Seismic exploration deep learning multichannel sample generation method and system

Abstract

The invention belongs to the technical field of oil-gas exploration data processing, and discloses a seismic exploration deep learning multichannel sample generation method and system. The method comprises the following steps: firstly, loading a plurality of pieces of three-dimensional seismic attribute data, setting a sampling direction, a sampling interval and a sampling size of a sample point according to a data range, respectively extracting two-dimensional matrix data from each piece of seismic attribute data along the sampling direction by taking the sample point as a center, and taking the two-dimensional matrix data of each attribute of a sample as channel information; samples are formed in a TIFF image format, and each sample point is traversed to form a sample data set of a whole region. According to the method, the spatial correlation of the geologic body can be better reflected, so that the deep learning algorithm can learn spatial correlation characteristics of seismic data, and various mature convolutional neural networks based on image data can be directly used to train and predict the formed sample set.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas exploration data processing technology, and in particular relates to a method and system for generating multi-channel samples using deep learning in seismic exploration. Background Technology

[0002] With the development of artificial intelligence in the field of petroleum geophysical exploration, deep learning methods have been widely applied in oil and gas seismic exploration, such as noise suppression, seismic velocity picking, fault identification, reservoir prediction, lithology identification, and physical property prediction. In these application scenarios, the general process is divided into the following steps: sample set construction, deep learning network structure design, model training, and model application. Among these, designing a reasonable deep learning network structure for the application scenario is the key technology. Based on this, the deep learning model is trained using a sample set, and the quantity and representativeness of the samples have a decisive impact on the model's accuracy.

[0003] In current applications of artificial intelligence in seismic exploration, the acquired samples are mainly of two types: digital samples and image samples. Due to format differences, digital samples generally require redesigning the network structure and validating its effectiveness when applying deep learning algorithms, and there are currently few successful cases. Image samples can use some validated classic deep learning network algorithms, but the content of current image samples is relatively simple and lacks the ability to fuse data with multiple different attributes.

[0004] The invention patent application (publication number CN113902769B, publication date 20240705) discloses an earthquake fault identification method based on deep learning semantic segmentation. The patent application extracts two-dimensional earthquake amplitude images from three-dimensional earthquake data by slicing them as samples for learning. The patent application mainly focuses on the identification method, and its sample generation method is a general and universal method.

[0005] Invention patent application (publication number CN113805235B, publication date 20231226) discloses a method and apparatus for three-dimensional seismic facies identification based on convolutional neural networks. The method includes: acquiring three-dimensional post-stack seismic data to be identified; acquiring model parameters obtained by training a training sub-network model based on a convolutional neural network, and assigning these parameters to a prediction sub-network model based on the convolutional neural network; inputting the three-dimensional post-stack seismic data to be identified into the prediction sub-network model using the model parameters, and outputting the seismic facies identification result of the three-dimensional post-stack seismic data to be identified. This patent application does not involve a sample generation method.

[0006] The invention patent application (publication number CN111126471A, publication date 20200508) relates to a method and system for detecting microseismic events. The method involves the following steps: acquiring microseismic signals from multiple monitoring stations during fracturing, establishing a training dataset and a test dataset; establishing a convolutional neural network (CNN) model, inputting training dataset samples into the CNN model for training, and inputting test dataset samples into the trained CNN model to verify its performance; storing the parameters of the trained CNN model; acquiring microseismic signals from multiple monitoring stations during real-time fracturing, establishing a test dataset; inputting the test dataset data into the trained CNN model for detection, obtaining the sample classification results of waveform data from each monitoring station, and determining whether a microseismic event exists based on the sample classification results. The main sample data generation method involved in this patent is "selecting vertical component waveform data from microseismic signals for normalization and grouping processing, with each 256 sampling points serving as a data sample."

[0007] This invention patent application (Publication No. CN111382799B, Publication Date 20230428) relates to a method for processing seismic fault images, which solves the shortcomings of unclear seismic fault images and difficulty in interpreting seismic faults compared with existing technologies. The invention includes the following steps: acquiring a seismic fault image dataset; preprocessing the seismic fault image dataset; constructing a seismic fault image processing network; training the seismic fault image processing network; acquiring the seismic fault image data to be processed; and processing the seismic fault images. This application generates geological data that closely resembles nature using a small amount of labeled seismic data, solving the problem of insufficient labeled seismic data, and utilizes a three-dimensional convolutional integral network to judge unknown data, improving the efficiency of seismic fault identification. The main focus of this application is the identification method.

[0008] An invention patent application (publication number CN115639592A, publication date 20230124) provides a method and apparatus for collecting and labeling big data samples in seismic exploration. The method includes: Step 1, loading seismic data from the work area and preprocessing the data; Step 2, interactively setting parameters for digital and image samples; Step 3, interactively drawing the boundaries for digital sample collection and interactively setting the range and size for image samples; Step 4, collecting both digital and image samples; Step 5, labeling the digital and image samples; and Step 6, publishing the seismic exploration big data samples. Steps 2 to 4 of this patent application only vaguely describe the interactive setting for image sample collection, without clearly specifying the sampling method. Furthermore, the application has certain technical deficiencies in its method of multi-data fusion sampling and outputting images as multi-channel samples.

[0009] Based on the above analysis, the problems and shortcomings of the existing technology are as follows: the existing technology cannot conveniently utilize the sample acquisition scheme of existing deep learning network algorithms in seismic exploration, and the use of multi-attribute data fusion and multi-channel image format storage methods results in poor quality of sample datasets for the entire area. Summary of the Invention

[0010] To overcome the problems existing in related technologies, the present invention discloses a method and system for generating multi-channel samples using deep learning in seismic exploration. The sample acquisition methods in existing technologies differ significantly from the sample acquisition and storage methods of the present invention. The present invention provides a better sample acquisition technology solution for deep learning-based artificial intelligence methods in seismic exploration.

[0011] The technical solution is as follows: A deep learning-based multi-channel sample generation method for seismic exploration, the method comprising the following steps:

[0012] S1 loads one or more three-dimensional seismic attribute data volumes within the study area;

[0013] S2 represents the three-dimensional spatial range of the study area. Sample points are set at certain sampling intervals to divide the entire area into multiple sample points, and the sampling direction and sampling size of the sample points are set.

[0014] S3, for each sample point in the study area, with the sample point as the center, extract data for each seismic attribute data according to the sampling direction and size to form a two-dimensional matrix data;

[0015] S4, each two-dimensional matrix is ​​treated as a channel data, and all channel data constitute a complete sample data, which is saved as a single sample data in the form of a multi-channel image;

[0016] S5. For each sample point set in step S2, repeat steps S3 and S4 to obtain the sample dataset of the study area.

[0017] Furthermore, in step S1, multiple three-dimensional seismic attribute data are selected, and the selection of attribute data volumes is based on the principle of having a significant characterizing effect on the research objectives.

[0018] Furthermore, in step S2, the sampling direction of the sample points is one of Inline, Crossline, or Time, and the sampling interval and size are set in the other two directions; the sampling size should be greater than or equal to the sampling interval.

[0019] Furthermore, in step S3, with each seismic sampling point as the center, data expansion is performed in two dimensions along the sampling direction and size, and two-dimensional matrix data is extracted from each attribute data volume.

[0020] Furthermore, in step S4, each two-dimensional matrix formed above is used as channel data and stored in multi-channel TIFF image format to realize multi-channel seismic exploration sample acquisition.

[0021] Furthermore, in step S4, multiple seismic attribute data are fused to form a multi-channel sample to characterize the research target, and a deep learning algorithm is used to learn the spatial correlation characteristics of the seismic data.

[0022] Furthermore, the seismic exploration deep learning multi-channel sample generation method is mounted on a computer device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor. When the processor executes the computer program, it implements the steps of the seismic exploration deep learning multi-channel sample generation method described above.

[0023] Another object of the present invention is to provide a deep learning multi-channel sample generation system for seismic exploration, comprising:

[0024] The seismic attribute data volume loading module is used to load one or more three-dimensional seismic attribute data volumes within the study area;

[0025] The sample point setting module is used to set sample points in the three-dimensional space of the study area at a certain sampling interval, thereby dividing the entire area into multiple sample points, and setting the sampling direction and sampling size of the sample points;

[0026] The two-dimensional matrix forming module is used to extract data from each seismic attribute data in the study area, with the sample point as the center, according to the sampling direction and size, and form two-dimensional matrix data.

[0027] The sample storage module is used to treat each two-dimensional matrix as a channel data, and all channel data constitute a complete sample data, which is saved as a single sample data in a multi-channel image mode.

[0028] The study area sample dataset acquisition module is used to repeat the two-dimensional matrix formation and sample storage steps for each set sample point to obtain the study area sample dataset.

[0029] Furthermore, the deep learning multi-channel sample generation system for seismic exploration is applied to noise suppression, seismic velocity picking, fault identification, reservoir prediction, lithology identification, and physical property prediction in oil and gas seismic exploration.

[0030] Combining all the above technical solutions, the beneficial effects of this invention are as follows: This invention loads one or more three-dimensional seismic attribute data volumes for the study area; it sets sample points at certain sampling intervals within the three-dimensional spatial range of the study area, thereby dividing the entire area into multiple sample points, and sets the sampling direction and sampling size of the sample points; for each sample point, with the sample point as the center, it extracts data from each seismic attribute data according to the sampling direction and size to form two-dimensional matrix data; it treats each two-dimensional matrix as a channel data, and all channel data constitute a complete sample data, which is saved as a single sample data in a multi-channel image manner. This process is repeated for each sample point to obtain the sample dataset for the entire area.

[0031] This deep learning-based multi-channel sample generation method for seismic exploration, on the one hand, forms multi-channel samples by fusing multiple seismic attribute data, which can better represent the research target; on the other hand, it expands the feature values ​​of seismic data sample points in two dimensions, which can better reflect the spatial correlation of geological bodies, thus making it more conducive for deep learning algorithms to learn the spatial correlation features of seismic data, and can directly use various mature convolutional neural networks based on image data to train and predict the formed sample set.

[0032] Compared to existing technologies, this invention addresses the lack of a sample acquisition scheme in deep learning-based seismic exploration artificial intelligence methods that can integrate various attribute data and conveniently utilize existing deep learning network algorithms. It provides a deep learning sample generation method by employing multi-attribute data fusion and multi-channel image format storage. This method first loads multiple three-dimensional seismic attribute data, sets the sampling direction, sampling interval, and sampling size of sample points according to the data range, and extracts a two-dimensional matrix data for each seismic attribute data along the sampling direction, centered on the sample point. The two-dimensional matrix data of each attribute of the sample is used as channel information to form a sample in TIFF image format. This process is repeated for each sample point to form a sample dataset covering the entire area.

[0033] This invention is innovative in the field of oil and gas exploration data processing. The method and system have advantages such as applicability to multiple application scenarios, high data acquisition efficiency, and ease of user operation. It has the following two significant effects:

[0034] (1) It can efficiently and conveniently extract big data samples from seismic exploration data volumes and has universal applicability in the field of intelligent seismic exploration.

[0035] (2) The format of big data samples in seismic exploration is unified by multi-channel image format, and the existing deep learning algorithms can be fully utilized for training and learning, laying a good foundation for intelligent multi-scenario application of seismic exploration. Attached Figure Description

[0036] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure;

[0037] Figure 1 This is a flowchart of the deep learning multi-channel sample generation method for seismic exploration provided in this embodiment of the invention;

[0038] Figure 2 This is an inline direction data slice of the seismic amplitude and seismic attributes of the study area provided in this embodiment of the invention;

[0039] Figure 3 This is an inline direction data slice of the instantaneous amplitude seismic attributes of the study area provided in this embodiment of the invention;

[0040] Figure 4 This is an inline direction data slice of the instantaneous phase seismic attributes of the study area provided in this embodiment of the invention;

[0041] Figure 5 This is an inline direction data slice of the instantaneous frequency seismic attributes of the study area provided in this embodiment of the invention;

[0042] Figure 6 This is a schematic diagram of the inline, crossline, and time directions of a three-dimensional seismic work area provided in this embodiment of the invention;

[0043] Figure 7 This is a schematic diagram of sample division on each inline data slice provided in an embodiment of the present invention;

[0044] Figure 8 This is a schematic diagram of code written in Python for storing and retrieving multi-channel sample data, provided in an embodiment of the present invention.

[0045] Figure 9 This is a system structure diagram of a specific embodiment of the seismic exploration deep learning multi-channel sample generation system of the present invention. Detailed Implementation

[0046] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0047] Example 1, such as Figure 1As shown, the seismic exploration deep learning multi-channel sample generation method provided in this embodiment of the invention can realize sample acquisition for multi-seismic attribute data fusion and storage of multi-channel image sample formats, specifically including:

[0048] S1 loads one or more three-dimensional seismic attribute data volumes within the study area;

[0049] S2 represents the three-dimensional spatial range of the study area. Sample points are set at certain sampling intervals to divide the entire area into multiple sample points, and the sampling direction and sampling size of the sample points are set.

[0050] S3, for each sample point in the study area, with the sample point as the center, extract data for each seismic attribute data according to the sampling direction and size to form a two-dimensional matrix data;

[0051] S4 treats each two-dimensional matrix as a channel data, and all channel data constitute a complete sample data, which is then saved as a single sample data in a multi-channel image manner.

[0052] S5. For each sample point set in step S2, repeat steps S3 and S4 to obtain the sample dataset of the study area.

[0053] For example, in step S1, multiple three-dimensional seismic attribute data can be selected to avoid omitting attribute data that may have a significant characterizing effect on the research target; the selection of attribute data volume is based on the principle of having a significant characterizing effect on the research target.

[0054] In step S2, the sampling direction of the sample points is selected from Inline, Crossline, and Time, and the sampling interval and size are set in the other two directions; the sampling size should be greater than or equal to the sampling interval, so as to ensure that the sample data collection fully covers the three-dimensional space (without missing data).

[0055] In step S3, with each seismic sampling point as the center, data expansion is performed in two dimensions along the sampling direction and size, and two-dimensional matrix data is extracted from each attribute data volume.

[0056] In step S4, each two-dimensional matrix formed above is used as channel data and stored in multi-channel TIFF image format, thereby realizing multi-channel seismic exploration sample acquisition.

[0057] Example 2, as another embodiment of the present invention, the seismic exploration deep learning multi-channel sample generation method provided in this embodiment includes:

[0058] Step 1: Load one or more 3D seismic attribute data volumes within the study area. The selection of attribute data is based on the principle of having a significant characterizing effect on the research target. In this embodiment, four 3D seismic attribute data volumes were selected: seismic amplitude, instantaneous amplitude, instantaneous phase, and instantaneous frequency. A data slice image of each data volume along a certain inline direction is shown below. Figures 2-5 As shown.

[0059] Step 2: Within the three-dimensional space of the study area, sample points are set at certain sampling intervals, thereby dividing the entire area into multiple sample points. The sampling direction and size of the sample points are then set. In this embodiment, the three coordinate axes of the three-dimensional work area are respectively set as: inline, crossline, and Time (e.g., ...). Figure 6 The sampling intervals are as follows: inline 2000-2100, interval 1; crossline 2000-2400, interval 1; Time 1000-3000, interval 1. The sampling direction is set along the inline, with a sampling interval and size of 20 in the crossline direction and 100 in the time direction. Based on this setting, a total of 400 samples (20x20 pixels) are collected on each inline slice. Figure 7 This is a sample distribution map on an inline slice of the earthquake amplitude data volume. Figure 7 The numbers 2000, 2020…2400 at the top represent the horizontal coordinates (crossline values) of the slice, while the numbers 1000 to 3000 on the left represent the vertical coordinates (time values). The slice is divided into 400 grids at intervals of 20 horizontally and 100 vertically, with each grid representing one sample. The study area contains 101 inline slices, and a total of 40,400 samples were collected.

[0060] Step 3: For each sample point, using the sample point as the center, extract data for each seismic attribute data according to the sampling direction and size to form a two-dimensional matrix data. In this embodiment, each inline slice of each data volume is processed according to... Figure 7 For each sample range shown, extract the two-dimensional matrix data of the above four attributes. The size of each two-dimensional matrix is ​​21x101, and each sample yields four two-dimensional matrix data.

[0061] Step 4: Treat each sample's two-dimensional matrix as a channel data, and all channel data constitute a complete sample data. Save this as a single sample data in a multi-channel image format. In this embodiment, the four two-dimensional matrix data obtained in the previous step are used, with each matrix serving as a channel data, and saved as a single sample data. Figure 8 The `write` function in the Python code shown saves the data as a TIFF image.

[0062] For each sample mentioned in step 2, repeat steps 3 and 4 to generate data for all samples in the study area.

[0063] It is understood that the seismic exploration deep learning multi-channel sample generation method of the present invention, by fusing multiple seismic attribute data to form multi-channel samples, can better characterize the research target and better reflect the spatial correlation of geological bodies, thus making it more conducive for deep learning algorithms to learn the spatial correlation features of seismic data, and can directly use relatively mature, image-based convolutional neural networks to train and predict the formed sample set.

[0064] The deep learning process consists of sample generation, labeling, training, and prediction. This invention only relates to sample generation.

[0065] Each seismic attribute is represented by a matrix, and each matrix by a channel, thus forming a multi-channel sample based on the TIFF image format.

[0066] In a specific embodiment 3 of the present invention, the invention is applied. Figure 9 This is a structural diagram of the deep learning-based multi-channel sample generation system for seismic exploration provided by the present invention. The system includes:

[0067] The seismic attribute data volume loading module 21 is used to load one or more three-dimensional seismic attribute data volumes within the study area; the sample point setting module 22 is used to set sample points at certain sampling intervals within the three-dimensional spatial range of the study area, thereby dividing the entire area into multiple sample points, and setting the sampling direction and sampling size of the sample points; the two-dimensional matrix forming module 23 is used to extract data from each seismic attribute data according to the sampling direction and size for each sample point within the study area, with the sample point as the center, to form two-dimensional matrix data; the sample storage module 24 is used to treat each two-dimensional matrix as a channel data, and all channel data constitute a complete sample data, which is saved as a single sample data in a multi-channel image mode; the sample dataset storage module 25 is used to repeat the sample point setting, two-dimensional matrix forming, and sample storage modules for each set sample point to obtain and store the sample dataset.

[0068] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A deep learning-based multi-channel sample generation method for seismic exploration, characterized in that, The method includes the following steps: S1 loads one or more three-dimensional seismic attribute data volumes within the study area; S2 represents the three-dimensional spatial range of the study area. Sample points are set at certain sampling intervals to divide the entire area into multiple sample points, and the sampling direction and sampling size of the sample points are set. S3, for each sample point in the study area, with the sample point as the center, extract data for each seismic attribute data according to the sampling direction and size to form a two-dimensional matrix data; S4, each two-dimensional matrix is ​​treated as a channel data, and all channel data constitute a complete sample data, which is saved as a single sample data in the form of a multi-channel image; S5. For each sample point set in step S2, repeat steps S3 and S4 to obtain the sample dataset of the study area.

2. The deep learning multi-channel sample generation method for seismic exploration according to claim 1, characterized in that, In step S1, multiple three-dimensional seismic attribute data are selected, and the selection of attribute data volumes is based on the principle of having a significant characterizing effect on the research objectives.

3. The deep learning multi-channel sample generation method for seismic exploration according to claim 1, characterized in that, In step S2, the sampling direction of the sample points is one of Inline, Crossline, or Time, and the sampling interval and size are set in the other two directions; the sampling size should be greater than or equal to the sampling interval.

4. The deep learning multi-channel sample generation method for seismic exploration according to claim 1, characterized in that, In step S3, with each seismic sampling point as the center, data expansion is performed in two dimensions along the sampling direction and size, and two-dimensional matrix data is extracted from each attribute data volume.

5. The deep learning multi-channel sample generation method for seismic exploration according to claim 1, characterized in that, In step S4, each two-dimensional matrix formed above is used as channel data and stored in multi-channel TIFF image format to achieve multi-channel seismic exploration sample acquisition.

6. The deep learning multi-channel sample generation method for seismic exploration according to claim 1, characterized in that, In step S4, each seismic attribute data is used as an image channel. Multiple seismic attribute data are merged into a multi-channel sample to represent the research target, and deep learning algorithms are used to learn the spatial correlation characteristics of the seismic data.

7. The deep learning multi-channel sample generation method for seismic exploration according to claim 1, characterized in that, The method is mounted on a computer device, which includes at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor. When the processor executes the computer program, it implements the steps of the above-described deep learning multi-channel sample generation method for seismic exploration.

8. A deep learning multi-channel sample generation system for seismic exploration, characterized in that, The system is implemented using the seismic exploration deep learning multi-channel sample generation method according to any one of claims 1-7, and the system includes: The seismic attribute data volume loading module is used to load one or more three-dimensional seismic attribute data volumes within the study area; The sample point setting module is used to set sample points in the three-dimensional space of the study area at a certain sampling interval, thereby dividing the entire area into multiple sample points, and setting the sampling direction and sampling size of the sample points; The two-dimensional matrix forming module is used to extract data from each seismic attribute data in the study area, with the sample point as the center, according to the sampling direction and size, and form two-dimensional matrix data. The sample storage module is used to treat each two-dimensional matrix as a channel data, and all channel data constitute a complete sample data, which is saved as a single sample data in a multi-channel image manner. The study area sample dataset acquisition module is used to repeat the two-dimensional matrix formation and sample storage steps for each set sample point to obtain the study area sample dataset.

9. The seismic exploration deep learning multi-channel sample generation system according to claim 8, characterized in that, The deep learning multi-channel sample generation system for seismic exploration is applied to noise suppression, seismic velocity picking, fault identification, reservoir prediction, lithology identification, and physical property prediction in oil and gas seismic exploration.

Images