A method and system for correcting a ground motion prediction model
By extracting strong earthquake observation data from the target area to construct multiple feature correction terms, adjusting parameter combinations, and fusing them with the ground motion prediction model, the problem of insufficient accuracy in existing ground motion attenuation relationship models is solved, achieving higher accuracy model matching and stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- YUNNAN POWER GRID CO LTD ELECTRIC POWER RES INST
- Filing Date
- 2026-06-04
- Publication Date
- 2026-07-24
AI Technical Summary
Existing technologies that indirectly model ground motion attenuation relationships based on seismic intensity suffer from insufficient accuracy and instability, making it difficult to accurately reflect the specific ground motion patterns of the target area.
By acquiring strong earthquake observation data of the target area, regional characteristic parameters such as inelastic attenuation coefficient, geometric diffusion coefficient, crustal amplification factor and stress drop are extracted to construct multiple feature correction terms of the benchmark ground motion prediction model. By adjusting the parameter combination of these feature correction terms, multiple candidate ground motion prediction models are generated. Finally, the modified ground motion prediction model of the target area is obtained through Bayesian fusion.
It improves the accuracy of the ground motion prediction model, ensures the matching degree between the model and the target area, reduces model error, and meets practical needs.
Smart Images
Figure CN122449618A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of earthquake engineering safety technology, and in particular to a method and system for correcting earthquake ground motion prediction models. Background Technology
[0002] Seismic motion attenuation relationships are fundamental to seismic hazard analysis and seismic design of power equipment. Due to the relative scarcity of early strong earthquake observation data and the abundance of seismic intensity data, the industry typically utilizes seismic intensity attenuation laws to indirectly construct seismic motion attenuation relationship models / equations by converting between reference and target areas. The core of this approach lies in using regions with abundant strong earthquake records and mature models as reference areas, and then applying their seismic motion attenuation relationships to target areas lacking strong earthquake data, thereby achieving the modeling of seismic motion attenuation relationships in the target areas.
[0003] However, the existing technique of indirectly modeling the seismic motion attenuation relationship through seismic intensity has obvious drawbacks: (1) Seismic intensity is not directly related to seismic motion characteristics. Its inherent uncertainty and the errors introduced by the conversion process will lead to a decrease in the final modeling accuracy and stability, resulting in the model's accuracy not meeting actual needs; (2) Due to the differences in tectonic features in different regions, the seismic motion attenuation relationship has significant regional characteristics. The model obtained by direct conversion modeling is difficult to truly reflect the specific seismic motion attenuation relationship law of the target area. Therefore, how to combine the specific laws of the target area to improve the accuracy of describing the seismic motion attenuation relationship of the target area is still a problem that needs to be solved by the existing technology. Summary of the Invention
[0004] This application provides a method and system for correcting seismic ground motion prediction models to solve the technical problem that the accuracy of existing seismic ground motion prediction models for specific areas cannot meet the requirements.
[0005] According to a first aspect of the embodiments of this application, a method for correcting a seismic motion prediction model is provided, comprising: Acquire strong earthquake observation data for the target area and a pre-selected benchmark ground motion prediction model corresponding to the target area; Based on the strong earthquake observation data, multiple regional characteristic parameters of the target area are extracted, and multiple feature correction terms of the benchmark ground motion prediction model are constructed based on the multiple regional characteristic parameters; wherein, the multiple regional characteristic parameters include at least the inelastic attenuation coefficient, geometric diffusion coefficient, crustal amplification factor, and stress drop; The baseline ground motion prediction model is modified based on the multiple feature correction terms. Multiple candidate ground motion prediction models are generated by adjusting the parameter combination of the multiple feature correction terms. The multiple candidate ground motion prediction models are then combined and fused to obtain the modified ground motion prediction model for the target area.
[0006] This application first acquires strong earthquake observation data and a pre-selected benchmark ground motion prediction model for the target area. Then, it extracts multiple regional characteristic parameters from the strong earthquake observation data, including the inelastic attenuation coefficient, geometric diffusion coefficient, crustal amplification factor, and stress drop. These parameters are then used to construct multiple feature correction terms for the benchmark ground motion prediction model. The benchmark ground motion prediction model is further modified based on these feature correction terms to generate multiple candidate ground motion prediction models. Finally, by combining and fusing these candidate models, a modified ground motion prediction model for the target area is obtained. Compared to existing technologies that use earthquake intensity as an intermediate quantity to convert a reference area prediction model into a target area prediction model, this application constructs a modified ground motion prediction model for the target area by extracting and using strong earthquake observation data from the target area. Multiple feature correction terms are used to modify the pre-selected benchmark ground motion prediction model, which can fully incorporate the regional characteristics of the target area to truly reflect the specific ground motion patterns of the target area, thereby improving the accuracy of the subsequently obtained modified ground motion prediction model. At the same time, by adjusting the parameter combination of multiple feature correction terms, multiple candidate ground motion prediction models are generated. By combining and fusing multiple candidate ground motion prediction models, the modified ground motion prediction model can be constructed. This can avoid the instability caused by uncertainties such as parameter errors in a single model. Furthermore, the combination of multiple models forms mutual error compensation, thereby reducing the error of the obtained modified ground motion prediction model and improving its accuracy to meet practical needs.
[0007] In some embodiments of this application, the acquisition of the reference seismic motion prediction model specifically involves: The strong earthquake observation data are input into multiple preset ground motion prediction models to obtain strong earthquake prediction data corresponding to each preset ground motion prediction model. Based on the strong earthquake observation data, residual analysis is performed on the strong earthquake prediction data of each preset ground motion prediction model to obtain the prediction residual of each preset ground motion prediction model. Based on the prediction residuals of each preset ground motion prediction model, a reference ground motion prediction model is selected from the plurality of preset ground motion prediction models.
[0008] This application uses strong earthquake observation data of the target area to perform residual analysis on the strong earthquake prediction data of each preset ground motion prediction model, and then selects a benchmark ground motion prediction model. This ensures that the benchmark ground motion prediction model, which serves as the basis for model correction, has the best matching degree with the target area, thereby reducing the difficulty of subsequent model correction from the source and reducing the accumulation of model errors during subsequent correction.
[0009] In some embodiments of this application, the extraction of the inelastic attenuation coefficient specifically involves: Based on the strong earthquake observation data, multiple sets of strong earthquake event records are determined; each set of strong earthquake event records corresponds to different earthquake events in the target area; each set of strong earthquake event records includes multiple observation records; each observation record in each set of strong earthquake event records comes from different stations in the target area and corresponds to the earthquake event in its respective set; Based on the records of the multiple strong earthquake events, attenuation correlation equations were constructed between the amplitude of the Fourier spectrum of the ground motion and the ground motion frequency, the quality factor related to the ground motion frequency, and the site amplification factor based on the spectral ratio method. Based on the records of the multiple strong earthquake events, the attenuation correlation equation is solved, the parameter range of the quality factor is calibrated, and the calibrated quality factor is used as the inelastic attenuation coefficient.
[0010] This application uses strong earthquake observation data to determine multiple sets of strong earthquake event records corresponding to different seismic events in the target area. Each set of strong earthquake records includes multiple observation records from different stations. Then, based on the spectral ratio method, it constructs attenuation correlation equations between the amplitude of the Fourier spectrum of the ground motion and the ground motion frequency, the quality factor related to the ground motion frequency, and the site amplification factor. Based on solving the attenuation correlation equations, it calibrates the parameter range of the quality factor and uses the calibrated quality factor as the inelastic attenuation coefficient. By leveraging the advantage of the spectral ratio method, which does not depend on the characteristics of the seismic source, it can achieve accurate separation and individual solution of the inelastic attenuation coefficient of the target area, thereby improving the accuracy of the obtained inelastic attenuation coefficient and providing a corresponding correction basis for subsequent correction of the benchmark ground motion prediction model.
[0011] In some embodiments of this application, the extraction of the geometric diffusion coefficient specifically involves: Based on the strong earthquake observation data, the crustal waveguide effect in the target area is segmented and modeled to construct a corresponding geometric diffusion model. Based on the strong earthquake observation data, the geometric diffusion model is solved, the parameter range of the segmented threshold of the geometric diffusion model is calibrated, and the corresponding geometric diffusion coefficient is obtained based on the calibrated segmented threshold.
[0012] This application first constructs a geometric diffusion model by segmenting the crustal waveguide effect in the target area, and then solves the parameter range of the segment threshold of the geometric diffusion model to obtain the geometric diffusion coefficient. The segmented model can accurately depict the actual propagation pattern of seismic waves in the crust. The geometric diffusion coefficient obtained by solving the geometric diffusion model can truly reflect the geometric diffusion characteristics of seismic waves in the target area, providing a basis for subsequent correction of the benchmark ground motion prediction model.
[0013] In some embodiments of this application, the extraction of the crustal amplification factor specifically includes: Based on the crustal thickness model of the target area and combined with the strong earthquake observation data, a linear regression correlation equation between crustal thickness and crustal amplification factor is constructed. Based on the strong earthquake observation data, the linear regression correlation equation is solved, and the corresponding crustal magnification factor is obtained based on the thickness parameter range of the crustal thickness model.
[0014] This application first constructs a linear regression correlation equation between crustal thickness and crustal amplification factor based on the crustal thickness model of the target area. Then, based on the thickness parameter range of the crustal thickness model, the corresponding crustal amplification factor is obtained. This linear regression can transform the crustal amplification effect, which is difficult to measure directly, into a linear statistical relationship related to crustal thickness that is easy to calculate and solve. This can truly reflect the relevant characteristics of the crustal amplification effect of seismic waves in the target area due to crustal thickness, and provide a corresponding basis for subsequent correction of the benchmark ground motion prediction model.
[0015] In some embodiments of this application, the extraction of stress drop specifically includes: Based on the strong earthquake observation data, and using finite fault random vibration simulation, the simulated ground motion time history of the target area is iteratively optimized. In each iteration, the parameter range of stress drop in the simulated ground motion time history is optimized and adjusted until the simulation residual of the simulated ground motion time history meets the preset threshold, and the iteration is completed to obtain the corresponding stress drop.
[0016] This application uses finite fault random vibration simulation to iteratively optimize the simulated ground motion time history of the target area, which can achieve regional accurate calibration of stress drop parameters and truly reflect the stress drop characteristics related to them in the target area, providing a corresponding basis for subsequent correction of the benchmark ground motion prediction model.
[0017] In some embodiments of this application, the step of modifying the baseline seismic motion prediction model based on the plurality of feature modification terms, and generating a plurality of candidate seismic motion prediction models by adjusting the parameter combination of the plurality of feature modification terms, specifically includes: According to the preset correction framework, the logarithmic spectral acceleration calculated by the feature coupling correction model of the target area is mapped to the superposition of the logarithmic spectral acceleration calculated by the benchmark ground motion prediction model and the multiple feature correction terms, and the feature coupling correction model is obtained. By adjusting the parameter combination of the multiple feature correction terms in the feature coupling correction model, multiple candidate ground motion prediction models are generated.
[0018] This application maps the logarithmic spectral acceleration calculated by the feature coupling correction model of the target area to the superposition of the logarithmic spectral acceleration calculated by the benchmark ground motion prediction model and multiple feature correction terms according to a preset correction framework, thereby obtaining the feature coupling correction model. The logarithmic spectral acceleration can accurately characterize the basic features of the ground motion model. By superimposing multiple feature correction terms, the regional features of the target area are fully incorporated, and the errors in the benchmark ground motion prediction model that do not match the target area are corrected and compensated to truly reflect the specific ground motion patterns of the target area, thereby improving the matching degree between the obtained feature coupling correction model and the target area. Furthermore, by adjusting the parameter combination of multiple feature correction terms in the feature coupling correction model, multiple candidate ground motion prediction models are generated, which can provide relevant model support for the subsequent combination and fusion of multiple candidate ground motion prediction models to avoid the instability of a single model.
[0019] In some embodiments of this application, the step of combining and fusing the plurality of candidate ground motion prediction models to obtain a modified ground motion prediction model for the target area specifically includes: Based on the log-likelihood value of each candidate ground motion prediction model, the Bayesian fusion weights of the corresponding candidate ground motion prediction models are calculated and determined. Based on the Bayesian fusion weights of each candidate ground motion prediction model, the multiple candidate ground motion prediction models are combined to obtain the modified ground motion prediction model for the target area.
[0020] This application determines the corresponding Bayesian fusion weight by calculating the log-likelihood value of each candidate ground motion prediction model, and then combines multiple candidate ground motion prediction models through the Bayesian fusion weight to obtain a modified ground motion prediction model for the target area. This avoids the instability caused by uncertainties such as parameter errors in a single model, and the combination of multiple models forms mutual error compensation, thereby reducing the error of the obtained modified ground motion prediction model and improving its accuracy to meet practical needs.
[0021] According to a second aspect of the embodiments of this application, a seismic motion prediction model correction system is provided, including a data model acquisition module, a correction item construction module, and a baseline model correction module; The data model acquisition module is used to acquire strong earthquake observation data of the target area and a pre-selected benchmark ground motion prediction model corresponding to the target area. The correction term construction module is used to extract multiple regional feature parameters of the target area based on the strong earthquake observation data, and to construct multiple feature correction terms of the benchmark ground motion prediction model based on the multiple regional feature parameters; wherein, the multiple regional feature parameters include at least the inelastic attenuation coefficient, geometric diffusion coefficient, crustal amplification factor, and stress drop; The baseline model correction module is used to correct the baseline ground motion prediction model according to the multiple feature correction terms, generate multiple candidate ground motion prediction models by adjusting the parameter combination of the multiple feature correction terms, and combine and fuse the multiple candidate ground motion prediction models to obtain the corrected ground motion prediction model for the target area.
[0022] In some embodiments of this application, the acquisition of the reference seismic motion prediction model specifically involves: The strong earthquake observation data are input into multiple preset ground motion prediction models to obtain strong earthquake prediction data corresponding to each preset ground motion prediction model. Based on the strong earthquake observation data, residual analysis is performed on the strong earthquake prediction data of each preset ground motion prediction model to obtain the prediction residual of each preset ground motion prediction model. Based on the prediction residuals of each preset ground motion prediction model, a reference ground motion prediction model is selected from the plurality of preset ground motion prediction models.
[0023] In some embodiments of this application, the extraction of the inelastic attenuation coefficient specifically involves: Based on the strong earthquake observation data, multiple sets of strong earthquake event records are determined; each set of strong earthquake event records corresponds to different earthquake events in the target area; each set of strong earthquake event records includes multiple observation records; each observation record in each set of strong earthquake event records comes from different stations in the target area and corresponds to the earthquake event in its respective set; Based on the records of the multiple strong earthquake events, attenuation correlation equations were constructed between the amplitude of the Fourier spectrum of the ground motion and the ground motion frequency, the quality factor related to the ground motion frequency, and the site amplification factor based on the spectral ratio method. Based on the records of the multiple strong earthquake events, the attenuation correlation equation is solved, the parameter range of the quality factor is calibrated, and the calibrated quality factor is used as the inelastic attenuation coefficient.
[0024] In some embodiments of this application, the extraction of the geometric diffusion coefficient specifically involves: Based on the strong earthquake observation data, the crustal waveguide effect in the target area is segmented and modeled to construct a corresponding geometric diffusion model. Based on the strong earthquake observation data, the geometric diffusion model is solved, the parameter range of the segmented threshold of the geometric diffusion model is calibrated, and the corresponding geometric diffusion coefficient is obtained based on the calibrated segmented threshold.
[0025] In some embodiments of this application, the extraction of the crustal amplification factor specifically includes: Based on the crustal thickness model of the target area and combined with the strong earthquake observation data, a linear regression correlation equation between crustal thickness and crustal amplification factor is constructed. Based on the strong earthquake observation data, the linear regression correlation equation is solved, and the corresponding crustal magnification factor is obtained based on the thickness parameter range of the crustal thickness model.
[0026] In some embodiments of this application, the extraction of stress drop specifically includes: Based on the strong earthquake observation data, and using finite fault random vibration simulation, the simulated ground motion time history of the target area is iteratively optimized. In each iteration, the parameter range of stress drop in the simulated ground motion time history is optimized and adjusted until the simulation residual of the simulated ground motion time history meets the preset threshold, and the iteration is completed to obtain the corresponding stress drop.
[0027] In some embodiments of this application, the baseline model correction module includes a model mapping conversion unit and a candidate model generation unit; The model mapping and conversion unit is used to map the logarithmic spectral acceleration calculated by the feature coupling correction model of the target area to the superposition of the logarithmic spectral acceleration calculated by the reference ground motion prediction model and the multiple feature correction terms according to the preset correction framework, and convert it to obtain the feature coupling correction model. The candidate model generation unit is used to generate multiple candidate ground motion prediction models by adjusting the parameter combination of the multiple feature correction terms in the feature coupling correction model.
[0028] In some embodiments of this application, the baseline model correction module includes a fusion weight determination unit and a model combination correction unit; The fusion weight determination unit is used to calculate and determine the Bayesian fusion weight of the corresponding candidate ground motion prediction model based on the log likelihood value of each candidate ground motion prediction model. The model combination correction unit is used to combine the multiple candidate ground motion prediction models according to the Bayesian fusion weight of each candidate ground motion prediction model to obtain the corrected ground motion prediction model for the target area.
[0029] This application first acquires strong earthquake observation data and a pre-selected benchmark ground motion prediction model for the target area. Then, it extracts multiple regional characteristic parameters from the strong earthquake observation data, including the inelastic attenuation coefficient, geometric diffusion coefficient, crustal amplification factor, and stress drop. These parameters are then used to construct multiple feature correction terms for the benchmark ground motion prediction model. The benchmark ground motion prediction model is further modified based on these feature correction terms to generate multiple candidate ground motion prediction models. Finally, by combining and fusing these candidate models, a modified ground motion prediction model for the target area is obtained. Compared to existing technologies that use earthquake intensity as an intermediate quantity to convert a reference area prediction model into a target area prediction model, this application constructs a modified ground motion prediction model for the target area by extracting and using strong earthquake observation data from the target area. Multiple feature correction terms are used to modify the pre-selected benchmark ground motion prediction model, which can fully incorporate the regional characteristics of the target area to truly reflect the specific ground motion patterns of the target area, thereby improving the accuracy of the subsequently obtained modified ground motion prediction model. At the same time, by adjusting the parameter combination of multiple feature correction terms, multiple candidate ground motion prediction models are generated. By combining and fusing multiple candidate ground motion prediction models, the modified ground motion prediction model can be constructed. This can avoid the instability caused by uncertainties such as parameter errors in a single model. Furthermore, the combination of multiple models forms mutual error compensation, thereby reducing the error of the obtained modified ground motion prediction model and improving its accuracy to meet practical needs. Attached Figure Description
[0030] Figure 1 This is a flowchart illustrating a method for correcting a seismic motion prediction model according to certain embodiments of this application. Figure 2 This is a block diagram of a seismic motion prediction model correction system shown in certain embodiments of this application. Figure 3 This is a graph showing the quality factor as a function of frequency, as illustrated in certain embodiments of this application. Detailed Implementation
[0031] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below in conjunction with the accompanying drawings are exemplary and are only used to explain some embodiments of this application, and should not be construed as limiting the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments shown in this application without inventive effort are within the protection scope of this application.
[0032] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, unless otherwise explicitly specified, "a plurality of" or "several" means two or more.
[0033] Please see Figure 1 This application provides a method for correcting a seismic motion prediction model, including steps S101 to S103, each step of which is as follows: Step S101: Obtain strong earthquake observation data of the target area and a pre-selected benchmark ground motion prediction model corresponding to the target area.
[0034] Specifically, the target area includes, but is not limited to, Yunnan Province; strong earthquakes refer to earthquakes of magnitude 5 or higher. More specifically, the strong earthquake observation data for the target area are earthquake data of magnitude 5 or higher in Yunnan Province, including earthquake data of magnitude 5.9 in Yingjiang, 5.9 in Shangri-La, 5.9 in Jinggu, 6.5 in Ludian, and 6.5 in Yangbi. After data acquisition, corresponding data screening is required, with the following criteria: focal depth not exceeding 30 km, and signal-to-noise ratio of station records not less than 3; all near-field saturated records with fault distance less than 5 km are excluded.
[0035] In some embodiments of this application, the acquisition of the reference seismic motion prediction model specifically involves: The strong earthquake observation data are input into multiple preset ground motion prediction models to obtain strong earthquake prediction data corresponding to each preset ground motion prediction model. Based on the strong earthquake observation data, residual analysis is performed on the strong earthquake prediction data of each preset ground motion prediction model to obtain the prediction residual of each preset ground motion prediction model. Based on the prediction residuals of each preset ground motion prediction model, a reference ground motion prediction model is selected from the plurality of preset ground motion prediction models.
[0036] Specifically, the pre-set ground motion prediction models include, but are not limited to: the Eastern China model, the Sichuan-Yunnan model, the Bindi European model, the Abrahamson global model, and the Campbell & Bozorgnia global model.
[0037] Specifically, in each preset seismic motion prediction model, the formula for calculating the prediction residual is as follows: ; in, For the first The first earthquake event Logarithmic residuals of each station For the first The first earthquake event Observations from individual stations This represents the predicted value of the corresponding model.
[0038] This application uses strong earthquake observation data of the target area to perform residual analysis on the strong earthquake prediction data of each preset ground motion prediction model, and then selects a benchmark ground motion prediction model. This ensures that the benchmark ground motion prediction model, which serves as the basis for model correction, has the best matching degree with the target area, thereby reducing the difficulty of subsequent model correction from the source and reducing the accumulation of model errors during subsequent correction.
[0039] Step S102: Based on the strong earthquake observation data, extract multiple regional characteristic parameters of the target area, and construct multiple feature correction terms of the benchmark ground motion prediction model based on the multiple regional characteristic parameters; wherein, the multiple regional characteristic parameters include at least the inelastic attenuation coefficient, geometric diffusion coefficient, crustal amplification factor, and stress drop.
[0040] In some embodiments of this application, the extraction of the inelastic attenuation coefficient specifically involves: Based on the strong earthquake observation data, multiple sets of strong earthquake event records are determined; each set of strong earthquake event records corresponds to different earthquake events in the target area; each set of strong earthquake event records includes multiple observation records; each observation record in each set of strong earthquake event records comes from different stations in the target area and corresponds to the earthquake event in its respective set; Based on the records of the multiple strong earthquake events, attenuation correlation equations were constructed between the amplitude of the Fourier spectrum of the ground motion and the ground motion frequency, the quality factor related to the ground motion frequency, and the site amplification factor based on the spectral ratio method. Based on the records of the multiple strong earthquake events, the attenuation correlation equation is solved, the parameter range of the quality factor is calibrated, and the calibrated quality factor is used as the inelastic attenuation coefficient.
[0041] Specifically, when selecting strong earthquake event records, it is necessary to ensure that the focal distance of each observation record in each group of strong earthquake event records is greater than 50 km, so as to ensure that inelastic attenuation is dominant.
[0042] Specifically, the attenuation correlation equation between the Fourier spectrum amplitude of the ground motion and the ground motion frequency, the quality factor related to the ground motion frequency, and the site amplification factor is as follows: ; in, For the first The earthquake event in the 1st The frequency recorded by each station The Fourier spectrum amplitude of the earthquake ground motion, in cm / s or gal; For the first The earthquake event in the 1st The frequency recorded by each station The amplitude of the Fourier spectrum of the seismic motion; The frequency of ground motion; For frequency The quality factor of the relevant seismic medium describes the degree of inelastic energy loss of seismic waves during propagation; For the first The epicenter of the first earthquake event was at the first... The distance between the seismic sources of each station, in km; For the first The epicenter of the first earthquake event was at the first... The distance from the seismic source to each station; For the first The site magnification factor of each station describes the amplification effect of local site conditions on ground motion. For reference, the first The site magnification factor of each station.
[0043] Specifically, after calibration, the quality factor is output. middle Expected range: , And output as follows Figure 3 The graph shows the quality factor as a function of frequency, expressed in discrete point form.
[0044] This application uses strong earthquake observation data to determine multiple sets of strong earthquake event records corresponding to different seismic events in the target area. Each set of strong earthquake records includes multiple observation records from different stations. Then, based on the spectral ratio method, it constructs attenuation correlation equations between the amplitude of the Fourier spectrum of the ground motion and the ground motion frequency, the quality factor related to the ground motion frequency, and the site amplification factor. Based on solving the attenuation correlation equations, it calibrates the parameter range of the quality factor and uses the calibrated quality factor as the inelastic attenuation coefficient. By leveraging the advantage of the spectral ratio method, which does not depend on the characteristics of the seismic source, it can achieve accurate separation and individual solution of the inelastic attenuation coefficient of the target area, thereby improving the accuracy of the obtained inelastic attenuation coefficient and providing a corresponding correction basis for subsequent correction of the benchmark ground motion prediction model.
[0045] In some embodiments of this application, the extraction of the geometric diffusion coefficient specifically involves: Based on the strong earthquake observation data, the crustal waveguide effect in the target area is segmented and modeled to construct a corresponding geometric diffusion model. Based on the strong earthquake observation data, the geometric diffusion model is solved, the parameter range of the segmented threshold of the geometric diffusion model is calibrated, and the corresponding geometric diffusion coefficient is obtained based on the calibrated segmented threshold.
[0046] Specifically, before performing segmented modeling of the crustal waveguide effect in the target area, it is necessary to screen near-field records with a source distance of less than 50km for segmented regression in order to accurately identify the anomalous amplification distance segment caused by the crustal waveguide effect. The resulting geometric diffusion model is as follows: ; in, The distance from the epicenter, The geometric attenuation function describes the amplitude attenuation effect caused by wavefront expansion during seismic wave propagation; The calibration yielded the segmented threshold. The parameter range (15-30km) and near-field / far-field attenuation index .
[0047] This application first constructs a geometric diffusion model by segmenting the crustal waveguide effect in the target area, and then solves the parameter range of the segment threshold of the geometric diffusion model to obtain the geometric diffusion coefficient. The segmented model can accurately depict the actual propagation pattern of seismic waves in the crust. The geometric diffusion coefficient obtained by solving the geometric diffusion model can truly reflect the geometric diffusion characteristics of seismic waves in the target area, providing a basis for subsequent correction of the benchmark ground motion prediction model.
[0048] In some embodiments of this application, the extraction of the crustal amplification factor specifically includes: Based on the crustal thickness model of the target area and combined with the strong earthquake observation data, a linear regression correlation equation between crustal thickness and crustal amplification factor is constructed. Based on the strong earthquake observation data, the linear regression correlation equation is solved, and the corresponding crustal magnification factor is obtained based on the thickness parameter range of the crustal thickness model.
[0049] Specifically, the linear regression equation relating crustal thickness to crustal amplification factor is as follows: ; in, The depth of the Mohorovičić discontinuity ranges from 40 to 60 km. The crustal magnification factor was obtained through calibration. .
[0050] This application first constructs a linear regression correlation equation between crustal thickness and crustal amplification factor based on the crustal thickness model of the target area. Then, based on the thickness parameter range of the crustal thickness model, the corresponding crustal amplification factor is obtained. This linear regression can transform the crustal amplification effect, which is difficult to measure directly, into a linear statistical relationship related to crustal thickness that is easy to calculate and solve. This can truly reflect the relevant characteristics of the crustal amplification effect of seismic waves in the target area due to crustal thickness, and provide a corresponding basis for subsequent correction of the benchmark ground motion prediction model.
[0051] In some embodiments of this application, the extraction of stress drop specifically includes: Based on the strong earthquake observation data, and using finite fault random vibration simulation, the simulated ground motion time history of the target area is iteratively optimized. In each iteration, the parameter range of stress drop in the simulated ground motion time history is optimized and adjusted until the simulation residual of the simulated ground motion time history meets the preset threshold, and the iteration is completed to obtain the corresponding stress drop.
[0052] Specifically, at the start of the iteration, the simulated seismic time history of the target area had an initial stress drop range of 30-150 bar, a rupture propagation velocity Vrupt=0.8β, and β=3.8 km / s; at the end of the iteration, the average stress drop was 70±20 bar, and the statistical relationship between stress drop and magnitude was as follows: , The magnitude is [magnitude].
[0053] This application uses finite fault random vibration simulation to iteratively optimize the simulated ground motion time history of the target area, which can achieve regional accurate calibration of stress drop parameters and truly reflect the stress drop characteristics related to them in the target area, providing a corresponding basis for subsequent correction of the benchmark ground motion prediction model.
[0054] Step S103: Based on the multiple feature correction terms, the baseline ground motion prediction model is corrected. By adjusting the parameter combination of the multiple feature correction terms, multiple candidate ground motion prediction models are generated. The multiple candidate ground motion prediction models are then combined and fused to obtain the corrected ground motion prediction model for the target area.
[0055] In some embodiments of this application, the step of modifying the baseline seismic motion prediction model based on the plurality of feature modification terms, and generating a plurality of candidate seismic motion prediction models by adjusting the parameter combination of the plurality of feature modification terms, specifically includes: According to the preset correction framework, the logarithmic spectral acceleration calculated by the feature coupling correction model of the target area is mapped to the superposition of the logarithmic spectral acceleration calculated by the benchmark ground motion prediction model and the multiple feature correction terms, and the feature coupling correction model is obtained. By adjusting the parameter combination of the multiple feature correction terms in the feature coupling correction model, multiple candidate ground motion prediction models are generated.
[0056] Specifically, the feature coupling correction model is as follows: ; in, The spectral acceleration calculated by the characteristic coupling correction model. The spectral acceleration calculated by the benchmark ground motion prediction model; The terms are, in order, regional attenuation correction terms based on inelastic attenuation coefficient, geometric diffusion coefficient and crustal amplification factor; site effect correction terms based on long-period amplification of deep sedimentary layers (>500m) in basins and topographic amplification effect of mountain stations through elevation slope correction; and source characteristic correction terms based on stress drop.
[0057] More specifically, the region attenuation correction term Specifically: ; in, Quality factor The amount of disturbance when the parameters change; Geometric decay function The amount of disturbance when the parameters change; This is the crustal amplification factor.
[0058] More specifically, the site effect correction term Specifically: ; in, For the first Site magnification factor for each station; For reference, the first Site magnification factor for each station; To be related to the cycle The long-period amplification factor of the relevant deep sedimentary layers in the basin, Effective immediately; The mountainous elevation slope topographic amplification factor is used to characterize the topographic amplification effect of mountain stations corrected by elevation slope.
[0059] More specifically, the source characteristic correction term Specifically: ; in, The stress drop obtained through calibration Source characteristic function relating to the rupture propagation velocity Vrupt.
[0060] Preferably, the number of candidate ground motion prediction models is 15.
[0061] This application maps the logarithmic spectral acceleration calculated by the feature coupling correction model of the target area to the superposition of the logarithmic spectral acceleration calculated by the benchmark ground motion prediction model and multiple feature correction terms according to a preset correction framework, thereby obtaining the feature coupling correction model. The logarithmic spectral acceleration can accurately characterize the basic features of the ground motion model. By superimposing multiple feature correction terms, the regional features of the target area are fully incorporated, and the errors in the benchmark ground motion prediction model that do not match the target area are corrected and compensated to truly reflect the specific ground motion patterns of the target area, thereby improving the matching degree between the obtained feature coupling correction model and the target area. Furthermore, by adjusting the parameter combination of multiple feature correction terms in the feature coupling correction model, multiple candidate ground motion prediction models are generated, which can provide relevant model support for the subsequent combination and fusion of multiple candidate ground motion prediction models to avoid the instability of a single model.
[0062] In some embodiments of this application, the step of combining and fusing the plurality of candidate ground motion prediction models to obtain a modified ground motion prediction model for the target area specifically includes: Based on the log-likelihood value of each candidate ground motion prediction model, the Bayesian fusion weights of the corresponding candidate ground motion prediction models are calculated and determined. Based on the Bayesian fusion weights of each candidate ground motion prediction model, the multiple candidate ground motion prediction models are combined to obtain the modified ground motion prediction model for the target area.
[0063] Specifically, when calculating the Bayesian fusion weight based on the log-likelihood value of each candidate ground motion prediction model, the Bayesian fusion weight of each candidate ground motion prediction model is kept proportional to its log-likelihood value.
[0064] Preferably, three candidate seismic motion prediction models are finally selected, with weight distributions of 45%, 30%, and 25%.
[0065] This application determines the corresponding Bayesian fusion weight by calculating the log-likelihood value of each candidate ground motion prediction model, and then combines multiple candidate ground motion prediction models through the Bayesian fusion weight to obtain a modified ground motion prediction model for the target area. This avoids the instability caused by uncertainties such as parameter errors in a single model, and the combination of multiple models forms mutual error compensation, thereby reducing the error of the obtained modified ground motion prediction model and improving its accuracy to meet practical needs.
[0066] Compared to existing technologies, this application first acquires strong earthquake observation data and a pre-selected benchmark ground motion prediction model for the target area. Then, it extracts multiple regional characteristic parameters, including inelastic attenuation coefficient, geometric diffusion coefficient, crustal amplification factor, and stress drop, based on the strong earthquake observation data. This allows for the construction of multiple feature correction terms for the benchmark ground motion prediction model. The benchmark ground motion prediction model is then corrected based on these feature correction terms to generate multiple candidate ground motion prediction models. Finally, by combining and fusing these candidate models, a corrected ground motion prediction model for the target area is obtained. Compared to existing technologies that use earthquake intensity as an intermediate quantity to convert a reference area prediction model into a target area prediction model, this application utilizes strong earthquake observation data from the target area... Extracting and constructing multiple feature correction terms to modify the pre-selected benchmark ground motion prediction model can fully incorporate the regional characteristics of the target area to truly reflect the specific ground motion patterns of the target area, thereby improving the accuracy of the subsequently obtained modified ground motion prediction model. At the same time, by adjusting the parameter combination of multiple feature correction terms, multiple candidate ground motion prediction models are generated. By combining and fusing multiple candidate ground motion prediction models to construct the modified ground motion prediction model, the instability caused by uncertainties such as parameter errors in a single model can be avoided. Furthermore, the combination of multiple models forms mutual error compensation, thereby reducing the error of the obtained modified ground motion prediction model and improving its accuracy to meet practical needs.
[0067] For a method corresponding to the one described above, please refer to [link / reference]. Figure 2 This application provides a seismic motion prediction model correction system, including a data model acquisition module 210, a correction item construction module 220, and a baseline model correction module 230. The data model acquisition module 210 is used to acquire strong earthquake observation data of the target area and a pre-selected benchmark ground motion prediction model corresponding to the target area. The correction term construction module 220 is used to extract multiple regional feature parameters of the target area based on the strong earthquake observation data, and to construct multiple feature correction terms of the benchmark ground motion prediction model based on the multiple regional feature parameters; wherein, the multiple regional feature parameters include at least the inelastic attenuation coefficient, geometric diffusion coefficient, crustal amplification factor, and stress drop; The baseline model correction module 230 is used to correct the baseline ground motion prediction model according to the plurality of feature correction terms, generate a plurality of candidate ground motion prediction models by adjusting the parameter combination of the plurality of feature correction terms, and combine and fuse the plurality of candidate ground motion prediction models to obtain the corrected ground motion prediction model for the target area.
[0068] In some embodiments of this application, the acquisition of the reference seismic motion prediction model specifically involves: The strong earthquake observation data are input into multiple preset ground motion prediction models to obtain strong earthquake prediction data corresponding to each preset ground motion prediction model. Based on the strong earthquake observation data, residual analysis is performed on the strong earthquake prediction data of each preset ground motion prediction model to obtain the prediction residual of each preset ground motion prediction model. Based on the prediction residuals of each preset ground motion prediction model, a reference ground motion prediction model is selected from the plurality of preset ground motion prediction models.
[0069] In some embodiments of this application, the extraction of the inelastic attenuation coefficient specifically involves: Based on the strong earthquake observation data, multiple sets of strong earthquake event records are determined; each set of strong earthquake event records corresponds to different earthquake events in the target area; each set of strong earthquake event records includes multiple observation records; each observation record in each set of strong earthquake event records comes from different stations in the target area and corresponds to the earthquake event in its respective set; Based on the records of the multiple strong earthquake events, attenuation correlation equations were constructed between the amplitude of the Fourier spectrum of the ground motion and the ground motion frequency, the quality factor related to the ground motion frequency, and the site amplification factor based on the spectral ratio method. Based on the records of the multiple strong earthquake events, the attenuation correlation equation is solved, the parameter range of the quality factor is calibrated, and the calibrated quality factor is used as the inelastic attenuation coefficient.
[0070] In some embodiments of this application, the extraction of the geometric diffusion coefficient specifically involves: Based on the strong earthquake observation data, the crustal waveguide effect in the target area is segmented and modeled to construct a corresponding geometric diffusion model. Based on the strong earthquake observation data, the geometric diffusion model is solved, the parameter range of the segmented threshold of the geometric diffusion model is calibrated, and the corresponding geometric diffusion coefficient is obtained based on the calibrated segmented threshold.
[0071] In some embodiments of this application, the extraction of the crustal amplification factor specifically includes: Based on the crustal thickness model of the target area and combined with the strong earthquake observation data, a linear regression correlation equation between crustal thickness and crustal amplification factor is constructed. Based on the strong earthquake observation data, the linear regression correlation equation is solved, and the corresponding crustal magnification factor is obtained based on the thickness parameter range of the crustal thickness model.
[0072] In some embodiments of this application, the extraction of stress drop specifically includes: Based on the strong earthquake observation data, and using finite fault random vibration simulation, the simulated ground motion time history of the target area is iteratively optimized. In each iteration, the parameter range of stress drop in the simulated ground motion time history is optimized and adjusted until the simulation residual of the simulated ground motion time history meets the preset threshold, and the iteration is completed to obtain the corresponding stress drop.
[0073] In some embodiments of this application, the baseline model correction module 230 includes a model mapping conversion unit and a candidate model generation unit; The model mapping and conversion unit is used to map the logarithmic spectral acceleration calculated by the feature coupling correction model of the target area to the superposition of the logarithmic spectral acceleration calculated by the reference ground motion prediction model and the multiple feature correction terms according to the preset correction framework, and convert it to obtain the feature coupling correction model. The candidate model generation unit is used to generate multiple candidate ground motion prediction models by adjusting the parameter combination of the multiple feature correction terms in the feature coupling correction model.
[0074] In some embodiments of this application, the baseline model correction module 230 includes a fusion weight determination unit and a model combination correction unit; The fusion weight determination unit is used to calculate and determine the Bayesian fusion weight of the corresponding candidate ground motion prediction model based on the log likelihood value of each candidate ground motion prediction model. The model combination correction unit is used to combine the multiple candidate ground motion prediction models according to the Bayesian fusion weight of each candidate ground motion prediction model to obtain the corrected ground motion prediction model for the target area.
[0075] This application first acquires strong earthquake observation data and a pre-selected benchmark ground motion prediction model for the target area. Then, it extracts multiple regional characteristic parameters from the strong earthquake observation data, including the inelastic attenuation coefficient, geometric diffusion coefficient, crustal amplification factor, and stress drop. These parameters are then used to construct multiple feature correction terms for the benchmark ground motion prediction model. The benchmark ground motion prediction model is further modified based on these feature correction terms to generate multiple candidate ground motion prediction models. Finally, by combining and fusing these candidate models, a modified ground motion prediction model for the target area is obtained. Compared to existing technologies that use earthquake intensity as an intermediate quantity to convert a reference area prediction model into a target area prediction model, this application constructs a modified ground motion prediction model for the target area by extracting and using strong earthquake observation data from the target area. Multiple feature correction terms are used to modify the pre-selected benchmark ground motion prediction model, which can fully incorporate the regional characteristics of the target area to truly reflect the specific ground motion patterns of the target area, thereby improving the accuracy of the subsequently obtained modified ground motion prediction model. At the same time, by adjusting the parameter combination of multiple feature correction terms, multiple candidate ground motion prediction models are generated. By combining and fusing multiple candidate ground motion prediction models, the modified ground motion prediction model can be constructed. This can avoid the instability caused by uncertainties such as parameter errors in a single model. Furthermore, the combination of multiple models forms mutual error compensation, thereby reducing the error of the obtained modified ground motion prediction model and improving its accuracy to meet practical needs.
[0076] It should be understood that the system provided in this application is corresponding to the aforementioned method. The seismic motion prediction model correction system provided in this application can implement the seismic motion prediction model correction method provided in any of the embodiments of this application.
[0077] Adaptively, embodiments of this application also provide a computer device and a computer-readable storage medium.
[0078] The computer device includes: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor; The processor executes the computer program to implement a seismic motion prediction model correction method of this application.
[0079] The computer-readable storage medium stores multiple instructions adapted for loading by a processor to execute a seismic motion prediction model correction method of this application.
[0080] The above description represents some embodiments of this application, providing a further detailed explanation of the purpose, technical solution, and beneficial effects of this application. It should be understood that the above-described embodiments of this application should not be construed as limiting this application. In particular, any changes, modifications, equivalent substitutions, and variations made by those skilled in the art within the spirit and principles of this application should be included within the scope of protection of this application.
Claims
1. A method for correcting a seismic motion prediction model, characterized in that, include: Acquire strong earthquake observation data for the target area and a pre-selected benchmark ground motion prediction model corresponding to the target area; Based on the strong earthquake observation data, multiple regional characteristic parameters of the target area are extracted, and multiple feature correction terms of the benchmark ground motion prediction model are constructed based on the multiple regional characteristic parameters; wherein, the multiple regional characteristic parameters include at least the inelastic attenuation coefficient, geometric diffusion coefficient, crustal amplification factor, and stress drop; The baseline ground motion prediction model is modified based on the multiple feature correction terms. Multiple candidate ground motion prediction models are generated by adjusting the parameter combination of the multiple feature correction terms. The multiple candidate ground motion prediction models are then combined and fused to obtain the modified ground motion prediction model for the target area.
2. The method for correcting a seismic motion prediction model according to claim 1, characterized in that, The acquisition of the benchmark ground motion prediction model is specifically as follows: The strong earthquake observation data are input into multiple preset ground motion prediction models to obtain strong earthquake prediction data corresponding to each preset ground motion prediction model. Based on the strong earthquake observation data, residual analysis is performed on the strong earthquake prediction data of each preset ground motion prediction model to obtain the prediction residual of each preset ground motion prediction model. Based on the prediction residuals of each preset ground motion prediction model, a reference ground motion prediction model is selected from the plurality of preset ground motion prediction models.
3. The method for correcting a seismic motion prediction model according to claim 1, characterized in that, The extraction of the inelastic attenuation coefficient is specifically as follows: Based on the strong earthquake observation data, multiple sets of strong earthquake event records are determined; each set of strong earthquake event records corresponds to different earthquake events in the target area; each set of strong earthquake event records includes multiple observation records; each observation record in each set of strong earthquake event records comes from different stations in the target area and corresponds to the earthquake event in its respective set; Based on the records of the multiple strong earthquake events, attenuation correlation equations were constructed between the amplitude of the Fourier spectrum of the ground motion and the ground motion frequency, the quality factor related to the ground motion frequency, and the site amplification factor based on the spectral ratio method. Based on the records of the multiple strong earthquake events, the attenuation correlation equation is solved, the parameter range of the quality factor is calibrated, and the calibrated quality factor is used as the inelastic attenuation coefficient.
4. The method for correcting a seismic motion prediction model according to claim 1, characterized in that, The extraction of the geometric diffusion coefficient is specifically as follows: Based on the strong earthquake observation data, the crustal waveguide effect in the target area is segmented and modeled to construct a corresponding geometric diffusion model. Based on the strong earthquake observation data, the geometric diffusion model is solved, the parameter range of the segmented threshold of the geometric diffusion model is calibrated, and the corresponding geometric diffusion coefficient is obtained based on the calibrated segmented threshold.
5. The method for correcting a seismic motion prediction model according to claim 1, characterized in that, The extraction of the crustal amplification factor is specifically as follows: Based on the crustal thickness model of the target area and combined with the strong earthquake observation data, a linear regression correlation equation between crustal thickness and crustal amplification factor is constructed. Based on the strong earthquake observation data, the linear regression correlation equation is solved, and the corresponding crustal magnification factor is obtained based on the thickness parameter range of the crustal thickness model.
6. The method for correcting a seismic motion prediction model according to claim 1, characterized in that, The extraction of the stress drop is specifically as follows: Based on the strong earthquake observation data, and using finite fault random vibration simulation, the simulated ground motion time history of the target area is iteratively optimized. In each iteration, the parameter range of stress drop in the simulated ground motion time history is optimized and adjusted until the simulation residual of the simulated ground motion time history meets the preset threshold, and the iteration is completed to obtain the corresponding stress drop.
7. A method for correcting a seismic motion prediction model according to any one of claims 1 to 6, characterized in that, The step of correcting the baseline ground motion prediction model based on the multiple feature correction terms, and generating multiple candidate ground motion prediction models by adjusting the parameter combination of the multiple feature correction terms, specifically includes: According to the preset correction framework, the logarithmic spectral acceleration calculated by the feature coupling correction model of the target area is mapped to the superposition of the logarithmic spectral acceleration calculated by the benchmark ground motion prediction model and the multiple feature correction terms, and the feature coupling correction model is obtained. By adjusting the parameter combination of the multiple feature correction terms in the feature coupling correction model, multiple candidate ground motion prediction models are generated.
8. A method for correcting a seismic motion prediction model according to any one of claims 1 to 6, characterized in that, The step of combining and fusing the multiple candidate ground motion prediction models to obtain the modified ground motion prediction model for the target area specifically includes: Based on the log-likelihood value of each candidate ground motion prediction model, the Bayesian fusion weights of the corresponding candidate ground motion prediction models are calculated and determined. Based on the Bayesian fusion weights of each candidate ground motion prediction model, the multiple candidate ground motion prediction models are combined to obtain the modified ground motion prediction model for the target area.
9. A seismic motion prediction model correction system, characterized in that, It includes a data model acquisition module, a correction item construction module, and a baseline model correction module; The data model acquisition module is used to acquire strong earthquake observation data of the target area and a pre-selected benchmark ground motion prediction model corresponding to the target area. The correction term construction module is used to extract multiple regional feature parameters of the target area based on the strong earthquake observation data, and to construct multiple feature correction terms of the benchmark ground motion prediction model based on the multiple regional feature parameters; wherein, the multiple regional feature parameters include at least the inelastic attenuation coefficient, geometric diffusion coefficient, crustal amplification factor, and stress drop; The baseline model correction module is used to correct the baseline ground motion prediction model according to the multiple feature correction terms, generate multiple candidate ground motion prediction models by adjusting the parameter combination of the multiple feature correction terms, and combine and fuse the multiple candidate ground motion prediction models to obtain the corrected ground motion prediction model for the target area.
10. A seismic motion prediction model correction system according to claim 9, characterized in that, The baseline model correction module includes a fusion weight determination unit and a model combination correction unit; The fusion weight determination unit is used to calculate and determine the Bayesian fusion weight of the corresponding candidate ground motion prediction model based on the log likelihood value of each candidate ground motion prediction model. The model combination correction unit is used to combine the multiple candidate ground motion prediction models according to the Bayesian fusion weight of each candidate ground motion prediction model to obtain the corrected ground motion prediction model for the target area.