By combining the physical mechanism model with artificial intelligence innovation, the model can perform high-precision prediction under different types of sites, improve the prediction fitness of the model, and incorporate aftershocks into the core consideration scope of the model. The conditional adversarial generation network is used to construct a ground motion generation model, avoiding overly complex formula calculations, improving the model construction efficiency, and further enhancing the prediction effect.

[0025] The following specifically elaborates on each step in detail: Please refer to Figure 2 、 Figure 3 As shown, step S201: Obtain historical earthquake records.

[0026] In this embodiment, it should be noted that a large number of sequential ground motion records can be collected from the databases PEER and KiK–net&K–NET as historical earthquake records, providing a data basis for subsequent modeling analysis. Among them, the sequential ground motion record refers to the ground motion time history record generated by multiple consecutive sub-events, such as main shocks, foreshocks, aftershocks, etc. or different rupture stages during an earthquake. Such records can more realistically reflect the complex characteristics of actual earthquakes and are especially common in strong earthquakes.

[0027] Please continue to refer to Figure 2 、 Figure 3 As shown, step S202: Construct a main shock ground motion random field model based on the physical mechanism of the ground motion field; where the physical mechanism includes the earthquake source, propagation path, and local site model.

[0028] In this embodiment, it should be noted that as Figure 4 shown, the physical mechanism is the core framework for understanding the characteristics of ground motion. The earthquake source mainly describes the dynamic process of earthquake rupture, the propagation path describes the attenuation that controls seismic waves from the earthquake source to the site, and the local site characterizes the modulation of ground motion by the surface sediment layer. By analyzing the earthquake source, propagation path, and local site, the three subsystems are coupled through the wave equation to construct a main shock ground motion random field model to analyze the characteristics of the main shock and provide a data basis for subsequent prediction and analysis.

[0029] Specifically, based on the one-dimensional wave equation, the generation and propagation process of the main shock ground motion is represented by the surface acceleration time history at a distance of R from the earthquake source and r from the center of the local site, and is described by formula (1): Based on the ground surface acceleration time history, the generation and propagation process of ground motion is decomposed according to the earthquake source, propagation path, and local site, and the amplitude spectra and phase spectra of the earthquake source, propagation path, and local site are constructed respectively, which are described by formulas (2) and (3): In the formulas, is the amplitude spectrum of the earthquake source, , and are the amplitude spectrum transfer functions of the propagation path, local site soil layer, and local site plane wave field respectively, , , and are the random vectors in the models of the earthquake source, propagation path, local site soil layer, and plane wave field respectively, is the phase spectrum of the earthquake source, , and are the phase spectrum transfer functions of the propagation path, local site soil layer, and local site plane wave field respectively; based on the amplitude spectra and phase spectra of the earthquake source, propagation path, and local site, the earthquake source, propagation path, and local site models are constructed respectively to obtain the model parameters that describe the characteristics of the earthquake source, propagation path, and local site, so as to construct the main shock ground motion random field model.

[0030] In an example, first for the earthquake source model, the dynamic process of earthquake rupture is described. The Brune disk dislocation model is adopted, where the dislocation is uniformly distributed and occurs instantaneously at the moment of fracture, and the shear wave propagates perpendicular to the fault plane. By assuming that the shear stress drop during the fault rupture process is a constant value, the expression of rock displacement on the fault plane is obtained, which is described by formula (4): In the formulas, is the shear stress drop of the earthquake source, and are the shear wave velocity and shear modulus of the rock near the fault respectively, is the Brune earthquake source coefficient. Then, through Fourier transform, the amplitude spectrum and phase spectrum of the displacement are obtained for the model to identify, as shown in the following formulas (5) and (6), providing the basic parameters for subsequent calculations.

[0031] In the formulas, the earthquake source amplitude coefficient , then the random vector of the earthquake source model.

[0032] For the propagation path model, it mainly describes the control of the attenuation of seismic waves from the earthquake source to the site. Considering the damping energy dissipation effect of the linearly elastic medium, the wave number-frequency curve is used to reflect the dispersion effect, and its amplitude spectrum transfer function is described by formula (7): In the formula, is the quality factor of the bedrock, is the group velocity of the seismic wave. To simplify the form of the model, a constant K is used to reflect the attenuation effect of the medium, and the empirical value is 10-5 s / km. The influence of the propagation path on the phase is quite complex. Here, the empirical formula of the phase spectrum transfer function is adopted and described by formula (8): In the formula, , and d are random variables describing the variability of the propagation path. Setting the amplitude spectrum parameter K as a constant, the random vector of the propagation path ; For the local site model, it characterizes the modulation of ground motion by the surface sediment layer. The local soil layer is simplified to an equivalent single-degree-of-freedom system, and the amplitude spectrum transfer function is obtained from the motion equation and described by formula (9): In the formula, is the equivalent damping ratio of the site, is the equivalent dominant circular frequency of the site. Since the distance from the bedrock to the surface is short, its influence on the phase of the seismic wave is ignored, that is, it can be expressed by formula (10) as Furthermore, the random vector of the local soil layer ; For the wave field model on the local site, considering that the local site size is much smaller than the epicentral distance, it is simplified to a plane wave field, and the amplitude spectrum transfer function on the local site is obtained and described by formula (11): In the formula, is the amplitude change coefficient of the site seismic wave, which reflects the attenuation effect of the ground motion amplitude along the direction of ground motion propagation. When calculating the phase spectrum transfer function, only the traveling wave effect is considered and the dispersion effect is not considered. The phase spectrum transfer function is:[[]] In the formula, is the average apparent wave velocity of the site. Based on this, the local wave field can be described by the random vector .

[0033] According to the above-mentioned random vectors in the models of the seismic source, propagation path, local site soil layer, and plane wave field , , and , the random variable vector in the single ground motion random field model, that is, the main shock ground motion random field model, is described by formula (13): In the formula, represents the seismic source amplitude coefficient, represents the seismic source coefficient, and the random vector of the seismic source model is , , , , are random variables describing the variability of the propagation path. The random vector of the propagation path model , is the equivalent damping ratio of the site, is the equivalent dominant circular frequency of the site. The random vector of the local soil layer , is the amplitude change coefficient of the seismic wave at the site, is the average apparent wave velocity of the site. The local wave field can be represented by the random vector .

[0034] Based on the amplitude spectra and phase spectra of the seismic source, propagation path, and local site calculated according to the above analysis, the random variables of the seismic source, propagation path, and local site are obtained, which respectively describe the characteristics of the seismic source, propagation path, and local site, comprehensively reflect various factors affecting earthquakes, and have relatively clear physical meanings. Based on the models of the seismic source, propagation path, and local site, the main shock ground motion random field model is constructed, and this model comprehensively reflects the dispersion effect, traveling wave effect, and attenuation effect of ground motion on the local site.

[0035] Please refer to Figure 3 , Figure 5 shown in step S203. Based on historical earthquake records, the main shock parameters are obtained according to the main shock ground motion random field model, including: Step S2031: Screen out the identification earthquake records based on historical earthquake records according to preset requirements; among them, the preset requirements include screening out the identification earthquake records with a seismic matrix magnitude not less than a specific magnitude and a peak acceleration not less than a specific acceleration according to the requirements.

[0036] In this embodiment, it should be noted that a large number of sequential ground motion records, i.e., historical earthquake records, are collected based on the datasets PEER and KiK–net&K–NET, and the obtained historical earthquake records are screened and processed. Since the modeling object is the strong ground motion time history, the original ground motion data is screened based on the preset requirements according to the seismic moment magnitude and the peak ground acceleration (PGA). Finally, the adopted ground motion acceleration time history satisfies that the seismic moment magnitude is not less than magnitude 5 and the peak acceleration is not less than 0.2g.

[0037] In one example, a total of 1014 groups of qualified sequential ground motions are screened. Among them, the sequential ground motions of site classes I, II, III, and IV are 130, 636, 226, and 22 groups respectively. In addition, five groups of ground motion field time histories recorded by the SMART–1 array are selected, and these five groups of ground motions only include the main shock. The basic information of the selected ground motions is shown in Table 1. Here, the horizontal component of the ground motion is mainly studied, so the E–W component and the N–S component in the records of each station are selected and regarded as independent ground motions for processing. Only the main shock in the identified sequential ground motions is analyzed in this part, and the aftershock part is reserved for subsequent prediction analysis.

[0038] Table 1 Measured sequential ground motion data obtained by screening In addition, as Figure 6 shown, in one example, there are differences in the sampling frequencies of the true ground motion time histories recorded by each station. For the convenience of analysis, the sampling frequencies are unified, and the ground motions are uniformly amplitude–scaled according to the PGA and amplitude–normalized. Specifically, here the time history is resampled at intervals of 0.02 s, and the corresponding frequency upper limit is 25 Hz. The difference between the resampled time history and the original time history in the time domain and the frequency domain is small, and the influence on the identification result can be basically ignored. In addition, the ground motions are uniformly amplitude–scaled to 0.1g according to the PGA and amplitude–normalized to provide a data basis for subsequent analysis and processing.

[0039] As Figure 5 、 Figure 6 shown, in step S2032, based on the main–shock ground motion random–field model, according to the Fourier transform, the amplitude–spectrum model and the phase–spectrum model of the main–shock ground motion random–field model are obtained; In this embodiment, it should be noted that the above–mentioned random variables of the earthquake source, propagation path, and local site are obtained based on the amplitude spectrum and phase spectrum of the earthquake source, propagation path, and local site: Combined with the generation and propagation process of the main earthquake ground motion represented by the acceleration time history, the Fourier amplitude spectrum and phase spectrum in the random field model of the main earthquake ground motion are described by formulas (14) and (15): Step S2033: Based on the seismic records for identification, the model parameters are identified according to the amplitude spectrum model and the phase spectrum model. , , , , , , , as well as , , including: according to the seismic records for identification that meet the conditions after screening, the model parameters are identified according to the amplitude spectrum model and the phase spectrum model, wherein the model parameters in the amplitude spectrum model are identified by fitting the amplitude spectrum model using the least squares criterion. , , and ; Based on the identification of earthquake records, the real ground motion phase spectrum is obtained; based on the model parameter source coefficient According to the phase spectrum model, the phase difference spectrum of the phase spectrum model is fitted with the real phase difference spectrum distribution using the least squares criterion, and the parameters in the phase spectrum model are identified using the genetic algorithm. , , , ; Based on the model parameters obtained by identification , , and , , , and , the model parameters in the plane wave field transfer function are identified by the least squares method ; Based on the time alignment and lag determination method, the apparent wave velocity in the model parameters is identified , where the apparent wave velocity is is the mean value over frequency, which is described by formula (16): In the formula, is the total number of stations in pairs, is the distance between the i-th group of stations along the wave propagation direction, For the Time lag between groups of samples. Based on the above parameter identification, the probability distribution of the model parameters is obtained. If there is no measured main shock record in the follow-up, the main shock record can be generated according to the probability distribution, so as to facilitate the subsequent aftershock prediction.

[0040] In this embodiment, it should be noted that specifically, among the parameters, in order to make the model approximate the true ground motion amplitude spectrum, the parameters corresponding to the amplitude spectrum 、 、 and are identified. The identification method uses the least squares method, and its objective function can be described by formula (17): In the formula, is the amplitude spectrum of the model, is the amplitude spectrum of the target ground motion.

[0041] In addition, for the parameters 、 、 and identification, since it is quite difficult to identify its parameters according to the phase spectrum of the true ground motion, the phase difference spectrum has relatively obvious characteristics, and its phase difference generally follows a normal distribution or an approximate normal distribution. Based on this, for the parameters corresponding to the phase spectrum 、 、 and identification, the genetic algorithm is used to make the phase difference distribution of the model close to the phase difference distribution of the true ground motion and improve the identification efficiency. Its objective function can be described by formula (18) In the formula, is the phase difference distribution of the model phase spectrum, is the phase difference distribution of the target ground motion phase spectrum; = is the difference between two adjacent phase angles in the frequency domain. Among them, the parameter settings of the genetic algorithm are shown in Table 2: Table 2 Parameter settings of the genetic algorithm Regarding the identification of the parameters and first, identify the physical parameters of the center point C00 of the site 、 、 and 、 、 、 and , according to the recognized parameters , , and , , , and , the parameters in the plane wave field transfer function are recognized by the least square method . The equivalent wave velocity of the site can be recognized according to the time alignment and lag determination method, and the apparent wave velocity obtained is the mean value in frequency, which can be described by formula (19): In the formula, N is the total number of groups of stations in pairs, is the distance of the i-th group of stations along the wave propagation direction, is the time lag between the i-th group of samples. Based on the obtained apparent wave velocity which is the mean value in frequency, a band-pass filter is used to filter the target ground motion group at different frequencies. Specifically, a Butterworth filter can be used, with the center frequency taken as , and the bandwidth is rad / s. The cross-correlation function of the filtered narrowband signal is obtained , and then the time lag corresponding to the maximum value of the cross-correlation function is obtained, which is the corresponding to ; substituting the station distance and the time lag into formula (19), based on the station distance and the obtained time lag , the apparent wave velocity corresponding to is obtained. Repeating the above process, the apparent wave velocities corresponding to different frequencies can be obtained. Using formula (18) as the objective function to fit the sample points, the equivalent quality factor of the target local site can be obtained.

[0042] Step S2034: Based on the recognized model parameters, the corresponding optimal probability model is obtained according to the Bayesian information criterion, and the marginal probability density function of each model parameter is determined to generate the main shock parameters of the model parameter probability distribution.

[0043] In this embodiment, it should be noted that the Bayesian Information Criterion is a method that, under incomplete information, subjectively estimates some unknown states using probabilities, then corrects the occurrence probabilities using Bayes' formula, and finally makes an optimal decision using the expected value and the corrected probabilities. It is an information metric that is being used more and more frequently. The Bayesian decision theory method is a fundamental method in statistical pattern recognition. The Bayesian decision criterion takes into account both the probabilities of various reference populations occurring and the losses caused by misclassification, and has strong discriminative ability. Specifically, here, based on the seismic records for identification, each group of sequential ground motions is identified one by one, their statistical distributions are examined, and the Bayesian Information Criterion (BIC) is used to obtain the corresponding optimal probability model. Furthermore, the marginal probability density functions of each parameter are determined and described by formula (20): In the formula, is the probability density function, is the number of sample points, , , are the parameters of the probability density function, is the number of distribution parameters. As shown in Table 3, the PDFs of the random variables of the earthquake source and the propagation path and their corresponding parameters are obtained. The meanings of the parameters and in the table are as follows: when the probability density function is Normal, they are the mean and the standard deviation respectively; when the probability density function is Lognormal, they are the mean of the logarithm and the standard deviation; when the probability density function is Weibull, they are the shape parameter and the scale parameter respectively.

[0044] Table 3 PDFs of the random variables of the earthquake source and the propagation path and their corresponding parameters Based on the marginal probability density functions of the model parameters, the probability distributions of the model parameters are obtained, and finally the main shock parameters are generated. This is to obtain the main shock ground motion time history based on the main shock parameters when the main shock time history cannot be obtained, or to obtain the main shock ground motion time history based on the actually obtained main shock records, providing a data basis for subsequent prediction of aftershock data.

[0045] Refer to Figure 2 , Figure 3As shown in the figure, in step S204, based on the main shock ground motion time history, a conditional adversarial network model is used to construct an aftershock ground motion generation model to predict the ground motion field, including: obtaining the main shock ground motion time history from historical earthquake records and selecting a training set and a test set; using the training set as input data to train the data according to the conditional adversarial network model to construct an aftershock ground motion generation model; using the test set as input data to predict the prediction accuracy of the model on unseen new data according to the aftershock ground motion generation model; comparing the predicted values and the actual values of the aftershock ground motion generation model to test the prediction accuracy of the aftershock ground motion generation model.

[0046] In this embodiment, it should be noted that due to the weight sharing and local connection characteristics of the convolutional layer, the number of model parameters can be greatly reduced, so a convolutional network is mainly used here. The operation of convolution is "circular multiplication and summation". The "convolution" in convolution refers to the flipping and translation operation, and the "product" refers to the integral operation. Two-dimensional convolution is mostly used in the field of images and is widely used in the current field of computer vision. Specifically, in this application, a reasonable method is first selected to convert the time series into a two-dimensional matrix, and then feature extraction is performed through two-dimensional convolution to achieve the purpose of predicting the aftershock time history from the main shock time history.

[0047] As Figure 7 shown, the Hilbert curve is a fractal curve (or space-filling curve) that can fill a plane square. This curve can uniquely convert the time series into a two-dimensional matrix without performing corresponding calculations, maximizing the retention of the original size and spatial position information, and can be reversibly converted back into a one-dimensional sequence, which is a more suitable conversion method in this application. Due to its special requirements for the side length of the square, in this application, the main and aftershock time histories are uniformly converted into square matrices, and the insufficient parts are padded with zeros. Referring to Figure 7 , it is the Hilbert curve of a square with a side length of 8.

[0048] In addition, GAN is an adversarial generation model. GAN mainly completes two tasks: manifold learning and probability distribution conversion. This is a strong support for the mathematical principle of GAN. The GAN adversarial generation network is an unsupervised model, that is, an unconditional constrained generation model that cannot control the generated samples. The conditional adversarial network (Conditional generative adversarial nets, CGAN) is a model whose structure adds additional information conditions y to both the generator and the discriminator, thereby adjusting the model, so as to guide the data generation process. Such a condition y can be any type of auxiliary information: it can be a class label or any data from a different distribution.

[0049] Referring to Figure 8As shown, CGAN establishes the correlation between conditional information and real data, and the discriminator in the CGAN network structure will improve the efficiency of each training. Since the purpose of the aftershock ground motion generation model is to generate the time history of aftershocks, the structure of the model adopts the CGAN generation model, and the generator adopts the U-net network. The specific network structures of the generator and the discriminator are as Figure 8 shown. On the left side of the generator in the figure is the convolution operation, and on the right side is the transposed convolution operation. Both the input graph and the output graph are . As the convolution progresses, the width and height of the input graph become smaller and smaller, but the number of corresponding convolution kernels increases accordingly, which can be regarded as feature extraction. The transposed convolution on the right side makes the width and height of the graph larger and larger, but essentially it is still a convolution operation. The matrix after the Hilbert curve transformation of the seismic acceleration time history is used as an additional condition for the model. In this model, except for the last convolutional layer of the discriminator with a stride of 4, the stride of the rest of the convolutional layers is 2. The size of all convolution kernels in the model is . In addition, for the training process of CGAN, in order to ensure the stability of model training and theoretically avoid the problems existing in the training of the original GAN, the gradient penalty WGANs-G algorithm is adopted, which can improve the training stability and the quality of training.

[0050] Specifically, in an example, as Figure 9 、 Figure 10 shown, the main shock ground motion time history is used to construct the aftershock ground motion generation model based on the conditional adversarial generation network model. First, 6 main shock time histories are randomly selected from the sequential ground motions in the historical earthquake records as the training set and the test set. According to the CGAN generation model, the training set is input into the conditional adversarial generation network model to train the two-dimensional convolutional model, and then the predicted two-dimensional matrix is transformed into a one-dimensional time series through the inverse Hilbert curve to obtain the finally predicted aftershock time history. Then, the test set is input into the aftershock ground motion generation model to train the two-dimensional convolutional model, and then the predicted two-dimensional matrix is transformed into a one-dimensional time series through the inverse Hilbert curve to obtain the finally predicted aftershock time history, and the prediction accuracy of the prediction model on unseen new data is evaluated. Finally, the predicted values obtained by using the training set and the test set as input data are compared with the real time history, that is, the actual value, and it is found that the prediction effect of the aftershock ground motion generation model is better.

[0051] In addition, in an example, as Figure 11 、 12As shown, the predicted values of the aftershock ground motion generation model are compared with the actual values to test the prediction accuracy of the aftershock ground motion generation model, which also includes: after adjusting the predicted values according to PGA, calculating the intensity parameters of the predicted values and comparing them with the intensity parameters of the actual values; as needed, randomly selecting a specific number of predicted aftershock time histories and comparing their response spectra with the true recorded response spectra. Specifically, for 3 randomly selected predicted aftershock time histories, their response spectra are compared with the true recorded response spectra. As Figure 13 shown, the mean response spectra of the predicted time histories obtained by using the training set and the test set as input data respectively can also be compared with the mean response spectra of the true records.

[0052] Specific application simulations are carried out according to the above earthquake prediction method. A region in Putuo District, Shanghai, with a length of 11.83 km in the east-west direction, a length of 9.27 km in the north-south direction, and an area of 55.53 km 2 , containing 29,461 individual buildings, is selected. Specifically, a building complex in a 6×6 km 2 area of this administrative region, which contains 8,219 individual buildings, is selected for earthquake disaster simulation, initially demonstrating the application of the proposed sequential ground motion random field model in earthquake disaster simulation of building complexes. According to the historical earthquake records in Shanghai and the results of seismic hazard analysis by the Shanghai Seismological Bureau, the intensity values with a 50-year exceedance probability of 10% and 5% in the center of Shanghai are 7.0 degrees and 7.3 degrees respectively, and the bedrock horizontal PGA with 50-year exceedance probabilities of 10%, 5%, and 2-3% are 0.86 m / s 2 , 1.05 m / s 2 and 1.17 m / s 2 respectively. Here, the Shanghai-Nantong potential seismic source area is selected for the simulation of the sequential ground motion field. The epicentral distance of the center point of this potential seismic source area is 105 km, and the focal depth is 12 km. Considering that there are many newly built residential buildings in the past 20 years and all have certain seismic fortification measures, the PGA of the main shock is specifically selected as 3.0 m / s2. The site type of Putuo District is type IV, and it can be considered that the equivalent quality factor of this area is less than the quality factor of SMART–1. Specifically, it is assumed that is 15, and other model parameters are determined by Table 3.

[0053] Referring to Figure 14 , the wave field of the sequential ground motion field at 13.5 s is generated according to the method proposed above. Different from the uniform excitation, affected by the traveling wave effect and the dispersion effect, there are both regions at the wave crest and regions at the wave trough in the wave field at a given moment, and affected by the traveling wave effect and the dispersion effect, the amplitudes of the accelerations at each point are not the same, showing a significant spatial variation effect.

[0054] Five samples of ground motion acceleration time histories taken from the wave field at 1000 m intervals are as follows Figure 15 As shown, although their waveforms are relatively similar, affected by the dispersion effect and traveling wave effect, the ground motion time histories in each time period of the five samples are different, and there is a certain time lag phenomenon, which cannot be achieved by models considering only a single effect. In addition, the aftershock time history in the sequence-type ground motion is slightly smaller in amplitude than the main shock time history and has a shorter duration, which is consistent with the statistical law in the measured ground motion samples.

[0055] As Figure 16 shown, it shows the comparison of the responses of the building group at 13.5 s under different actions of the uniform excitation and the sequence-type ground motion field. Among them, the ground motion at the center point of the site is selected for the uniform excitation, and the PGA is 3 m / s 2 . In Figure 16 (a) shown, in the response of the building group under the action of uniform excitation, the building groups in the entire area are simultaneously excited by the same amplitude, and the differences in the responses of each individual building are only caused by the different structural heights and restoring force parameters. For buildings with the same height and plane size, regardless of their positions in the area, their responses are the same. This obviously does not conform to the law of ground motion propagation and the actual earthquake damage situation. While in Figure 16 (b) shown in the response of the building group under the action of the ground motion field, it can be clearly observed that there are significant differences in the responses of buildings at different positions, and the propagation process and spatial variation effect of the ground motion are quite clearly reflected in the figure. At this time, the response of the building is not only related to the parameters of the building itself, but also closely related to the position where the building is located. Different seismic excitations at different positions make the responses of completely identical buildings still different.

[0056] The implementation principle of this embodiment: The above steps mainly construct the main shock ground motion random field model based on the physical mechanism of the ground motion field. According to the obtained historical earthquake records, various parameters of the main shock are generated based on the main shock ground motion random field model, so as to comprehensively reflect various factors affecting the earthquake and provide an accurate data basis for subsequent aftershock prediction. Then, based on the obtained main shock parameters, a ground motion generation model is constructed based on the conditional adversarial generation network to predict the aftershock time history. By combining the physical mechanism model with artificial intelligence innovation, the model can perform high-precision prediction under different types of sites, improve the prediction fitness of the model, and include the aftershock in the core consideration range of the model. A ground motion generation model is constructed using the conditional adversarial generation network, and overly complex formula calculations are avoided, improving the model construction efficiency and further enhancing the prediction effect.

[0057] Refer to Figure 17As shown in the figure, the present application also provides a physical-data-driven sequential ground motion generation system, which may include: an acquisition module 301, a main shock model module 302, a main shock ground motion time history generation module 303, and an aftershock model prediction module 304. The main functions of each component module are as follows: The acquisition module 301 is configured to acquire historical earthquake records; The main shock model module 302 is configured to construct a main shock ground motion random field model based on the physical mechanism of the ground motion field; wherein, the physical mechanism includes the earthquake source, the propagation path, and the local site model; The main shock time history generation module 303 is configured to obtain main shock parameters based on the historical earthquake records according to the main shock ground motion random field model; wherein, the main shock parameters are used to predict unknown main shock earthquake records according to the main shock ground motion random field model to obtain the main shock ground motion time history; or, obtain measured earthquake records to obtain the main shock ground motion time history; The aftershock model prediction module 304 is configured to construct an aftershock ground motion generation model based on the main shock ground motion time history based on the conditional adversarial generation network model to realize the prediction of the ground motion field.

[0058] As Figure 18 shown, it is a block diagram of a computer device according to an embodiment of the present application. The computer device is intended to represent various forms of digital computers or mobile devices. Among them, the digital computer may include a desktop computer, a portable computer, a workbench, a personal digital assistant, a server, a mainframe computer, and other suitable computers. The mobile device may include a tablet computer, a smart phone, a wearable device, etc.

[0059] As Figure 18 shown, the device 600 includes a computing unit 601, a ROM 602, a RAM 603, a bus 604, and an input / output (I / O) interface 605. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other through the bus 604. The input / output (I / O) interface 605 is also connected to the bus 604.

[0060] The computing unit 601 can perform various processes in the method embodiments of the present application according to computer instructions stored in the read-only memory (ROM) 602 or computer instructions loaded from the storage unit 608 into the random access memory (RAM) 603. The computing unit 601 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. The computing unit 601 can include, but is not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. In some embodiments, the method provided by the embodiments of the present application can be implemented as a computer software program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 608.

[0061] The RAM 603 can also store various programs and data required for the operation of the device 600. Part or all of the computer programs can be loaded and / or installed onto the device 600 via the ROM 602 and / or the communication unit 609.

[0062] The input unit 606, output unit 607, storage unit 608, and communication unit 609 in the device 600 can be connected to the I / O interface 605. Among them, the input unit 606 can be, for example, a keyboard, a mouse, a touch screen, a microphone, etc.; the output unit 607 can be, for example, a display, a speaker, an indicator light, etc. The device 600 can exchange information, data, etc. with other devices through the communication unit 609.

[0063] Various embodiments of the systems and technologies described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGA), application-specific integrated circuits (ASIC), application-specific standard products (ASSP), system-on-chip systems (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof.

[0064] The computer instructions for implementing the method of the present application can be written in any combination of one or more programming languages. These computer instructions can be provided to the computing unit 601, such that when the computer instructions are executed by a computing unit 601 such as a processor, the steps involved in the method embodiments of the present application are executed.

[0065] The computer-readable storage medium provided by the present application can be a tangible medium that can contain or store computer instructions for executing the steps involved in the method embodiments of the present application. The computer-readable storage medium can include, but is not limited to, storage media in the forms of electronic, magnetic, optical, electromagnetic, etc.

[0066] The above specific embodiments do not constitute a limitation on the protection scope of this application. Those skilled in the art should understand that various modifications, combinations, sub - combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims

1. A physical-data-driven sequential ground motion generation method, characterized in that, It includes the following steps: Obtain historical earthquake records; Construct a main shock ground motion random field model based on the physical mechanism of the ground motion field; wherein, The physical mechanism includes the earthquake source, the propagation path, and the local site model; Based on the historical earthquake records, obtain main shock parameters according to the main shock ground motion random field model; wherein, The main shock parameters are used to predict unknown main shock earthquake records according to the main shock ground motion random field model to obtain the main shock ground motion time history; or, obtain measured earthquake records to obtain the main shock ground motion time history; Based on the main shock ground motion time history, construct an aftershock ground motion generation model based on the conditional adversarial generation network model to realize the prediction of the ground motion field.

2. The method according to claim 1, wherein, The constructing the main shock ground motion random field model based on the physical mechanism includes: Based on the one-dimensional wave equation, represent the generation and propagation process of the main shock ground motion by the surface acceleration time history at a distance of R from the earthquake source and r from the center of the local site, and describe it by formula (1): Based on the surface acceleration time history, decompose the generation and propagation process of the ground motion according to the earthquake source, the propagation path, and the local site, and respectively construct the amplitude spectrum and phase spectrum of the earthquake source, the propagation path, and the local site, and describe them by formulas (2) and (3): In the formula, is the amplitude spectrum of the seismic source, , and are the amplitude spectrum transfer functions of the propagation path, local site soil layer, and local site plane wave field, respectively, , , and are the random vectors in the models of the seismic source, propagation path, local site soil layer, and plane wave field, respectively, is the phase spectrum of the seismic source, , and are the phase spectrum transfer functions of the propagation path, local site soil layer, and local site plane wave field, respectively; Based on the amplitude spectra and phase spectra of the earthquake source, the propagation path, and the local site, respectively construct the earthquake source, the propagation path, and the local site models to obtain model parameters that respectively describe the characteristics of the earthquake source, the propagation path, and the local site, so as to construct the main shock ground motion random field model.

3. The method according to claim 2, characterized in that The constructing the earthquake source, the propagation path, and the local site models based on the amplitude spectra and phase spectra of the earthquake source, the propagation path, and the local site respectively to obtain model parameters that respectively describe the characteristics of the earthquake source, the propagation path, and the local site, and constructing the main shock ground motion random field model includes: Based on the amplitude spectra and phase spectra of the earthquake source, the propagation path, and the local site, obtain the random variables of the earthquake source, the propagation path, and the local site, which are used to describe the characteristics of the earthquake source, the propagation path, and the local site to construct the earthquake source, the propagation path, and the local site models; wherein, The random variable, that is, the model parameter, is described by formula (13): In the formula, represents the source amplitude coefficient, represents the source coefficient, and the random vector of the source model is , , , , are random variables describing the variability of the propagation path, and the random vector of the propagation path model is , is the equivalent damping ratio of the site, is the equivalent dominant circular frequency of the site, and the random vector of the local soil layer is , is the amplitude change coefficient of the site seismic wave, is the average apparent wave velocity of the site, and the local wave field can be represented by the random vector ; Based on the earthquake source model, the propagation path model, and the local site model, construct the main shock ground motion random field model.

4. The method according to claim 3, wherein Based on the historical earthquake records, obtaining main shock parameters according to the main shock ground motion random field model includes: Based on the historical earthquake records, screen out the earthquake records for identification according to preset requirements; wherein, The preset requirements include screening earthquake records for identification with a seismic matrix level not less than a specific order of magnitude and a peak acceleration not less than a specific acceleration according to the requirements; Based on the main shock ground motion random field model, obtain the amplitude spectrum model and phase spectrum model of the main shock ground motion random field model according to the Fourier transform; Based on the seismic records for identification, according to the amplitude spectrum model and the phase spectrum model, the model parameters are respectively identified , , , , , , , and , ; Based on the identified model parameters, obtain the corresponding optimal probability model according to the Bayesian information criterion, determine the marginal probability density function of each model parameter, so as to obtain the model parameter probability distribution to generate the main shock parameters.

5. The method according to claim 4, wherein, The screening of the seismic records for identification based on the historical seismic records according to preset requirements further includes: The ground motions of the seismic records for identification are uniformly amplitude - adjusted to 0.1g according to PGA and subjected to amplitude normalization processing.

6. The method according to claim 4 or 5, characterized in that Based on the identified seismic records, respectively identify the model parameters according to the amplitude spectrum model and the phase spectrum model and and and , and and and and and , specifically including: Based on the amplitude spectrum model, perform least squares criterion fitting to identify the model parameters in the amplitude spectrum model , , and ; Based on the seismic records for identification, the true ground motion phase spectrum is obtained; Based on the source coefficient of the model parameters , according to the phase spectrum model, the phase difference spectrum of the phase spectrum model is fitted to the true phase difference spectrum distribution by using the least square criterion, and the parameters in the phase spectrum model are identified by using the genetic algorithm , , and ; Based on the recognized model parameters , , and , , , and , the model parameters in the plane wave field transfer function are identified by the least squares method ; The apparent wave velocity in the model parameters is identified based on the time alignment and lag determination method , where the obtained apparent wave velocity c g is the mean value in frequency and is described by the formula: In the formula, is the total number of groups of stations in pairs, is the distance of the i-th group of stations along the wave propagation direction, is the time lag between the i-th group of samples.

7. The method according to claim 6, characterized in that, The apparent wave velocity is identified by the time alignment and lag determination method described above , and further includes: Based on the obtained apparent wave velocity that is the mean in frequency , a band - pass filter is used to filter the target ground motion group at different frequencies, and the cross - correlation function of the filtered narrow - band signals is calculated , and then the time lag corresponding to the maximum value of the cross - correlation function is obtained , which is the corresponding to ; Based on the station distance and the obtained time lag , obtain the corresponding apparent wave velocity ; Repeat the above process to obtain the apparent wave velocities corresponding to different frequencies .

8. The method according to any one of claims 1 to 7, characterized in that, The construction of the aftershock ground motion generation model based on the main - shock ground motion time history according to the conditional adversarial generation network model to realize the prediction of the ground motion field includes: Based on the historical seismic records, the main - shock ground motion time history is obtained and the training set and the test set are selected; Using the training set as input data, data training is carried out according to the conditional adversarial generation network model to construct the aftershock ground motion generation model; Using the test set as input data, the prediction accuracy of the prediction model on unseen new data is predicted according to the aftershock ground motion generation model; According to the comparison between the predicted value and the actual value of the aftershock ground motion generation model, the prediction accuracy of the aftershock ground motion generation model is verified.

9. The method according to claim 8, wherein The comparison between the predicted value and the actual value of the aftershock ground motion generation model to verify the prediction accuracy of the aftershock ground motion generation model further includes: After the predicted value is amplitude - adjusted according to PGA, the intensity parameters of the predicted value are calculated and compared with the intensity parameters of the actual value; According to needs, a specific number of predicted aftershock time histories are randomly selected, and their response spectra are compared with the true record response spectra; The mean response spectra of the predicted time histories obtained by using the training set and the test set as input data respectively are compared with the mean response spectra of the true records.

Images