Regional seismic oscillation time history generation method and system

Abstract

The invention discloses a regional seismic oscillation time history generation method and system. The method comprises the steps that existing seismic animal physical simulation data information is acquired and preprocessed to construct a training data set; constructing a regional seismic oscillation time history generation primary model based on an XGBoost model and a physical information neural network model; training the constructed regional seismic oscillation time history generation primary model by using the constructed training data set to obtain a regional seismic oscillation time history generation model; and completing the generation of the seismic oscillation time history of the target region by adopting the obtained region seismic oscillation time history generation model. The regional seismic oscillation time history generation model is constructed and trained based on the XGBoost model and the physical information neural network model, the generation of regional seismic oscillation time history data is realized, the reliability is higher, the accuracy is better, and the effect is better.

Description

Technical Field

[0001] This invention belongs to the field of seismology, specifically relating to a method and system for generating regional ground motion time histories. Background Technology

[0002] Research on earthquakes helps people understand the earthquake process, thereby greatly reducing the disasters caused by earthquakes. Therefore, earthquake research is of great significance.

[0003] In earthquake research, accurately and reliably generating ground motions for specific scenarios has always been fundamental to earthquake hazard analysis, risk assessment, seismic structural design, and disaster prevention. Currently, traditional regional ground motion time history generation schemes mainly include physics-based deterministic simulation schemes, stochastic schemes, hybrid schemes, and empirical ground motion prediction equation schemes. Physics-based deterministic simulation schemes generally achieve mechanical fidelity through three-dimensional wave propagation modeling. However, due to the deterministic solution paradigm, unresolved source physics characterization, and the need to address small-scale crustal heterogeneity, these schemes require massive computational processes, are extremely time-consuming and labor-intensive, and their performance is relatively poor in integrating uncertainty analysis and high-frequency (>1 Hz) waveform synthesis. Stochastic schemes assume ground motions as Gaussian white noise and, based on stochastic vibration theory, iteratively adjust the matching degree between the simulated response spectrum and the target response spectrum to generate a synthetic ground motion record that conforms to the characteristics of the target spectrum. While these schemes have good computational efficiency, they require stochastic phase assumptions and sacrifice physical coherence, limiting their ability to accurately simulate long-period motions. Hybrid schemes typically combine deterministic low-frequency (<1 Hz) simulations with stochastic high-frequency (≥1 Hz) simulations to achieve full waveform synthesis; however, these schemes rely on source model calibration, making them impractical in real-world applications. Furthermore, this process requires extensive parameterization, resulting in high complexity and poor scalability. Empirical ground motion prediction equation schemes allow researchers to select seismic motions from databases (e.g., PEER NGA-West2) whose spectral characteristics match those near the structure's natural frequencies; however, this scaling introduces waveform distortion, disrupting the characteristics of unstable ground motions. Moreover, these schemes introduce mismatches in energy temporal distributions or other characteristics. Additionally, when structures exhibit nonlinear behavior, the time-history analysis of scaled motions used in these schemes may deviate from the actual structural response because spectral matching is limited to the vicinity of the natural frequencies, significantly impacting the accuracy and reliability of the scheme. Summary of the Invention

[0004] One of the objectives of this invention is to provide a method for generating regional ground motion time histories that is highly reliable, accurate, and effective.

[0005] The second objective of this invention is to provide a system for implementing the method for generating time histories of regional ground motions.

[0006] The method for generating regional ground motion time histories provided by this invention includes the following steps:

[0007] S1. Obtain existing physical simulation data of seismic motion;

[0008] S2. Preprocess the data obtained in step S1 to construct a training dataset;

[0009] S3. Based on the XGBoost model and the physical information neural network model, a primary model for generating regional ground motion time histories is constructed; wherein, the XGBoost model is used to predict the Fourier amplitude spectrum, pseudospectral acceleration and Arias intensity, and the physical information neural network model is used to generate seismic wave data;

[0010] S4. Using the training dataset constructed in step S2, train the primary model for generating regional ground motion time histories constructed in step S3 to obtain the regional ground motion time histories generation model.

[0011] S5. Using the regional ground motion time history generation model obtained in step S4, the ground motion time history of the target area is generated.

[0012] Step S1 specifically includes the following steps:

[0013] Seismic motion physical simulation data information was obtained from the CyberShake seismic motion physical simulation dataset;

[0014] The acquired data includes source parameters, path parameters, and site parameters;

[0015] Among them, the source parameters include magnitude. Fault width W, fault length L, dip angle Dip, slip angle Rake, focal depth and the depth of the fracture top ;

[0016] Path parameters include rupture distance Joyner-Boore distance Epicenter distance Fault normal coordinates Parallel coordinates of the fault ;

[0017] Site parameters include the time-averaged shear wave velocity over the upper 30 meters. , depth up to 1.0 km / s shear wave velocity level and depth up to 2.5 km / s shear wave velocity level .

[0018] Step S3 includes the following steps:

[0019] A primary model for generating regional ground motion time histories is constructed based on the XGBoost model and the physical information neural network model.

[0020] Weak learners are iteratively constructed based on the XGBoost model to correct the residuals generated in previous iterations; the goal of each new weak learner is to optimize the prediction along the steepest descent direction of the loss function.

[0021] A time series generation module is constructed based on an information neural network model. The time series generation module receives the output of the weak learner and, after nonlinear mapping through several hidden layers, outputs the distribution sampling of the ground motion acceleration time series at the output layer to generate seismic wave data.

[0022] The method for iteratively constructing a weak learner based on the XGBoost model includes the following steps:

[0023] The update rule for the predicted value in the t-th iteration is expressed as follows:

[0024] In the formula Let be the predicted value for the t-th iteration; The learning rate is used to control the step size. Let be the feature vector of the i-th sample; Let be the weak learner trained in the t-th iteration, and n is the number of training samples; It is a differentiable loss function; Let be the truth label of the i-th sample; This is the predicted value of the weak model for the i-th sample in this round; Let be the feature vector of the i-th sample;

[0025] The following formula is used as the objective function:

[0026] In the formula Let the loss function be the first-order gradient of the current prediction; Let the loss function be the second-order gradient of the current prediction; It is a regularization term, and , The complexity cost of adding a new leaf, where T is the number of leaves in the current tree. The L2 regularization coefficient for the leaf weights is... A vector of leaf weights. The square of the L2 norm;

[0027] weight of leaf node j By taking the derivative and setting it to 0, the optimal leaf weights can be obtained. for ,in Let be the set of indices of all samples in the j-th leaf node;

[0028] Substituting the optimal leaf weights into the objective function, the difference between the objective function before and after splitting is calculated, and the splitting gain G is obtained. ,in For the set of instances of the left child node, I is the set of instances of the right child node, and I is the set of instances of the parent node;

[0029] By maximizing the splitting gain G, the optimal splitting features and splitting points are identified;

[0030] The eigenvalues ​​are discretized into K bins using the following binning function to reduce the number of split points that need to be traversed:

[0031] In the formula It is the bin number of sample i on feature j. It is the boundary of the quantile point;

[0032] If missing values ​​exist, the split gain is first calculated for the non-missing samples. Then, the missing values ​​are directed to the left and right child nodes respectively, and the gains of both are calculated. Finally, the direction with the largest gain is selected as the "default direction".

[0033] The following formula is used as the loss function:

[0034] In the formula y represents the loss function value; y represents the observed value. This is a predicted value; The target quantile;

[0035] The following formula is used as the covariance matrix learning formula:

[0036] In the formula Regularization for correlation between outputs; The tracking operator is used to calculate the sum of the diagonal elements in a matrix; This is a multi-task weight matrix; Let be the covariance matrix between targets.

[0037] The time series generation module based on the information neural network model includes the following steps:

[0038] The constructed time series generation module is used to generate Fourier amplitude spectra, pseudospectral accelerations, and Arias intensities;

[0039] The Fourier amplitude spectrum is calculated as follows:

[0040] In the formula The frequency domain complex value spectrum; The Fourier amplitude spectrum, that is The modulus; N is the total number of sampling points of the signal; For a discrete-time signal; k is the frequency index; n is the time index; and k corresponds to the physical frequency. for , The sampling frequency;

[0041] The formula for calculating pseudospectral acceleration is:

[0042] In the formula Pseudospectral acceleration; The natural period of the structure; The damping ratio; This represents the time-domain acceleration response at the m-th sampling point; Frequency domain acceleration response; The acceleration transfer function of a single-degree-of-freedom system at discrete frequency points The value at; Let be the natural angular frequency of the target, and ;

[0043] The formula for calculating the intensity of Arias is:

[0044] In the formula Let be the cumulative Arias intensity at step p, and when hour The total Arias intensity is given by g; g is the acceleration due to gravity. It is a time-domain discrete signal; For time step.

[0045] The training described in step S4 includes the following steps:

[0046] Visualize the iterative waveforms to more easily observe the model's behavior patterns and performance characteristics;

[0047] For the generated Fourier amplitude spectrum, pseudospectral acceleration, and Arias intensity, a loss function is constructed, and the loss weights corresponding to the Fourier amplitude spectrum, pseudospectral acceleration, and Arias intensity are dynamically adjusted to optimize the training effect;

[0048] The AdamW optimizer was used to train the model.

[0049] The training specifically includes the following steps:

[0050] The following formula is used as the loss function:

[0051] In the formula, LOSS is the value of the loss function; For Fourier amplitude spectrum loss; Weights for Fourier amplitude spectrum loss; For pseudospectral acceleration loss; Weights for pseudospectral acceleration loss; For Arias intensity loss; The Arias intensity loss weights; The calculation formula is , This represents the total number of samples. For the true value of the i-th sample, This is the predicted value for the i-th sample;

[0052] At the start of training, the weights were set to satisfy... During training, when When the value is less than the set threshold, reduce... The value increases simultaneously and The value;

[0053] When training a model using the AdamW optimizer, the update rules include:

[0054] The gradient momentum is estimated using the following formula. and gradient square momentum :

[0055] In the formula The first exponential decay rate; Let be the gradient at step t; This is the second exponential decay rate;

[0056] The following formula is used for deviation correction:

[0057] In the formula This is the first-order moment estimate after bias correction; This is the second-order moment estimate after bias correction; The decay rate of the first moment; The decay rate of the second moment;

[0058] The following formula is used for parameter updates with decoupled weight decay:

[0059] In the formula These are the model parameters after optimization in step t; This is a set constant used for numerical stability; This is the weight decay coefficient.

[0060] This invention also provides a system for implementing the regional seismic motion time history generation method, comprising a data acquisition module, a data processing module, a model building module, a model training module, and a data generation module; the data acquisition module, data processing module, model building module, model training module, and data generation module are connected in series; the data acquisition module is used to acquire existing seismic motion physical simulation data and upload the data to the data processing module; the data processing module is used to preprocess the acquired data based on the received data to construct a training dataset and upload the data to the model building module; the model building module is used to, based on the received data, perform a simulation based on the XGBoost model and physical simulation. An information neural network model is used to construct a primary model for generating regional ground motion time histories and upload the data to the model training module. Specifically, the XGBoost model is used to predict the Fourier amplitude spectrum, pseudospectral acceleration, and Arias intensity, while the physical information neural network model is used to generate seismic wave data. The model training module trains the primary model using the received data and a constructed training dataset to obtain a regional ground motion time history generation model, and uploads the data to the data generation module. The data generation module then uses the obtained regional ground motion time history generation model to generate the ground motion time histories for the target region based on the received data.

[0061] The regional ground motion time history generation method and system provided by this invention constructs and trains a regional ground motion time history generation model based on the XGBoost model and the physical information neural network model. This not only realizes the generation of regional ground motion time history data, but also has higher reliability, better accuracy, and better results. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0063] Figure 2 This is a schematic diagram illustrating the data volume distribution of the database under different parameters in an embodiment of the method of the present invention; wherein, Figure 2 (a) is a schematic diagram showing the data distribution of source parameters and site parameters. Figure 2 (b) is a schematic diagram of the data distribution of path parameters and site parameters.

[0064] Figure 3 This is a schematic diagram of the representative waveform evolution sequence of randomly sampled acceleration records in an embodiment of the method of the present invention.

[0065] Figure 4 This is a schematic diagram illustrating the synchronous evolution of multiple index features of acceleration records in an embodiment of the method of the present invention; wherein, Figure 4 (a) is a schematic diagram of the target pseudospectral acceleration; Figure 4 (b) is a schematic diagram of the target Arias intensity; Figure 4 (c) is a schematic diagram of the target Fourier amplitude spectrum; Figure 4 (d) is a diagram illustrating the weighted total loss; Figure 4 (e) is a schematic diagram of the loss of each target index.

[0066] Figure 5 This is a schematic diagram of the FAS (Flat Earthquake Assay) of ground motion including basin effects under different moment magnitudes and fault distances, according to an embodiment of the method of the present invention; wherein, Figure 5 (a) is R rup =20km, M w A diagram showing the value at 7.5. Figure 5 (b) is R rup =20km, M w A diagram showing the value at 7.95. Figure 5 (c) is R rup =20km, M w A diagram showing the value at 8.4. Figure 5 (d) is R rup =40km, M w A diagram showing the value at 7.5. Figure 5 (e) is R rup =40km, M w A diagram showing the value at 7.95. Figure 5 (f) is R rup =40km, M w A diagram showing the value at 8.4. Figure 5 (g) is R rup =60km, M w A diagram showing the value at 7.5. Figure 5 (h) is Rrup =60km, M w A diagram showing the value at 7.95. Figure 5 (i) is R rup =60km, M w A diagram showing the value at 8.4. Figure 5 (j) is R rup =80km, M w A diagram showing the value at 7.5. Figure 5 (k) is R rup =80km, M w A diagram showing the value at 7.95. Figure 5 (l) is R rup =80km, M w A diagram showing the value at 8.4.

[0067] Figure 6 This is a schematic diagram of PSA (Plasma Seismic Amplitude) of ground motion including basin effects under different moment magnitudes and fault distances, according to an embodiment of the method of the present invention; wherein, Figure 6 (a) is R rup =20km, M w A diagram showing the value at 7.5. Figure 6 (b) is R rup =20km, M w A diagram showing the value at 7.95. Figure 6 (c) is R rup =20km, M w A diagram showing the value at 8.4. Figure 6 (d) is R rup =40km, M w A diagram showing the value at 7.5. Figure 6 (e) is R rup =40km, M w A diagram showing the value at 7.95. Figure 6 (f) is R rup =40km, M w A diagram showing the value at 8.4. Figure 6 (g) is R rup =60km, M w A diagram showing the value at 7.5. Figure 6 (h) is R rup =60km, M w A diagram showing the value at 7.95. Figure 6 (i) is R rup =60km, M w A diagram showing the value at 8.4. Figure 6 (j) is R rup =80km, M w A diagram showing the value at 7.5. Figure 6(k) is R rup =80km, M w A diagram showing the value at 7.95. Figure 6 (l) is R rup =80km, M w A diagram showing the value at 8.4.

[0068] Figure 7 M is an embodiment of the method of the present invention. w A schematic diagram of the probabilistic acceleration at 7.45; where, Figure 7 (a1) is R rup A schematic diagram of acceleration observed at 20km. Figure 7 (a2) is R rup A schematic diagram of acceleration observed at 40km / h. Figure 7 (a3) is R rup A schematic diagram of acceleration observed at 60km / h. Figure 7 (b1) is R rup The first schematic diagram of acceleration generated in the 5%-95% confidence interval at 20km. Figure 7 (b2) is R rup First schematic diagram of acceleration generated in the 5%-95% confidence interval at 40km. Figure 7 (b3) is R rup First schematic diagram of acceleration generated in the 5%-95% confidence interval at 60km. Figure 7 (b4) is R rup The second schematic diagram of acceleration generated in the 5%-95% confidence interval at 20km. Figure 7 (b5) is R rup The second schematic diagram of acceleration generated in the 5%-95% confidence interval at 40km. Figure 7 (b6) is R rup The second schematic diagram of acceleration generated in the 5%-95% confidence interval at 60km. Figure 7 (b7) is R rup The third schematic diagram of acceleration generated in the 5%-95% confidence interval at 20km. Figure 7 (b8) is R rup The third schematic diagram of acceleration generated in the 5%-95% confidence interval at 40km. Figure 7 (b9) is R rup The third schematic diagram of acceleration generated in the 5%-95% confidence interval at 60km. Figure 7 (b10) is R rup The fourth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 20km. Figure 7 (b11) is R rup The fourth schematic diagram shows the acceleration generated at 40km / h within the 5%-95% confidence interval. Figure 7 (b12) is R rup The fourth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 60km. Figure 7 (b13) is R rup The fifth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 20km. Figure 7 (b14) is R rup The fifth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 40km. Figure 7 (b15) is R rup The fifth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 60km. Figure 7 (b16) is R rup The sixth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 20km. Figure 7 (b17) is R rup The sixth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 40km. Figure 7 (b18) is R rup The sixth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 60km. Figure 7 (b19) is R rup The seventh diagram illustrates the acceleration generated in the 5%-95% confidence interval at 20km. Figure 7 (b20) is R rup The seventh diagram illustrates the acceleration generated within the 5%-95% confidence interval at 40km. Figure 7 (b21) is R rup The seventh diagram illustrates the acceleration generated in the 5%-95% confidence interval at 60km. Figure 7 (c1) is R rup The pseudo-acceleration spectrum and Arias intensity diagram generated at 20km. Figure 7 (c2) is R rup The pseudo-acceleration spectrum and Arias intensity diagram generated at 40km. Figure 7 (c3) is R rup The pseudo-acceleration spectrum and Arias intensity diagram generated at 60km. Figure 7 (c4) is R rup A schematic diagram of the Fourier amplitude spectrum of the acceleration diagram generated at 20km. Figure 7 (c5) is R rup A schematic diagram of the Fourier amplitude spectrum of the acceleration diagram generated at 40km. Figure 7 (c6) is R rup A schematic diagram of the Fourier amplitude spectrum for generating an acceleration diagram at 60km.

[0069] Figure 8 M is an embodiment of the method of the present invention. w A schematic diagram of the probabilistic acceleration at 7.45; where, Figure 8 (a1) is R rup A schematic diagram of acceleration observed at 20km. Figure 8 (a2) is R rup A schematic diagram of acceleration observed at 40km / h. Figure 8 (a3) is R rup A schematic diagram of acceleration observed at 60km / h. Figure 8 (b1) is R rup The first schematic diagram of acceleration generated in the 5%-95% confidence interval at 20km. Figure 8 (b2) is R rup First schematic diagram of acceleration generated in the 5%-95% confidence interval at 40km. Figure 8 (b3) is R rup First schematic diagram of acceleration generated in the 5%-95% confidence interval at 60km. Figure 8 (b4) is R rup The second schematic diagram of acceleration generated in the 5%-95% confidence interval at 20km. Figure 8 (b5) is R rup The second schematic diagram of acceleration generated in the 5%-95% confidence interval at 40km. Figure 8 (b6) is R rup The second schematic diagram of acceleration generated in the 5%-95% confidence interval at 60km. Figure 8 (b7) is R rup The third schematic diagram of acceleration generated in the 5%-95% confidence interval at 20km. Figure 8 (b8) is R rup The third schematic diagram of acceleration generated in the 5%-95% confidence interval at 40km. Figure 8 (b9) is R rup The third schematic diagram of acceleration generated in the 5%-95% confidence interval at 60km. Figure 8 (b10) is R rup The fourth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 20km. Figure 8 (b11) is R rup The fourth schematic diagram shows the acceleration generated at 40km / h within the 5%-95% confidence interval. Figure 8 (b12) is R rup The fourth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 60km. Figure 8 (b13) is R rup The fifth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 20km. Figure 8(b14) is R rup The fifth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 40km. Figure 8 (b15) is R rup The fifth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 60km. Figure 8 (b16) is R rup The sixth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 20km. Figure 8 (b17) is R rup The sixth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 40km. Figure 8 (b18) is R rup The sixth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 60km. Figure 8 (b19) is R rup The seventh diagram illustrates the acceleration generated in the 5%-95% confidence interval at 20km. Figure 8 (b20) is R rup The seventh diagram illustrates the acceleration generated within the 5%-95% confidence interval at 40km. Figure 8 (b21) is R rup The seventh diagram illustrates the acceleration generated in the 5%-95% confidence interval at 60km. Figure 8 (c1) is R rup The pseudo-acceleration spectrum and Arias intensity diagram generated at 20km. Figure 8 (c2) is R rup The pseudo-acceleration spectrum and Arias intensity diagram generated at 40km. Figure 8 (c3) is R rup The pseudo-acceleration spectrum and Arias intensity diagram generated at 60km. Figure 8 (c4) is R rup A schematic diagram of the Fourier amplitude spectrum of the acceleration diagram generated at 20km. Figure 8 (c5) is R rup A schematic diagram of the Fourier amplitude spectrum of the acceleration diagram generated at 40km. Figure 8 (c6) is R rup A schematic diagram of the Fourier amplitude spectrum for generating an acceleration diagram at 60km.

[0070] Figure 9 M is an embodiment of the method of the present invention. w A schematic diagram of the probabilistic acceleration at 7.45; where, Figure 9 (a1) is R rup A schematic diagram of acceleration observed at 20km. Figure 9 (a2) is R rupA schematic diagram of acceleration observed at 40km / h. Figure 9 (a3) is R rup A schematic diagram of acceleration observed at 60km / h. Figure 9 (b1) is R rup The first schematic diagram of acceleration generated in the 5%-95% confidence interval at 20km. Figure 9 (b2) is R rup First schematic diagram of acceleration generated in the 5%-95% confidence interval at 40km. Figure 9 (b3) is R rup First schematic diagram of acceleration generated in the 5%-95% confidence interval at 60km. Figure 9 (b4) is R rup The second schematic diagram of acceleration generated in the 5%-95% confidence interval at 20km. Figure 9 (b5) is R rup The second schematic diagram of acceleration generated in the 5%-95% confidence interval at 40km. Figure 9 (b6) is R rup The second schematic diagram of acceleration generated in the 5%-95% confidence interval at 60km. Figure 9 (b7) is R rup The third schematic diagram of acceleration generated in the 5%-95% confidence interval at 20km. Figure 9 (b8) is R rup The third schematic diagram of acceleration generated in the 5%-95% confidence interval at 40km. Figure 9 (b9) is R rup The third schematic diagram of acceleration generated in the 5%-95% confidence interval at 60km. Figure 9 (b10) is R rup The fourth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 20km. Figure 9 (b11) is R rup The fourth schematic diagram shows the acceleration generated at 40km / h within the 5%-95% confidence interval. Figure 9 (b12) is R rup The fourth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 60km. Figure 9 (b13) is R rup The fifth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 20km. Figure 9 (b14) is R rup The fifth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 40km. Figure 9 (b15) is R rup The fifth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 60km. Figure 9 (b16) is Rrup The sixth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 20km. Figure 9 (b17) is R rup The sixth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 40km. Figure 9 (b18) is R rup The sixth schematic diagram shows the acceleration generated in the 5%-95% confidence interval at 60km. Figure 9 (b19) is R rup The seventh diagram illustrates the acceleration generated in the 5%-95% confidence interval at 20km. Figure 9 (b20) is R rup The seventh diagram illustrates the acceleration generated within the 5%-95% confidence interval at 40km. Figure 9 (b21) is R rup The seventh diagram illustrates the acceleration generated in the 5%-95% confidence interval at 60km. Figure 9 (c1) is R rup The pseudo-acceleration spectrum and Arias intensity diagram generated at 20km. Figure 9 (c2) is R rup The pseudo-acceleration spectrum and Arias intensity diagram generated at 40km. Figure 9 (c3) is R rup The pseudo-acceleration spectrum and Arias intensity diagram generated at 60km. Figure 9 (c4) is R rup A schematic diagram of the Fourier amplitude spectrum of the acceleration diagram generated at 20km. Figure 9 (c5) is R rup A schematic diagram of the Fourier amplitude spectrum of the acceleration diagram generated at 40km. Figure 9 (c6) is R rup A schematic diagram of the Fourier amplitude spectrum for generating an acceleration diagram at 60km.

[0071] Figure 10 This document provides key dimensions, key component cross-sections, and a three-dimensional schematic diagram of the RC frame in an embodiment of the method of the present invention; wherein, Figure 10 (a) is an elevation view showing the overall dimensions and the height of each floor. Figure 10 (b) Plan view of beams and columns, Figure 10 (c) shows the detailed cross-sectional views of the beams and columns. Figure 10 (d) is a schematic diagram of the three-dimensional finite element model for dynamic analysis.

[0072] Figure 11 This is a schematic diagram illustrating the verification of probabilistic acceleration diagrams through structural displacement response in an embodiment of the method of the present invention; wherein, Figure 11 (a) is a comparison chart of peak displacements for each floor. Figure 11 (b) is a comparison of the peak values ​​of inter-story drift angles.

[0073] Figure 12 These are acceleration diagrams and schematic diagrams of base shear force distribution of the structure under actual ground motion, generated as embodiments of the method of the present invention; wherein, Figure 12 (a) is a schematic diagram of the acceleration-base shear force distribution of column 1. Figure 12 (b) is a schematic diagram of the acceleration-base shear force distribution of column 2. Figure 12 (c) is a schematic diagram of the acceleration-base shear force distribution of column 3. Figure 12 (d) is a schematic diagram of the acceleration-base shear force distribution of column 4. Figure 12 (e) is a schematic diagram of the acceleration-base shear force distribution of column 5. Figure 12 (f) is a schematic diagram of the acceleration-base shear force distribution of column 6. Figure 12 (g) is a schematic diagram of the acceleration-base shear force distribution of column 7. Figure 12 (h) is a schematic diagram of the acceleration-base shear force distribution of column 8. Figure 12 (i) is a schematic diagram of the acceleration-base shear force distribution of column 9.

[0074] Figure 13 This is a schematic diagram of the multi-component internal force response in a representative column under actual ground motion, generated for an embodiment of the method of the present invention; wherein, Figure 13 (a) is a schematic diagram of axial force P. Figure 13 (b) is the bending moment M z A schematic diagram, Figure 13 (c) represents the bending moment M y A schematic diagram, Figure 13 (d) is a schematic diagram of torque T.

[0075] Figure 14 This is a schematic diagram of the functional modules of the system of the present invention. Detailed Implementation

[0076] like Figure 1 The diagram shown is a flowchart of the method of the present invention: The regional ground motion time history generation method disclosed in this invention includes the following steps:

[0077] S1. Obtain existing seismic motion physical simulation data; specifically including the following steps:

[0078] Ground motion physical simulation data was obtained from the CyberShake ground motion physical simulation dataset; specifically, a subset of the CyberShake ground motion physical simulation version 15.12 dataset; the main velocity models used in this model are CVM-S4.26.M01 and Northridge 1D; the former was obtained through full 3D waveform tomography, with borehole information used as a supplement to near-surface velocities, and shear wave velocities ( The minimum value is 500 m / s; the latter represents a hierarchical model constructed by spatial averaging of CVM-S4.26.M01; M01 has a 100 m grid resolution and a smooth interface; a hybrid approach is used to combine low-frequency (1 Hz) CyberShake 15.4 seismograms with high-frequency components of the Graves-Pitarka broadband platform to synthesize broadband ground motion.

[0079] The acquired data includes source parameters, path parameters, and site parameters;

[0080] Among them, the source parameters include magnitude. Fault width W, fault length L, dip angle Dip, slip angle Rake, focal depth and the depth of the fracture top ;

[0081] Path parameters include rupture distance Joyner-Boore distance Epicenter distance Fault normal coordinates Parallel coordinates of the fault When the dip angle and slip angle indicate a pure strike-slip mechanism, the vertical coordinates of the fault ( ) and fault parallel coordinates ( It exhibits geometric symmetry, and due to the lack of directional meaning, it needs to be represented by absolute value;

[0082] Site parameters include the time-averaged shear wave velocity over the upper 30 meters. , depth up to 1.0 km / s shear wave velocity level and depth up to 2.5 km / s shear wave velocity level .

[0083] S2. Preprocess the data obtained in step S1 to construct a training dataset.

[0084] S3. Based on the XGBoost model and the physical information neural network model, a primary model for generating regional ground motion time histories is constructed; wherein, the XGBoost model is used to predict the Fourier amplitude spectrum, pseudospectral acceleration, and Arias intensity, and the physical information neural network model is used to generate seismic wave data; including the following steps:

[0085] A primary model for generating regional ground motion time histories is constructed based on the XGBoost model and the physical information neural network model.

[0086] Weak learners are iteratively constructed based on the XGBoost model to correct the residuals generated in previous iterations; the goal of each new weak learner is to optimize the prediction along the steepest descent direction of the loss function.

[0087] A time series generation module is constructed based on an information neural network model. The time series generation module receives the output of the weak learner and, after nonlinear mapping through several hidden layers, outputs the distribution sampling of the ground motion acceleration time series at the output layer to generate seismic wave data.

[0088] In specific implementation, the iterative construction of weak learners based on the XGBoost model includes the following steps:

[0089] The update rule for the predicted value in the t-th iteration is expressed as follows:

[0090] In the formula Let be the predicted value for the t-th iteration; The learning rate is used to control the step size. Let be the feature vector of the i-th sample; Let be the weak learner trained in the t-th iteration, and n is the number of training samples; It is a differentiable loss function; Let be the truth label of the i-th sample; This is the predicted value of the weak model for the i-th sample in this round; Let be the feature vector of the i-th sample;

[0091] The following formula is used as the objective function:

[0092] In the formula Let the loss function be the first-order gradient of the current prediction; Let the loss function be the second-order gradient of the current prediction; It is a regularization term, and , The complexity cost of adding a new leaf, where T is the number of leaves in the current tree. The L2 regularization coefficient for the leaf weights is... A vector of leaf weights. The square of the L2 norm;

[0093] weight of leaf node j By taking the derivative and setting it to 0, the optimal leaf weights can be obtained. for ,in Let be the set of indices of all samples in the j-th leaf node;

[0094] Substituting the optimal leaf weights into the objective function, the difference between the objective function before and after splitting is calculated, and the splitting gain G is obtained. ,in For the set of instances of the left child node, I is the set of instances of the right child node, and I is the set of instances of the parent node;

[0095] By maximizing the split gain G to identify the optimal split features and split points, this method constitutes the basic mechanism for decision tree construction.

[0096] The eigenvalues ​​are discretized into K bins using the following binning function to greatly reduce the number of split points that need to be traversed:

[0097] In the formula b ij τ is the bin number of sample i on feature j, and τ is the quantile boundary;

[0098] If missing values ​​exist, first calculate the split gain for non-missing samples, then try to guide the missing values ​​to the left and right child nodes respectively and calculate the gains of both, and finally select the direction with the largest gain as the "default direction".

[0099] The following formula is used as the quantile loss function:

[0100] In the formula y represents the loss function value; y represents the observed value. This is a predicted value; The target quantile;

[0101] The following formula is used as the covariance matrix learning formula:

[0102] In the formula Regularization for correlation between outputs; The tracking operator is used to calculate the sum of the diagonal elements in a matrix; This is a multi-task weight matrix; Let be the covariance matrix between targets.

[0103] In the above scheme, a second-order Taylor expansion is introduced into the objective function to improve optimization accuracy by more accurately capturing the curvature information of the loss function; explicit integration of regularization terms achieves a dynamic balance between model complexity and generalization ability; sparse-aware segmentation algorithm dynamically learns the optimal allocation direction of missing values ​​to enhance robustness to incomplete data;

[0104] Meanwhile, the histogram-based binning strategy accelerates feature splitting calculation; native quantile regression support is achieved through quantile loss function optimization, thereby enabling rigorous prediction interval estimation; the vector leaf tree model explicitly models the dependencies between output variables through covariance matrix learning, thereby enhancing the synergistic effect of multi-task prediction.

[0105] These technical features are highly compatible with the constraint prediction requirements of this invention: the multi-output dependency modeling capability enables synchronous optimization of logical consistency among constraints; the histogram acceleration algorithm significantly improves the model optimization efficiency; and the quantile regression function provides a key probabilistic intensity measure estimate for probabilistic seismic hazard analysis.

[0106] In specific implementation, the time series generation module based on the information neural network model includes the following steps:

[0107] The mapping process of hidden layers in a neural network can be described as first performing a linear transformation, followed by a nonlinear transformation, the latter being achieved through activation functions acting on neurons; in this invention, all nonlinear activation functions are Gaussian Error Linear Unit (GELU) functions, which show significant performance advantages compared to other activation functions such as ReLU, ELU, and LeakyReLU;

[0108] The constructed time series generation module is used to generate Fourier amplitude spectra, pseudospectral accelerations, and Arias intensities;

[0109] The Fourier amplitude spectrum is calculated as follows:

[0110] In the formula The frequency domain complex value spectrum; The Fourier amplitude spectrum, that is The modulus; N is the total number of sampling points of the signal; For a discrete-time signal; k is the frequency index; n is the time index; and k corresponds to the physical frequency. for , The sampling frequency;

[0111] The formula for calculating pseudospectral acceleration is:

[0112] In the formula Pseudospectral acceleration; The natural period of the structure; The damping ratio; This represents the time-domain acceleration response at the m-th sampling point; Frequency domain acceleration response; The acceleration transfer function of a single-degree-of-freedom system at discrete frequency points The value at; Let be the natural angular frequency of the target, and ;

[0113] The formula for calculating the intensity of Arias is:

[0114] In the formula Let be the cumulative Arias intensity at step p, and when hour The total Arias intensity is given by g; g is the acceleration due to gravity. It is a time-domain discrete signal; For time step.

[0115] The time histories generated by the above scheme achieve a balance between the physical nature of earthquake motion and its engineering applicability, ensuring physical rationality and the reliability of structural analysis.

[0116] S4. Using the training dataset constructed in step S2, train the primary model for generating regional ground motion time histories constructed in step S3 to obtain the regional ground motion time histories generation model.

[0117] The training process includes the following steps:

[0118] Visualizing the iterative waveforms facilitates observation of the model's behavior and performance characteristics: In the early training phase, the generated signal appears as unstructured white noise lacking typical seismic features (such as time-varying energy distribution, envelope morphology, or high-frequency oscillation details); the waveform is chaotic, with no obvious energy accumulation / attenuation trend, uniform amplitude distribution, and inability to distinguish between foreshocks, mainshocks, and aftershocks; in the mid-training phase, as the number of iterations increases, the signal begins to form a preliminary envelope outline and exhibits a non-uniform energy distribution; the energy accumulation trend becomes identifiable, but the envelope morphology remains incomplete and high-frequency oscillations are still present. The signal becomes sparse; a weak peak acceleration prototype begins to appear, but the peak position is unstable; local periodic oscillations appear, but the overall frequency distribution is still relatively broad and lacks true seismic spectrum characteristics; in the later training stage, the generated signal significantly converges to the physical seismic characteristics: the waveform forms a complete envelope structure, and the energy follows a three-stage evolution pattern: initial accumulation, PGA peak stage and subsequent decay, the envelope curve gradually becomes smoother and the symmetry increases; the high-frequency components become more refined, short-period oscillations appear and the matching degree with seismic spectrum characteristics improves; finally, the signal stability is enhanced, and the amplitude fluctuations gradually converge to a physically reasonable range.

[0119] For the generated Fourier amplitude spectrum (FAS), pseudospectral acceleration, and Arias intensity (PSA), a loss function is constructed, and the Fourier amplitude spectrum, pseudospectral acceleration, and Arias intensity are dynamically adjusted. The corresponding loss weights are assigned to optimize training performance; in the initial training phase, the adaptive system assigns... The highest weight is prioritized to optimize the time-domain energy distribution, establishing the overall waveform shape by controlling the energy accumulation / decrease mode; simultaneously, the PSA is assigned a slightly lower weight to ensure continuous attention to frequency-domain characteristics and maintain cross-cycle consistency with the target spectral amplitude; the FAS, as a supplementary constraint, refines high-spectral details under the basic frequency control provided by the PSA; once the time-domain mode stabilizes, the adaptive mechanism initiates weight reallocation according to the preset decay rate: systematically releasing part of the weight. Weights are assigned to PSA and FAS according to the instantaneous loss ratio; this dynamic rebalancing strategy achieves two key objectives: (1) maintaining sufficient weights during the transition phase. (1) Maintaining macroscopic waveform accuracy; (2) Gradually optimizing spectral features by enhancing PSA / FAS contribution; Through dynamic weight adjustment throughout training, the system avoids single index dominating optimization, ensuring the coordinated convergence of time-domain energy dynamics, frequency-domain spectral features, and high-resolution waveform details; This hierarchical priority mechanism first emphasizes the global structure and then shifts to local feature optimization. It effectively avoids the risk of premature convergence to suboptimal solutions while maintaining the physical rationality of generating acceleration records.

[0120] The AdamW optimizer is used for model training: The AdamW optimizer extends the traditional Adam algorithm by decoupling weight decay from gradient updates. In customized computation graphs, adaptive balancing of multiple loss components can lead to significant differences in the gradient magnitudes of each loss term. The traditional Adam algorithm entangles weight decay with gradient updates, causing the regularization strength to be scaled by the adaptive learning rate. This coupling can destroy the regularization effect when loss coefficients change abruptly, as effective weight decay is affected by gradient noise. AdamW addresses this issue by directly applying weight decay after gradient updates, ensuring that the regularization effect is independent of gradient statistics. This stability is crucial for multi-objective optimization. When competing loss terms lead to ill-conditioned optimization terrain, by separating gradient adaptation and parameter shrinkage, AdamW can maintain robust regularization constraints and adaptively traverse complex loss topologies, ultimately achieving better convergence performance.

[0121] In practice, the following steps can be used for training:

[0122] The following formula is used as the loss function:

[0123] In the formula, LOSS is the value of the loss function; For Fourier amplitude spectrum loss; Weights for Fourier amplitude spectrum loss; For pseudospectral acceleration loss; Weights for pseudospectral acceleration loss; For Arias intensity loss; The Arias intensity loss weights; The calculation formula is , This represents the total number of samples. For the true value of the i-th sample, This is the predicted value for the i-th sample;

[0124] At the start of training, the weights were set to satisfy... During training, when When the value is less than the set threshold, reduce... The value increases simultaneously and The value;

[0125] When training a model using the AdamW optimizer, the update rules include:

[0126] The gradient momentum is estimated using the following formula. and gradient square momentum :

[0127] In the formula The first exponential decay rate; Let be the gradient at step t; This is the second exponential decay rate;

[0128] The following formula is used for deviation correction:

[0129] In the formula This is the first-order moment estimate after bias correction; This is the second-order moment estimate after bias correction; The decay rate of the first moment; The decay rate of the second moment;

[0130] The following formula is used for parameter updates with decoupled weight decay:

[0131] In the formula These are the model parameters after optimization in step t; This is a set constant used for numerical stability; This is the weight decay coefficient.

[0132] S5. Using the regional ground motion time history generation model obtained in step S4, the ground motion time history of the target area is generated.

[0133] The effects of the method of the present invention will be illustrated below with reference to an embodiment:

[0134] A subset of seismic motion simulation data generated using CyberShake was employed. Two key data sources—accelerometer file header metadata obtained from the Globus platform and project-specific information—were integrated through a MySQL database query system to obtain data including M... w R rup V s30 Z hyp Z1 and Z 2.5 Basic scenario parameters, etc. Z was extracted by parsing the event-specific rupture configuration file. tor Rake, Dip, W, L, and other fracture geometry parameters. Source-site geometry parameters (R epi R jb R x R yThe coordinates were calculated through a transformation between the WGS-84 geographic system and the Universal Transverse Mercator (UTM) grid, a process requiring detailed geometric analysis of the source-site relationship. After acquiring the acceleration records, an energy-based window function method was used to truncate them. All waveforms underwent initial energy analysis to determine their 1-99% energy concentration range. Records containing more than 3,500 sampling points within these ranges (representing 1.0% of the original dataset; for details on length distribution and rejection thresholds, see [link to relevant documentation]). Figure 2 The retained records were uniformly processed using a fixed window length of 3,500 sampling points centered on the energy range. When the window position exceeded the boundaries of the original record, it was adjusted to ensure the window was entirely within the usable data range. The truncated waveforms underwent baseline correction (third-order polynomial removal) and fourth-order Butterworth bandpass filtering (0.1–10 Hz) to obtain the final acceleration time histories. Key ground motion parameters (IMs) were then extracted from these processed records for analysis. The final, filtered database contained 157,905 validated acceleration time histories.

[0135] Figure 2 This shows the data volume distribution of the database in (a) source parameters and (b) path parameters (bottom left) and site parameters (top right). The red line represents the linear relationship between the features. Source parameters include moment magnitude (M), fault width (W) and length (L), dip angle (Dip), slip angle (Rake), and focal depth (Z). hyp ) and the depth of the fracture top surface (Z) tor The path parameters include the rupture distance (R). rup Joyner-Boore distance (R) jb ), epicentral distance (R) epi ), fault normal coordinates (R) x ) and fault parallel coordinates (R y Site parameters are determined by the average shear wave velocity (V) at a surface depth of 30 meters. s30 ) and burial depths (Z1 / Z) where shear wave velocities reach 1.0 / 2.5 km / s. 2.5 Composed of ) . As shown in the figure, the moment magnitude M in the dataset w The range of 7.4-8.5 is specifically designed to emphasize the dominant role of large earthquakes in structural failure mechanisms and earthquake risk assessment frameworks; furthermore, the model's superior performance on highly complex samples (i.e., large earthquake events) naturally demonstrates its ability to handle low-complexity scenarios (such as smaller magnitude events). Seismogenic depth Z tor Limited to within 0.41 kilometers, focal depth Z hyp Extending to 29 kilometers, this collectively confirms the classification of shallow-source crustal earthquakes. Rupture distance R rupThe constraints are limited to a 200 km radius, with dense near-field sampling (<30 km) being crucial for seismic structural analysis, while data density decreases progressively beyond 80 km. s30 The value range is 240-720 m / s, effectively characterizing site conditions from compacted soil to weathered bedrock. The Z1 depth extends to 1120 meters, reflecting significant differences in site conditions from exposed bedrock outcrops to deep sedimentary basins. Source parameters W and L, path parameter R... rup With R jb Site parameters Z1 and Z 2.5 There is a strong correlation between them, V s30 With Z1 / Z 2.5 Although the correlation was weak, it was still statistically significant.

[0136] Figure 3 This paper presents a representative waveform evolution sequence of a randomly sampled acceleration record. As shown in the figure, in the early training phase, the generated signal exhibits unstructured white noise, lacking typical seismic features (e.g., non-stationary features such as temporal energy variation, envelope shape, or high-frequency oscillation details). The waveform appears chaotic, lacking a clear energy accumulation / decrease trend, with a uniform amplitude distribution, making it impossible to distinguish between foreshocks, mainshocks, and aftershocks. In the middle training phase, with increasing iterations, the signal begins to form a preliminary envelope profile, with a non-uniform energy distribution. An energy accumulation trend can be observed, but the envelope shape remains incomplete, and high-frequency components are sparse. A weak peak acceleration prototype appears, but the peak position is unstable. Local periodic oscillations emerge, although the overall frequency distribution remains broad, lacking true seismic spectral features. In the later training phase, the generated signal significantly converges to physical seismic features. The waveform develops a complete envelope structure, with energy following a three-stage evolution pattern: initial accumulation, followed by a PGA peak phase, and then decay, with the envelope curve gradually becoming smoother and more symmetrical. Furthermore, the high-frequency components become more refined, and the alignment with seismic spectral features improves with the appearance of short-period oscillations. Finally, the signal exhibited greater stability, with amplitude fluctuations gradually converging to a physically reasonable range.

[0137] Figure 4 exhibit Figure 3 The synchronous evolution of multiple index features of acceleration records in the data. Figure 4 (a) shows that in the early iterations, there is a significant difference between the generated response spectrum and the target, which gradually converges to the target PSA as training progresses. Figure 4 (b) Display the generated acceleration record I a The energy accumulation curve is close to the target value, confirming the successful modeling of the seismic energy release pattern. Figure 4 (c) reveals that spectral features are learned stepwise through FAS evolution. Figure 4(d) shows that all loss components decrease monotonically with iteration, indicating continuous optimization. a The loss-weighted loss converges the fastest, followed by PSA, while FAS converges the slowest. The total weighted loss is close to zero, verifying global convergence. Figure 4 (e) Display I a Initially, the weights were highest for I, followed by PSA, while FAS remained the lowest. However, as training progressed, I... a The dynamic decrease in weights, while the increases in PSA and FAS weights, indicates a prioritization of time-domain and frequency-domain feature refinement in later stages. This behavior is due to a carefully designed adaptive loss coefficient adjuster with multi-objective optimization. Specifically, during the initial training phase, the adaptive system assigns I... a The highest weight is given to prioritize the temporal energy distribution, establishing the overall waveform shape by controlling the energy accumulation / decrease mode. Simultaneously, a slightly lower weight assigned to the PSA ensures continued attention to frequency domain characteristics, maintaining alignment with the target spectral amplitude across periods. The FAS component, as a supplementary constraint, refines high-spectral details based on the fundamental frequency control provided by the PSA. Once the temporal pattern stabilizes in the later stages, the adaptive mechanism begins to reallocate weights according to a predefined decay rate: it systematically releases a portion of the Ig... a Weights are assigned, and these resources are allocated proportionally to PSA and FAS based on their immediate loss ratios. This dynamic rebalancing strategy achieves two key objectives: (1) maintaining sufficient IT during the transition phase. a (1) Influence, maintain the accuracy of macroscopic waveforms; (2) Achieve gradual refinement of spectral features by enhancing the contribution of PSA / FAS. By dynamically adjusting the weights throughout the training process, the system prevents any single index from dominating the optimization, thereby ensuring the coordinated convergence of time-domain energy dynamics, frequency-domain spectral features, and high-resolution waveform details. This hierarchical priority mechanism—initially emphasizing global structure and then transitioning to local feature optimization—effectively reduces the risk of premature convergence to suboptimal solutions while maintaining the physical rationality of the generated acceleration records.

[0138] Figure 5 and Figure 6 This demonstrates the physical trends exhibited by the proposed model after training. Specifically, Figure 5 The FAS (Flat Earthquake Assay) of ground motion including basin effects (quantified by parameter Z1) is shown at different moment magnitudes (M). w ) and fault distance (R rup The following situation was observed: (1) As M... w(1) With the increase of Z1, the FAS amplitude increases significantly, and large earthquakes release more low-frequency energy proportionally; (2) the spectral amplitude generally decreases with increasing distance, with the high-frequency components decreasing faster than the low-frequency components; (3) with the increase of Z1 value, the basin amplification effect is enhanced, and thick sedimentary basins show significant low-frequency amplification (long-period effect). These observations are consistent with established empirical knowledge about the changes in various parameters. Notably, the model also successfully captures the complex interaction effects between the source, path, and basin parameters. Figure 6 Showing different M w and R rup Under these conditions, a PSA including basin effects (Z1 parameter) was used. Key observations include: (1) Total seismic energy increases significantly with increasing magnitude, particularly enhancing the long-period (>1s) component. This is consistent with Brune-type source models, where large earthquakes exhibit lower corner frequencies and longer energy release periods, shifting the spectral peak to longer periods. Short-period amplification remains moderate or even stable due to the more dispersed distribution of high-frequency energy. (2) Increased distance leads to a decrease in the overall spectrum due to geometric diffusion and inelastic decay. High-frequency components decay faster, significantly reducing short-period values, while long-period components decay more slowly, producing a flatter far-field spectrum with a slight rightward shift in peak value. (3) Larger Z1 values ​​(deeper basins) enhance amplification from low-velocity sediments. Multiple basin reflections prolong wave duration, significantly increasing long-period spectral values ​​(>1s), exhibiting a “basin pulsation” effect. These single-parameter trends are consistent with empirical expectations, while the model also accurately captures complex interaction effects.

[0139] Figures 7-9 The above figure compares the actual acceleration records in the test set with those in R. rup The figure below shows probabilistic acceleration maps generated under the scenario of approximately [20, 40, 60] km, while the figure below shows a comparison of their corresponding IMs (only a portion of the generated acceleration maps are shown due to space limitations). The generated probabilistic acceleration maps were constrained to within ±1 standard deviation of the target IM distribution using 300 random samples. Observations show that the generated waveforms exhibit strong consistency with actual records in terms of peak ground acceleration (PGA) amplitude and main waveform patterns at different magnitudes and distances. The envelope of the generated waveforms effectively covers the rise and decay phases of the actual waveforms, indicating that the model can capture the temporal characteristics of actual ground motion. Notably, although the PGA is not explicitly constrained, the generated results still show strong consistency with the observed data. This indicates that the model can learn the implicit correlations between seismic parameters from the training data, demonstrating excellent generalization ability. This implicit correlation further suggests that the selected constraints adequately characterize key waveform features, especially engineering-related features such as seismic design parameters, thus eliminating the need for explicit PGA constraints.

[0140] To further verify the effectiveness of the generated probabilistic ground motion, Figure 8 Chinese R rup Synthetic seismic waves at a range of approximately 20 km were input into a 6-story reinforced concrete (RC) frame structure. The peak structural responses, including force and displacement responses, induced by generated and actual ground motions were compared. The frame model was implemented in OpenSeesPy. Figure 10 The key dimensions, key component sections, and 3D schematics of the RC frame are shown. Three material models were used for the column and beam elements: unconfined concrete (Concrete01) for the overlying concrete, Kent-Park confined concrete (Concrete01) for the core concrete, and steel reinforcement (Steel01) for the longitudinal reinforcement. The concrete confinement used the Kent-Scott-Park model. The steel reinforcement parameters include: yield strength F. y =400MPa, elastic modulus E=206,000MPa, strain hardening ratio b=0.005. Figure 11 (a) Peak displacements at different floors were compared. The results show that peak displacement increases significantly with floor height, indicating that high-rise buildings are more prone to larger seismic displacements. This trend was consistently observed in both recorded and generated ground motions. The envelope region represents the area generated by the wave in R... rup The induced response boundary (i.e., uncertainty range) under the ≈20km scenario is the result of mapping the waveform-level confidence interval to the structural response level. All responses induced by actual waves fall within this range, confirming that the uncertainty of generated waves can directly provide information for structural seismic risk assessment. The confidence interval widens with increasing floor height, revealing the increased sensitivity of high-rise structures to changes in ground motion. Figure 11 (b) The inter-story drift ratios between generated and recorded ground motions were compared. Larger inter-story drift ratios occurred in lower floors, with smaller values ​​in upper floors—a pattern consistently captured by both observed and generated waves. The red observed curves are densely covered by the gray generated curve set, indicating that the model can reproduce realistic structural responses.

[0141] Figure 12 The base shear force distribution of the structure under generated acceleration maps and actual ground motion was compared. The results show that the base shear force distribution induced by the generated seismic waves (colored shaded areas) completely covers all observations (red dots within the shaded areas), and most observation data points are closely aligned with the central trend of the generated results (blue curve). This indicates that the model can reproduce the realistic base shear force distribution. Figure 13The multi-component internal force responses—including axial force, bending moment, and torque—in a representative column were compared under generated and actual ground motions. The force responses induced by the generated waves consistently covered the actual values ​​across all components, confirming the model's effectiveness in reconstructing the key internal force distributions.

[0142] like Figure 14 The diagram shows the functional modules of the system of the present invention: The system for generating regional seismic time histories disclosed in this invention includes a data acquisition module, a data processing module, a model building module, a model training module, and a data generation module; the data acquisition module, data processing module, model building module, model training module, and data generation module are connected in series; the data acquisition module is used to acquire existing seismic physical simulation data and upload the data to the data processing module; the data processing module is used to preprocess the acquired data based on the received data to construct a training dataset and upload the data to the model building module; the model building module is used to, based on the received data, perform XGBo... The XGBoost model and the Physical Information Neural Network (PIN) model are used to construct a primary model for generating regional ground motion time histories, and the data information is uploaded to the model training module. Specifically, the XGBoost model is used to predict Fourier amplitude spectrum, pseudospectral acceleration, and Arias intensity, while the PIN model is used to generate seismic wave data. The model training module trains the primary model using the received data and a pre-constructed training dataset to obtain a regional ground motion time history generation model, and uploads the data to the data generation module. The data generation module then uses the obtained regional ground motion time history generation model to generate the ground motion time histories for the target region based on the received data.

Claims

1. A method for generating regional ground motion time histories, characterized in that... Includes the following steps: S1. Obtain existing physical simulation data of seismic motion; S2. Preprocess the data obtained in step S1 to construct a training dataset; S3. Based on the XGBoost model and the physical information neural network model, a primary model for generating regional ground motion time histories is constructed; wherein, the XGBoost model is used to predict the Fourier amplitude spectrum, pseudospectral acceleration and Arias intensity, and the physical information neural network model is used to generate seismic wave data; S4. Using the training dataset constructed in step S2, train the primary model for generating regional ground motion time histories constructed in step S3 to obtain the regional ground motion time histories generation model. S5. Using the regional ground motion time history generation model obtained in step S4, the ground motion time history of the target area is generated.

2. The method for generating regional ground motion time histories according to claim 1, characterized in that... Step S1 specifically includes the following steps: Seismic motion physical simulation data information was obtained from the CyberShake seismic motion physical simulation dataset; The acquired data includes source parameters, path parameters, and site parameters; Among them, the source parameters include magnitude. Fault width W, fault length L, dip angle Dip, slip angle Rake, focal depth and the depth of the fracture top ; Path parameters include rupture distance Joyner-Boore distance Epicenter distance Fault normal coordinates Parallel coordinates of the fault ; Site parameters include the time-averaged shear wave velocity over the upper 30 meters. , depth up to 1.0 km / s shear wave velocity level and depth up to 2.5 km / s shear wave velocity level .

3. The method for generating regional ground motion time histories according to claim 2, characterized in that... Step S3 includes the following steps: A primary model for generating regional ground motion time histories is constructed based on the XGBoost model and the physical information neural network model. Weak learners are iteratively constructed based on the XGBoost model to correct the residuals generated in previous iterations; the goal of each new weak learner is to optimize the prediction along the steepest descent direction of the loss function. A time series generation module is constructed based on an information neural network model. The time series generation module receives the output of the weak learner and, after nonlinear mapping through several hidden layers, outputs the distribution sampling of the ground motion acceleration time series at the output layer to generate seismic wave data.

4. The method for generating regional ground motion time histories according to claim 3, characterized in that... The method for iteratively constructing a weak learner based on the XGBoost model includes the following steps: The update rule for the predicted value in the t-th iteration is expressed as follows: In the formula Let be the predicted value for the t-th iteration; The learning rate is used to control the step size. Let be the feature vector of the i-th sample; Let be the weak learner trained in the t-th iteration, and n is the number of training samples; It is a differentiable loss function; Let be the truth label of the i-th sample; This is the predicted value of the weak model for the i-th sample in this round; Let be the feature vector of the i-th sample; The following formula is used as the objective function: In the formula Let the loss function be the first-order gradient of the current prediction; Let the loss function be the second-order gradient of the current prediction; It is a regularization term, and , The complexity cost of adding a new leaf, where T is the number of leaves in the current tree. The L2 regularization coefficient for the leaf weights is... A vector of leaf weights. The square of the L2 norm; weight of leaf node j By taking the derivative and setting it to 0, the optimal leaf weights can be obtained. for ,in Let be the set of indices of all samples in the j-th leaf node; Substituting the optimal leaf weights into the objective function, the difference between the objective function before and after splitting is calculated, and the splitting gain G is obtained. ,in For the set of instances of the left child node, I is the set of instances of the right child node, and I is the set of instances of the parent node; By maximizing the splitting gain G, the optimal splitting features and splitting points are identified; The eigenvalues ​​are discretized into K bins using the following binning function to reduce the number of split points that need to be traversed: In the formula It is the bin number of sample i on feature j. It is the boundary of the quantile point; If there are missing values, first calculate the split gain for the non-missing samples, then guide the missing values ​​to the left and right child nodes respectively and calculate the gains of both, and finally select the direction that maximizes the gain as the "default direction". The following formula is used as the loss function: In the formula y represents the loss function value; y represents the observed value. This is a predicted value; The target quantile; The following formula is used as the covariance matrix learning formula: In the formula Regularization for correlation between outputs; The tracking operator is used to calculate the sum of the diagonal elements in a matrix; This is a multi-task weight matrix; Let be the covariance matrix between targets.

5. The method for generating regional ground motion time histories according to claim 4, characterized in that... The time series generation module based on the information neural network model includes the following steps: The constructed time series generation module is used to generate Fourier amplitude spectra, pseudospectral accelerations, and Arias intensities; The Fourier amplitude spectrum is calculated as follows: In the formula The frequency domain complex value spectrum; The Fourier amplitude spectrum, that is The modulus; N is the total number of sampling points of the signal; For a discrete-time signal; k is the frequency index; n is the time index; and k corresponds to the physical frequency. for , The sampling frequency; The formula for calculating pseudospectral acceleration is: In the formula Pseudospectral acceleration; The natural period of the structure; The damping ratio; This represents the time-domain acceleration response at the m-th sampling point; Frequency domain acceleration response; The acceleration transfer function of a single-degree-of-freedom system at discrete frequency points The value at; Let be the natural angular frequency of the target, and ; The formula for calculating the intensity of Arias is: In the formula Let be the cumulative Arias intensity at step p, and when hour The total Arias intensity is given by g; g is the acceleration due to gravity. It is a time-domain discrete signal; For time step.

6. The method for generating regional ground motion time histories according to claim 5, characterized in that... The training described in step S4 includes the following steps: Visualize the iterative waveforms to more easily observe the model's behavior patterns and performance characteristics; For the generated Fourier amplitude spectrum, pseudospectral acceleration, and Arias intensity, a loss function is constructed, and the loss weights corresponding to the Fourier amplitude spectrum, pseudospectral acceleration, and Arias intensity are dynamically adjusted to optimize the training effect; The AdamW optimizer was used to train the model.

7. The method for generating regional seismic time histories according to claim 6, characterized in that... The training specifically includes the following steps: The following formula is used as the loss function: In the formula, LOSS is the value of the loss function; For Fourier amplitude spectrum loss; Weights for Fourier amplitude spectrum loss; For pseudospectral acceleration loss; Weights for pseudospectral acceleration loss; For Arias intensity loss; The Arias intensity loss weights; The calculation formula is , This represents the total number of samples. For the true value of the i-th sample, This is the predicted value for the i-th sample; At the start of training, the weights were set to satisfy... ; During training, when When the value is less than the set threshold, reduce... The value increases simultaneously and The value; When training a model using the AdamW optimizer, the update rules include: The gradient momentum is estimated using the following formula. and gradient square momentum : In the formula The first exponential decay rate; Let be the gradient at step t; This is the second exponential decay rate; The following formula is used for deviation correction: In the formula This is the first-order moment estimate after bias correction; This is the second-order moment estimate after bias correction; The decay rate of the first moment; The decay rate of the second moment; The following formula is used for parameter updates with decoupled weight decay: In the formula These are the model parameters after optimization in step t; This is a set constant used for numerical stability; This is the weight decay coefficient.

8. A system for implementing the regional ground motion time history generation method according to any one of claims 1 to 7, characterized in that... It includes a data acquisition module, a data processing module, a model building module, a model training module, and a data generation module; the data acquisition module, data processing module, model building module, model training module, and data generation module are connected in series; the data acquisition module is used to acquire existing seismic motion physical simulation data and upload the data to the data processing module; The data processing module is used to preprocess the acquired data information based on the received data information in order to build a training dataset and upload the data information to the model building module; The model building module constructs a primary model for generating regional ground motion time histories based on the received data, using the XGBoost model and the Physical Information Neural Network (PIN) model, and uploads the data to the model training module. The XGBoost model predicts the Fourier amplitude spectrum, pseudospectral acceleration, and Arias intensity, while the PIN model generates seismic wave data. The model training module trains the primary model using the constructed training dataset to obtain a regional ground motion time history generation model, and uploads the data to the data generation module. The data generation module uses the obtained regional ground motion time history generation model to generate the ground motion time histories for the target region based on the received data.

Images