Multi-layer automatic picking method based on sparse two-stage dynamic normalization algorithm

Abstract

The application discloses a kind of multilayer automatic picking methods based on sparse two-stage dynamic normalization algorithm, including the following steps in turn: S1.similarity and backtracking path of two sequences are obtained;S2.seismic trace clustering;S3.multilayer automatic picking.The application provides multilayer automatic picking method, utilizes sparse two-stage dynamic normalization algorithm to carry out layer position tracking work, on the basis of traditional dynamic normalization algorithm, join the similarity of data itself and join constraint to reduce search range and second-order distance formula, solve the problem of low efficiency and precision in the application of classic dynamic time normalization algorithm in layer position tracking.The application is suitable for multilayer automatic picking based on sparse two-stage dynamic normalization algorithm, can obtain accurate layer position tracking work in complex geological environment, improve the efficiency and accuracy of layer picking.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of geophysical exploration, and relates to horizon tracking of two-dimensional seismic profiles and three-dimensional seismic bodies, in particular to a multi-horizon automatic picking method based on a sparse two-stage dynamic warping algorithm. BACKGROUND

[0002] Horizon tracking is a process of extracting events from reflection layers in obtained seismic images, and is an important part of seismic interpretation work and a basic work of seismic image interpretation. The earliest horizon tracking work relies on manual calibration, but it has great limitations because of large and complex workload and the need for personnel with certain geological experience. With the development of technology, the horizon automatic tracking technology is currently a commonly used method, which can not only identify horizons on two-dimensional profiles, but also obtain three-dimensional horizon tracking results for research.

[0003] Currently, the horizon automatic tracking technology includes methods based on coherence, edge detection operators, waveform similarity and deep learning. Anderson (1983) et al. proposed applying waveform matching to horizon tracking; Li Peng (2010) proposed applying five edge detection operator methods to horizon tracking work and made a comparative analysis to draw a conclusion; Li Xuefeng (2011) applied a method based on coherence and analyzed the horizon tracking interpretation and workflow of coherence profiles. The above are related researches on horizon tracking using other related algorithms by domestic and foreign scholars. However, there are relatively few related researches on the application of dynamic warping algorithm. Song Jin (2017) proposed introducing the dynamic time warping algorithm into seismic horizon interpretation work, combining the algorithm with regularization to process complex seismic data; Aina Juell Bugge (2019) proposed using the combination of classic DTW and instantaneous phase to extract target horizons to realize horizon tracking on three-dimensional seismic images. According to the previous research, the application of dynamic time warping algorithm for horizon tracking is relatively small, and the method considering time and accuracy in complex geological areas is even less. SUMMARY

[0004] The present application aims to provide a multi-horizon automatic picking method based on a sparse two-stage dynamic warping algorithm. The improved dynamic warping algorithm is used for horizon tracking work. On the basis of the traditional dynamic warping algorithm, the similarity of the data itself is added as a constraint to reduce the search range, so as to solve the problems of low efficiency and accuracy of the classic dynamic time warping algorithm in horizon tracking application. The space is saved to save time. And considering the existence of faults in the actual data tracking process, the idea of hypothesis testing is added to improve the accuracy of horizon tracking, so as to obtain better tracking effect.

[0005] To achieve the above object, the technical scheme adopted by the present application is:

[0006] A multi-layer bit automatic picking method based on a sparse two-stage dynamic normalization algorithm, the search path is obtained in the last distance matrix, so the idea of this improved algorithm is to first set a sparse matrix (this matrix can be considered as a matrix in which each position is empty), and then further set the positions in the matrix according to the sequence similarity, that is, the example given below (the non-empty positions in the matrix are the search path range), the method comprises the following steps performed in turn:

[0007] S1. Obtain the similarity of two sequences and the backtracking path

[0008] S11. Normalize the original sequence X and the original sequence Y

[0009] The original sequence X and the original sequence Y are normalized to obtain the sequence X' and the sequence Y';

[0010]

[0011] where S' is the normalized sequence, S i is the value of the i-th point of the sequence, and max and min are the maximum and minimum values of the sequence, respectively;

[0012] S12. Set the window range and number

[0013] Set the parameter θ, the range is 0-1, set the window range and number according to the set parameter θ, determine the number and range of windows and the points on the sequence in each window, the points of the sequence belonging to the same window are similar points, and then obtain the search range of the original sequence X and the original sequence Y;

[0014] S13. Set the backtracking path range according to the similar points

[0015] S131. Set the initial sparse search path matrix M, the size of M is n*m, n is the length of the original sequence X, m is the length of the original sequence Y, and the value of each cell position in the matrix is 0, i.e. a sparse cell;

[0016] A sparse matrix of m*n size is a rectangular array arranged by m rows and n columns of elements, and the elements in the matrix are all zero;

[0017] Because the initial distance matrix we set as sparse matrix (the value in the matrix is all 0), equivalent to the initial no search path, in the later according to the similar point and distance formula, the value of a certain coordinate (the position in the matrix) is obtained, the position with value is the search path, and the position without value is still 0 (the value in the sparse matrix is 0), which is not the final search path range;

[0018] S132. According to step S12, the search range of the original sequence X and the original sequence Y is obtained, and the corresponding positions of the similar points in the sequence X' and the sequence Y' in each window correspond to the positions of the initial matrix M, that is, the initial search path range obtained according to the sequence similarity;

[0019] S133. Since the initial search path range involves the surrounding of the corresponding positions in the matrix which are still original sparse cells, in order to ensure the subsequent minimum path backtracking, it is necessary to set these sparse cells as the search path range;

[0020] S14. Calculate the distance matrix according to the second-order formula

[0021] According to the alignment distance formula, the alignment distance of the search range of the sequence X and the sequence Y obtained in step S13 is calculated, and then the cumulative distance is calculated according to the cumulative distance formula, and the optimal alignment path, that is, the backtracking path range, is found;

[0022]

[0023] In the formula, d ij is the alignment distance, x i is the i-th value of the sequence X, y j is the j-th value of the sequence Y, and i and j respectively represent the i-th and j-th corresponding matrix cells of the sequence X and the sequence Y;

[0024]

[0025] In the formula, D ij is the cumulative distance, d ij is the alignment distance, and i and j respectively represent the i-th and j-th corresponding matrix cells of the sequence X and the sequence Y;

[0026] S15. According to the distance matrix, the similarity of the two sequences and the backtracking path (i.e. the corresponding coordinates of the two sequences) are obtained;

[0027] S2. Seismic trace clustering

[0028] According to the idea of statistical clustering, the similarity between seismic traces is obtained by the sparse two-stage dynamic regularization algorithm, similar seismic traces are classified into one class, and the clustering result of the seismic traces is obtained. One seismic trace is matched with a seismic trace of a different class;

[0029] S3. Multi-layer automatic picking

[0030] The matching result is obtained according to an interpolation method to obtain the layer position, that is, the multi-layer automatic picking is completed.

[0031] That is, the matching result and the similarity clustering of the seismic trace data are obtained through S1, and then the layer position tracking is performed, that is, a grid is set according to the existing algorithm, and then there is a reference trace when the sequence in the grid is matched, multiple matching traces, after clustering, the matching is not performed on the reference trace and each matching trace, only one trace belonging to a class and the reference trace need to be matched, and the results can be obtained, finally, the matching result is obtained according to the interpolation method to obtain the layer position.

[0032] As a limitation, when performing the multi-layer automatic picking, the improved dynamic time warping algorithm can also be used to track the layer position in the seismic data with faults, and the hypothesis testing principle is used to determine whether the layer positions on both sides of the fault belong to a unified layer position.

[0033] The layer position tracking needs to set a grid according to the existing method after obtaining the three-dimensional seismic data, and the matching result of two seismic trace data is obtained by using the improved algorithm of the patent one by one, and the layer position is obtained by using the interpolation method of the previous research according to the matching result (some coordinates are obtained, that is, some points on the seismic image).

[0034] As a further limitation, the specific steps of the hypothesis testing principle are: according to the waveform similarity principle, when judging whether the Ith layer and the Jth layer belong to the same layer position, if they belong to the same layer position, the distribution is the same or similar, that is, the mean values of the two distributions are equal.

[0035] As another limitation, in the step S12, the method for setting the window range and the number is: the difference between the upper limit of the window and the upper limit of the adjacent window is set to , and the difference between the lower limit and the upper limit in a window is set to .

[0036] According to the method for setting the window range and the number and the numerical size of the sequence itself, the number and range of the window and the points on the sequence in each window are determined.

[0037] As a third limitation, in the step S2, the method for obtaining the clustering result of the seismic trace is: the clustering similarity size sim and the boundary range width are set, the seismic traces smaller than sim and width are clustered into a class, and the clustering result of the seismic trace is obtained, and the seismic traces in each class have high similarity.

[0038] When the reference trace and the matching trace of the horizon tracking are matched, the results of the seismic traces belonging to the same category in all the matching traces can be matched with one of the traces in the category and the reference trace by using the matching sparse two-stage dynamic warping, so as to save time.

[0039] Compared with the prior art, the technical progress achieved by the present application lies in that:

[0040] ① The multi-horizon automatic picking method based on the sparse two-stage dynamic warping algorithm provided by the present application uses the improved dynamic warping algorithm to perform horizon tracking, and adds a constraint according to the similarity of the data itself to reduce the search range on the basis of the traditional dynamic warping algorithm, so as to solve the problems of low efficiency and low accuracy of the classic dynamic time warping algorithm in horizon tracking application.

[0041] ② The multi-horizon automatic picking method based on the sparse two-stage dynamic warping algorithm provided by the present application sets the search path range according to the similarity between the seismic trace data to improve the matching efficiency, and uses a second-order formula for the cumulative distance formula to improve the matching accuracy.

[0042] ③ The multi-horizon automatic picking method based on the sparse two-stage dynamic warping algorithm provided by the present application saves time by saving space.

[0043] ④ The multi-horizon automatic picking method based on the sparse two-stage dynamic warping algorithm provided by the present application considers the existence of faults in the actual data tracking process, adds the idea of hypothesis testing, and improves the accuracy of horizon tracking.

[0044] The present application is suitable for multi-horizon automatic picking based on the sparse two-stage dynamic warping algorithm, can obtain accurate horizon tracking in complex geological environments, and improves the efficiency and accuracy of horizon picking. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 is the flow chart of the multi-horizon automatic picking method based on the sparse two-stage dynamic warping algorithm designed in the embodiment;

[0046] Figure 2 is the flow chart of the sparse two-stage dynamic warping algorithm in the embodiment;

[0047] Figure 3 is the schematic diagram of the horizon tracking result of one seismic profile in the embodiment;

[0048] Figure 4 is the three-dimensional horizon tracking effect diagram of the sparse two-stage dynamic warping algorithm in the embodiment;

[0049] Figure 5 is the schematic diagram of the actual application of horizon tracking of seismic trace clustering in the embodiment. DETAILED DESCRIPTION

[0050] The application will be further described in detail by specific examples. It should be understood that the described examples are only for explaining the application and do not limit the application.

[0051] Embodiment 1: A multi-layer automatic picking method based on a sparse two-stage dynamic normalization algorithm

[0052] The embodiment is a multi-layer automatic picking method based on a sparse two-stage dynamic normalization algorithm. The embodiment is performed in a seismic simulation data set. A workflow diagram of the embodiment is shown in FIG. 1. According to the diagram, the multi-layer automatic picking method based on the sparse two-stage dynamic normalization algorithm can be implemented. The method includes the following steps performed in sequence: Figure 1

[0053] S1. Setting a search path range using the similarity of the sequence itself. A flowchart of the sparse two-stage dynamic normalization algorithm is shown in FIG. 2. Figure 2

[0054] S11. Normalizing the original sequence X and the original sequence Y

[0055] The original sequence X is [4, 0, 1, 2, 1], and the original sequence Y is [3, 5, 3, 2, 1].

[0056] The original sequence X and the original sequence Y are normalized to obtain the sequence X' and the sequence Y'.

[0057]

[0058] where S' is the normalized sequence, S i is the value of the i-th point of the sequence, max and min are the maximum and minimum values in the sequence, respectively.

[0059] The sequence X' is [1, 0, 0.25, 0.5, 0], and the sequence Y' is [0.5, 1, 0.5, 0.25, 0].

[0060] S12. Setting the window range and the number

[0061] The parameter θ is set to 1. The window range and the number are set according to the set parameter θ. The number and range of the windows and the points on the sequence in each window are determined. The points of the sequence belonging to the same window are similar points. The search range of the original sequence X and the original sequence Y is obtained, as shown in Table 1.

[0062] The method for setting the window range and the number is that the difference between the upper bound of a window and the upper bound of the adjacent window is set to , and the difference between the lower bound and the upper bound in a window is set to ​​;

[0063] According to the above requirements and the numerical value of the sequence itself, the number and range of windows and the points on the sequence in each window are determined;

[0064]

[0065] The search path range coordinates obtained above are the coordinates of the distance matrix, which correspond to the positions of the initial matrix. Since the above may only obtain part of the positions of the m*n matrix, the corresponding positions are still in the original sparse state (value 0, which can be understood as empty), and the non-empty positions are the range of the subsequent search path. When performing path search subsequently, we can only find the non-empty coordinate positions, so this is the search path range set according to the sequence itself.

[0066] S13. Setting the backtracking path range according to similar points

[0067] S131. Setting an initial sparse search path matrix M, the size of M is n*m, n is the length of the original sequence X, 5, m is the length of the original sequence Y, 5, and the value of each cell position in the matrix is 0, that is, a sparse cell;

[0068] S132. According to step S12, the search range of the original sequence X and the original sequence Y is obtained, and the corresponding positions of the similar points in the sequence X' and the sequence Y' in each window can correspond to the positions of the initial matrix M. That is, the initial search path range obtained according to the sequence similarity is [1, 0] [1, 2] [1, 3] [1, 4] [2, 0] [2, 2] [2, 3] [2, 4] [3, 0] [3, 2] [3, 3] [3, 4] [4, 0] [4, 2] [4, 3] [4, 4] [0, 1] [3, 1];

[0069] S133. Since the initial search path range involves the surrounding of the corresponding positions in the matrix which are still original sparse cells, in order to ensure the subsequent minimum path backtracking, it is necessary to set these sparse cells as search path ranges;

[0070] S14. Calculating the distance matrix according to the second-order formula

[0071] The search range of the sequence X and the sequence Y obtained in step S13 is used to calculate the alignment distance according to the alignment distance formula, and then the cumulative distance is calculated according to the cumulative distance formula to find the optimal alignment path, that is, the backtracking path range;

[0072]

[0073] In the formula, d ij is the alignment distance, x i is the i-th value of the sequence X, and yj is the jth value of sequence Y, i, j represent the ith and jth corresponding matrix unit of sequence X and sequence Y respectively;

[0074]

[0075] where D ij is the cumulative distance, d ij is the alignment distance, i, j represent the ith and jth corresponding matrix unit of sequence X and sequence Y respectively;

[0076] S15. According to the distance matrix, the similarity of the two sequences is 5 and the backtracking path (0, 2) (1, 3) (2, 4) (3, 4) (4, 4), that is, the corresponding coordinates of the two sequences;

[0077] A horizon tracking result of one seismic profile in the embodiment is shown in Figure 3 It can be seen from Figure 3 that the improved algorithm can obtain an accurate horizon in a seismic image, Figure 3 represents a horizon tracking result of one seismic profile;

[0078] S2. The sparse two-stage dynamic regularization algorithm is used for horizon tracking in seismic data with faults. The hypothesis testing principle is used to determine whether the horizons on both sides of the fault belong to a unified horizon. The specific method is as follows: according to the waveform similarity principle, when determining whether the Ith horizon and the Jth horizon belong to the same horizon, if they belong to the same horizon, the distributions are the same or similar, that is, the means of the two distributions are equal;

[0079] The distributions of the horizons on the left and right sides of the fault can be constructed by some amplitude parameter values of the horizon waveforms, that is, the mean and variance of the statistics can be calculated. According to the mathematical statistics theory, the statistics of the new F distribution can be obtained according to the statistics of the two distributions. Finally, according to the significance and critical value, a conclusion can be drawn whether they belong to the same horizon.

[0080] A three-dimensional horizon tracking effect diagram of the sparse two-stage dynamic regularization algorithm in the embodiment is shown in Figure 4 It can be seen from Figure 4 that the multi-horizon tracking method based on the sparse two-stage dynamic regularization algorithm can obtain continuous horizon tracking results, and the horizon continuity results in the fault part are also good;

[0081] S3. Seismic trace clustering

[0082] According to the idea of statistical clustering, the similarity between seismic traces is obtained by the sparse two-stage dynamic regularization algorithm. Similar seismic traces are classified into one class to obtain the clustering result of the seismic traces. One seismic trace is matched with a seismic trace of a different class.

[0083] The actual application diagram of the seismic trace clustering and horizon tracking in the embodiment is shown in Figure 5 Figure 5 It can be seen that after adding the clustering idea, the horizon tracking result is still good, and the time for ensuring the accuracy of the horizon tracking is reduced to obtain the horizon tracking result.​

Claims

1. A multi-lane auto pick method based on a sparse two-stage dynamic sizing algorithm, characterized in that, The method comprises the following steps performed in sequence: S1. Obtain the similarity and backtracking path of two sequences S11. Normalize the original sequence X and the original sequence Y The original sequence X and the original sequence Y are normalized to obtain the sequence X' and the sequence Y'; where S' is the normalized sequence, S i is the value of the i-th point of the sequence, max and min are the maximum and minimum values of the sequence, respectively. S12. Set the window range and quantity Set the parameter θ in the range of 0-1, set the window range and quantity according to the set parameter θ, determine the number and range of windows and the points on the sequence in each window, the points of the sequence belonging to the same window are similar points, and then obtain the search range of the original sequence X and the original sequence Y; S13. Set the backtracking path range according to the similar points S131. Set the initial sparse search path matrix M, the size of M is n*m, n is the length of the original sequence X, m is the length of the original sequence Y, and the value of each cell position in the matrix is 0, that is, a sparse cell; S132. According to the search range of the original sequence X and the original sequence Y obtained in step S12, the positions of the similar points in each window in the sequence X' and the sequence Y' can correspond to the positions of the initial matrix M, that is, the initial search path range obtained according to the sequence similarity; S133. Since the initial search path range involves the surrounding of the corresponding position in the matrix which is still the original sparse cell, in order to ensure the subsequent minimum path backtracking, it is necessary to set these sparse cells as the search path range; S14. Calculate the distance matrix according to the second-order formula The search range of the sequence X and the sequence Y obtained in step S13 is used to calculate the alignment distance according to the alignment distance formula, and then the cumulative distance is calculated according to the cumulative distance formula, and the optimal alignment path, that is, the backtracking path range, is found; where d ij is the alignment distance, x i is the ith value of sequence X, y j is the jth value of sequence Y, i, j represent the ith and jth corresponding matrix elements of sequence X and sequence Y, respectively; where D ij is the accumulated distance, d ij is the alignment distance, i and j represent the i-th and j-th corresponding matrix elements of sequences X and Y, respectively. S15. Obtain the similarity and backtracking path of two sequences according to the distance matrix; S2. Seismic trace clustering According to the idea of statistical clustering, the similarity between seismic traces is obtained by a two-stage dynamic warping algorithm, similar seismic traces are classified into the same class, and the clustering result of the seismic traces is obtained. One of the seismic traces is matched with the seismic traces of different classes; S3. Multi-layer automatic picking The matching result is obtained according to the interpolation method to obtain the horizon, that is, the multi-layer automatic picking is completed.

2. The multi-lane auto pick method based on the sparse two-stage dynamic normalization algorithm according to claim 1, characterized in that, When performing the multi-layer automatic picking, the improved dynamic time warping algorithm can be used for horizon tracking in seismic data with faults. According to the hypothesis testing principle, it is judged whether the horizons on both sides of the fault belong to the same horizon.

3. The multi-lane auto pick method based on the sparse two-stage dynamic sizing algorithm of claim 2, wherein, The specific steps of the hypothesis testing principle are: according to the waveform similarity principle, when judging whether the Ith layer and the Jth layer belong to the same horizon, if they belong to the same horizon, the distributions are the same or similar, that is, the means of the two distributions are equal.

4. The multi-lane auto pick method based on the sparse two-stage dynamic normalization algorithm of claim 1, wherein, In the step S12, the method of setting the window range and the number is that the difference between the upper limit of a window and the upper limit of the adjacent window is set as And the difference between the lower limit and the upper limit in a window is set as According to the method of setting the window range and quantity and the numerical size of the sequence itself, the number and range of windows and the points on the sequence in each window are determined.

5. The multi-lane auto pick method based on the sparse two-stage dynamic normalization algorithm according to any one of claims 1-4, characterized in that, In step S2, the method for obtaining the clustering result of the seismic traces is: setting the clustering similarity size sim and the boundary range width, and clustering the seismic traces smaller than sim and width into a class.

Images