TWI687711B - Epicenter distance estimation device, epicenter distance estimation method, and computer-readable recording medium - Google Patents

Abstract

無。

Description

Epicenter distance estimation device, epicenter distance estimation method, and computer-readable recording medium

The invention relates to an epicenter distance estimation device and an epicenter distance estimation method used for estimating the epicenter distance when an earthquake occurs, and also relates to a computer-readable recording medium for recording a program for implementing the aforementioned method.

When an earthquake occurs, in order to estimate the magnitude of each location and the arrival time of the main vibration, the epicentral distance must be quickly determined. Generally, the depth of the epicentral distance will be determined by the magnitude detected by the seismometers at multiple locations.

However, if the location of the source is on the seabed or in an area where the seismometers are installed at a low density, it takes too much time to obtain the seismicity measured by a plurality of seismometers, which makes it impossible to determine the epicenter distance in time. Therefore, in recent years, techniques have been developed to determine the epicenter distance using only the magnitude measured by a single seismometer.

As this technique, there is known a technique for estimating the epicentral distance by using "the intensity of seismic waveform data when the earthquake arrives is closer to the source, and the farther the source is, the smoother the earthquake is." Literature 1).

Specifically, in the technique disclosed in Patent Document 1, the absolute value of the time series data obtained from the seismometer is y(t), the time is t, and the time when the seismometer detects an earthquake is t= 0, the waveform shape of the initial part of the earthquake obtained by the seismometer is fitted by the function shown in the following mathematical formula 1. In the following mathematical formula 1, A is a parameter related to the maximum amplitude of the initial motion part, and B is a parameter related to the time change of the initial motion amplitude of the seismic waveform. Moreover, in fact, in the fitting process, one must rely on human experience and intuition in order to reflect the complex individual location attributes in the parameters A and B of unknown factors and methods.

[Mathematical Formula 1] y ( t )= Bte ^{- At}

Then, in the technique disclosed in Patent Document 1, the parameters A and B are obtained by the least square method. Among them, it can be seen that although the parameter B and the epicentral distance have a mutual relationship, this mutual relationship will not be affected by the magnitude of the earthquake. Therefore, if the relationship between the parameter B and the epicentral distance is formulated in advance, the epicentral distance can be determined by calculating the parameter B from the waveform shape of the initial part of the earthquake using Mathematical Formula 1. According to the technique disclosed in Patent Document 1, the epicentral distance can be quickly determined from the waveform shape of the initial part of the earthquake.

[Previous Technical Literature]

[Patent Literature]

[Patent Document 1] Japanese Unexamined Patent Publication No. 2002-277557

However, the technology disclosed in Patent Document 1 may cause the situation that the coefficients A and B cannot be calculated under certain circumstances, and there is also a problem of insufficient reliability. In addition, the technique disclosed in Patent Document 1 has a problem that it is not easy to shorten the time required to calculate the epicenter distance.

An example of an object of the present invention is to provide an epicenter distance estimation device, an epicenter distance estimation method, and a computer-readable recording medium that can solve the above problems and stably calculate the epicenter distance and shorten the calculation time.

In order to achieve the above object, a type of epicenter distance estimation device of the present invention is characterized by: an earthquake information acquisition unit that acquires waveform data of an earthquake that has occurred; and an estimation processing unit that learns the waveform data of an earthquake and The learning model obtained from the relationship between epicenter distances applies the aforementioned waveform data to estimate the epicenter distance.

In addition, in order to achieve the above object, a type of epicenter distance estimation method of the present invention is characterized by the following steps: (a) a step of acquiring waveform data of an existing earthquake; (b) learning waveform data of an earthquake and The learning model obtained from the relationship between epicenter distances applies the aforementioned waveform data to estimate the epicenter distance.

In addition, in order to achieve the above purpose, a type of computer-readable recording medium of the present invention is characterized by recording a program that includes commands to perform the following steps in the computer: (a) Obtain the occurrence of an earthquake Waveform data steps; (b) For the learning model obtained by learning the relationship between the waveform data of the earthquake and the epicentral distance, the aforementioned waveform data has been applied to estimate the epicentral distance.

As described above, according to the present invention, the epicentral distance can be calculated stably, and the calculation time can be shortened.

10: Device for estimating the epicentral distance (implementation type 1)

11: Earthquake Information Acquisition Department

12: Estimation processing department

13: Learning Information Acquisition Department

14: Learning Department

15: Storage Department

16: learning model

20: Earthquake detection device

30: Integrated monitoring system for seismic activity

40: Epicenter distance estimation device (implementation type 2)

41: Waveform pre-processing section

110: computer

111: CPU

112: main memory

113: Storage device

114: Input interface

115: display controller

116: data reader/writer

117: Communication interface

118: Enter the machine

119: display device

120: Recording media

121: busbar

[FIG. 1] FIG. 1 is a block diagram showing a schematic configuration of an epicenter distance estimating device according to Embodiment 1 of the present invention.

[FIG. 2] FIG. 2 is a block diagram specifically showing the configuration of the epicenter distance estimating device according to Embodiment 1 of the present invention.

[FIG. 3] FIG. 3 is a diagram showing an example of input data and positive solution data for learning in Embodiment 1 of the present invention.

[FIG. 4] FIG. 4 is a flowchart showing the operation when the learning process of the epicenter distance estimating device in Embodiment 1 of the present invention is executed.

[Fig. 5] Fig. 5 is a flowchart showing the operation when the estimation process of the epicentral distance estimation device of Embodiment 1 of the present invention is executed.

[FIG. 6] FIG. 6 is a block diagram specifically showing the configuration of the epicenter distance estimating device according to Embodiment 2 of the present invention.

[FIG. 7] FIG. 7 is a flowchart showing the operation when the learning process of the epicentral distance estimation device of Embodiment 2 of the present invention is executed.

[FIG. 8] FIG. 8 is a flowchart showing the operation when the estimation process of the epicentral distance estimation device of Embodiment 2 of the present invention is executed.

[FIG. 9] FIG. 9 is a block diagram showing an example of a computer that realizes the epicenter distance estimation device of Embodiments 1 and 2 of the present invention.

(Implementation type 1)

Hereinafter, the epicenter distance estimation device, epicenter distance estimation method, and program according to Embodiment 1 of the present invention will be described with reference to FIGS. 1 to 5.

[Device configuration]

First, the schematic configuration of the epicenter distance estimation device of the first embodiment will be described using FIG. 1. FIG. 1 is a block diagram showing a schematic configuration of an epicenter distance estimation device according to Embodiment 1 of the present invention.

As shown in FIG. 1, the epicentral distance estimation device 10 of the first embodiment is an apparatus for estimating the epicentral distance from waveform data measured when an earthquake occurs. As shown in FIG. 1, the epicenter distance estimation device 10 includes an earthquake data acquisition unit 11 and an estimation processing unit 12.

The seismic data acquisition unit 11 acquires waveform data of an earthquake that has occurred. The estimation processing unit 12 applies the waveform data acquired by the seismic data acquisition unit 11 to the learning model, and estimates the epicenter distance. The learning model is obtained in advance by learning the relationship between the waveform data of the earthquake and the epicentral distance.

In this way, the first type of the present embodiment is different from the conventional technology, and the epicenter distance can be estimated without fitting the waveform data to the function, so the epicenter distance can be calculated stably. In addition, in the first embodiment, calculation processing by the least square method is unnecessary, so the calculation time can be shortened.

Next, the configuration of the epicenter distance estimation device of the first embodiment will be described more specifically using FIG. 2. FIG. 2 is a block diagram specifically showing the configuration of the epicenter distance estimating device according to Embodiment 1 of the present invention.

As shown in FIG. 2, in the first embodiment, the epicenter distance estimation device 10 is connected to the earthquake detection device 20 and the integrated monitoring system 30 such as the seismic activity via the network. Among them, the seismic detection device 20 is equipped with a seismometer. When the seismic wave is detected by the seismometer, the waveform data of the detected seismic wave is transmitted to the epicenter distance estimation device 10. In the first embodiment, the seismic detection device 20 is the source of waveform data acquisition by the seismic data acquisition unit 11.

In addition, in the example of FIG. 2, although only a single seismic detection device 20 is illustrated, the number of seismic detection devices 20 to which the epicenter distance estimation device 10 is connected is not particularly limited. However, the seismic detection device 20 as the source of the seismic data acquisition unit 11 may be any one of them.

The integrated monitoring system 30 for seismic activity and other systems in Japan is maintained by the Meteorological Agency. When an earthquake occurs, the system calculates the magnitude of the earthquake of the Meteorological Agency and then predicts the tsunami height based on the calculated magnitude of the earthquake of the Meteorological Agency. In addition, the integrated monitoring system 30 for earthquake activity and the like transmits the calculated magnitude of the earthquake of the Meteorological Agency and the predicted height of the tsunami to various media in the form of an emergency earthquake quick report.

In addition, in the first embodiment, the epicenter distance estimation device 10 inputs the estimated epicenter distance to the integrated monitoring system 30 such as seismic activity. Therefore, the integrated monitoring system 30 such as seismic activity uses the epicentral distance estimated by the epicentral distance estimation device 10 to calculate the earthquake scale and the predicted tsunami height of the Meteorological Agency.

In addition, as shown in FIG. 2, in the present embodiment 1, the epicenter distance estimation device 10 includes a learning data acquisition unit 13, a learning unit 14, and a storage unit 15 in addition to the seismic data acquisition unit 11 and the estimation processing unit 12 described above. Furthermore, FIG. 2 shows an example of the epicenter distance estimation device 10, and the learning data acquisition unit 13, the learning unit 14, and the storage unit 15 may be located in devices other than the epicenter distance estimation device 10.

The learning material acquisition unit 13 acquires the waveform data that becomes the input data in the learning of the learning unit 14 described later, and the epicentral distance that is also the positive solution data during the learning, and inputs it to the learning unit 14. Moreover, the source of the input data and correct solution data is not particularly limited to this.

The learning section 14 uses the waveform data of the earthquake as input data and the epicenter distance of the earthquake as positive solution data, and learns the relationship between the waveform data and the epicenter distance, and then generates a learning model 16 representing the learning result. In addition, the learning unit 14 stores the generated learning model 16 in the storage unit 15.

FIG. 3 is a diagram showing an example of input data and positive solution data for learning in the first embodiment. Figure 3 shows a plurality of waveform data with different epicenter distances. The waveform data shown in Figure 3 is the waveform data of earthquakes observed in the past. In addition, the epicentral distance corresponding to each waveform data is positive solution data. The learning unit 14 uses the waveform data shown in FIG. 3 as input data and the epicentral distance as positive solution data for learning.

In addition, in the present embodiment 1, as the positive solution data, the data published by the Meteorological Agency can be used. The data published by the Meteorological Agency contains the epicentral distance and source elements of each observation point. The detection value calculated by the chemical system (http://www.data.jma.go.jp/svd/eqev/data/bulletin/deck.html) is obtained. In addition, in the first embodiment, the input data and the correct data for learning are preferably calculated from earthquakes having a magnitude greater than a set value (for example, magnitude 4).

In addition, in the first embodiment, the learning unit 14 can construct a neural network by mechanical learning, for example, and then set the neural network as the learning model 16. Specifically, the learning unit 14 uses input data and positive solution data in a hierarchical neural network having an input layer, an intermediate layer, and an output layer to generate learning by adjusting the coupling value between nodes of adjacent layers model.

In addition, in the present embodiment 1, the "learning" performed by the learning unit 14 refers to so-called "mechanical learning". In addition, the "learning" performed by the learning unit 14 is not limited to the deep learning using the aforementioned neural network, but may be learning using logistic regression, learning using support vector machines, learning using decision trees, and heterogeneous hybrid learning.

The seismic data acquisition unit 11 in the present embodiment 1 receives waveform data of an earthquake that has occurred from a single seismic detection device 20. In addition, the seismic data acquisition unit 11 transmits the received waveform data to the estimation processing unit 12.

The estimation processing unit 12 accesses the storage unit 15 in the present embodiment 1, acquires the learning model 16, and then applies the waveform data transmitted from the learning data acquisition unit 13 to the acquired learning model 16 to estimate the epicenter distance.

In addition, in the first embodiment, the learning unit 14 can additionally use the depth of the seismic source as the positive solution data, so as to learn the relationship between the waveform data, the epicentral distance, and the depth of the focal source to generate the learning model 16. At this time, the estimation processing unit 12 can estimate the depth of the seismic source in addition to the epicenter distance.

In addition, in the first embodiment, in addition to the waveform data as the input data, the learning unit 14 can also use the location data of the location where the waveform data has been obtained as the input data. At this time, the learning unit 14 will learn the relationship between the waveform data and location data, and the epicenter distance (or epicenter distance and depth of the epicenter), and generate a learning model 16.

Here, the location where the waveform data has been obtained refers to the location where the seismic wave as the source of the waveform data has been observed. In addition, as the location data, for example, the surface site increase rate, the data indicating the state of the plate, the data indicating the presence of volcanoes near the location where the seismic wave has been observed, the thickness of the crust, the thickness of the lithosphere, and the like. In this way, if both the waveform data and the location data are used as input data to generate the learning model 16, the accuracy of the estimation process can be improved.

When using location data as input data for learning, the seismic data acquisition unit 11 not only acquires waveform data of an earthquake that has occurred, but also acquires location data of a location where the waveform data has been acquired, that is, a location where the seismic detection device 20 is installed Location information.

In addition, the location data may be stored in the storage unit 15 of each seismic detection device 20 in advance. In this state, whenever the seismic data acquisition unit 11 acquires waveform data, the corresponding location data will be acquired from the storage unit 15. In addition, the location data can be transmitted from the seismic detection device 20 together with the waveform data. In this state, the seismic data acquisition unit 11 will acquire the location data together with the waveform data.

In addition, when using the location data as the input data for learning, the estimation processing unit 12 applies the acquired waveform data and location data to the learning model 16 generated by the learning unit 14 and estimates the epicenter distance (or epicenter distance and the source depth).

Furthermore, in the first embodiment, the input data and the positive solution data are not limited to the above examples. Data other than waveform data and location data can be used as input data. In addition, data other than the epicentral distance and the depth of the source can be used as positive solution data.

[Device operation]

Next, the operation of the epicentral distance estimation device 10 according to the first embodiment will be described using FIGS. 4 and 5. In the following description, reference will be made to Figures 1 to 3 as appropriate. In addition, in the first embodiment, the epicenter distance estimation method is implemented by operating the epicenter distance estimation device 10. Therefore, the description of the epicenter distance estimation method in the first embodiment will replace the following operational description of the epicenter distance estimation device 10.

In the first embodiment, the epicentral distance estimation device 10 mainly performs learning processing and estimation processing. First, the learning process will be explained. FIG. 4 is a flowchart showing the operation when the learning process of the epicentral distance estimation device of Embodiment 1 of the present invention is executed.

As shown in FIG. 4, first, the learning material acquisition unit 13 acquires input data and correct solution data (step A1). Specifically, in step A1, the learning data acquisition unit 13 not only uses waveform data as input data, but also acquires location data as input data, in addition to epicentral distance as positive solution data, and further obtains the depth of the source as positive solution data.

Next, the learning unit 14 determines whether the learning model 16 already exists (step A2). Specifically, the learning unit 14 determines whether the learning model 16 is stored in the storage unit 15.

When the result of the determination in step A2 is that the learning model 16 does not yet exist, the learning unit 14 will learn the relationship between the waveform data and location data, the epicentral distance and the depth of the source, and then newly generate a learning model 16 representing the learning result (step A3 ).

Specifically, in step A3, the learning unit 14 constructs a neural network through learning and uses it as a learning model 16. In addition, the learning unit 14 stores the created learning model 16 in the storage unit 15.

In addition, when the result of the determination in step A2 is that the learning model 16 already exists, the learning unit 14 uses the input data and the positive solution data obtained in step A1 to update the existing learning model 16 (step A4). Specifically, the learning unit 14 uses the input data and the positive solution data obtained in step A1 to update the coupling value between the nodes.

By executing steps A1~A4, a learning model can be created or updated. Then, the estimation process is performed using the learning model that has been created or updated. FIG. 5 is a flowchart showing the operation when the estimation process of the epicentral distance estimation device of Embodiment 1 of the present invention is executed.

As shown in FIG. 5, first, if the waveform data of the earthquake that has occurred is transmitted from the earthquake detection device 20, the seismic data acquisition unit 11 receives the transmitted waveform data (step B1).

Next, the seismic data acquisition unit 11 acquires the location data of the location where the seismic detection device 20 having transmitted waveform data is installed from the storage unit 15 (step B2). Moreover, when the location data and the waveform data are transmitted, the seismic data acquisition unit 11 receives the transmitted location data.

Next, the estimation processing unit 12 applies the waveform data received in step B1 and the location data obtained in step B2 to the learning model 16 created or updated by the learning process shown in FIG. 4 to estimate the epicentral distance and The depth of the source (step B3).

By performing steps B1 to B3, the epicenter distance and the depth of the source are estimated based on the waveform data obtained from the single seismic detection device 20.

[Effects of Embodiment 1]

As described above, according to the first embodiment, it is possible to estimate the epicenter distance and the depth of the source from the single waveform data without fitting the waveform data to the function. In addition, the estimation process is performed by learning the model 16, so the epicentral distance and the depth of the source can be stably calculated in a short time.

In other words, in the first embodiment, unlike the conventional method disclosed in Patent Document 1, it is not necessary to manually perform the operation of reflecting the complex individual location attributes in the influencing factors and parameters of unknown methods. According to the first embodiment, the available accuracy data can be obtained only by objective waveform data and machine learning. The epicentral distance estimation device 10 of the first embodiment can be introduced to multiple locations and multiple regions.

In addition, as described above, in the present embodiment, the depth of the source can also be estimated from single waveform data, but the technique disclosed in Patent Document 1 above cannot estimate the depth of the source. When using the technique disclosed in the above-mentioned Patent Document 1, in order to measure the depth of the seismic source, it is necessary to use the measurement results of a plurality of seismometers.

[Modification 1]

Next, a modification of the first embodiment will be described. First, in Modification 1, the learning unit 14 generates a learning model 16 for each of the set waveform amounts. Specifically, the amount of waveform is displayed as the time elapsed since the earthquake occurred. Therefore, the learning unit 14 cuts out the waveform data corresponding to the elapsed time length from the waveform data acquired by the learning data acquisition unit 13 at each set elapsed time, and then uses the cut waveform data as input data, and then performs Learning to produce a learning model 16. With this, the learning model 16 is generated for each of the waveform amounts.

In addition, in Modification 1, the estimation processing unit 12 calculates the waveform amount of the waveform data of the earthquake that has occurred, and then selects the learning model to be used from the generated plurality of learning models 16 based on the calculated waveform amount . Then, the estimation processing unit 12 applies the waveform data of the earthquake that has occurred to the selected learning model 16 and estimates the epicenter distance (or epicenter distance and depth of the epicenter).

In general, the estimation of the epicentral distance and the depth of the focal point used in the emergency quick report can be performed even when the amount of waveform data is small. Therefore, the waveform data obtained by the seismic data acquisition unit 11 may not be constant, and the waveform amount of the waveform data used for generating the learning model may not match the waveform amount of the waveform data of the earthquake that has occurred, making the estimation accuracy reduce. However, according to the first modification, the learning model 16 is selected according to the waveform amount of the waveform data of the earthquake that has occurred, so that the above-mentioned problem of reduced estimation accuracy can be avoided.

[Modification 2]

In Modification 2, the learning unit 14 generates a learning model 16 for each observation point of the waveform data to be input data. Specifically, the learning unit 14 uses only the waveform data acquired by each of the seismic detection devices (each of the seismometers) to generate the learning model 16.

In addition, in Modification 2, the estimation processing unit 12 confirms the observation point of the waveform data of the earthquake that has occurred (that is, the seismic detection device 20 as the transmission source of the waveform data), and then based on the confirmed observation point, generates Among the plurality of learning models 16, select the learning model to be used. Then, the estimation processing unit 12 applies the waveform data of the earthquake that has occurred to the selected learning model 16 and estimates the epicenter distance (or epicenter distance and depth of the epicenter).

According to Modification 2, even if the characteristics of each observation point (that is, location data) are not used for learning, it is possible to perform estimation processing that conforms to the characteristics of the observation point. Moreover, it is difficult to perform complete learning at observation points where input data cannot be sufficiently secured, resulting in difficulty in generating a learning model for such observation points.

[Modification 3]

In Modification 3, the learning unit 14 generates a learning model 16 for each of the site characteristics of the observation points of the waveform data to be input data. Specifically, for example, the observation points (earthquake detection device 20) are classified into different groups in accordance with site characteristics (value of location data) such as site increase rate. At this time, the learning unit 14 uses only the waveform data obtained in the group for each group to generate the learning model 16.

In addition, in Modification 3, the estimation processing unit 12 confirms the site characteristics of the observation point of the waveform data of the earthquake that has occurred, and then selects the desired learning model 16 from the plurality of generated learning models 16 based on the confirmed site characteristics Learning model. Then, the estimation processing unit 12 applies the waveform data of the earthquake that has occurred to the selected learning model 16 and estimates the epicenter distance (or epicenter distance and depth of the epicenter).

According to Modification 3, even if there are observation points where the input data are not completely obtained, it is not necessary to learn the use location data, that is, to perform the estimation processing conforming to the characteristics of the observation points.

〔Program〕

The program of this embodiment 1 may be a program that executes steps A1 to A4 shown in FIG. 4 and steps B1 to B3 shown in FIG. 5 in a computer. By installing the program to a computer and executing it, the epicenter distance estimation device 10 and the epicenter distance estimation method of the first embodiment can be realized. At this time, the CPU (Central Processing Unit) of the computer can perform functions by performing functions of the seismic data acquisition unit 11, the estimation processing unit 12, the learning data acquisition unit 13, and the learning unit 14.

In addition, the program of this embodiment 12 can be executed by a computer system constructed by a plurality of computers. At this time, each computer can play the functions of any one of the seismic data acquisition unit 11, the estimation processing unit 12, the learning data acquisition unit 13, and the learning unit 14, respectively. In addition, the storage unit 15 may be provided on a computer different from the computer that executes the program of this embodiment.

(Implementation type 2)

Next, the epicentral distance estimation device, epicentral distance estimation method and formula of Embodiment 2 of the present invention will be described with reference to FIGS. 6 to 8.

[Device configuration]

First, the configuration of the epicenter distance estimation device of the second embodiment will be described using FIG. 6. 6 is a block diagram specifically showing the configuration of an epicenter distance estimating device according to Embodiment 2 of the present invention.

As shown in FIG. 6, the epicentral distance estimation device 40 of the present embodiment 2 includes the waveform pre-processing unit 41. This point is different from the epicenter distance estimation device 10 of the embodiment 1 shown in FIGS. 1 and 2. . In the following, the differences from Embodiment 1 are mainly explained.

The waveform preprocessing unit 41 performs preprocessing on the waveform data used as the input data in the learning unit 14 and the waveform data obtained in the seismic data acquisition unit 11. Examples of pre-processing include image conversion processing, envelope conversion processing, band-pass conversion processing, differential conversion processing, and Fourier conversion processing.

Specifically, the image conversion process is a process of converting waveform data into image data of an image displayed as a graph. It is generally believed that if the image conversion process is followed, the learning unit 14 will perform learning based on the image data, making the learning process easier.

In addition, the envelope conversion processing is processing to smooth the waveform of the waveform data. If the envelope processing is performed, it is easy to confirm the rising characteristic of the seismic wave, and thus a learning model 16 reflecting the rising characteristic of the seismic wave is generated.

The band-pass conversion process is a process of highlighting the waveform of a specific period. According to the band-pass conversion process, the characteristics of the seismic wave will be highlighted, and thus a learning model 16 reflecting the characteristics of the seismic wave will be generated.

In addition, the differential conversion process is a process of differentiating the waveform data and then converting it into acceleration data. When performing the differential conversion process, it is also easy to confirm the rising characteristics of the seismic wave, so that a learning model 16 reflecting the rising characteristics of the seismic wave is generated.

In addition, the Fourier transform processing is processing to obtain the frequency distribution of the waveform data. According to the Fourier transform, the period difference of each of the waveform data will be prominent, and thus a learning model 16 reflecting the period of the seismic wave will be generated.

The waveform pre-processing section 41 may perform any one or more of any one of image conversion processing, envelope conversion processing, band-pass conversion processing, differential conversion processing, and Fourier conversion processing.

[Device operation]

Next, the operation of the epicentral distance estimation device 40 of the second embodiment will be described using FIGS. 7 and 8. In the following description, refer to FIGS. 1 to 6 as appropriate. In addition, in the second embodiment, the epicenter distance estimation method is implemented by operating the epicenter distance estimation device 40. Therefore, the description of the epicenter distance estimation method of the present embodiment 2 replaces the following operational description of the epicenter distance estimation device 40.

First, the learning process will be explained. FIG. 7 is a flowchart showing the operation when the learning process of the epicentral distance estimation device of Embodiment 2 of the present invention is executed.

As shown in FIG. 7, first, the learning material acquisition unit 13 acquires input data and correct solution data (step A11). In addition, the learning data acquisition unit 13 inputs the acquired data to the waveform pre-processing unit 41.

Next, the waveform preprocessing section 41 performs preprocessing on the waveform data included in the input data acquired in step A11 (step A12). Then, the waveform pre-processing section 41 inputs the pre-processed waveform data, other input data (location data) and correct solution data to the learning section 14.

Next, the learning unit 14 determines whether the learning model 16 already exists (step A13).

As a result of the determination in step A13, when the learning model 16 does not yet exist, the learning unit 14 will learn the relationship between the waveform data and location data, the epicentral distance and the depth of the epicenter, and then newly generate a learning model 16 representing the learning result (step A14 ).

In addition, when the learning model 16 already exists as a result of the determination in step A13, the learning unit 14 updates the existing learning model 16 using the input data and the correct solution data (step A15).

By performing steps A11 to A15, the learning model 16 is updated or created. After that, the learning model 16 that has been created or updated is used to execute the estimation process. FIG. 8 is a flowchart showing the operation when the learning process of the epicentral distance estimation device of Embodiment 2 of the present invention is executed.

As shown in FIG. 8, first, if the waveform data of the earthquake that has occurred is transmitted from the earthquake detection device 20, the seismic data acquisition unit 11 receives the transmitted waveform data (step B11).

Next, the seismic data acquisition unit 11 acquires the location data of the location where the seismic detection device 20 that has transmitted the waveform data is installed from the storage unit 15 (step B12). Moreover, when the location data and the waveform data are transmitted, the seismic data acquisition unit 11 receives the transmitted location data.

Next, the waveform preprocessing section 41 performs preprocessing on the waveform data received in step B11 (step B13). Then, the waveform pre-processing section 41 inputs the pre-processed waveform data and location data to the estimation processing section 12.

Next, the estimation processing unit 12 applies the waveform data after the pre-processing in step B13 and the location data obtained in step B12 to the learning model 16 created or updated by the learning process shown in FIG. 7 to estimate the epicenter Distance and source depth (step B14).

By performing steps B11 to B14, the present embodiment 2 is also the same as the embodiment 1, based on the waveform data obtained from the single seismic detection device 20 to estimate the epicentral distance and the depth of the source.

[Effects of Embodiment 2]

As described above, in the second embodiment, the pre-processing performed by the waveform pre-processing unit 41 suppresses noise and highlights features of the waveform data used for learning. Therefore, according to the second embodiment, the accuracy of the learning model is improved, and as a result, the estimation accuracy can be improved.

(Physical composition)

Here, by executing the programs of the implementation modes 1 and 2, the computer for realizing the epicenter distance estimation device will be described using FIG. 9. 9 is a block diagram showing an example of a computer that realizes the epicentral distance estimation device of Embodiments 1 and 2 of the present invention.

As shown in FIG. 9, the computer 110 includes a CPU 111, a main memory 112, a storage device 113, an input interface 114, a display controller 115, a data reader/writer 116 and a communication interface 117. The above-mentioned parts are connected via the bus 121 in such a way that data can be transferred to each other.

The CPU 111 is stored in the storage device 113, expands the programs (codes) of the present embodiment in the main memory 112, and then executes the programs in a predetermined order to implement various calculations. The main memory 112 is generally a volatile storage device such as DRAM (Dynamic Random Access Memory). In addition, the program of the present embodiment is provided in a state of being stored in a computer-readable recording medium 120. Moreover, the program of the present embodiment may be a program that circulates on the network connected through the communication interface 117.

In addition, as a specific example of the storage device 113, in addition to a hard disk, a semiconductor storage device such as a flash memory can also be mentioned. The input interface 114 serves as an intermediary to assist data transfer between the input devices 118 such as the CPU 111, keyboard, and mouse. The display controller 115 is connected to the display device 119 and controls the display of the display device 119.

The data reader/writer 116 acts as an intermediary to assist data transfer between the CPU 111 and the recording medium 120, read programs from the recording medium 120, and write processing results of the computer 110 to the recording medium 120. The communication interface 117 serves as an intermediary to assist data transfer between the CPU 111 and other computers.

In addition, specific examples of the recording medium 120 include general-purpose semiconductor recording devices such as CF (Compact Flash (registered trademark)) and SD (Secure Digital), magnetic recording media such as flexible disks, and CD-ROM ( Compact Disk Read Only Memory) and other optical recording media.

In addition, the epicentral distance estimation devices 10 and 40 of the first and second embodiments can be implemented not only by a computer with a program installed, but also by using a hard disk corresponding to each part. In addition, the epicentral distance estimation devices 10 and 40 may also be configured to be implemented partly by a program, and the remaining part by a hard disk.

Part or all of the above-mentioned embodiments can be expressed by (Supplement 1) to (Supplement 15) described below, but is not limited to the following description.

(Remark 1)

A device for estimating the epicentral distance, which is characterized by having: an earthquake information acquisition unit that acquires waveform data of an earthquake that has occurred; and The estimation processing unit applies the aforementioned waveform data to the learning model obtained by learning the relationship between the waveform data of the earthquake and the epicentral distance to estimate the epicentral distance.

(Remark 2)

For example, the device for estimating the epicentral distance in Note 1 further includes a learning unit, which uses the waveform data of the earthquake as input data and the epicentral distance of the aforementioned earthquake as the positive solution data, and learns the relationship between the waveform data and the epicentral distance, Then, a learning model representing the learning result is generated, and the estimation processing unit applies the acquired waveform data to the learning model generated by the learning unit to estimate the epicenter distance.

(Remark 3)

For example, in the device for estimating the epicentral distance in Note 2, in addition to using the waveform data as input data, the learning section also uses the location data of the location where the waveform data has been obtained as input data, and learns the waveform data and the location data and the foregoing The relationship between the epicentral distances, and then the aforementioned learning model is generated. In addition to acquiring the aforementioned waveform data, the aforementioned seismic information acquiring unit also acquires the location data of the location where the waveform data of the earthquake that has occurred has been acquired. Apply the aforementioned waveform data, and apply the aforementioned location data already obtained, and estimate the aforementioned epicentral distance.

(Remark 4)

If the device for estimating the epicentral distance of remarks 2 or 3, where The learning unit further uses the depth of the seismic source as the positive solution data, and learns the relationship between the waveform data and the epicenter distance and the depth of the epicenter, and then generates the learning model. The estimation processing unit estimates the epicenter distance. , And the depth of the source is presumed.

(Remark 5)

The device for estimating the epicentral distance according to any one of Remarks 2 to 4, wherein the learning unit constructs a neural network by learning, and uses the neural network as a learning model.

(Note 6)

For example, in the device for estimating the epicentral distance according to any one of Remarks 2 to 5, the learning unit generates the learning model for each of the waveform quantities of the waveform data to be input data, and the estimation processing unit calculates the acquired waveform data. The amount of waveforms, based on the calculated amount of waveforms, select the learning model to be used from among the generated learning models, and then apply the acquired waveform data to the selected learning model to estimate The aforementioned epicentral distance.

(Note 7)

An epicenter distance estimation device according to any one of remarks 2 to 5, wherein the learning section generates the learning model for each observation point of the waveform data to be input data, and the estimation processing section confirms the observation of the acquired waveform data Point, based on the confirmed observation point, select the learning model to be used from each of the generated learning models, and then apply the acquired waveform data to the selected learning model to estimate the aforementioned Epicenter distance.

(Note 8)

For example, in the device for estimating the epicentral distance according to any one of Remarks 2 to 5, the learning unit generates the learning model for each of the site characteristics of the observation point of the waveform data to be input data, and the estimation processing unit confirms the acquired waveform The site characteristics of the observation point of the data, and then based on the confirmed site characteristics, from each of the aforementioned learning models, select the aforementioned learning model to be used, and then apply the acquired aforementioned to the selected learning model Waveform data, and the aforementioned epicentral distance is estimated.

(Note 9)

As described in any one of the remarks 1 to 8, the epicentral distance estimation device further includes a waveform pre-processing unit for the waveform data used as the input data in the learning unit and the seismic information obtaining unit Perform pre-processing on the acquired waveform data.

(Note 10)

As in the epicentral distance estimation device of note 9, the waveform pre-processing section executes at least one process from image conversion processing, envelope conversion processing, band-pass conversion processing, differential conversion processing, and Fourier conversion processing as the pre-processing.

(Remark 11)

A method for estimating the epicentral distance, which is characterized by the following steps: (a) the step of obtaining waveform data of an existing earthquake; (b) the learning model obtained by learning the relationship between the seismic waveform data and epicentral distance, Apply the aforementioned waveform data and estimate the epicentral distance.

(Note 12)

For example, the method of estimating the epicentral distance in Note 11 has the following steps: (c) Use the waveform data of the earthquake as input data and the epicentral distance of the aforementioned earthquake as the positive solution data, and learn the relationship between the waveform data and epicentral distance, and then To generate a learning model representing the learning result, in the step (b), apply the acquired waveform data to the learning model generated in the step (c), and estimate the epicenter distance.

(Remark 13)

For example, the method of estimating the epicentral distance in Note 12, where: in the step (c), in addition to the aforementioned waveform data as input data, the location data of the location where the aforementioned waveform data has been obtained is also used as input data, and the aforementioned waveform data and The relationship between the aforementioned location data and the aforementioned epicentral distance, and then the aforementioned learning model is generated. In step (a), in addition to obtaining the aforementioned waveform data, the location data of the location where the waveform data of the earthquake that has occurred has been obtained. In the step (b), in addition to applying the waveform data to the learning model, the acquired location data is applied to estimate the epicenter distance.

(Remark 14)

For example, the method for estimating the epicentral distance of Note 12 or 13, wherein: in the step (c) above, the depth of the epicenter of the earthquake is used as the positive solution data, and the waveform data is learned between the epicentral distance and the depth of the epicenter Then, the aforementioned learning model is generated. In the aforementioned step (b), in addition to estimating the aforementioned epicentral distance, the depth of the source is also estimated.

(Note 15)

The method for estimating the epicentral distance according to any one of Remarks 12 to 14, wherein: in the step (c), a neural network is constructed by learning, and the neural network is used as a learning model.

(Note 16)

For example, the method for estimating the epicentral distance according to any one of Remarks 12 to 15, wherein: in the step (c) above, the aforementioned learning model is generated for each waveform amount of the waveform data to be input data, and in the step (b) above , Calculate the waveform amount of the acquired waveform data, and then select the learning model to be used from among the generated learning models based on the calculated waveform amount, and then apply the selected learning model The aforementioned waveform data is obtained, and the aforementioned epicentral distance is estimated.

(Note 17)

For example, the method for estimating the epicentral distance according to any one of Remarks 12 to 15, wherein: in the step (c) above, the aforementioned learning model is generated for each observation point of the waveform data that becomes the input data, and in the step (b) above , Confirm the observation points of the acquired waveform data, and then select the learning model to be used from among the generated learning models based on the confirmed observation points, and then apply the selected learning model The aforementioned waveform data is obtained, and the aforementioned epicentral distance is estimated.

(Note 18)

For example, the method for estimating the epicentral distance according to any one of Remarks 12 to 15, wherein: in the step (c), the learning model is generated for each of the site characteristics of the observation point of the waveform data that becomes the input data. ), confirm the site characteristics of the observation points of the acquired waveform data, and then, based on the confirmed site characteristics, select the learning model to be used from each of the generated learning models, and then select The aforementioned learning model applies the acquired waveform data to estimate the epicentral distance.

(Remark 19)

If the method for estimating the epicentral distance according to any one of Remarks 11 to 18, it also has the following steps: (d) in the step (c), the waveform data used as the input data, and the (a) Perform the preprocessing on the waveform data obtained in the step.

(Remark 20)

In the step (d), at least one of image conversion processing, envelope conversion processing, band-pass conversion processing, differential conversion processing, and Fourier conversion processing is executed as the aforementioned pre-processing.

(Note 21)

A computer-readable recording medium for recording programs, which contains commands to execute the following steps in a computer: (a) the step of obtaining waveform data of an earthquake that has occurred; (b) the distance between the waveform data of a learned earthquake and the epicenter The learning model obtained from the relationship between the two is applied to the aforementioned waveform data to estimate the epicentral distance.

(Remark 22)

If the computer-readable recording medium of Note 21, it also performs the following steps in the aforementioned computer: (c) Use the waveform data of the earthquake as input data, the epicenter distance of the aforementioned earthquake as the positive solution data, and learn the waveform data and the epicenter The relationship between the distances, and a learning model representing the learning result is regenerated. In the step (b), the learning model generated by the step (c) is applied with the waveform data obtained to estimate the epicenter distance. .

(Remark 23)

For example, the computer-readable recording medium of Note 22, where: in the step (c) above, in addition to the aforementioned waveform data as input data, the location data of the location obtained from the aforementioned waveform data will also be used as input data, and learn The aforementioned waveform data and the relationship between the aforementioned location data and the aforementioned epicentral distance, and then the aforementioned learning model is generated. In the step (a), in addition to acquiring the aforementioned waveform data, the location of the waveform data of the earthquake that has occurred has been obtained. For the location data, in the step (b), in addition to the waveform data, the acquired location data is applied to the learning model, and the epicenter distance is estimated.

(Remark 24)

For example, the computer-readable recording medium of note 22 or 23, wherein: in the step (c), the depth of the earthquake source is used as the positive solution data, and the distance between the waveform data and the epicenter and the source The relationship between depth, and then generate the aforementioned learning model, In the aforementioned step (b), in addition to estimating the aforementioned epicentral distance, the depth of the source is also estimated.

(Note 25)

A computer-readable recording medium according to any one of remarks 22 to 24, wherein: in the step (c), a neural network is constructed by learning, and the aforementioned neural network is used as a learning model.

(Remark 26)

A computer-readable recording medium as described in any one of Remarks 22 to 25, in which: in the step (c) above, the aforementioned learning model is generated for each of the waveform quantities of the waveform data to be input data, in the aforementioned (b ), calculate the waveform amount of the acquired waveform data, and then select the learning model to be used from among the generated learning models based on the calculated waveform amount, and then select the learning The model applies the acquired waveform data and estimates the epicentral distance.

(Remark 27)

A computer-readable recording medium according to any one of remarks 22 to 25, wherein: in the step (c) above, the aforementioned learning model is generated for each observation point of the waveform data that becomes the input data, in the aforementioned (b ), confirm the observation points of the acquired waveform data, and then, based on the confirmed observation points, select the learning model to be used from among the generated learning models, and then select the selected learning model. The model applies the acquired waveform data and estimates the epicentral distance.

(Note 28)

A computer-readable recording medium as described in any one of Remarks 22 to 25, in which: in the step (c) above, the aforementioned learning model is generated for each of the site characteristics of the observation point of the waveform data that becomes the input data, in In the step (b) above, confirm the site characteristics of the observation points of the acquired waveform data, and then, based on the confirmed site characteristics, select the learning model to be used from each of the generated learning models, and then For the selected learning model, the acquired waveform data is applied to estimate the epicenter distance.

(Note 29)

If the computer-readable recording medium of any one of Remarks 21 to 28, the computer also performs the following steps: (d) In the step (c), the waveform data used for the input data and the Perform the pre-processing on the waveform data obtained in the aforementioned step (a).

(Note 30)

A computer-readable recording medium as described in any one of Note 29, wherein: in the step (d) above, image conversion processing, envelope conversion processing, band-pass conversion processing, differential conversion processing, and Fourier conversion processing are performed At least one process is used as the aforementioned pre-processing.

Although the invention of the present application has been described above with reference to the embodiment mode, the invention of the present application is not limited to the above embodiment mode. Various changes that can be understood by those in the relevant field within the scope of the invention of the present application can be made to the structure or details of the invention of the present application.

This application claims priority based on the Japanese petition 2016-136310 filed on July 8, 2016, and the entire disclosure content is cited here.

[Industry availability]

As described above, according to the present invention, the epicentral distance can be calculated stably, and the calculation time can be shortened. The invention can be used in a system that must transmit earthquake related data as early as possible when an earthquake occurs.

10‧‧‧Estimation of epicentral distance (implementation type 1)

11‧‧‧Earthquake Information Acquisition Department

12‧‧‧Estimated Processing Department

13‧‧‧ Learning Information Acquisition Department

14‧‧‧Learning Department

15‧‧‧Storage Department

16‧‧‧ learning model

20‧‧‧Earthquake detection device

30‧‧‧Earthquake activity monitoring system

Claims (10)

A device for estimating epicenter distance, which is characterized by having: a learning unit that takes the waveform data of an earthquake as input data, and uses the epicenter distance of the earthquake as positive solution data, and learns the relationship between the waveform data and the epicenter distance, and then generates representation learning A learning model of results; an earthquake information acquisition unit that acquires waveform data of an earthquake that has occurred; and an estimation processing unit that learns a model obtained by learning the relationship between the waveform data of an earthquake and the epicentral distance by the learning unit, Apply the acquired waveform data to estimate the epicenter distance; and the learning section cuts out the waveform data corresponding to the elapsed time length at each set elapsed time from the waveform data that becomes the input data, and then cuts the cut out waveform data As input data, the learning model is generated for each of the waveform amounts displayed from the time elapsed when the earthquake occurred, and the estimation processing section calculates the acquired waveform amount of the waveform data, and then based on the calculated waveform amount, From each of the generated learning models, select the learning model to be used, and then apply the acquired waveform data to the selected learning model to estimate the epicentral distance. For example, the device for estimating the epicentral distance according to item 1 of the patent scope, in which the learning section uses the waveform data as input data, and also uses the location data of the location where the waveform data has been obtained as input data to learn the waveform data and The relationship between the location data and the epicentral distance generates the learning model. In addition to acquiring the waveform data, the seismic information acquisition unit also acquires the location data of the location where the waveform data of the earthquake that has occurred has been obtained. In addition to applying the waveform data to the learning model, the acquired location data is also applied, and the epicentral distance is estimated. For example, the device for estimating the epicentral distance according to item 1 or 2 of the patent scope, in which the learning part further uses the depth of the epicenter of the earthquake as the positive solution data to learn between the waveform data and the epicentral distance and the depth of the epicenter In order to generate the learning model, in addition to estimating the epicenter distance, the estimation processing unit also estimates the depth of the source. For example, the device for estimating the epicentral distance according to item 1 or item 2 of the patent scope, in which the learning unit constructs a neural network by learning and uses the neural network as a learning model. For example, the device for estimating the epicentral distance according to item 1 or item 2 of the patent scope, wherein the learning unit generates the learning model for each observation point of the waveform data that becomes the input data, and the estimation processing unit confirms the acquired waveform data The observation point, based on the confirmed observation point, select the learning model to be used from each of the generated learning models, and then apply the acquired waveform data to the selected learning model, and The epicentral distance is presumed. For example, the device for estimating the epicentral distance according to item 1 or item 2 of the patent scope, in which the learning unit generates the learning model for each of the site characteristics of the observation point of the waveform data that becomes the input data, and the estimation processing unit confirms the acquired The site characteristics of the observation point of the waveform data, and then based on the confirmed site characteristics, from each of the generated learning models, select the learning model to be used, and then apply the acquired learning model to the acquired The waveform data, and the epicenter distance is estimated. For example, the device for estimating the epicentral distance according to item 1 or item 2 of the patent application scope further includes a waveform pre-processing unit for the waveform data used as the input data in the learning unit and the seismic information The waveform data obtained by the acquisition section is pre-processed. An apparatus for estimating the epicentral distance as claimed in item 7 of the patent scope, wherein the waveform pre-processing section executes at least one process from the image conversion process, the envelope conversion process, the band-pass conversion process, the differential conversion process, and the Fourier conversion process, as The pre-processing. A method for estimating the epicentral distance, which is characterized by the following steps: (a) The waveform data of the earthquake is used as input data, and the epicentral distance of the earthquake is used as the positive solution data, and the relationship between the waveform data and the epicentral distance is learned, and then generated A learning model representing the learning result; (b) the step of obtaining waveform data of an earthquake that has occurred; (c) the learning model obtained by learning the relationship between the waveform data of the earthquake and the epicentral distance in the step (a), Apply the acquired waveform data to estimate the epicenter distance; and in step (a), from the waveform data that becomes the input data, at each set elapsed time, cut out the waveform data corresponding to the elapsed time length, and then Cut the waveform data as input data, generate the learning model for each of the waveform amounts displayed from the time elapsed when the earthquake occurred, and in step (c), calculate the waveform amount of the acquired waveform data Then, based on the calculated waveform amount, select the learning model to be used from each of the generated learning models, and then apply the acquired waveform data to the selected learning model to estimate the epicenter distance. A computer-readable recording medium that records a program that contains commands that perform the following steps in the computer: (a) Use the waveform data of the earthquake as input data, and use the epicenter distance of the earthquake as the positive solution data, and learn the relationship between the waveform data and epicenter distance, and then generate a learning model representing the learning results; (b) Obtain the occurrence. Steps of waveform data of earthquakes; (c) Apply the obtained waveform data to the learned model obtained by learning the relationship between the waveform data of the earthquake and the epicentral distance in step (a); and estimate the epicentral distance; And in step (a), from the waveform data that becomes the input data, at each set elapsed time, cut out the waveform data corresponding to the elapsed time length, and then use the cut waveform data as the input data. Each of the waveform amounts displayed at the elapsed time of occurrence generates the learning model. In step (c), the waveform amount of the acquired waveform data is calculated, and then based on the calculated waveform amount, the generated Among each of the learning models, select the learning model to be used, and then apply the acquired waveform data to the selected learning model to estimate the epicenter distance.

Images