Distributed Optical Fiber Seismic First Arrival Picking Method Based on Enhanced Fractional Low-Order Moments

Abstract

The present invention discloses a distributed fiber optic seismic first arrival picking method based on enhanced fractional low-order moments, comprising the steps of obtaining distributed fiber optic seismic data to be picked, sampling channel by channel and detection window by detection window, obtaining a data segment each time sampling; estimating the characteristic index and scale parameter of each data segment that obeys a Levy symmetric stable distribution using a parameter estimation method based on sample quantiles; and obtaining a characteristic matrix of the distributed fiber optic seismic data using the characteristic index, scale parameter, and enhanced fractional low-order moments. CF d ;pass CF d Determine a first arrival picking interval; then sample the distributed fiber optic seismic data within the first arrival picking interval channel by channel and window by window, and obtain the characteristic matrix of the first arrival picking interval CF P ,pass CF P Determine the time of the first arrival of an earthquake. The present invention can accurately pick up the first arrival of a distributed optical fiber earthquake and has good noise robustness.

Description

Technical Field

[0001] The present invention relates to a distributed optical fiber seismic first-break picking method, in particular to a distributed optical fiber seismic first-break picking method based on enhanced fractional low-order moments. Background Art

[0002] Distributed fiber-optic seismic sensors have received increasing attention and use in earthquake observations. Their intensive observations generate massive amounts of seismic data, posing significant challenges to earthquake detection and first-arrival picking. Earthquake first-arrival picking is a fundamental and critical step in seismic data processing, and the accuracy of the picked results is crucial for the subsequent calculation of earthquake source information. Currently, deep learning-based earthquake arrival time picking methods have achieved significant success. However, due to differences in sensing modes, acquisition parameters, and data signal-to-noise ratios between distributed fiber-optic seismic sensors and conventional geophones, deep learning-based earthquake arrival time picking models trained on conventional geophone seismic data struggle to obtain accurate results on distributed fiber-optic seismic data. Furthermore, due to the sheer volume of distributed fiber-optic seismic data, manual first-arrival picking is difficult, and training a deep neural network-based earthquake first-arrival picking model for distributed fiber-optic seismic sensors using supervised learning is extremely challenging. However, the more commonly used traditional first-arrival picking methods based on AR-AIC+STA / LTA and high-order statistics-kurtosis are easily interfered by complex noise on distributed fiber optic seismic data, and the picking results are not accurate.

[0003] Glossary:

[0004] Levy symmetric stable distribution: Levy stable distribution is a family of stable distributions, of which Gaussian distribution is a special case. The characteristic function of Levy stable distribution is:

[0005]

[0006]

[0007] Where t is a real number, α is the characteristic exponent of the Levy stable distribution, γ is the scale parameter of the Levy stable distribution, β is the skew parameter of the Levy stable distribution, and δ is the location parameter of the Levy stable distribution. When the skew parameter β is 0, the distribution is a Levy symmetric stable distribution. When α is 2, the distribution is a Gaussian distribution.

[0008] Common parameter estimation of Levy stable distribution includes a method based on sample quantiles and a method based on maximum likelihood estimation. The present invention adopts a method based on sample quantiles with less computational complexity. Summary of the Invention

[0009] The purpose of the present invention is to provide a distributed fiber optic seismic first arrival picking method based on enhanced fractional low-order moments, which can solve the above problems, effectively improve the accuracy of distributed fiber optic seismic first arrival picking, and has a certain noise robustness.

[0010] In order to achieve the above object, the technical solution adopted by the present invention is as follows: a distributed optical fiber seismic first arrival picking method based on enhanced fractional low-order moments comprises the following steps;

[0011] (1) Obtain the number of channels M and the data length N of each channel of the distributed fiber optic seismic data to be picked up, and set a detection window with a length of W d , the sliding step is S d ;

[0012] (2) Sliding sampling is performed on each channel of the distributed fiber optic seismic data. Each sampling generates a data segment, where the data segment of the jth sampling of the i-th channel is D ij , i=1~M, j=1~S, S is the sampling times of each channel;

[0013] (3) Using the parameter estimation method based on sample quantiles, the characteristic index and scale parameter of each data segment that obeys the Levy symmetric stable distribution are estimated, D ij The corresponding characteristic index is α ij , the scale parameter is γ ij ;

[0014] (4) Construct the expression of enhanced fractional low-order moments, and obtain the characteristic matrix CF of distributed fiber optic seismic data with a size of M×S from the characteristic index, scale parameter and enhanced fractional low-order moments. d ;

[0015] The expression of the enhanced fractional low-order moment is:

[0016]

[0017] Where, CF(D ij ) is CF d The element in row i and column j, m is the enhancement parameter, P ij D ij is the order of the enhanced fractional low-order moment, Γ(·) is the gamma function;

[0018] (5) CF d For each row of elements, search for the maximum value, record the index of the maximum value, and count the mode d of all indexes;

[0019] (6) Determine the first arrival picking interval of the distributed fiber optic seismic data as [d·S d , d·S d +Wd ];

[0020] (7) Set a picking window with a length of W p , the sliding step is S p For the distributed fiber seismic data within the first arrival picking interval, the characteristic matrix CF of the first arrival picking interval is obtained according to the method of steps (2)-(4). p ;

[0021] (8) Get the earthquake first arrival time of each channel, where the earthquake first arrival time t of the i-th channel is obtained i The method comprises steps (81)-(83);

[0022] (81) CF p For the element in row i, calculate the first-order difference of adjacent elements, search for the maximum value of the first-order difference, and record its index P i ;

[0023] (82)P i Mapped to the two data segments corresponding to the i-th channel of the distributed fiber optic seismic data;

[0024] (83) Select the latter of the two data segments and count its sampling time as the earthquake first arrival time t i .

[0025] As a preference: in step (2),

[0026] As a preference: in step (5), obtain CF d The method for indexing the maximum value in row i is:

[0027] (51) CF d Normalize the elements in the i-th row to obtain the normalized elements;

[0028] (52) Search for the normalized element with the largest value in the row and record its index as d i .

[0029] As a preference: in step (5), when counting the modes of all indexes, if there are multiple modes, the minimum value of all modes is selected as the mode d.

[0030] The idea of the present invention is:

[0031] Through step (3), the distributed fiber optic seismic data is modeled one by one using the Levian symmetric stable distribution to obtain the characteristic index and scale parameter of the Levian symmetric stable distribution, thereby obtaining the characteristic index matrix and scale parameter matrix of the Levian symmetric stable distribution.

[0032] Then use steps (4)-(6) to obtain the first arrival picking interval of the distributed fiber optic seismic data.

[0033] Then, using step (7), the distributed fiber optic seismic data in the first arrival picking interval are modeled one by one using the Levy symmetric stable distribution.

[0034] Finally, use step (8) to determine the earthquake first arrival time in the first arrival picking interval.

[0035] Compared with the prior art, the advantages of the present invention are:

[0036] 1. A new distributed fiber-optic seismic first-arrival picking method based on enhanced fractional low-order moments is proposed. This method is based on the difference in probability density distribution between noise and seismic waveforms. It uses the Levy symmetric distribution to model random noise (generally Gaussian) and seismic waveforms (non-Gaussian). The enhanced fractional low-order moments under the Levy symmetric distribution assumption are used as characteristic functions for first-arrival picking.

[0037] 2. The first arrival picking method provided by the present invention first models the distributed fiber optic seismic data using the Levy symmetric stable distribution to obtain the characteristic index, scale parameter, and the enhanced fractional low-order moment to obtain the characteristic matrix CF d , combined with CF d The processing method is used to determine a first arrival picking interval in the distributed fiber optic seismic data and narrow the picking range. Then, the distributed fiber optic seismic data in the first arrival picking interval are modeled using the Levy symmetric stable distribution to obtain the characteristic index, scale parameter, and the enhanced fractional low-order moment to obtain the characteristic matrix CF p , combined with first-order difference and other processing methods, the first arrival time of the earthquake in each channel is finally obtained. Therefore, the present invention can accurately pick up the first arrival of distributed optical fiber earthquakes and has good noise robustness.

[0038] 3. Compared with deep learning-based seismic first-break picking methods, this method does not require pre-training with a large amount of distributed fiber-optic seismic data with first-break labels, which makes it more practical. Compared with several traditional seismic first-break picking methods based on AR-AIC+STA / LTA and high-order statistics (kurtosis), the proposed method is more robust to noise in distributed fiber-optic seismic data. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Flowchart of the present invention;

[0040] Figure 2 This is the distributed optical fiber seismic data diagram of Example 2;

[0041] Figure 3 This is the characteristic graph obtained after processing in step (5) of Example 2;

[0042] Figure 4 In Example 2, Figure 3 The obtained first arrival picking interval diagram;

[0043] Figure 5 for Figure 4 The first arrival picking interval contains the distributed fiber optic seismic data segments at the first arrival time of the earthquake;

[0044] Figure 6 The following is a comparison of the first arrival picking results of distributed fiber optic seismic data using four methods;

[0045] Figure 7a Comparison chart of the picking results of four methods at CH0, CH50, CH100, CH150, CH200, and CH250;

[0046] Figure 7b Comparison chart of the picking results of four methods on CH300, CH350, CH400, CH450, CH500 and CH550;

[0047] Figure 7c Comparison chart of the picking results of four methods on CH600, CH650, CH700, CH750, CH800 and CH850;

[0048] Figure 7d Comparison chart of the picking results of four methods on CH900, CH950, CH1000, CH1050, CH1100 and CH1149;

[0049] Figure 8a The histogram of the residuals between the STA / LTA method picking results and the reference results;

[0050] Figure 8b The histogram of the residuals between the HOS-Kurt method picking results and the reference results;

[0051] Figure 8c The histogram of the residuals between the picking results of the method of the present invention and the reference results. DETAILED DESCRIPTION

[0052] The present invention will be further described below with reference to the accompanying drawings.

[0053] Example 1: See Figure 1 ,A distributed optical fiber seismic first arrival picking method based on enhanced fractional low-order moments, comprising the following steps;

[0054] (1) Obtain the number of channels M and the data length N of each channel of the distributed fiber optic seismic data to be picked up, and set a detection window with a length of W d , the sliding step is S d ;

[0055] (2) Sliding sampling is performed on each channel of the distributed fiber optic seismic data. Each sampling generates a data segment, where the data segment of the jth sampling of the i-th channel is D ij , i=1~M, j=1~S, S is the sampling times of each channel;

[0056] (3) Using the parameter estimation method based on sample quantiles, the characteristic index and scale parameter of each data segment that obeys the Levy symmetric stable distribution are estimated, D ij The corresponding characteristic index is α ij , the scale parameter is γ ij ;

[0057] (4) Construct the expression of enhanced fractional low-order moments, and obtain the characteristic matrix CF of distributed fiber optic seismic data with a size of M×S from the characteristic index, scale parameter and enhanced fractional low-order moments. d ;

[0058] The expression of the enhanced fractional low-order moment is:

[0059]

[0060] Where, CF(D ij ) is CF d The element in row i and column j, m is the enhancement parameter, P ij D ij is the order of the enhanced fractional low-order moment, Γ(·) is the gamma function;

[0061] (5) CF d For each row of elements, search for the maximum value, record the index of the maximum value, and count the mode d of all indexes;

[0062] (6) Determine the first arrival picking interval of the distributed fiber optic seismic data as [d·S d , d·S d +W d ];

[0063] (7) Set a picking window with a length of W p , the sliding step is S p For the distributed fiber seismic data within the first arrival picking interval, the characteristic matrix CF of the first arrival picking interval is obtained according to the method of steps (2)-(4). p ;

[0064] (8) Get the earthquake first arrival time of each channel, where the earthquake first arrival time t of the i-th channel is obtained i The method comprises steps (81)-(83);

[0065] (81) CFp For the element in row i, calculate the first-order difference of adjacent elements, search for the maximum value of the first-order difference, and record its index P i ;

[0066] (82)P i Mapped to the two data segments corresponding to the i-th channel of the distributed fiber optic seismic data;

[0067] (83) Select the latter of the two data segments and count its sampling time as the earthquake first arrival time t i .

[0068] In this embodiment, in step (2),

[0069] In step (5), CF is obtained d The method for indexing the maximum value in row i is:

[0070] (51) CF d Normalize the elements in the i-th row to obtain the normalized elements;

[0071] (52) Search for the normalized element with the largest value in the row and record its index as d i ;

[0072] In step (5), when counting the modes of all indexes, if there are multiple modes, the minimum value of all modes is selected as the mode d.

[0073] Example 2: See Figures 1 to 8c In order to better illustrate the effect of the present invention, based on Example 1, a distributed optical fiber seismic first arrival picking method based on enhanced fractional low-order moment is provided, which includes the following steps:

[0074] (1) Obtain the distributed optical fiber seismic data to be picked up, which has been filtered by a bandpass filter, such as Figure 2 As shown, the number of channels M is 1150, the data length N is 12500, and the sampling frequency is 250Hz; a detection window is set with a length of W d is 1000, and the sliding step size is S d is 500.

[0075] (2) Same as step (2) of Example 1, when sampling the distributed optical fiber seismic data channel by channel and window by window, a data segment is obtained, and the sampling times of each channel is

[0076] (3) For each data segment, a parameter estimation method based on sample quantiles is used to estimate the characteristic index and scale parameter of the Levy symmetric stable distribution that the data segment obeys, D ij The corresponding characteristic index is αij , the scale parameter is γ ij ;

[0077] (4) Same as step (4) of Example 1, obtain the characteristic matrix CF of distributed fiber optic seismic data with a size of M×S=1150×23 d ;

[0078] (5) CF d For each row of elements, search for the maximum value, record the index of the maximum value, and count the mode d of all indexes; at this time, CF d Each row of elements is normalized, and then the index is recorded and the mode is counted. The result is as follows Figure 3 As shown, Figure 3 , where the white dashed line is the mode of the maximum value index of each row of the feature matrix.

[0079] (6) Determine the first arrival picking interval of the distributed fiber optic seismic data as [d·S d , d·S d +W d ],like Figure 4 As shown, the two white dashed lines are the first arrival picking intervals.

[0080] (7) Then set a picking window with a length of W p is 250, sliding step S p =1. For the sake of distinction, the distributed fiber seismic data within the first arrival picking interval is referred to as interval DAS data. This step uses the picking window to get the characteristic matrix CF of the first arrival picking interval according to the method of steps (2)-(4). p .

[0081] (8) Same as step (8) in Example 1. This step obtains the earthquake first arrival time of each channel, such as Figure 5 shown.

[0082] To illustrate the effectiveness of the present invention, we use the theoretical first arrival time calculated by TauP, a seismic travel time calculation tool commonly used in the field of seismology, as a reference to evaluate the first arrival picking performance of the four methods.

[0083] Method 1: The method of the present invention is also called Enhanced Fractional Lower Order Moment, abbreviated as Enhanced FLOM.

[0084] Method 2: Short–Time Average / Long–Time Average method. This method is a conventional picking method, abbreviated as STA / LTA.

[0085] Method 3: Autoregressive AIC combined method, abbreviated as AR-AIC+STA / LTA.

[0086] Method 4: High-order statistics-kurtosis method, abbreviated as HOS-kurt.

[0087] The comparison of the first arrival picking results of distributed fiber optic seismic data using the four methods is shown in the figure. Figures 7a-7d As shown. Figure 6 It can be seen that the method of the present invention rarely has obvious errors in picking. In the case of no coherent noise interference or less interference, the picking result is closer to the reference result. Figure 7c CH700 to CH850 in Figure 1, and CH400 to CH500 in Figure 2.

[0088] Similar to the fourth method: the method based on high-order statistics - kurtosis, the method of the present invention is more sensitive to small amplitude changes, resulting in the picking results of a considerable number of traces being earlier than the reference results, such as Figure 7a-7c CH150 to CH750 in this range. Figure 7a-7d The residuals of the picking results shown and the reference results are statistically shown, and the mean of the residuals based on the method of the present invention is less than zero.

[0089] The third method: the method based on AR-AIC+STA / LTA has more errors, such as Figure 7d As shown in Figure 9, the standard deviation of the residuals between its picking results and the reference results is the largest among the three picking methods.

[0090] Also, see Figure 8a-8c , the histogram of the residuals between the picking results of different methods and the reference results. Figure 8a-8c In this equation, μ is the mean of the residuals and σ is the standard deviation of the residuals. STA / LTA , t HOS , t FLOM Respectively represent the picking results of STA / LTA method, HOS-kurt method, and the method of the present invention, t TauP is the reference picking result obtained by theoretical arrival time calculation, that is, the reference result.

[0091] See also Figure 8a-8cCompared to the reference picking results, both the high-order statistics-kurtosis-based method and the enhanced fractional low-order-based method tend to pick first arrival times earlier and closer to the reference picking results, as their μ values are both less than zero and their absolute values are not very large. In contrast, the conventional AR-AIC+STA / LTA method tends to pick first arrival times later, significantly later than the reference picking results, as its μ values are greater than zero and their absolute values are larger. This indicates that the high-order statistics-kurtosis-based method and the enhanced fractional low-order-based method perform better than the conventional AR-AIC+STA / LTA method. Further comparison of the high-order statistics-kurtosis-based method and the enhanced fractional low-order-based method reveals that the enhanced fractional low-order-based method exhibits fewer obvious false positives and has a smaller σ value, indicating that the enhanced fractional low-order-based method is slightly more resistant to noise than the high-order statistics-kurtosis-based method.

[0092] The above analysis shows that the method of the present invention and the method based on higher-order statistics-kurtosis have better picking effects than the conventional AR-AIC+STA / LTA, indicating that the method of the present invention and the method based on higher-order statistics-kurtosis are more robust to noise interference. Compared with the method based on higher-order statistics-kurtosis, the picking effect of the present invention is comparable, both tend to pick earlier, and the method of the present invention has fewer obvious wrong picks. This shows that the noise interference resistance of the present invention is slightly stronger than that of the method based on higher-order statistics-kurtosis.

[0093] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A distributed fiber optic seismic first-break picking method based on enhanced fractional low-order moments, characterized by: The following steps are included: (1) Obtain the number of channels M and the data length N of each channel of the distributed fiber optic seismic data to be picked up, and set a detection window with a length of W d , the sliding step is S d ; (2) Sliding sampling is performed on each channel of the distributed fiber optic seismic data. Each sampling generates a data segment, where the data segment of the jth sampling of the i-th channel is D ij , i=1~M, j=1~S, S is the sampling times of each channel; (3) Using the parameter estimation method based on sample quantiles, the characteristic index and scale parameter of each data segment that obeys the Levy symmetric stable distribution are estimated, D ij The corresponding characteristic index is α ij , the scale parameter is γ ij ; (4) Construct the expression of enhanced fractional low-order moments, and obtain the characteristic matrix CF of distributed fiber optic seismic data with a size of M×S from the characteristic index, scale parameter and enhanced fractional low-order moments. d ; The expression of the enhanced fractional low-order moment is: Where, CF(D ij ) is CF d The element in row i and column j, m is the enhancement parameter, p ij D ij is the order of the enhanced fractional low-order moment, Γ(·) is the gamma function; (5) CF d For each row of elements, search for the maximum value, record the index of the maximum value, and count the mode d of all indexes; (6) Determine the first arrival picking interval of the distributed fiber optic seismic data as [d·S d , d·S d +W d ]; (7) Set a picking window with a length of W p , the sliding step is S p For the distributed fiber seismic data within the first arrival picking interval, the characteristic matrix CF of the first arrival picking interval is obtained according to the method of steps (2)-(4). p ; (8) Get the earthquake first arrival time of each channel, where the earthquake first arrival time t of the i-th channel is obtained i The method comprises steps (81)-(83); (81) CF p For the element in row i, calculate the first-order difference of adjacent elements, search for the maximum value of the first-order difference, and record its index P i ; (82)P i Mapped to the two data segments corresponding to the i-th channel of the distributed fiber optic seismic data; (83) Select the latter of the two data segments and count its sampling time as the earthquake first arrival time t i .

2. The distributed fiber optic seismic first-break picking method based on enhanced fractional low-order moments according to claim 1, characterized in that: In step (2), 3. The distributed fiber optic seismic first-break picking method based on enhanced fractional low-order moments according to claim 1, characterized in that: In step (5), CF is obtained d The method for indexing the maximum value in row i is: (51) CF d Normalize the elements in the i-th row to obtain the normalized elements; (52) Search for the normalized element with the largest value in the row and record its index as d i .

4. The distributed fiber optic seismic first-break picking method based on enhanced fractional low-order moments according to claim 1 or 3, characterized in that: In step (5), when counting the modes of all indexes, if there are multiple modes, the minimum value of all modes is selected as the mode d.

Images