EARTHQUAKE EARLY WARNING METHOD BASED ON SUPPORT VECTOR REGRESSION

Abstract

An earthquake early warning method for an earthquake detecting system includes utilizing support vector regression (SVR) method to build an earthquake detecting model according to the a plurality of vectors, wherein each of the vectors is corresponding to an earthquake information and a ground motion intensity; detecting a new earthquake information of a new earthquake and generating a specific vector according to the new earthquake information when the new earthquake occurs; and predicting a new ground motion intensity of the new earthquake according to the specific vector and the earthquake detecting model.

Description

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates to an earthquake early warning method based on support vector regression, and more particularly, to an earthquake early warning method capable of predicting the ground motion intensity of an earthquake according to earthquake initial waves a few seconds after the earthquake arrives and an earthquake detecting model built by a support vector regression method.

2. Description of the Prior Art

When an earthquake occurs, a regional earthquake early warning technology can disseminate earthquake warning seconds to tens of seconds before the strong shaking wave of the earthquake arrives via predicting the effects (intensities) and the onset time of the earthquake (e.g. the onset time of the earthquake is later if the distance from the epicenter is farther), so as to adopt essential contingency measures. Since the conventional regional earthquake early warning technology requires multiple earthquake detecting stations and earthquake initial waves information a few seconds after the earthquake arrives for completing predictions, the conventional regional earthquake early warning technology needs 20 seconds for disseminating the earthquake warning. If the velocity of the earthquake waves is 3 km per second, the region located from the epicenter within radius of at least 60 km will be the blind zone.

On the other hand, an on-site earthquake early warning technology utilizes earthquake initial waves a few seconds after the earthquake arrives at a single site to predict the earthquake intensity. However, the model to predict to final earthquake intensity is either too simple to acquire accurate earthquake intensity or too complex to be employed. A predict model with better accuracy and efficiency is needed.

SUMMARY OF THE INVENTION

Therefore, the present invention provides an earthquake early warning method capable of predicting the ground motion intensity of an earthquake according to earthquake initial waves a few seconds after the earthquake arrives and an earthquake detecting model built by a support vector regression method.

The present invention discloses an earthquake early warning method for an earthquake detecting system. The earthquake early warning method comprises utilizing support vector regression (SVR) method to build an earthquake detecting model according to the a plurality of vectors, wherein each of the vectors is corresponding to an earthquake information and a ground motion intensity; detecting new earthquake information of a new earthquake and generating a specific vector according to the new earthquake information when the new earthquake occurs; and predicting a new ground motion intensity of the new earthquake according to the specific vector and the earthquake detecting model.

The present invention further discloses an earthquake predicting system. The earthquake predicting system comprises a computing module, for utilizing support vector regression (SVR) method to build an earthquake detecting model according to a plurality of vectors; an earthquake detecting module, for detecting new earthquake information of a new earthquake and generating a specific vector according to the new earthquake information when the new earthquake occurs; and a ground motion intensity coefficient calculating module, for predicting a new ground motion intensity of the new earthquake according to the specific vector and the earthquake detecting model.

These and other objectives of the present invention will no doubt become obvious to those of ordinary skill in the art after reading the following detailed description of the preferred embodiment that is illustrated in the various figures and drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic diagram of an earthquake detecting system according to an embodiment of the present invention.

FIG. 2 is a comparison chart of true peak ground acceleration and a peak ground accelerations calculated by an earthquake detecting model.

FIG. 3 is a chart of the percentage of the prediction with intensity errors within plus one level or minus one level and the time of observing the vertical initial waves

FIG. 4 is a flowchart of an earthquake detecting process according to an embodiment of the present invention.

DETAILED DESCRIPTION

Please refer to FIG. 1, which is a schematic diagram of an earthquake detecting system 10 according to an embodiment of the present invention. The earthquake detecting system 10 comprises a computing module 100, an earthquake detecting module 110, a ground motion intensity coefficient calculating module 120 and an earthquake warning module 150. The computing module 100 utilizes a support vector regression (SVR) method to build an earthquake detecting model f(x) according to vectors x₁-x_m, and then stores the earthquake detecting model f(x) in the ground motion intensity calculating module 120. Each vector x_n (1≦n≦m) of the vectors x₁-x_m consists of earthquake information Q_n1-Q_nk and corresponds to an intensity y_n. The input of the earthquake detecting model f(x) is a vector with the same dimension of the vectors x₁-x_m. When a new earthquake NE occurs, the earthquake detecting module 110 detects new earthquake information N₁-N_k of the new earthquake NE, generates a specific vector x_new according to the new earthquake information N₁-N_k and transmits the specific vector x_new to the ground motion intensity coefficient calculating module 120. The ground motion intensity coefficient calculating module 120 utilizes the specific vector x_new as the input of the earthquake detecting model f(x) and calculates anew intensity y_new of the new earthquake NE. The earthquake warning module 150 then determines whether to disseminate an earthquake warning according to the new intensity y_new.

In detail, the support vector regression method projects the vectors x₁-x_m to a high-dimensional characteristic space H and performs a linear regression in the high-dimensional characteristic space H. The computing module 100 therefore assumes the earthquake detecting model f(x) as the following:

f(x)=w,g(x)+b   (1)

wherein “w” is a vector of the high-dimensional characteristic space H, “g(x)” is a nonlinear projection which projects the input x to the high-dimensional characteristic space H, “” is an inner product mathematical operation and “b” is a constant. Via substituting slack variables and regression risks, the computing module 100 derives the equation (1) to a target function shown as the following:

$\begin{array}{cc}\underset{\omega ,b,\xi ,{\xi }^{*}}{\min }\left[\frac{1}{2}{w}^2+\frac{C}{m}\sum _{i=1}^m\left({\xi }_i+{\xi }^{*}\right)\right]& \left(2\right)\end{array}$

which is limited by

w,x _i +b−y _i≦ε+ξ_i, y_i−w,x−b≦ε+ξ_i* and ξ_i,ξ_i*≧0,i=1, . . . , m;

wherein “C” is a constant, and “ξ₁−ξ_m”, “ξ₁*−ξ_m*” and “b” are variables of the target function (i.e. the equation (2)). The computing module 100 further utilizes Lagrange multipliers α₁-α_m, β₁-β_m for converting the equation (2) to a quadratic programming shown as the following:

$\begin{array}{cc}\underset{\alpha ,\beta }{\max }\left[\begin{array}{c}-\frac{1}{2}\sum _{i=1}^m\sum _{j=1}^m\left({\alpha }_i-{\beta }_i\right)\left({\alpha }_j-{\beta }_j\right)k\left(x_i,x_j\right)-\\ \varepsilon \sum _{i=1}^m\left({\alpha }_i+{\beta }_i\right)+\sum _{i=1}^my_i\left({\alpha }_i-{\beta }_i\right)\end{array}\right]& \left(3\right)\end{array}$

The equation (3) is constrained by

$\sum _i^m\left({\alpha }_i-{\beta }_i\right)=0\mathrm{and}0\le {\alpha }_i,{\beta }_i\le \frac{C}{m};$

wherein “ε” is a constant. Similarly, the computing module 100 utilizes the Lagrange multipliers α₁-α_m, β₁-β_m to convert the equation (1) to the following equation:

$\begin{array}{cc}f\left(x\right)=\left[\sum _{i=1}^m\left({\alpha }_i-{\beta }_i\right)k\left(x,x_i\right)\right]+b^{*}& \left(4\right)\end{array}$

Wherein “b*” is a constant, function k is a kernel function shown as the following:

k(x_i,x_j)=exp(−∥x_i−x_j∥²/2σ²)   (5)

Wherein “σ” is a constant, “∥ ∥” represents the Euclidean norm. The computing module 100 substitutes the vectors x₁-x_m (i.e. the earthquake information Q_n1-Q_nk) and the intensities y₁-y_m corresponding to the vectors x₁-x_m into the equation (3) and the equation (4), and then the Lagrange multipliers α₁-α_m, β₁-β_m can be acquired. Next, the constant b* is acquired by substituting the vector x_n and the intensity y_n, which fits α_n#0 and

$\begin{array}{cc}{b}^{*}=y_n-\sum _{i=1}^m\left({\alpha }_i-{\beta }_i\right)k\left(x_i,x_n\right)-\varepsilon & \left(6\right)\end{array}$

into the following equation:

${\alpha }_n\ne \frac{C}{m},$

Finally, the computing module 100 determines the constants C, ε and σ according to the vectors x₁-x_m.

According to the above embodiment, the computing module 100 builds the earthquake detecting model f(x) (i.e. the equation (4)) according to the vectors x₁-x_m (i.e. the earthquake information Q_n1-Q_nk) and the intensities y₁-y_m thereof via the support vector regression method, and then stores the earthquake detecting model f(x) in the ground motion intensity coefficient calculation module 120. When the new earthquake NE occurs, the earthquake detecting module 110 detects the new earthquake information N₁-N_k of the new earthquake NE, to generate the specific vector x_new. The ground motion intensity coefficient calculating module 120 then substitutes the specific vector x_new into the earthquake detecting model f(x) to acquire the new intensity y_new of the new earthquake (i.e. y_new=f(x_new)).

Please note that, the computing module 100, the earthquake detecting module 110 and the ground motion intensity coefficient calculating module 120 of the earthquake early warning system 10 may locate at different locations and may communicate with each other via a wired network or a wireless network. For example, the computing module 100 may locate at the earthquake information centre for acquiring massive earthquakes Q_n1-Q_nk, to improve the accuracy of the earthquake detecting model f(x). The earthquake detecting module 110 may locate at the region where the earthquake occurs frequently, for detecting the new earthquake information N₁-N_k rapidly. The ground motion intensity coefficient calculating module 120 may locate at the warning centre, so as to determine whether to disseminate the earthquake warning immediately. Besides, the intensity is utilized for illustrating the concept of the above embodiment and is not utilized for limiting the scope of the present invention. For example, the method of the above embodiment can be utilized for predicting coefficients related to the intensity and the damage degree of the earthquake, such as a peak ground acceleration (PGA), a peak ground velocity (PGV), a peak ground displacement (PGD), a cumulative absolute velocity (CAV), an Arias intensity, a spectral intensity and a response spectra.

In addition, the solution of the support vector regression method (i.e. the equations (1)-(6)) is utilized for illustrating the concept of the above embodiment and is not utilized for limiting the scope of the present invention. Those skilled in the art may use other mathematical equations in cooperation with the support vector regression method to acquire equations for calculating the new ground motion intensity of the new earthquake NE.

In order to illustrate performance of the above embodiment, the new ground motion intensity y_new of the new earthquake NE acquired by analyzed simulations is compared with the actual intensity of the new earthquake NE. Please note that, the intensity (I) of an earthquake is converted from the peak ground acceleration, thus the comparisons of the peak ground acceleration also can be utilized for illustrating the performance of the above embodiment. In this embodiment, the conversion equation between the intensity and the PGA is

$\log \left(\mathrm{PGA}\right)=\left(\frac{I}{2}\right)-0.6$

and the conversion equation between the intensity and the PGA may be different in different countries. The conversion equations between the intensity and the PGA in other countries are also suitable to the above embodiment and do not affect the performance.

Please refer to FIG. 2, which is a comparison chart between the peak ground accelerations calculated according to the earthquake detecting model f(x) and the actual peak ground accelerations. The testing conditions in FIG. 2 comprise: vectors x₁-x₉₁₁₄₂ generated according to 91142 data of the earthquake records provided by Taiwan Central Weather Bureau and intensities y₁-y₉₁₁₄₂ corresponding to the vectors x₁-x₉₁₁₄₂, wherein each vector x_n (1≦n≦91142) comprises a peak acceleration, a peak velocity, a peak displacement, a predominate period, a cumulative absolute velocity and integral squared velocity (acquired from initial 3 seconds of the vertical initial waves). As shown in FIG. 2, the peak ground accelerations calculated according to the earthquake detecting model f(x) are highly close to the actual peak ground accelerations and the percentage of the prediction with intensity errors within plus one level or minus one level (i.e. a predict ratio) reaches up to 99.22%.

Please refer to FIG. 3, which is a chart between the percentage of the prediction with intensity errors within plus one level or minus one level and the time of observing the vertical initial waves. The range of the time of observing the vertical initial waves is from 0.1 to 10 seconds, and interval between two sample points is 0.1 seconds. As shown in FIG. 3, the predict ratio increases as the time of observing the vertical initial waves. The earthquake detecting model f(x) in the above embodiment has a high accuracy even the time of observing the vertical initial waves is only 0.1 seconds, however. As shown in FIG. 3, the predict ratio reaches up to 91.24% when the time of observing the vertical initial waves is 0.1 seconds. Moreover, when the time of observing the vertical initial waves increases to 0.4 seconds, the prediction ratio arises to 97.19%. As can be seen from the above, the method disclosed in the above embodiment can utilize extremely short time of observing only the vertical initial waves for achieving high prediction accuracy. The warning blind zone can be shrunk effectively and valuable reaction time can be acquired.

The above-mentioned operations of the earthquake detecting system 10 can be further summarized into an earthquake detecting process 40 as shown in FIG. 4. The earthquake detecting process 40 comprises the following steps:

Step 400: Start.

Step 402: Utilize a support vector regression (SVR) method to build an earthquake detecting model according to a plurality of vectors, wherein each of the vectors is corresponding to earthquake information and a ground motion intensity.

Step 404: Detect new earthquake information of the new earthquake and generating a specific vector according to the new earthquake information when a new earthquake occurs.

Step 406: Predict a new ground motion intensity of the new earthquake according to the specific vector and the earthquake detecting model.

Step 408: End.

The detailed operations and related alternations can be known by referring to the above, and are not narrated herein for brevity.

In the prior art, since the conventional regional earthquake early warning technology requires multiple earthquake detecting stations and earthquake initial waves information a few seconds after the earthquake arrives for completing predictions, the conventional regional earthquake early warning technology needs 20 seconds for disseminating the earthquake warning. If the velocity of the earthquake waves is 3 km per second, the region located from the epicenter within radius of at least 60 km will be the blind zone. On the other hand, the conventional earthquake on-site early warning technology utilizes earthquake records of few representative earthquakes for verifying the usability thereof. The performance of the conventional earthquake on-site early warning technology in practical applications is unknown, thus a considerable misjudging possibility may be obtained when utilizing the conventional earthquake on-site early warning technology in the practical applications. Some of the conventional earthquake on-site early warning technology use different method to build a predict model, but the methods are too complex to be employed. In comparison, the earthquake early warning method of the above embodiments can accurately predict the intensity of the earthquake via using the earthquake initial waves a few seconds after the earthquake arrives and the earthquake detecting model built by the support vector regression method with ease.

To sum up, the earthquake early warning method of the above embodiments utilizes the support vector regression method to build the earthquake detecting model according to the established earthquake information and the intensities thereof, and utilizes the earthquake detecting model to early detect the intensity of the new earthquake according to the earthquake initial waves a few seconds before the new earthquake occurs when the new earthquake occurs.

Those skilled in the art will readily observe that numerous modifications and alterations of the device and method may be made while retaining the teachings of the invention. Accordingly, the above disclosure should be construed as limited only by the metes and bounds of the appended claims.

Claims

1. An earthquake early warning method for an earthquake detecting system comprising: utilizing a support vector regression (SVR) method to build an earthquake detecting model according to a plurality of vectors, wherein each of the vectors is corresponding to an earthquake information and a ground motion intensity;detecting new earthquake information of a new earthquake and generating a specific vector according to the new earthquake information when the new earthquake occurs; andpredicting a new ground motion intensity of the new earthquake according to the specific vector and the earthquake detecting model.

2. The earthquake early warning method of claim 1, wherein the step of predicting the new ground motion intensity of the new earthquake according to the specific vector and the earthquake detecting model is performed according to the following equation:

3. The earthquake early warning method of claim 2, wherein the kernel function is k(xi,xj)=exp(−∥xi−xj∥2/2σ2), where σ is a constant.

4. The earthquake early warning method of claim 2, wherein the coefficients α1-αm, β1-βm are acquired via solving a quadratic programming and the quadratic programming is

5. The earthquake early warning method of claim 4, wherein the quadratic programming is acquired according to a target function with Lagrange multipliers, and the target function is

6. The earthquake early warning method of claim 1, wherein the earthquake information corresponding to each vector of the plurality of vectors are characteristic values of a plurality of earthquake initial waves.

7. The earthquake early warning method of claim 1, wherein the step of detecting the new earthquake information of the new earthquake and generating a specific vector according to the new earthquake information when the new earthquake occurs comprises: detecting the new earthquake information of the new earthquake and generating a specific vector according to the new earthquake information in a specific time when the new earthquake occurs.

8. The earthquake early warning method of claim 1, wherein the ground motion intensity and the new ground motion intensity are one of an intensity, a peak ground acceleration, a peak ground velocity, a peak ground displacement, a cumulative absolute velocity, an Arias intensity, a spectral intensity and a response spectra.

9. The earthquake early warning method of claim 1 further comprising: determining whether to disseminate an earthquake warning according to the new ground intensity.

10. An earthquake predicting system, comprising: a computing module, for utilizing a support vector regression (SVR) method to build an earthquake detecting model according to a plurality of vectors;an earthquake detecting module, for detecting new earthquake information of a new earthquake and generating a specific vector according to the new earthquake information when the new earthquake occurs; anda ground motion intensity coefficient calculating module, for predicting a new ground motion intensity of the new earthquake according to the specific vector and the earthquake detecting model.

11. The earthquake predicting system of claim 10, wherein the ground motion intensity coefficient calculating module predicts the new ground motion intensity of the new earthquake according to the specific vector and the earthquake detecting model according to the following equation:

12. The earthquake predicting system of claim 11, wherein the kernel function is k(xi,xj)=exp(−ƒxi−xj∥2/2σ2), where σ is a constant.

13. The earthquake predicting system of claim 11, wherein the coefficients α1-αm, β1-βm are acquired via solving a quadratic programming and the quadratic programming is

14. The earthquake predicting system of claim 13, wherein the quadratic programming is acquired according to a target function with Lagrange multipliers, and the target function is

15. The earthquake predicting system of claim 10, wherein the earthquake information corresponding to each vector of the plurality of vectors are characteristic values of a plurality of earthquake initial waves.

16. The earthquake predicting system of claim 10, wherein the earthquake detecting module detects the new earthquake information of the new earthquake and generating a specific vector according to the new earthquake information in a specific time when the new earthquake occurs.

17. The earthquake predicting system of claim 10, wherein the ground motion intensity and the new ground motion intensity are one of the an intensity, a peak ground acceleration, a peak ground velocity, a peak ground displacement, a cumulative absolute velocity, an Arias intensity, a spectral intensity and a response spectra.

18. The earthquake predicting system of claim 10 further comprising: an earthquake warning module, for determining whether to disseminate an earthquake warning according to the new ground intensity.