Seismic waveform generation method, learning model generation method, seismic waveform generation device, and learning model generation method

Abstract

To provide a seismic waveform generation method which enables easy generation of seismic waveforms without having to perform complex computations and analyses.SOLUTION: A seismic waveform generation method disclosed herein uses a computer to generate seismic waveform data representing a time history waveform of seismic motion caused by an earthquake. The seismic waveform generation method comprises: a reception step of receiving a group of seismic motion characteristic parameters indicative of characteristics of prediction target seismic motion; and a generation step of generating seismic waveform data for the prediction target seismic motion, using a machine-trained learning model 13.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present invention relates to a seismic wave form generation method, a learning model creation method, a seismic wave form generation device, and a learning model creation device.

Background Art

[0002] The response at a receiving point when a force is applied at a vibration point is called a Green's function. In an earthquake, the vibration point is the earthquake source, the receiving point is the earthquake observation point (hereinafter referred to as the "observation point"), and the response corresponds to the seismic wave form observed at the observation point as an earthquake observation record (hereinafter referred to as the "observation record"). In particular, the seismic wave form observed at the observation point in a past earthquake is called an empirical Green's function and can be used for evaluating and predicting various seismic motions. Also, when predicting the seismic wave form at an arbitrary point where no observation record exists, a statistical Green's function that calculates the seismic wave form based on the statistical analysis results of a large number of past observation records (see, for example, Patent Document 1), or a theoretical Green's function that calculates the seismic wave form based on a subsurface structure model that models the subsurface structure from the earthquake source to the observation point (see, for example, Patent Document 2) has been proposed.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] The empirical Green's function has the limitation that it can only be applied to the observation points where observation records have been obtained in the past. As statistical Green's functions without such limitations, various regression formulas and models have been proposed, but it has been pointed out that variations in calculation results occur due to differences in methods. For example, in the case of creating a ground motion waveform by inverse Fourier transform using two modeled Fourier amplitudes and Fourier phases, the correlation between the amplitude and phase of the original ground motion may be disrupted. Furthermore, the application of the theoretical Green's function requires complex and high-precision modeling of the underground structure. In particular, for the evaluation of short-period components, a high-resolution model corresponding to the wavelength is required, but since such models can be prepared in extremely few scenarios, it is often applied only to long-period components, and the application range is limited.

[0005] The present invention has been made in consideration of such circumstances, and an object thereof is to provide a ground motion waveform generation method, a learning model creation method, a ground motion waveform generation device, and a learning model creation device that enable easy generation of ground motion waveforms without performing complex calculations and analyses.

Means for Solving the Problems

[0006] A ground motion waveform generation method according to an embodiment of the present invention is a ground motion waveform generation method for generating ground motion waveform data representing the time history waveform of ground motion caused by an earthquake using a computer, a receiving step of receiving a group of ground motion characteristic parameters representing the characteristics of the ground motion to be predicted; and a generating step of generating the ground motion waveform data for the ground motion to be predicted using a machine-learned learning model.

[0007] A learning model creation method according to an embodiment of the present invention is a learning model creation method for creating a learning model for generating ground motion waveform data representing the time history waveform of ground motion caused by an earthquake using a computer by machine learning, For each combination of a plurality of earthquakes and at least one observation point where ground motions caused by the plurality of earthquakes are observed, from a database that stores in association the group of ground motion characteristic parameters and the ground motion waveform data when the ground motion was observed, using the group of ground motion characteristic parameters as feature quantities and the ground motion waveform data associated with the group of ground motion characteristic parameters as target variables, an acquisition step of acquiring a plurality of pieces of learning data composed of the feature quantities and the target variables; A creation step of creating a ground motion waveform generation model as the learned learning model by learning the correlation between the feature quantities and the target variables by the machine learning based on the plurality of pieces of learning data acquired in the acquisition step.

[0008] A ground motion waveform generation device according to an embodiment of the present invention is A computer including a control unit that executes each step included in the above ground motion waveform generation method.

[0009] A learning model creation device according to an embodiment of the present invention is A computer including a control unit that executes each step included in the above learning model creation method.

Advantages of the Invention

[0010] According to the ground motion waveform generation method and the ground motion waveform generation device according to an embodiment of the present invention, ground motion waveform data for a ground motion to be predicted is generated using a learned learning model. Therefore, the ground motion waveform can be easily generated without performing complicated calculations and analyses.

[0011] According to the learning model creation method and the learning model creation device according to an embodiment of the present invention, a learning model that enables generation of ground motion waveform data for a ground motion to be predicted is created. Therefore, it is possible to create a learning model that enables easy generation of ground motion waveforms without performing complicated calculations and analyses.

Brief Description of the Drawings

[0012]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Embodiments for Carrying Out the Invention

[0013] Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings.

[0014] (First Embodiment) FIG. 1 is a schematic configuration diagram showing an example of the ground motion waveform generation system 1A according to the first embodiment. FIG. 2 is a block diagram showing an example of the ground motion waveform generation system 1A according to the first embodiment.

[0015] The ground motion waveform generation system 1A includes an observation data providing device 2A that collects ground motion waveform data as ground motion observation records obtained by observing ground motion caused by an earthquake at a plurality of observation points and provides the observation results to the outside, an analysis data providing device 2B that analyzes the earthquake source and scale based on the ground motion waveform data and provides the analysis results to the outside, a subsurface structure data providing device 2C that provides subsurface structure parameters related to the subsurface structure to the outside, and a ground motion simulation data providing device 2D that executes a ground motion simulation according to a predetermined simulation method and provides the simulation conditions and simulation results at that time to the outside as simulation data.

[0016] Further, the ground motion waveform generation system 1A includes a learning model creation device 3A that registers the data provided by the data providing devices 2A to 2D in the database 10 and creates a learning model 13 based on machine learning using the database 10, a ground motion waveform generation device 4A that generates a ground motion waveform using the learning model 13 generated by the learning model creation device 3A, and a network 5 that connects the devices.

[0017] In the present embodiment, the learning model 13 generated by the learning model creation device 3A will be described as being used in the ground motion waveform generation device 4A. However, for example, it can be widely used in society, from the analysis and interpretation of ground motion characteristics to the prediction of the behavior of buildings and structures, structural design, and even earthquake disaster prevention.

[0018] When an earthquake occurs, the observation data providing device 2A collects ground motion waveform data representing the three-component time history waveforms in the north-south direction, east-west direction, and vertical direction measured by seismographs (not shown) installed at a plurality of observation points, and provides the observation data including the ground motion waveform data and additional information such as the position of the observation point (e.g., the position of the observation point) to the outside. As the observation data, for example, data provided by the K-NET strong motion observation network of the National Research and Development Agency for Earthquake Science and Disaster Prevention (hereinafter referred to as "NIED") is used. In FIG. 1, 138 observation points (see FIG. 1) installed in one metropolis and six prefectures (Tokyo, Kanagawa, Chiba, Saitama, Ibaraki, Tochigi, Gunma) in the Kanto region are shown as the observation points, but the positions and numbers of the observation points are not limited to these.

[0019] The analysis data providing device 2B analyzes the earthquake source and scale based on the ground motion waveform data, and provides the analysis data including, for example, moment magnitude, Japan Meteorological Agency magnitude, epicenter position, focal depth, earthquake type, fault type, and focal mechanism solution to the outside as the analysis result. As the analysis data, for example, data provided by the broadband earthquake observation network F-net of NIED or the Japan Meteorological Agency is used.

[0020] The subsurface structure data providing device 2C provides subsurface structure data including, for example, earthquake basement depth, engineering basement depth, layer thickness, density, seismic wave propagation velocity, Q value, attenuation constant, etc. as subsurface structure parameters to the outside. As the subsurface structure data, for example, data provided by the earthquake hazard station J-SHIS of NIED is used.

[0021] Note that when an earthquake occurs, the observation data providing device 2A, the analysis data providing device 2B, and the underground structure data providing device 2C may provide the data to be provided regarding the earthquake to the learning model creation device 3A in real time, or when receiving a data request from the learning model creation device 3A, they may provide the data to be provided regarding the request (which may be the data to be provided regarding a plurality of earthquakes that meet predetermined conditions among the earthquakes that have occurred in the past) to the learning model creation device 3A. In the present embodiment, the data providing devices 2A to 2C are described as three separate devices, but the present invention is not limited to this, and they may be configured as one device, or other data providing devices may be further added.

[0022] The ground motion simulation data providing device 2D analyzes the ground motion caused by an earthquake or when assuming that a virtual earthquake has occurred according to a predetermined simulation method, and provides the simulation conditions and simulation results at that time to the outside as simulation data. At that time, any method can be adopted as the predetermined simulation method, and a plurality of methods may be adopted. In addition, the ground motion simulation data providing device 2D may provide simulation data by executing a ground motion simulation in its own device, or may provide simulation data when a ground motion simulation is executed by another device.

[0023] Note that when a new ground motion simulation is executed, the ground motion simulation data providing device 2D may provide the simulation data regarding the ground motion simulation to the learning model creation device 3A at any time, or when receiving a data request from the learning model creation device 3A, it may provide the simulation data regarding the request (when receiving simulation conditions from the learning model creation device 3A, the simulation data including the simulation results when a ground motion simulation is executed based on the simulation conditions may also be used) to the learning model creation device 3A.

[0024] The network 5 communicates various types of data and signals through wireless communication or wired communication, and any communication standard can be used.

[0025] (Regarding the learning model creation device 3A) The learning model creation device 3A collects the provided data (observation data, analysis data, underground structure data, simulation data) provided by the data providing devices 2A to 2D and registers it in the database 10. The learning model creation device 3A uses the database 10 to execute machine learning algorithms such as gradient boosting trees, random forests, neural networks, convolutional neural networks (CNNs), diffusion models, and adversarial generative networks (GANs), for example, to create the learning model 13 as a trained model of machine learning.

[0026] The learning model creation device 3A is composed of a general-purpose or dedicated computer. As shown in FIG. 2, it includes a storage unit 30 composed of an HDD, a memory, etc., a control unit 31 composed of a processor such as a CPU and a GPU, a communication unit 32 which is a communication interface with the network 5, an input unit 33 composed of a keyboard, a mouse, etc., and a display unit 34 composed of a display, a touch panel, etc.

[0027] The storage unit 30 stores the database 10 in which the data provided by the data providing devices 2A to 2D is registered and updated, the learning model 13 which is a trained model, and the learning model creation program 300A which controls the operation of the learning model creation device 3A to implement the learning model creation method. In this embodiment, the learning model 13 consists of a reference waveform output model 13A and a correction value output model 13B. Note that the database 10 may be stored in an external storage device instead of the storage unit 30. In that case, the learning model creation device 3A may communicate with the external storage device via the network 5 and access the database 10.

[0028] By executing the learning model creation program 300A, the control unit 31 functions as a DB management unit 310, a first acquisition unit 311A, a second acquisition unit 311B, a first creation unit 312A, and a second creation unit 312B.

[0029] FIG. 3 is a functional explanatory diagram showing an example of the learning model creation device 3A and the learning model creation method according to the first embodiment. FIG. 4 is a data configuration diagram showing an example of the database 10.

[0030] (Regarding the database management process by the DB management unit 310) Based on the provided data (observation data, analysis data, underground structure data, simulation data) provided by the data providing devices 2A to 2D, the DB management unit 310 collects the ground motion data 11 and registers it in the database 10, thereby managing the database 10.

[0031] As shown in FIG. 4, in the database 10, for each combination of a plurality of earthquakes and at least one observation point where ground motions caused by the plurality of earthquakes were observed, a group of ground motion characteristic parameters when the ground motion was observed at the observation point, and a plurality of ground motion data 11 associated with the ground motion waveform data are registered and stored. When the simulation data includes simulation conditions and simulation results when calculating the ground motion caused by an actual earthquake or a virtual earthquake by ground motion simulation, by treating the simulation conditions as a group of ground motion characteristic parameters and the simulation results as ground motion waveform data, the simulation data is registered and stored in the database 10 as the ground motion data 11.

[0032] The group of ground motion characteristic parameters consists of various ground motion characteristic parameters that describe the characteristics of ground motion. The characteristics of ground motion include, for example, the source characteristics and propagation characteristics of ground motion, and further include site characteristics, azimuth characteristics, ground motion observation characteristics, and ground motion prediction characteristics. In this embodiment, the ground motion characteristic parameters are obtained by classifying and recording the provided data provided by the data providing devices 2A to 2D according to the above-mentioned characteristics of ground motion. Hereinafter, the source characteristic parameters related to the source characteristics included in the ground motion characteristic parameters, the propagation characteristic parameters related to the propagation characteristics, the site characteristic parameters related to the site characteristics, the azimuth characteristic parameters related to the azimuth characteristics, the observation characteristic parameters related to the ground motion observation characteristics, and the prediction characteristic parameters related to the ground motion prediction characteristics will be described.

[0033] The source characteristic parameters related to the source characteristics are, for example, at least one of magnitude (moment magnitude Mw, Japan Meteorological Agency magnitude MJ, etc.), source depth H, source location (latitude lat_eq, longitude lon_eq), earthquake type Type (intracontinental crustal earthquake, plate boundary earthquake, slab earthquake), fault type Mech (normal fault, reverse fault, strike-slip fault), source mechanism solution (strike Strike1, dip angle dip1, slip angle rake1), conjugate solution of the source mechanism solution (strike Strike2, dip angle dip2, slip angle rake2), six moment tensors Mxx, Mxy, Mxz, Myy, Myz, Mzz, and the source fracture propagation effect coefficient Dir, etc. In the example of FIG. 4, as the source characteristics, the moment magnitude Mw, the Japan Meteorological Agency magnitude MJ, the source depth H, the latitude and longitude of the source, the source mechanism solution, and the conjugate solution of the source mechanism solution, the moment tensor, and the source fracture propagation effect coefficient are illustrated.

[0034] The propagation characteristic parameters related to the propagation characteristics are, for example, at least one of the source distance X, the shortest fault distance, the epicentral distance, and the determination flag for passing through the volcanic front. In the example of FIG. 4, as the propagation characteristics, the source distance X and the determination flag Xvflg for passing through the volcanic front are illustrated, and the source distance X is calculated as the distance between the observation point position and the source.

[0035] The site characteristic parameters related to site characteristics are, for example, at least one of the observation point position (latitude lat_site, longitude lon_site), microtopography classification JCODE, seismic bedrock depth, engineering bedrock depth, layer thickness, density, seismic wave propagation velocity, Q value, and attenuation constant. In the example of FIG. 4, as site characteristics, the latitude and longitude of the observation point, the S-wave velocity VS1 of the uppermost layer, the average S-wave velocity AVS10 of the upper 10 m of the surface layer, the average S-wave velocity AVS30 of the upper 30 m of the surface layer, and the seismic bedrock depth Dbase are shown. The seismic bedrock depth Dbase is the depth of the lower surface of the 28th layer of the deep ground model of the mesh including the target observation point position published by the Earthquake Research Institute's seismic hazard station J-SHIS (equal to the depth of the upper surface of the 29th layer corresponding to the seismic bedrock in the same model with a P-wave velocity of 5000 m / s and an S-wave velocity of 2700 m / s).

[0036] The azimuth characteristic parameters related to azimuth characteristics are, for example, the epicenter azimuth Λ indicating the azimuth where the epicenter is located with respect to the observation point. Therefore, the epicenter azimuth Λ is calculated as the azimuth where the epicenter position exists with respect to the observation point position. At that time, the epicenter azimuth Λ is defined clockwise with true north as 0°, and since it becomes a discontinuous quantity across true north, in the example of FIG. 4, the case of using a pair of sin Λ and cos Λ representing the epicenter azimuth Λ as azimuth characteristics is shown.

[0037] The observation characteristic parameters related to ground motion observation characteristics are, for example, the direction component Comp of the ground motion indicating that the time history waveform represented by the ground motion waveform data is in any of the north-south direction, east-west direction, and vertical direction. In the example of FIG. 4, the ground motion observation characteristics are omitted.

[0038] The prediction characteristic parameters related to ground motion prediction characteristics are, for example, the prediction result MF13 of the ground motion index based on the past ground motion waveform generation formula f. In the example of FIG. 4, the prediction result MF13 of the ground motion index is shown, and the prediction result MF13 is calculated by substituting, for example, four types of ground motion characteristic parameter groups, namely, the moment magnitude Mw, the focal depth H, the source distance X, and the earthquake type Type, into the ground motion waveform generation formula f.

[0039] The ground motion waveform data is, for example, the time history waveforms of acceleration, velocity, and displacement for three components in the north-south direction, east-west direction, and vertical direction. The ground motion waveform data is evaluated and analyzed by a predetermined evaluation method to obtain various ground motion indices. The ground motion waveform data according to the present embodiment is the time history waveforms for the horizontal two components (north-south direction and east-west direction) observed at 138 observation points (see FIG. 1) installed in six prefectures and one metropolis in the Kanto region (Tokyo, Kanagawa, Chiba, Saitama, Ibaraki, Tochigi, Gunma) of the strong motion observation network K-NET. Note that the positions and numbers of the observation points are not limited to these.

[0040] Any data format can be used for the ground motion waveform data. The ground motion waveform data may be, for example, in an image data format or a point sequence data format.

[0041] The ground motion index is an index obtained from the ground motion waveform data and includes at least one of the amplitude characteristics, period characteristics, and time history characteristics of the ground motion.

[0042] The amplitude characteristic is at least one of the maximum acceleration PGA, maximum velocity, and maximum displacement of the ground motion obtained from the ground motion waveform data (time history waveforms of acceleration, velocity, and displacement). The amplitude characteristic according to the present embodiment is the maximum acceleration PGA.

[0043] The period characteristic is a response value for at least one period in a response spectrum or Fourier spectrum obtained from the ground motion waveform data (time history waveforms of acceleration, velocity, and displacement). The response spectrum is, for example, an acceleration response spectrum, pseudo velocity response spectrum pSv, velocity response spectrum, and displacement response spectrum for a predetermined damping constant (e.g., 5%). The Fourier spectrum is, for example, an acceleration Fourier spectrum, velocity Fourier spectrum, and displacement Fourier spectrum. The period characteristics according to the present embodiment are five pseudo velocity response spectra pSv(0.1s), pSv(0.5s), pSv(1s), pSv(3s), pSv(5s) with a damping constant of 5% at each of the periods of 0.1 second, 0.5 second, 1 second, 3 seconds, and 5 seconds.

[0044] The time-dependent characteristics are, for example, the response duration for at least one period in the response duration spectrum obtained from seismic motion waveform data (acceleration, velocity, and displacement time history waveforms). The response duration spectrum is, for example, the acceleration response duration spectrum for a predetermined damping constant (e.g., 5%), the velocity response duration spectrum TSv, and the displacement response duration spectrum, etc. The time-dependent characteristics according to the present embodiment are five of the velocity response duration spectra TSv(0.1s), TSv(0.5s), TSv(1s), TSv(3s), and TSv(5s) with a damping constant of 5% at each of the periods of 0.1 second, 0.5 second, 1 second, 3 seconds, and 5 seconds. Note that the parameters defining the start and end of the response duration are p1 = 0.03 and p2 = 0.95.

[0045] (Regarding the first acquisition step by the first acquisition unit 311A and the first learning data 12A) As shown in FIG. 3, the first acquisition unit 311A acquires a plurality of first learning data 12A from the database 10. The first learning data 12A is composed of the first group of seismic motion characteristic parameters included in the group of seismic motion characteristic parameters as feature amounts and the reference waveform of the time history waveform represented by the seismic motion waveform data associated with the group of seismic motion characteristic parameters as target variables. The first learning data 12A is data used as learning data (training data), verification data, and test data in supervised learning.

[0046] (Regarding the second acquisition step by the second acquisition unit 311B and the second learning data 12B) As shown in FIG. 3, the second acquisition unit 311B acquires a plurality of pieces of second learning data 12B from the database 10. The second learning data 12B has, as a feature amount, a second group of earthquake motion characteristic parameters included in the group of earthquake motion characteristic parameters, and has, as an objective variable, a correction value of a reference waveform of a time history waveform represented by earthquake motion waveform data associated with the group of earthquake motion characteristic parameters, and is composed of the feature amount and the objective variable. The second learning data 12B is data used as learning data (training data), verification data, and test data in supervised learning.

[0047] The first group of earthquake motion characteristic parameters and the second group of earthquake motion characteristic parameters include earthquake motion characteristic parameters that overlap with each other. In the present embodiment, as shown in FIG. 3, the first learning data 12A includes, as a feature amount, a first group of earthquake motion characteristic parameters composed of two types of earthquake motion characteristic parameters, the Japan Meteorological Agency magnitude MJ and the hypocentral distance X, but is not limited to these examples. Further, as shown in FIG. 3, the second learning data 12B includes, as a feature amount, a second group of earthquake motion characteristic parameters composed of five types of earthquake motion characteristic parameters, the Japan Meteorological Agency magnitude MJ, the focal depth H, the hypocentral distance X, the average S-wave velocity AVS30 in the upper 30 m of the surface layer, and the depth Dbase of the upper surface of the earthquake base, but is not limited to these examples. That is, in the present embodiment, the Japan Meteorological Agency magnitude MJ and the hypocentral distance X correspond to earthquake motion characteristic parameters that overlap with each other, but are not limited to these examples. Note that the first group of earthquake motion characteristic parameters and the second group of earthquake motion characteristic parameters may not include earthquake motion characteristic parameters that overlap with each other.

[0048] The reference waveform of the time history waveform is shaped from the time history waveform, for example, by performing a normalization process of normalizing the maximum amplitude value in the amplitude axis direction in the time history waveform to a predetermined reference amplitude value and normalizing the maximum continuous time in the time axis direction in the time history waveform to a predetermined reference continuous time. At that time, as the correction value of the reference waveform, for example, an amplitude value ratio indicating the ratio of the maximum amplitude value to the reference amplitude value in the amplitude axis direction and a continuous time ratio indicating the ratio of the maximum continuous time to the reference continuous time in the time axis direction are used.

[0049] The reference waveform of the time history waveform and the correction value of the reference waveform are obtained by performing normalization processing on the ground motion waveform represented by the ground motion waveform data registered in the database 10 by at least one of the first acquisition unit 311A and the second acquisition unit 311B. Note that the reference waveform of the time history waveform and its correction value may be obtained at the time when the ground motion waveform data is registered in the database 10 and registered in the database 10. In that case, the first acquisition unit 311A and the second acquisition unit 311B may obtain the reference waveform of the time history waveform and the correction value of the reference waveform from the database 10.

[0050] (Regarding the first creation step by the first creation unit 312A and the reference waveform output model 13A) As shown in FIG. 3, the first creation unit 312A learns the correlation between the feature amount and the target variable by machine learning based on a plurality of first learning data 12A acquired by the first acquisition unit 311A, and creates a reference waveform output model 13A as a learned learning model 13, and stores it in the storage unit 30.

[0051] The reference waveform output model 13A is generated by learning the correlation between the first ground motion characteristic parameter group as the feature amount and the reference waveform of the time history waveform as the target variable by machine learning.

[0052] (Regarding the second creation step by the second creation unit 312B and the correction value output model 13B) As shown in FIG. 3, the second creation unit 312B learns the correlation between the feature amount and the target variable by machine learning based on a plurality of second learning data 12B acquired by the second acquisition unit 311B, and creates a correction value output model 13B as a learned learning model 13, and stores it in the storage unit 30.

[0053] The correction value output model 13B is generated by using, as feature quantities, a second group of earthquake motion characteristic parameters and, as an objective variable, the correction value of the reference waveform, and learning the correlation between the feature quantities and the objective variable by machine learning.

[0054] As a machine learning algorithm in machine learning when creating the learning model 13 (in this embodiment, the reference waveform output model 13A and the correction value output model 13B), for example, gradient boosting tree, random forest, neural network, convolutional neural network (CNN), diffusion model, generative adversarial network (GAN), etc. can be used. At that time, the machine learning algorithm may be appropriately selected according to the feature quantities and objective variables of the reference waveform output model 13A and the correction value output model 13B. Also, different machine learning algorithms may be used for the reference waveform output model 13A created by the first creation unit 312A and the correction value output model 13B created by the second creation unit 312B, or the same machine learning algorithm may be used.

[0055] In the database management process by the DB management unit 310, when earthquake motion data 11 is registered in the database 10 at any time and a predetermined number of earthquake motion data 11 is newly accumulated, based on a plurality of earthquake motion data 11 including the new earthquake motion data 11, the first acquisition process by the first acquisition unit 311A, the second acquisition process by the second acquisition unit 311B, the first creation process by the first creation unit 312A, and the second creation process by the second creation unit 312B may be performed again to update the learning model 13.

[0056] (Regarding the earthquake motion waveform generation device 4A) The ground motion waveform generation device 4A uses the learning models 13 (in this embodiment, the reference waveform output model 13A and the correction value output model 13B) generated by the learning model creation device 3A to generate ground motion waveform data for the ground motion to be predicted, and outputs the result to an output medium such as a storage medium, a display medium, or a paper medium. That is, the ground motion waveform generation device 4A functions as a device that generates a machine-learned Green's function by using the machine-learned learning model 13.

[0057] Similar to the learning model creation device 3A, the ground motion waveform generation device 4A is composed of a general-purpose or dedicated computer. As shown in FIG. 2, it includes a storage unit 40 composed of an HDD, a memory, etc., a control unit 41 composed of a processor such as a CPU or a GPU, a communication unit 42 which is a communication interface with the network 5, an input unit 43 composed of a keyboard, a mouse, etc., and a display unit 44 composed of a display, a touch panel, etc.

[0058] The storage unit 40 stores the machine-learned learning model 13 created by the learning model creation device 3A and a ground motion waveform generation program 400A that controls the operation of the ground motion waveform generation device 4A to implement the ground motion waveform generation method.

[0059] By executing the ground motion waveform generation program 400B, the control unit 41 functions as a reception unit 410, a generation unit 411A, and an output processing unit 412.

[0060] FIG. 5 is a functional explanatory diagram showing an example of the ground motion waveform generation device 4A and the ground motion waveform generation method according to the first embodiment.

[0061] (Regarding the reception process by the reception unit 410) The reception unit 410 receives a group of earthquake motion characteristic parameters to be predicted. Specifically, the reception unit 410 receives, via the input unit 33, for example, a group of earthquake motion characteristic parameters of an earthquake (hereinafter referred to as the "assumed earthquake") that the user of the earthquake motion waveform generation device 4A assumes as the prediction target, and also receives, via the input unit 33, the prediction point position indicating the position of the point where it is desired to predict the magnitude of the earthquake motion generated by the assumed earthquake. Note that the prediction point position may be any position, or may be the same as the observation point position. Also, there may be a plurality of prediction point positions, for example, each grid point at a predetermined grid interval (for example, 5 km interval).

[0062] (Regarding the generation process by the generation unit 411A) The generation unit 411A inputs the first group of earthquake motion characteristic parameters included in the group of earthquake motion characteristic parameters received by the reception unit 410 into the reference waveform output model 13A, and corrects the reference waveform output from the reference waveform output model 13A with the correction value of the reference waveform output by inputting the second group of earthquake motion characteristic parameters included in the group of earthquake motion characteristic parameters received by the reception unit 410 into the correction value output model 13B, thereby generating earthquake motion waveform data for the earthquake motion to be predicted.

[0063] At this time, the generation unit 411A generates earthquake motion waveform data for the earthquake motion to be predicted by expanding and contracting the reference waveform in the amplitude axis direction by the amplitude value ratio as the correction value of the reference waveform. That is, the reference waveform is expanded and contracted in the amplitude axis direction by multiplying by the amplitude value ratio.

[0064] Also, the generation unit 411A generates earthquake motion waveform data for the earthquake motion to be predicted by expanding and contracting the reference waveform in the time axis direction by the duration ratio as the correction value of the reference waveform. That is, the reference waveform is expanded and contracted in the time axis direction by multiplying by the duration ratio.

[0065] Note that when the reception unit 410 receives a plurality of assumed earthquakes or a plurality of predicted point positions, and thus receives a plurality of groups of seismic motion characteristic parameters as prediction targets, the generation unit 411B generates seismic motion waveform data for each of the plurality of groups of seismic motion characteristic parameters, respectively.

[0066] In addition, when the generation unit 411A corrects the reference waveform with the correction value of the reference waveform, the reference waveform may be expanded or contracted only in the amplitude axis direction by the amplitude value ratio, or the reference waveform may be expanded or contracted only in the time axis direction by the duration ratio. In that case, the correction value output model 13B in the learning model 13 may output either the amplitude value ratio or the duration ratio as the correction value of the reference waveform. Also, the normalization process for the time history waveform may be performed only by normalizing in the amplitude axis direction or only by normalizing in the time axis direction.

[0067] (Regarding the output processing step by the output processing unit 412) The output processing unit 412 outputs the seismic motion waveform data generated by the generation unit 411A to the output medium. For example, when the output medium is a storage medium such as the storage unit 40, the output processing unit 412 stores and outputs the seismic motion waveform data (output data) to the storage medium. Also, when the output medium is a display medium such as the display unit 44, the output processing unit 412 generates display data (output data) for display on the display medium and outputs it to the display medium. Further, when the output medium is a paper medium, the output processing unit 412 generates print data (output data) for printing on the paper medium and outputs it to the paper medium.

[0068] As described above, according to the learning model creation device 3A and the learning model creation method according to the present embodiment, it is possible to create a reference waveform output model 13A and a correction value output model 13B as a learning model 13 that can generate seismic motion waveform data for the seismic motion to be predicted.

[0069] Further, according to the ground motion waveform generation device 4A and the ground motion waveform generation method according to the present embodiment, by using the reference waveform output model 13A and the correction value output model 13B as the learning model 13 created by the learning model creation device 3A and the learning model creation method, ground motion waveform data for the ground motion to be predicted can be generated.

[0070] At this time, the generation unit 411A generates ground motion waveform data for the ground motion to be predicted by correcting the reference waveform output from the reference waveform output model 13A with the correction value of the reference waveform output from the correction value output model 13B. That is, since the reference waveform is a waveform standardized with respect to at least one of the amplitude axis direction and the time axis direction, it is possible to generate a reference waveform that includes the characteristics of the waveform shape of the ground motion waveform while suppressing the influence of the maximum amplitude value and the maximum duration. Further, since the correction value of the reference waveform is a correction value for expanding and contracting the reference waveform with respect to at least one of the amplitude axis direction and the time axis direction, the ground motion waveform can be reproduced with high accuracy by deforming the reference waveform into the ground motion waveform.

[0071] (Second Embodiment) FIG. 6 is a block diagram showing an example of the ground motion waveform generation system 1B according to the second embodiment. The ground motion waveform generation system 1B according to the second embodiment is different from the first embodiment in that a ground motion waveform generation model 13C is used as the learning model 13 instead of the reference waveform output model 13A and the correction value output model 13B. Since the other basic configurations and operations are the same as those of the first embodiment, the differences between the two will be mainly described below.

[0072] The ground motion waveform generation model 13C is generated by learning the correlation between the ground motion characteristic parameter group as a feature amount and the ground motion waveform data as a target variable by machine learning.

[0073] (Regarding the learning model creation device 3B) As shown in FIG. 6, the storage unit 30 of the learning model creation device 3B stores a database 10, a learning model 13 that is a learned model (in this embodiment, a ground motion waveform generation model 13C), and a learning model creation program 300B that controls the operation of the learning model creation device 3B to implement a learning model creation method.

[0074] By executing the learning model creation program 300B, the control unit 31 of the learning model creation device 3B functions as a DB management unit 310, an acquisition unit 311, and a creation unit 312.

[0075] FIG. 7 is a functional explanatory diagram showing an example of the learning model creation device 3B and the learning model creation method according to the second embodiment.

[0076] As shown in FIG. 7, the acquisition unit 311 acquires a plurality of pieces of learning data 12 from the database 10. The learning data 12 is composed of a group of ground motion characteristic parameters as feature amounts and ground motion waveform data associated with the group of ground motion characteristic parameters as target variables, and is used as learning data (training data), verification data, and test data in supervised learning. In this embodiment, the learning data 12 includes, as feature amounts, a group of ground motion characteristic parameters consisting of five types of ground motion characteristic parameters: Japan Meteorological Agency magnitude MJ, focal depth H, epicentral distance X, average S-wave velocity AVS30 in the upper 30 m of the surface layer, and depth Dbase of the upper surface of the seismic bedrock, but is not limited to these examples.

[0077] As shown in FIG. 7, the creation unit 312 creates a ground motion waveform generation model 13C as a machine-learned learning model 13 based on the plurality of pieces of learning data 12 acquired by the acquisition unit 311 by learning the correlation between the feature amounts and the target variables by machine learning, and stores it in the storage unit 30.

[0078] (Regarding the ground motion waveform generation device 4B) As shown in FIG. 6, the storage unit 40 of the ground motion waveform generation device 4B stores a learning model 13 (in this embodiment, the ground motion waveform generation model 13C) created by the learning model creation device 3B and a ground motion waveform generation program 400B that controls the operation of the ground motion waveform generation device 4B to implement the ground motion waveform generation method.

[0079] The control unit 41 of the ground motion waveform generation device 4B functions as a reception unit 410, a generation unit 411B, and an output processing unit 412 by executing the ground motion waveform generation program 400B.

[0080] FIG. 8 is a functional explanatory diagram showing an example of the ground motion waveform generation device 4B and the ground motion waveform generation method according to the second embodiment.

[0081] The generation unit 411B generates ground motion waveform data for the ground motion to be predicted by inputting the group of ground motion characteristic parameters received by the reception unit 410 into the ground motion waveform generation model 13C and outputting the ground motion waveform data from the ground motion waveform generation model 13C. At that time, when the reception unit 410 receives a plurality of assumed earthquakes or receives a plurality of prediction point positions, and thus receives a plurality of groups of ground motion characteristic parameters as the prediction target, the generation unit 411B generates ground motion waveform data for each of the plurality of groups of ground motion characteristic parameters.

[0082] As described above, according to the learning model creation device 3B and the learning model creation method according to this embodiment, a ground motion waveform generation model 13C can be created as a learning model 13 that enables generation of ground motion waveform data for the ground motion to be predicted.

[0083] Also, according to the ground motion waveform generation device 4B and the ground motion waveform generation method according to this embodiment, by using the ground motion waveform generation model 13C as the learning model 13 created by the learning model creation device 3B and the learning model creation method, ground motion waveform data for the ground motion to be predicted can be generated.

[0084] (Other Embodiments) As described above, the embodiments of the present invention have been explained. However, the present invention is not limited to the above-described embodiments, and can be appropriately modified without departing from the technical idea of the present invention.

[0085] For example, in each of the above embodiments, the region targeted by the learning model 13 can be appropriately set. For example, the range and shape of the targeted region can be arbitrarily changed. Further, the learning model 13 may target not only a region but also an arbitrary point, or a group of points grouped according to a predetermined classification criterion.

[0086] Also, a predetermined region may be divided into a plurality of sections, for example, in a mesh shape, and the learning model creation devices 3A and 3B may create the learning model 13 for each section. Then, the generation units 411A and 411B of the ground motion waveform generation devices 4A and 4B may generate ground motion waveform data for the ground motion to be predicted for each section using the learning model 13 for each section. Further, the generation units 411A and 411B may synthesize the time history waveforms represented by the ground motion waveform data for each section according to the time difference of the propagation time when the seismic wave recorded as the time history waveform propagates, and generate composite waveform data representing the synthesized time history waveform. Since the ground motion data 11 of small and medium earthquakes with a magnitude of about M6 or less is relatively abundant, by using the learning model 13 for each section created with reference to the database 10 in which these ground motion data 11 are registered, the ground motion waveform of a large earthquake in the magnitude M7 - M8 class can be generated as composite waveform data.

[0087] In each of the above-described embodiments, the learning model creation programs 300A and 300B and the ground motion waveform generation programs 400A and 400B have been described as being stored in the storage units 30 and 40, respectively. However, they may be provided by being recorded on a computer-readable recording medium such as a CD-ROM, DVD, or USB memory in an installable or executable file format. Further, the learning model creation programs 300A and 300B and the ground motion waveform generation programs 400A and 400B may be provided by being stored on a computer connected to a network such as the Internet and downloaded via the network.

Explanation of Reference Numerals

[0088] 1A, 1B... Ground motion waveform generation system, 2A... Observation data providing device, 2B... Analysis data providing device, 2C... Subsurface structure data providing device, 2D... Ground motion simulation data providing device, 3A, 3B... Learning model creation device, 4A, 4B... Ground motion waveform generation device, 5... Network, 10... Database, 11... Ground motion data, 12... Learning data, 12A... First learning data, 12B... Second learning data, 13... Learning model, 13A... Reference waveform output model, 13B... Correction value output model, 13C... Ground motion waveform generation model, 30... Storage unit, 31... Control unit, 32... Communication unit, 33... Input unit, 34... Display unit, 40... Storage unit, 41... Control unit, 42... Communication unit, 43... Input unit, 44... Display unit, 300A, 300B... Learning model creation program, 310... DB management unit, 311... Acquisition unit, 311A... First acquisition unit, 311B... Second acquisition unit, 312... Creation unit, 312A... First creation unit, 312B... Second creation unit, 400A, 400B... Ground motion waveform generation program, 410... Reception unit, 411A, 411B... Generation unit, 412... Output processing unit

Claims

1. A ground motion waveform generation method for generating ground motion waveform data representing a time history waveform of ground motion caused by an earthquake using a computer, comprising: a receiving step of receiving a group of ground motion characteristic parameters representing the characteristics of the ground motion to be predicted; a generating step of generating the ground motion waveform data for the ground motion to be predicted using a machine-learned learning model; A ground motion waveform generation method.

2. The learning model is a ground motion waveform generation model generated by learning the correlation between the group of ground motion characteristic parameters as feature amounts and the ground motion waveform data as target variables by machine learning, The generating step is generating the ground motion waveform data for the ground motion to be predicted by inputting the group of ground motion characteristic parameters received in the receiving step into the ground motion waveform generation model and outputting the ground motion waveform data from the ground motion waveform generation model; The ground motion waveform generation method according to Claim 1.

3. The learning model is a reference waveform output model generated by learning the correlation between a first group of ground motion characteristic parameters included in the group of ground motion characteristic parameters as feature amounts and a reference waveform of the time history waveform as a target variable by machine learning, and a correction value output model generated by learning the correlation between a second group of ground motion characteristic parameters included in the group of ground motion characteristic parameters as feature amounts and a correction value of the reference waveform as a target variable by machine learning, The generating step is generating the ground motion waveform data for the ground motion to be predicted by correcting the reference waveform output from the reference waveform output model by inputting the first group of ground motion characteristic parameters included in the group of ground motion characteristic parameters received in the receiving step into the reference waveform output model with the correction value output by inputting the second group of ground motion characteristic parameters included in the group of ground motion characteristic parameters received in the receiving step into the correction value output model; The ground motion waveform generation method according to Claim 1.

4. The generating step is generating the ground motion waveform data for the ground motion to be predicted by stretching and shrinking the reference waveform in the amplitude axis direction by the correction value. The ground motion waveform generation method according to claim 3.

5. The generation step is generating the ground motion waveform data for the ground motion to be predicted by stretching and shrinking the reference waveform in the time axis direction by the correction value. The ground motion waveform generation method according to claim 3.

6. The first group of ground motion characteristic parameters and the second group of ground motion characteristic parameters include ground motion characteristic parameters that overlap with each other. The ground motion waveform generation method according to claim 3.

7. The learning model is created for each area into which a predetermined area is divided, The generation step is generating the ground motion waveform data for the ground motion to be predicted for each area using the learning model for each area, synthesizing the time history waveforms respectively represented by the ground motion waveform data for each area according to the time difference of the propagation time when seismic waves propagate, and generating composite waveform data representing the synthesized time history waveform. The ground motion waveform generation method according to claim 1.

8. A learning model creation method for creating, by machine learning, a learning model for generating ground motion waveform data representing the time history waveform of ground motion caused by an earthquake using a computer, comprising: an acquisition step of acquiring a plurality of pieces of learning data composed of the feature amount and the objective variable, with the group of ground motion characteristic parameters as the feature amount and the ground motion waveform data associated with the group of ground motion characteristic parameters as the objective variable, from a database that stores in association the group of ground motion characteristic parameters and the ground motion waveform data for each combination of a plurality of earthquakes and at least one observation point where the ground motion caused by the plurality of earthquakes is observed; a creation step of creating a ground motion waveform generation model as the learned learning model by learning the correlation between the feature amount and the objective variable by the machine learning based on the plurality of pieces of learning data acquired in the acquisition step. Learning model creation method.

9. A learning model creation method for creating, by machine learning, a learning model for generating ground motion waveform data representing the time history waveform of ground motion caused by an earthquake using a computer, comprising: For each combination of a plurality of earthquakes and at least one observation point where ground motions caused by the plurality of earthquakes are observed, from a database that stores the ground motion characteristic parameter groups at the time when the ground motions are observed and the ground motion waveform data in association with each other, using the first ground motion characteristic parameter group included in the ground motion characteristic parameter groups as a feature quantity, and using the reference waveform of the time history waveform represented by the ground motion waveform data associated with the ground motion characteristic parameter groups as an objective variable, a first acquisition step of acquiring a plurality of first learning data constituted by the feature quantity and the objective variable; From the database, using the second ground motion characteristic parameter group included in the ground motion characteristic parameter groups as a feature quantity, and using the correction value of the reference waveform of the time history waveform represented by the ground motion waveform data associated with the ground motion characteristic parameter groups as an objective variable, a second acquisition step of acquiring a plurality of second learning data constituted by the feature quantity and the objective variable; A first creation step of creating a reference waveform output model as the learned learning model by learning the correlation between the feature quantity and the objective variable constituting the first learning data by the machine learning based on the plurality of first learning data acquired in the first acquisition step; A second creation step of creating a correction value output model as the learned learning model by learning the correlation between the feature quantity and the objective variable constituting the second learning data by the machine learning based on the plurality of second learning data acquired in the second acquisition step, including: A learning model creation method.

10. A computer, comprising a control unit that executes each step included in the ground motion waveform generation method according to any one of Claims 1 to 7. A ground motion waveform generation device.

11. A computer, comprising a control unit that executes each step included in the learning model creation method according to Claim 8 or Claim 9. A learning model creation device.

Images