Shale fracture seismic identification method based on 3D U-Net convolutional neural network combined with ant tracking

Abstract

The present application relates to a kind of mud shale fracture seismic identification method based on 3D U-Net convolutional neural network combined with ant tracking.The technical scheme is: based on single well data to establish the geological model of mud shale layer fracture, its forward modeling is clear mud shale fracture identification advantage frequency band;Frequency band data body is obtained using frequency spectrum peak value, and 3D U-Net convolutional neural network combined with ant tracking calculation is carried out, 3D U-Net ant body is obtained, and is verified using microseismic data, extracts the attribute of 3D U-Net ant body along layer, and the regional plane distribution characteristics of mud shale fracture are clear.The beneficial effects are as follows: single well and seismic data are combined to predict fracture, and the effect of exploration cost reduction is remarkable;Seismic forward-spectrum peak decomposition-advantage frequency band data calculation comprehensive combination significantly improves the seismic resolution of mud shale fracture;Using data body extraction along layer slice, the fracture identification range is expanded.

Description

Technical Field

[0001] This invention relates to the field of seismic attribute identification of shale and mudstone fractures, specifically a seismic identification method for shale and mudstone fractures based on a 3D U-Net convolutional neural network combined with ant tracking. Background Technology

[0002] Currently, the identification of fractures in shale and mudstone remains a key challenge in shale formation risk exploration, severely restricting the exploration and development of shale gas resources. Besides core CT scanning, other methods for identifying fractures in shale and mudstone, such as single-attribute prediction, anisotropy, and fracture parameter inversion, have unsatisfactory prediction accuracy. CT scanning relies on well cores and can only identify fractures within a small area of ​​the well core; its results are insufficient for understanding the fractures in the entire shale and mudstone region. In recent years, geologists have successfully identified small faults using convolutional neural network image processing technology from the biomedical field, but effective research on fracture identification in shale and mudstone has yet to be conducted. Ant tracking is highly sensitive to fracture differences in data volumes and is usually combined with attributes such as coherence volume and curvature for fracture identification. While it has achieved accurate identification of carbonate faults with high fracture reflection characteristics, its results are subject to excessive interference in the identification of fractures in shale and mudstone where fracture reflection characteristics are not obvious, and the fracture identification effect needs improvement. Therefore, developing a new method that can efficiently and accurately identify fractures in shale is of great significance for promoting the in-depth exploration and development of shale gas resources.

[0003] Well 215 is located in the Zigong configuration area of ​​the southwestern Sichuan low-fold tectonic belt in the Sichuan Basin. The high-yield oil and gas flow obtained by Well 215 in the Longmaxi Formation confirms the good production potential of shale gas in the Longmaxi Formation. However, the H1 drilling platform in Well 215 has experienced a casing deformation accident during the production of shale gas in the Longmaxi Formation. Through microseismic signal monitoring and analysis, it is preliminarily suspected that this accident was caused by hydraulic fracturing activating the northwest-trending fracture.

[0004] Because the Longmaxi Formation consists of shale and mudstone layers buried at depths greater than 3500m, the seismic data resolution is low, and the seismic response of fractures is weak, making it impossible to identify micro-faults and fractures using conventional methods. Furthermore, the fracture prediction techniques used for pre- and post-stack seismic events in the Zi 215 well area are limited in scope and accuracy, and validation standards have not yet been established, making it difficult to adequately support fracturing optimization design and hindering subsequent production operations in the Zi 215 well area. To address this issue, this study utilizes a 3DU-Net convolutional neural network combined with ant tracking technology to improve the accuracy of shale and mudstone fracture prediction, aiming to clarify the fracture distribution patterns and provide strong support for safe production and efficient gas extraction in the Zi 215 well area. Summary of the Invention

[0005] The purpose of this invention is to address the aforementioned deficiencies in existing technologies by providing a seismic identification method for shale and mudstone fractures based on a 3D U-Net convolutional neural network combined with ant tracking. Considering the impact of seismic data resolution on fracture identification accuracy, this invention first utilizes seismic forward modeling to identify the dominant spectral peak frequency band for fracture identification. Then, based on the dominant frequency band from the forward modeling, the seismic data is spectrally decomposed to obtain seismic data with the optimal resolution for shale and mudstone fracture identification. Finally, the method is validated through a combination of a 3D U-Net convolutional neural network and ant tracking technology, along with microseismic monitoring. This shale and mudstone fracture identification method offers advantages such as low exploration cost, high identification accuracy, and wide identification range. It can effectively guide well location deployment in shale and mudstone formations, directly impacting shale gas extraction costs and increasing production capacity.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A seismic identification method for shale and mudstone fractures based on 3D U-Net convolutional neural network and ant tracking includes the following steps:

[0008] (1) Main data preparation for fracture identification: collect logging data (including lithology, sonic transit time and density) and 3D seismic data of single wells, and understand the spectral peak characteristics and effective frequency band range of 3D seismic data;

[0009] (2) Establish a forward model of shale fractures using well logging data, and perform seismic forward modeling on different peak frequency bands based on actual seismic parameters to identify the dominant frequency band for shale fracture identification;

[0010] (3) Use spectral decomposition technology to perform frequency division processing on seismic data so that its frequency band is consistent with the dominant frequency band for identifying shale fractures;

[0011] (4) Perform 3D U-Net convolutional neural network calculation on the seismic data volume of the dominant frequency band for identifying shale fractures to obtain the 3D U-Net data volume;

[0012] (5) Ant tracking calculations were performed on the 3D U-Net volume to obtain the final mudstone and shale fracture identification data volume. Microseismic data was used to verify the accuracy of the mudstone and shale fracture prediction of the 3D U-Net ant data volume.

[0013] (6) Regional mudstone and shale fracture prediction results can be obtained by extracting the data volume of fracture identification along the mudstone and shale layers.

[0014] The advantages of this invention compared to the prior art are as follows:

[0015] (1) The combination of single-well data and seismic data can be used to predict mudstone and shale fractures, which significantly reduces exploration costs.

[0016] (2) The seismic resolution of shale fractures is maximized by using a three-in-one method of seismic forward modeling, peak spectral decomposition, and dominant frequency band data calculation.

[0017] (3) The accuracy of crack identification is greatly improved by combining 3D U-Net convolutional neural network with ant tracking and verification by microseismic data profile.

[0018] (4) Using data volume to extract slices along the layer, the crack identification range is wider. Attached Figure Description

[0019] Appendix Figure 1 This is an analysis diagram of the peak values ​​and effective frequency bands of the three-dimensional data from well area 215 in this invention example;

[0020] Appendix Figure 2 This is a fracture geological model diagram established based on well data from well 215 in this invention example;

[0021] Appendix Figure 3 This invention presents seismic forward modeling results of different peak frequency band signals based on the actual seismic dominant frequency wavelet of a geological model of mudstone and shale fractures.

[0022] Appendix Figure 4 This is a seismic data profile of the dominant frequency band identified from shale fractures in well 215 in this invention example;

[0023] Appendix Figure 5 This is an overlay image of the 3D U-Net data volume and seismic volume from well area 215 in this invention example;

[0024] Appendix Figure 6 This is a 3D U-Net ant body profile and microseismic overlay image from well 215 through wells H59 and H3 in an example of the present invention;

[0025] Appendix Figure 7 This is an example of the invention showing the overlay of variance volume, 3D U-Net ant volume, and 3DU-Net ant volume along the formation near well H1 in the Longmaxi Formation of the 215 well area with microseismic data. Detailed Implementation

[0026] To make the technical means and objectives of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0027] Unless otherwise specified, the methods used in the following embodiments are conventional methods.

[0028] Combined with appendix Figure 1-7The present invention will be further described below:

[0029] Step 1: Preparation of key data for earthquake prediction of shale and mudstone fractures;

[0030] This invention takes the Zigong 215 well area in southern Sichuan as an example;

[0031] Three-dimensional seismic data collected from Well 215, along with dynamic and static data such as lithology, sonic transit time, density logging, and microseismic data, were used to understand the spectral characteristics of the block's seismic bodies, such as... Figure 1 As shown, it is clear that the effective frequency band for earthquake data is when the earthquake amplitude energy is greater than 0.3, i.e., 10-50Hz. Within the effective frequency band, the spectrum exhibits multiple peak characteristics, with 17Hz, 25Hz, 32Hz, and 43Hz being its spectral peaks.

[0032] Step 2: As Figure 2 As shown, based on well logging data of lithology, sonic transit time, and density from Well 215, a geological model of fractured mudstone and shale layers in the Well 215 area was established. The fractures were treated as small-displacement faults, and geological models with fault displacements of 0.2m, 0.4m, 0.6m, 0.8m, 1.0m, and 2.0m were established. Figure 3 As shown, seismic forward modeling of different peak frequency bands within the effective frequency band of seismic data was performed. It was found that cracks with a displacement of less than 2m could not be identified from the seismic reflection axis characteristics. However, when the dominant frequency was 25Hz, the brightness and darkness of the amplitude corresponded well with the location of crack development. Therefore, the dominant frequency of 25Hz is the dominant frequency band for identifying mudstone and shale cracks in the 215 well area.

[0033] Step 3: As Figure 4 As shown, based on the seismic forward modeling results and the spectral characteristics of the 3D seismic data, the seismic data is spectrally decomposed to obtain the dominant frequency band data volume for identifying shale fractures with a center frequency of 25Hz.

[0034] Step 4: As Figure 5 As shown, the 3DU-Net data volume is obtained by performing 3D U-Net convolutional neural network calculation on the dominant frequency band data volume for identifying shale fractures;

[0035] Step 5: As Figure 6 As shown, ant tracking calculations were performed on the 3DU-Net data volume to obtain the 3DU-Net ant volume, and the accuracy of the 3DU-Net ant volume was verified using microseismic data;

[0036] Step 6: As Figure 7As shown, regional fracture prediction results can be obtained by extracting the 3DU-Net ant body attributes along the layer. Analysis of fracture identification results from the H1 level in the 215 well area shows poor performance on conventional seismic attributes such as variance volume. However, the ant algorithm combined with the 3D U-Net convolutional neural network not only identified the NW-trending fractures from the H1 platform, but also identified two NE-trending large faults and one near-E-W micro-fault running through the entire H1 platform, as well as a series of NW-trending fractures. The microseismic signals showed a high degree of agreement with the results, proving that it has high accuracy in predicting shale fractures.

[0037] The beneficial effects of this invention are:

[0038] (1) The combination of single-well data and seismic data can be used to predict mudstone and shale fractures, which significantly reduces exploration costs.

[0039] (2) The seismic resolution of shale fractures is maximized by using a three-in-one method of seismic forward modeling, peak spectral decomposition, and dominant frequency band data calculation.

[0040] (3) The accuracy of crack identification is greatly improved by combining 3D U-Net convolutional neural network with ant tracking and verification by microseismic data profile.

[0041] (4) Using data volume to extract slices along the layer, the crack identification range is wider.

Claims

1. A seismic identification method for shale and mudstone fractures based on 3D U-Net convolutional neural network and ant tracking, characterized by: Considering the impact of seismic data resolution on the accuracy of shale fracture identification, this paper first establishes a geological model of underground shale fractures using lithology, sonic transit time, and density data from a single well. Seismic forward modeling is then used to identify the dominant peak frequency band for shale fracture identification. Next, spectral peak decomposition technology is employed to obtain seismic data with the optimal resolution for shale fracture identification. Finally, a seismic identification method for shale fractures based on 3D U-Net convolutional neural network combined with ant tracking technology and microseismic monitoring is established for verification. The method includes the following steps: (1) Preparation of main data for earthquake prediction of mudstone and shale fractures: Taking the Zigong area of ​​southern Sichuan, specifically the Zi 215 well area, as an example, this invention collects three-dimensional seismic data from the Zi 215 well, as well as dynamic and static data such as lithology, sonic transit time and density logging data, and microseismic data; and understands the peak characteristics and effective frequency bands of the seismic body spectrum in the block. (2) Based on the logging data of lithology, sonic transit time and density of well 215, establish a geological model of the mudstone and shale layer fracture in the well 215 area, and perform seismic forward modeling of different frequency band signals within the effective frequency band of the seismic data in step (1) to clarify the seismic response characteristics of mudstone and shale fractures and select the optimal frequency band for seismic identification of mudstone and shale fractures. (3) Based on the seismic forward modeling mudstone and shale fracture response characteristics obtained in step (2) and the spectral characteristics of the three-dimensional seismic data in step (1), the seismic data is decomposed into spectral peak values ​​to obtain the dominant frequency band data volume for mudstone and shale fracture identification. (4) Perform 3D U-Net convolutional neural network calculations on the dominant frequency band data volume of mudstone and shale fracture identification obtained in step (3) to obtain 3DU-Net data volume; (5) Perform ant tracking calculations on the 3DU-Net data volume from step (4) to obtain the 3DU-Net ant volume, and verify the accuracy of the 3DU-Net ant volume using microseismic data. (6) By extracting the layer-by-layer attributes of the 3DU-Net ant bodies in step (5), the regional mudstone and shale fracture prediction results can be obtained, and the intuitive fracture plane distribution results can be obtained.

2. The seismic identification method for shale and mudstone fractures based on 3D U-Net convolutional neural network and ant tracking as described in claim 1, characterized in that: The aforementioned peak frequency decomposition refers to performing seismic forward modeling based on the peak frequency characteristics and effective frequency bands of seismic data, selecting the peak frequency band of the optimal seismic response of shale and mudstone fractures, and using spectral decomposition to decompose the peak frequency band to obtain advantageous data for identifying shale and mudstone fractures.

3. The seismic identification method for shale and mudstone fractures based on 3D U-Net convolutional neural network and ant tracking as described in claim 1, is characterized in that: The 3D U-Net convolutional neural network combined with ant tracking technology is used to calculate the 3D U-Net data volume of the dominant frequency band data volume of shale fracture identification using the 3D U-Net convolutional neural network, and then ant tracking calculation is performed on it to obtain the 3D U-Net ant body.

4. The seismic identification method for shale and mudstone fractures based on 3D U-Net convolutional neural network and ant tracking as described in claim 1, characterized in that: The accuracy of the microseismic data verification was verified by overlaying profile and planar microseismic data with 3D U-Net ant-like data.

Images