A hidden fault identification method based on fault strike feature analysis

Abstract

This invention relates to a method for identifying hidden faults based on fracture strike characteristic analysis, belonging to the field of fault identification technology in seismic exploration of oil and gas fields. The method first selects the overlying strata with relatively continuous phase axes or good seismic imaging of the adjacent target layer as the reference layer, and then analyzes the development characteristics of the fault by angular scanning based on the coherence attribute data volume of the reference layer and the three-dimensional post-stack seismic data of the study area. Then, the fault development characteristics of the corresponding target layer are calculated using the fault angle parameters of the reference layer, so as to realize the identification of hidden faults in the target layer with weak continuous seismic reflection characteristics or poor seismic imaging.

Description

Technical Field

[0001] This invention relates to the field of fault identification technology in seismic exploration of oil and gas fields, and in particular to a method for identifying hidden faults based on fault strike characteristic analysis. Background Technology

[0002] A fault is a geological structure formed when underground rock strata are fractured due to compression or tension caused by crustal movement, and displaced along the fracture surface or zone. Fault structures have significant geological importance; oil and gas are often found near faults, which serve as major pathways for oil and gas migration and accumulation. In oil and gas exploration and development, fault interpretation is crucial for seismic data interpretation. Clearly locating faults and understanding their distribution is of paramount importance. The results of fault interpretation directly affect the accuracy of reservoir prediction and the efficiency and profitability of oil and gas extraction.

[0003] Currently, the commonly used fault prediction methods at home and abroad are mainly: fault identification methods based on seismic attribute fusion, automatic fault identification methods, and fault identification methods based on deep learning, among which the fault identification method based on seismic attribute fusion is the most widely used.

[0004] I. Principle of Fault Identification Method Based on Seismic Attribute Fusion

[0005] Seismic attribute analysis methods utilize the displacement or interruption of reflection layers exhibited by faults in seismic data for analysis, and achieve this by enhancing seismic attributes. The main seismic attribute-based methods include coherence volume technology and curvature attributes. In fault identification methods, coherence volume technology is widely used. It utilizes coherence volume attributes to represent the similarity of seismic data; that is, it converts three-dimensional amplitude data volumes into coherence attribute volumes through the calculation of attributes such as inter-trace similarity, thereby highlighting the development characteristics of faults and geological structures. Curvature attributes have good noise resistance; this method mainly identifies faults through changes in the undulation (curvature) of strata. Furthermore, using fusion attributes can fully tap into the information contained in the data, remove duplicate and confounding information, reduce the ambiguity of anomaly predictions, and further improve the accuracy of geological anomaly predictions.

[0006] II. Principle of Automatic Fault Identification Method

[0007] Automatic fault identification methods can automatically identify faults in seismic data volumes, effectively overcoming errors caused by subjective human experience. For example, ant tracking technology identifies fault planes and removes the influence of noise and non-fault elements. This method generally uses high-resolution processing, noise reduction, enhancement of phase axis continuity, and elimination of false structures to suppress noise and non-fault responses, and then utilizes ant tracking technology to identify faults.

[0008] III. Principles of Deep Learning-Based Tomography Recognition Methods

[0009] Deep learning is currently widely used in various fields and has achieved excellent results, demonstrating enormous potential. Earthquake data is massive, perfectly meeting the requirement of deep learning being driven by big data. Furthermore, the applications of deep learning in various fields share a similarity with fault identification, both requiring the extraction of image features. Traditional fault interpretation methods first require determining the fault location, which is very similar to the process of extracting regions of interest in image segmentation. Therefore, using deep learning for fault identification is a feasible and efficient method. Although fault identification methods and deep learning-based fault identification methods are gradually being implemented, some challenges and limitations still exist.

[0010] IV. Deficiencies of Existing Methods

[0011] 1. Among the multi-attribute fusion techniques, the representative RGB attribute fusion technique has certain shortcomings. This technique can only be applied to three or at most four seismic attributes, which cannot meet the requirements for analyzing a larger number of seismic attributes. Therefore, there is a problem of insufficient analysis of seismic attribute information.

[0012] 2. In the actual workspace, different models require different parameters. These model-specific hyperparameters will greatly affect the final recognition effect of the model. If we simply rely on manually adjusting the parameters, it will take a lot of time and it will be difficult to try as many parameter combinations as possible. Therefore, how to optimize the hyperparameters of different models in a more reasonable and convenient way is also an important problem that needs to be solved.

[0013] 3. Although automatic fault identification methods are gradually becoming automated, they still face some challenges. One major issue is that the accuracy of the identification results depends on multiple parameters. The selection of these parameters is crucial to the final identification effect and requires careful manual adjustment and optimization. Different parameter choices lead to different results, thus requiring experienced experts to adjust the parameters, which to some extent limits the full automation of the method. Deep learning-based fault identification methods use convolutional neural network models to predict faults. This method achieves automated and intelligent fault identification and improves the accuracy and efficiency of fault identification. However, some challenges and problems remain for fault prediction in actual work areas, such as: the lack of geological knowledge guidance when using seismic data for learning, the difficulty of model training, and low identification accuracy when the seismic data imaging quality is poor.

[0014] 4. Existing technologies are limited by data quality. Coherence and curvature attributes require good continuity of the phase axis. However, the accuracy of fault identification based on these two attributes is not high when dealing with problems such as poor quality deep seismic data and unclear imaging of the phase axis. Furthermore, the accuracy and efficiency of automatic fault identification methods and deep learning-based fault identification methods are significantly reduced in this situation.

[0015] Therefore, in cases where the quality of seismic data for the target layer is poor, the continuity of the phase axis is poor, or the quality of seismic imaging is poor, the above-mentioned conventional methods and techniques are difficult to achieve good results. This invention proposes a method for identifying hidden faults based on the analysis of fault strike characteristics. It uses known reference horizons and seismic data with good imaging quality to identify hidden faults in the strata of the target layer, thereby improving the reliability of fault identification. Summary of the Invention

[0016] This invention relates to a method for identifying hidden faults based on fracture strike characteristic analysis, belonging to the field of fault identification technology in oil and gas field seismic exploration. The method first inputs a 3D post-stack seismic data volume of the study area, the seismic horizon of the target segment, and a reference horizon. The reference horizon is selected from the overlying strata with relatively continuous phase axes or good seismic imaging adjacent to the target layer. Coherence analysis is performed on the 3D post-stack seismic data volume to obtain a 3D coherence data volume. Then, a coherence profile of a single main seismic line is extracted along the main seismic survey line, and a certain horizon point of the reference horizon on this profile is set as the analysis point. Fault development at various angles within the rectangular analysis window centered on the analysis point is analyzed by setting a rectangular analysis window and a multi-angle scanning analysis window (calculating all samples at a certain angle within the rectangular window). The root mean square (RMS) of the coherence values ​​of the points is calculated. If the RMS is less than a set threshold, it is determined that a fault has developed at that angle. For angles with fault development, the fault extension position (target point) of the target layer is determined based on the angle value and the position of the analysis points of the reference layer. The value of the target point is then modified to the RMS value corresponding to that angle. The calculation is repeated for all analysis points of the reference layer on the profile, and then for all main survey lines in the study area to obtain the fault analysis array of the target layer in the study area. Finally, the fault analysis array of the target layer is normalized to characterize the fault development features of the target layer.

[0017] The specific steps of this invention include:

[0018] (1) Input the three-dimensional post-stack seismic data volume of the study area, the reference layer data volume (select the overlying strata with relatively continuous seismic phase axes or good seismic imaging adjacent to the target layer) and the target layer layer data volume (with poor seismic phase axis continuity).

[0019] (2) Calculate the coherence properties of the three-dimensional post-stack data of the study area (A. Gersztenkorn and KJ Marfurt, 1999) to obtain the three-dimensional coherence volume attribute data of the study area;

[0020] (3) Extract the coherent body profile of a single main seismic survey line along the direction of the main seismic survey line, and calculate the distance between the reference layer and the target layer;

[0021] (4) For a single seismic coherence profile, a single CMP point on the reference horizon:

[0022] (4-1) Open an N×N rectangular time window for a single CMP point (the value of N is twice the distance between the reference layer and the target layer, obtained from step 3), and the CMP point is located at the center of the time window;

[0023] (4-2) Extract N×N data within the time window, and scan and calculate the root mean square value of the coherent volume attribute in this direction within the rectangular time window every 10° (total angle range 0-180°). When the root mean square value of the coherent volume attribute is less than the threshold value, it indicates that a fault has developed in this direction.

[0024] (4-3) For the N×N time window in step (4-1), extract the discrete data within the time window along the fault development direction in step (4-2), and fit the discrete data using the least squares fitting method to obtain a two-dimensional straight line characterizing the discrete points of the fault data of the reference layer. The horizontal axis of this line is the distance from the CMP point (the difference between the CMP point trace number) and the vertical axis is time.

[0025] (4-4) Extend the seismic fault data according to the two-dimensional straight line trend obtained in step (4-3) until it passes through the target layer, and obtain the intersection point of the fitted straight line and the target layer. The location of this intersection point is the development location of the fault in the target layer.

[0026] (4-5) Use the root mean square value of the coherence attribute in the N×N rectangular time window of the reference layer to replace the root mean square value of the coherence attribute in the N×N rectangular time window of the target layer intersection point, and set the coherence value of the angle direction of the target layer without fault development to 1.

[0027] (5) Iteratively calculate all seismic reference horizon CMP points in the current profile to obtain the target fault prediction data volume in the profile, i.e. the predicted hidden fault data volume.

[0028] (6) Calculate all main lateral line directions in a loop to obtain the three-dimensional hidden fault prediction data volume of the study area, and normalize the fault analysis array of the target layer to characterize the fault development characteristics of the target layer.

[0029] A method for identifying hidden faults based on fracture strike characteristic analysis has the following characteristics, mainly manifested as follows:

[0030] (1) This invention utilizes the advantages of continuous reference layer, phase axis and clear seismic fault section to establish the connection between the reference layer fault and the hidden fault of the target layer. By calculating the intersection of the fitted line of the reference layer fault and the target layer, the development location of the hidden fault of the target layer is obtained. Thus, it is not limited by the seismic data quality of the target layer and has strong innovation.

[0031] (2) The core basis of this invention is whether a connection can be established between the reference fault and the hidden fault of the target layer? Can the intersection of the fitted straight line of the reference fault and the target layer be used to indicate the development location of the hidden fault of the target layer? After consulting a large amount of data, the inventors found that earthquake sources are often deep, earthquake fault planes mostly extend from bottom to top, and the fault plane can be fitted with two-dimensional straight lines within a small longitudinal range. By using the results of two-dimensional straight line fitting within the longitudinal range of the fault, a connection can be established between the reference fault and the hidden fault of the target layer, thereby identifying the hidden fault of the target layer. Moreover, there is a methodological theoretical support: ① Earthquakes are mainly caused by the mutual compression and collision between tectonic plates on Earth. At this time, the Earth's interior releases a large amount of energy to the surface, causing the plate edges and interiors to move and fracture, thus forming a fault plane from bottom to top; ② The straight line obtained by fitting the earthquake fault plane using the least squares method within a small range can be equivalent to the extension trend of the fault plane within this range; Therefore, this invention has a very good methodological theoretical basis. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating the technical process of the present invention.

[0033] Figure 2 This is a plan view of the hidden fault in the target layer obtained using conventional methods;

[0034] Figure 3 This is a predicted planar map of the hidden fault in the target layer obtained using the method of this invention; Detailed Implementation

[0035] Example 1

[0036] A method for identifying hidden faults based on fracture strike characteristic analysis, comprising the following steps:

[0037] Step 1: Input the 3D seismic data volume of the study area, the reference layer and the target layer horizon data volumes. The reference layer horizon data volume is denoted as H. 上 (line, xline, t), the target layer's data volume is denoted as H. 下 (line,xline,t), where line is the line number, xline is the track number, and t is the time.

[0038] Step 2: Calculate the coherence properties of the three-dimensional data volume of the study area to obtain the three-dimensional coherence property data volume of the reference layer of the study area. The calculation method is as follows.

[0039] Given the covariance matrix of a 3D data volume within an analysis window,

[0040]

[0041] The coherence value of the C3 coherence algorithm based on eigenvalues ​​is defined as follows:

[0042]

[0043] Where, the denominator Tr(C) is the trace of the matrix, representing the energy of the covariance matrix, and the numerator λ max It is the largest eigenvalue, generally the first eigenvalue λ1, representing the dominant energy. j (j=1,2,…,J) are the eigenvalues ​​of the covariance matrix C, arranged in descending order. When the waveforms of all traces within the analysis window are completely consistent, i.e., no faults are developed, the eigenvalue coherence coefficient is equal to 1.

[0044] Step 3: Extract attribute profiles from the 3D coherence data along the main survey line, i.e., pr(xline,t,value), where value is the seismic coherence attribute value, and calculate the distance N = 2 × (t) between the reference layer and the target layer. 下 -t 上 ), t 下 It is the time of the target layer, t 上 It is the time of the reference layer;

[0045] Step 4: For a single CMP point within a single seismic attribute profile, the coordinates are (xline0, t0).

[0046] Step 4-1: Open an N×N time window for a single CMP point, and the CMP point is located at the center of the time window. The value of N is twice the distance between the reference layer and the target layer, which is obtained from step 3.

[0047] Step 4-2: Extract N×N data within the time window. Every 10°, with a total angle range of 0-180°, calculate the root mean square of the coherent volume attribute value in this direction within the time window. When the root mean square of the coherent volume attribute value is <= 0.3, 0.3 is the set threshold value. That is, assuming that the coherent volume attribute value is lower than 0.3, it is a fault development point. Then, this direction is a fault development direction within this time window.

[0048] Step 4-3: For the N×N time window in Step 4-1, extract all discrete data within this direction along the fault development direction in Step 4-2. That is, extract the x-coordinate (trace number xline) and ordinate (time t) of the discrete data, denoted as the reference layer fault data f(xline-xline0,t-t0). Fit the earthquake fault data f(xline-xline0,t-t0) using the least squares fitting method. The fitting x-coordinate range is from xline to xline0, resulting in a two-dimensional straight line T(xline1) = a + b × xline1 representing the discrete points of the fault data f(xline-xline0,t-t0). The x-coordinate of this line is the distance between the earthquake fault f(xline-xline0,t-t0) and the CMP point (xline0,t0), where xline1 = xline0. n -xline0,n=1,2,…,n, where the ordinate is time t, and the least squares formula is as follows.

[0049] The fitted straight line is

[0050] y = a + bx (3)

[0051] and

[0052]

[0053] Where a and b are the parameters of the fitted line, which are obtained by combining formulas 4 and 5; x is the difference in channel number on the horizontal axis; y is the difference in time on the vertical axis; and n is the number of discrete data points to be fitted.

[0054] Step 4-4: Extend the earthquake fault data f(xline-xline0,t-t0) according to the trend of the two-dimensional straight line T(xline1) = a + b × xline1 obtained in Step 4-2. The extension point data is the value of the two-dimensional straight line data at the extension point, until it passes through the target layer. The extended two-dimensional straight line data volume is used as the target layer hidden fault data fpre(xline,t) corresponding to the reference layer fault data f(xline-xline0,t-t0).

[0055] Steps 4-5: Replace the root mean square value of the coherence attribute in the N×N rectangular time window of the target layer with the root mean square value of the coherence attribute in the N×N rectangular time window of the reference layer, and set the angular direction coherence value of the target layer without fault development to 1.

[0056] Step 5: Iteratively calculate all CMP points within the current profile to obtain the hidden fault prediction data volume fpre(xline,t) of the target layer within the profile, which is the hidden fault prediction data volume of the profile.

[0057] Step 6: Iteratively calculate all main survey line directions to obtain the three-dimensional hidden fault prediction data volume Fre(line,xline,t) of the study area, and perform normalization processing to obtain the coherent data volume characterizing the fault development features of the target layer.

[0058] Example 2

[0059] Figure 2 and Figure 3 These are examples of hidden fault prediction for the target segment in a certain study area. Figure 2 This is a plan view of the predicted results of conventional overlapping concealed faults in the study area. Figure 3 The hidden fault prediction planar map obtained using the present invention shows that the crack prediction effect using the present invention is better.

[0060] 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. The specific steps of a method for identifying hidden faults based on fault strike characteristic analysis include: Step 1: Input the 3D seismic data volume of the study area, the reference layer and the target layer layer data volume. The reference layer data volume is denoted as H_upper(line,xline,t), and the target layer layer data volume is denoted as H_lower(line,xline,t), where line is the line number, xline is the trace number, and t is the time. Step 2: Calculate the coherence properties of the three-dimensional data volume of the study area to obtain the three-dimensional coherence property data volume of the reference layer of the study area. The calculation method is as follows. Given the covariance matrix of a 3D data volume within an analysis window, The coherence value of the C3 coherence algorithm based on eigenvalues ​​is defined as follows: Where, the denominator Tr(C) is the trace of the matrix, representing the energy of the covariance matrix, and the numerator λ max It is the largest eigenvalue, generally the first eigenvalue λ1, representing the dominant energy. j (j=1,2,…,J) are the eigenvalues ​​of the covariance matrix C, arranged in descending order. When the waveforms of all traces within the analysis window are completely consistent, i.e., no faults are developed, the eigenvalue coherence coefficient is equal to 1. Step 3: Extract the attribute profile from the three-dimensional coherence data along the main survey line direction, which is pr(xline,t,value), where value is the seismic coherence attribute value, and calculate the distance between the reference layer and the target layer N = 2 × (t_lower - t_upper), where t_lower is the time of the target layer and t_upper is the time of the reference layer. Step 4: For a single CMP point within a single seismic attribute profile, the coordinates are (xline0, t0). Step 4-1: Open an N×N time window for a single CMP point, and the CMP point is located at the center of the time window. The value of N is twice the distance between the reference layer and the target layer, which is obtained from step 3. Step 4-2: Extract N×N data within the time window. Every 10°, with a total angle range of 0-180°, calculate the root mean square of the coherent volume attribute value in this direction within the time window. When the root mean square of the coherent volume attribute value is <= 0.3, 0.3 is the set threshold value. That is, assuming that the coherent volume attribute value is lower than 0.3, it is a fault development point. Then, this direction is a fault development direction within this time window. Step 4-3: For the N×N time window in Step 4-1, extract all discrete data within this direction along the fault development direction in Step 4-2. That is, extract the x-coordinate (trace number xline) and y-coordinate (time t) of the discrete data, denoted as the reference layer fault data f(xline-xline0,t-t0). Fit the earthquake fault data f(xline-xline0,t-t0) using the least squares fitting method. The fitting x-coordinate range is from xline to xline0, resulting in a two-dimensional straight line T(xline1) = a + b × xline1 representing the discrete points of the fault data f(xline-xline0,t-t0). The x-coordinate of this line is the distance between the earthquake fault f(xline-xline0,t-t0) and the CMP point (xline0,t0), where xline1 = xlinen - xline0, n = 1, 2, ..., n. The y-coordinate is time t. The least squares formula is as follows. The fitted straight line is y = a + bx (3) and Where a and b are the parameters of the fitted line, which are obtained by combining formulas 4 and 5; x is the difference in channel number on the horizontal axis; y is the difference in time on the vertical axis; and n is the number of discrete data points to be fitted. Step 4-4: Extend the earthquake fault data f(xline-xline0,t-t0) according to the trend of the two-dimensional straight line T(xline1) = a + b × xline1 obtained in Step 4-2. The extension point data is the value of the two-dimensional straight line data at the extension point, until it passes through the target layer. The extended two-dimensional straight line data volume is used as the target layer hidden fault data fpre(xline,t) corresponding to the reference layer fault data f(xline-xline0,t-t0). Steps 4-5: Replace the root mean square value of the coherence attribute in the N×N rectangular time window of the target layer with the root mean square value of the coherence attribute in the N×N rectangular time window of the reference layer, and set the angular direction coherence value of the target layer without fault development to 1. Step 5: Iteratively calculate all CMP points within the current profile to obtain the hidden fault prediction data volume fpre(xline,t) of the target layer within the profile, which is the hidden fault prediction data volume of the profile. Step 6: Iteratively calculate all main survey line directions to obtain the three-dimensional hidden fault prediction data volume Fre(line,xline,t) of the study area, and perform normalization processing to obtain the coherent data volume characterizing the fault development features of the target layer.

Images