JP2012127929A - P-wave/s-wave arrival time of day automatic reading method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve the following problem: automatic reading by a conventional method is performed by applying information criterion by Akaike to waveform data filter processed, where the method is completely mathematical and does not assemble seismological perspective, so that there can occur a case in which reading noises is mistaken as earthquake waves or, in contrast, the earthquake waves as noises; when seismological specialists or the like read arrival times of day of P-waves and S-waves, they determine them generally--for example, a tremor of an earthquake tends to last long in comparison with noises, and there is difference in frequency between the noises and earthquakes--and considering such seismological perspective, difference between the two is distinguished even when noise amplitude is large and an arrival time of day is accurately read, but the conventional technology does not develop a method for incorporating the seismological perspective.SOLUTION: A method for incorporating the seismological perspective and a function for performing general check as a human does into software for automatic reading is developed and, by incorporating the function, precision of automatic reading is improved.

Description

  In Japan, more than 1000 high-sensitivity seismometers are installed, and about 400 earthquakes occur every day. The National Research Institute for Earth Science and Disaster Prevention, the Japan Meteorological Agency, universities, etc. are conducting research on seismic activity and underground structures, but these researches require reading the arrival times of P waves and S waves. Due to the large number of observation points and the number of earthquakes, the number of earthquake records that need to be read is enormous, and a lot of expenses are required for the reading work. In addition, it is necessary to set up more observation points in order to improve the accuracy of survey and research on seismic activity and underground structures. As the number of observation points increases, the amount of reading increases, so reading is becoming a bottleneck for advanced research. The present invention relates to a method for accurately reading the arrival times of P waves and S waves, and the fields of use include automatic analysis equipment for earthquake observation data, etc. by the Japan Meteorological Agency, National Research Institute for Disaster Prevention, universities, electric power companies, etc. Equipment.

  Research on automatic reading of earthquake P wave and S wave arrival time started in the latter half of the 1970s, and since the 1980s, many researchers have developed systems for automatic reading (eg, non-patented). Reference 1). The prior art uses Akaike's information criterion (for example, see Non-Patent Document 2) to determine the boundary between the noise time interval and the P wave or S wave time interval, This is the arrival time of the S wave. However, the amplitude of noise varies greatly with time, and noise from an automobile or the like may be mixed just before the arrival of the P wave. In the conventional method, noise is often mistakenly read as an earthquake wave. For this reason, organizations engaged in earthquake research and operations such as the National Research Institute for Earth Science and Disaster Prevention, universities, and the Japan Meteorological Agency have judged that automatic reading results using conventional technology cannot be adopted as the final result, and many reading work is still in progress. It employs human resources to perform reading by humans.

  In addition, a technique for reading an arrival time by applying an artificial intelligence algorithm to seismic waveform data and read data has been proposed (see Non-Patent Document 3). However, the accuracy of this method is low and it has not been put into practical use.

  Research on automatic P-wave and S-wave reading of AE generated in the laboratory (see, for example, Patent Document 1) has also been conducted, but noise from automobiles and the like does not enter the experimental data. There is no built-in algorithm.

Patent Publication No. 2004-286527, National Institute of Advanced Industrial Science and Technology, March 20, 2003 (2003. 3.20) Multi-component AE waveform initial motion detection method.

Takashi Yokota, Katsumi Zhou, Megumi Mizokami, Isao Nakamura (1981): Automatic test method for seismic wave data and on-line processing system, Bull. Earthq. Res. Inst. , 55, 449-484. Shosuke Awashima, Genshiro Kitagawa, Shunichi Amari, Koji Akaike (2007): Akaike Information Criterion AIC-Modeling, Prediction, Knowledge Discovery, Kyoritsu Publishing. Zhao Y. and K.K. Takano, An Artificial Neural Network Approach for Broadband Seismic Phase Picking, Bull. Seism. Soc. Am. 89, 3, 670--680, 1999.

  In reading by the conventional method, the time position where the information amount standard (AIC) [Non-Patent Document 1] is maximized is mainly used as the arrival time. This is a method of obtaining the arrival time stochastically assuming that the noise is stationary, and seismological knowledge is not incorporated. For this reason, in many cases, the phase after arrival of the P wave and S wave is read, or conversely, noise before arrival is read as the arrival time. In reading, if it is noise, it is necessary not to read even if the amplitude is large. Conversely, if it is a seismic wave, it is necessary to read even if the amplitude is small. If it is possible to automatically determine whether it is an earthquake or noise, an improvement in reading accuracy can be expected. By developing a method to distinguish between the two based on the same judgment as a seismologist, the accurate arrival time can be automatically read.

  When an earthquake expert or an operator engaged in reading reads the arrival time, it is possible to determine whether the amplitude has increased due to the arrival of the seismic wave, or whether it has increased due to noise. Seeing it, it judges comprehensively and distinguishes both. If an automatic reading system can incorporate a function for comprehensive judgment similar to that performed by earthquake specialists, it will be possible to increase the accuracy.

  On the other hand, an earthquake expert or the like reads the time when the amplitude is increased or the frequency is temporally changed when the arrival time is read. When passed through an appropriate filter, the amplitude changes at the time when the frequency changes. Therefore, an earthquake expert or the like can consider that the time when the amplitude of the waveform data that has passed through an appropriate filter increases is read as the arrival time.

  Even when noise due to vibration of a car or the like is mixed, the amplitude increases, so there may be a plurality of times when the amplitude increases. Therefore, reading the arrival time can be considered to select a candidate whose amplitude increases due to the arrival of the seismic wave from among a plurality of arrival time candidates whose amplitude increases.

When an earthquake expert or the like reads the arrival time, it seems that the frequency, amplitude, time difference, characteristics of noise before arrival, and the like are comprehensively determined by visually checking the arrival times. The method according to the present invention, as shown in FIG.
1) Using a large number of seismic waveform data and P wave and S wave arrival time data read by earthquake experts, 2) From the seismic waveform data, calculate the time when the amplitude becomes large and use it as a candidate for arrival time ,
3) Calculate values of parameters such as dominant frequency, time difference, amplitude ratio, etc. in the interval between candidates, the interval before the first candidate, and the interval after the last candidate, and save the values To do.
4) Create an initial model of the evaluation function for selecting the arrival time of the P wave and S wave from the arrival times using parameters such as the dominant frequency according to 3),
5) The initial model of 4) is improved so that the time difference between the arrival time reading by the earthquake expert and the arrival time of the candidate selected by the evaluation function is minimized.

  Because the accuracy of automatic P-wave and S-wave reading by the conventional method is low, the Meteorological Agency, universities, disaster prevention science and technology research institutes, etc. engaged in earthquake research employ a total of about 100 operators for reading work. Is going. If automatic reading is improved in the present invention, this work can be reduced, and it is expected that significant cost reduction and rapid earthquake data analysis can be performed. Currently, it takes time and labor to read, making it difficult to develop a higher-density seismic observation network. The present invention can be expected to increase the density of the observation network and promote earthquake research.

Conceptual diagram of how to create P wave and S wave automatic reading system

The seismic waveform data of the i-th earthquake and j-th observation point through an appropriate filter is F _ij (t), and the time variation of the amplitude is below the threshold in the interval t = 0 to t _{0. It} is assumed that the amplitude becomes greater than or equal to the threshold due to the arrival of seismic waves or noise mixing after ₀ . Large amplitude, sets the time t ₄ to the time interval considered seismic wave is clearly arrival, arrival time is assumed to be included in the interval t ₀ -t _{4. Let} A _{0 be} the average of the absolute values of F _ij (t) in the interval 0-t ₀ .

Ｒ_ｍは、１より大きい定数で、例えば、

と置く。Next, an example of a P wave arrival time candidate estimation method will be described. The function Z (t) is defined as follows.

R _m is a constant greater than 1, for example,

Put it.

_Let C _ijm be the time at which the expression (1) in the section 0-T ₄ is minimum. C _ijm obtained by the equations (1) and (2) when m = 1 matches the time position where the average amplitude is 10% larger than the noise. This is because C _ijm is the time at which Equation (1) has the minimum value, and the amplitude after time C _ijm is 1.1 times or more the average amplitude of noise. When the amplitude suddenly increases, the time _Cijm is constant or only slightly changed even if m is changed. Therefore, as candidates for arrival time, even if m is changed, if C _ijm is constant or slightly changed, the arrival time candidates are combined into a single candidate.

Even in the noise section before time t ₀ , the amplitude may increase in a short section. When obtaining candidates for arrival time using the equations (1) and (2), even if the amplitude increases in a short time interval, if the amplitude after that time is small, (1 ) Is never minimized. On the other hand, the seismic waveform has the property that the amplitude continuously increases for a longer time period than the noise. For this reason, in the case of a shake due to an earthquake, Equation (1) is minimized at the arrival time. The advantage of using the formula (1) is that the calculation is very simple, the time of amplitude change of about 10% can be detected by the value of m in the formula (2), and (3) noise with short-term amplitude change is detected. It can be removed.

と置ける。ここに、Ｂ_ｋは、到着時刻を選ぶための未知パラメータ、ε_ｉｊは、地震の専門家等による読み取り値との差である。評価関数の形や、未知パラメータＢ_ｋを、

が最小となるよう決定する。Next, a method for selecting the arrival time of the P wave or S wave from the arrival time candidates C _ijm in claim 1 will be described. In the time interval between the m-th candidate and the m + 1-th candidate, that is, C _ijm -C _{ijm + 1} , the dominant frequency, the amplitude ratio with the noise part, the amplitude ratio between the vertical movement and the horizontal movement, and the amplitude parameter Calculate changes over time. _Let A _{ijmn be} the n th measured value of the i th earthquake, the j th observation point, and the m th candidate. Let H be an evaluation function, that is, a function that selects the arrival time of a P wave or S wave from a plurality of arrival time candidates using A _ijmn . When the reading of the arrival time of the j-th observation point of the i-th earthquake by an earthquake expert or the like is P _ij ,

I can put it. Here, B _k is an unknown parameter for selecting an arrival time, and ε _ij is a difference from a reading value by an earthquake expert or the like. The form of the evaluation function and the unknown parameter B _k

Is determined to be minimal.

In general, equation (4) is a special function, and it is difficult to obtain B _k by the method of least squares. However, since the equation (4) does not include earthquake waveform data, the amount of calculation is not large. The inventor created Formula (4) corresponding to thousands of read data and calculated it using a personal computer. As a result, the calculation time for several thousand was about 0.1 seconds. Since the calculation time is short, B _k can be obtained numerically. Further, since a record having a large difference between P and H in the expression (4) can be output, the cause of the large difference can be pursued and the evaluation function can be enhanced. When the unknown parameter _Bk is changed, the influence on all the recordings can be instantaneously calculated, so that the unknown parameter can be set relatively easily.

とする、即ち、Ｍ番目の候補の到着時刻を採用するというものである。この場合は、ノイズとのＳ／Ｎ比が１．１^ｍ以上となる時刻を到着時刻であるとする、ということと等価である。評価関数の初期モデルの作成方法の例は、卓越周波数等の測定値と、未知パラメータＢ_ｋを用いて、スコアを定義し、スコアがある値以上となるＣ_ｉｊｍを到着時刻とするようにすることである。To make the evaluation function image easier to understand, the simplest form of the evaluation function is shown. that is,

That is, the arrival time of the Mth candidate is adopted. In this case, it is equivalent to assuming that the time when the S / N ratio to noise is 1.1 ^m or more is the arrival time. As an example of a method for creating an initial model of an evaluation function, a score is defined using a measured value such as a dominant frequency and an unknown parameter B _k , and C _ijm having a score greater than or equal to a certain value is set as an arrival time That is.

Claims (2)

A large number of waveform data generated in the past, P wave and S wave arrival time reading data of those waveform data by earthquake specialists and operators, and a plurality of waveform data whose amplitude and frequency change. Arrival of P wave and S wave characterized by using data such as S / N ratio, dominant frequency, amplitude ratio between horizontal and vertical movement, time length of section, amplitude change in section in time section A method for determining an evaluation function to distinguish time from noise arrival time. An automatic determination method for arrival times of P-waves and S-waves using the evaluation function of [Claim 1].

Images