River identification method, deep learning model training method and device, and related equipment
By preprocessing 3D post-stack seismic data and training a convolutional neural network model, the problem of relying on manual qualitative analysis for river channel identification in existing technologies has been solved, and efficient quantitative identification of complex river systems and small rivers has been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PETROCHINA CO LTD
- Filing Date
- 2022-01-19
- Publication Date
- 2026-08-04
AI Technical Summary
Existing methods for identifying river channels based on seismic data attributes rely heavily on human intervention, lack sufficient qualitative analysis, and are difficult to accurately identify complex river systems and small channels, resulting in low computational efficiency.
A convolutional neural network model based on the maximum amplitude preprocessing of three-dimensional post-stack seismic data and a training sample set generated by simulation was used to train the convolutional neural network to identify and recover river channel data.
It enables accurate quantitative identification of complex river systems and small rivers, improves identification accuracy and efficiency, reduces the workload of manual annotation, and provides efficient technical tool support.
Smart Images
Figure CN116524342B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of seismic data processing technology, and in particular to a river channel identification method, a deep learning model training method, a device, and related equipment. Background Technology
[0002] River channels are an important geological target in oil and gas exploration, and accurately identifying them and revealing their development processes is crucial. For example, in basin and prospective scale analysis, understanding river channel characteristics helps in the quantitative analysis of geomorphology and accurate understanding of sedimentary processes, and helps in understanding the sedimentary environment and the main direction of sediment migration over time. In reservoir scale analysis, determining the volume of river sandstone and the properties of the fluids filling it is essential. In the oil and gas field development stage, accurately depicting the macroscopic planar distribution characteristics of river channels, sedimentation and its variation patterns can provide important basis for the selection of development target blocks, the preparation of oil and gas development plans, and the design of well locations and well trajectories.
[0003] Currently, attribute identification of rivers based on seismic data remains the main technical means. Commonly used attributes for river identification include coherence, spectral decomposition, curvature, texture, edge detection, and shearlet transform. Coherence is an edge-sensitive property for determining channel width (Bahorich and Farmer, 1995); spectral decomposition, which complements coherence and other edge-sensitive properties, is sensitive to channel thickness and can detect small channels (Partyka, Gridley, and Lopez, 1999); curvature quantifies the degree of curvature of a surface, with the most positive and negative curvatures used to detect small channels that other properties cannot identify (Chopra and Marfurt, 2008); texture properties are derived from statistical measures of the gray-level co-occurrence matrix (Eichkitz et al., 2015), and are used to identify channel systems within the Vienna sedimentary basin; edge detection operators commonly used in image processing, such as the Canny and Sobel operators, are often used to identify channels in seismic data (Marfurt, 2014); Karbalaali et al. (2017, 2018) used shearlet transform to determine channel boundaries in seismic data. Summary of the Invention
[0004] The inventors found that although the above-mentioned seismic attributes have been widely used in river channel identification and have achieved good results, the following problems also exist: (1) They are all attributes that indirectly reflect the characteristics of the river channel. The analysis of the attributes is very important and requires a lot of human participation and expert experience; (2) They are all qualitative attributes and cannot quantitatively characterize the river channel, resulting in a weak ability to identify complex river systems and small rivers; (3) The calculation process of these attributes is complicated and the calculation efficiency is low. It is time-consuming and labor-intensive to identify the river channel on a large data volume.
[0005] In view of the above problems, the present invention is proposed to provide a river identification method, a deep learning model training method, an apparatus and related equipment that overcome or at least partially solve the above problems.
[0006] In a first aspect, embodiments of the present invention provide a river channel identification method, which may include:
[0007] Based on the maximum amplitude value of the acquired three-dimensional post-stack seismic data, the three-dimensional post-stack seismic data is preprocessed.
[0008] The preprocessed 3D post-stack seismic data is input into a pre-trained convolutional neural network model to identify river channel data in the 3D post-stack seismic data.
[0009] Based on the maximum amplitude, the river channel data in the identified three-dimensional post-stack seismic data is recovered to obtain the river channel identification result in the three-dimensional post-stack seismic data.
[0010] Optionally, the preprocessing of the three-dimensional post-stack seismic data based on the maximum amplitude of the acquired three-dimensional post-stack seismic data may include:
[0011] The amplitude of the three-dimensional post-stack seismic data is divided by the maximum amplitude value so that the amplitude of the preprocessed three-dimensional post-stack seismic data is between -1 and 1.
[0012] Optionally, the convolutional neural network model is pre-trained through the following steps:
[0013] Obtain a training sample set, wherein each sample in the training sample set includes a synthetic seismic data volume and a river channel data volume annotated in the synthetic seismic data volume;
[0014] The convolutional neural network model is trained using samples from the training sample set. The synthetic seismic data volume is input into the convolutional neural network model, and the river channel data volume in the seismic data volume is output to determine the model training parameters in the convolutional neural network model.
[0015] Secondly, embodiments of the present invention provide a deep learning model training method, which may include:
[0016] Obtain a training sample set, wherein each sample in the training sample set includes a synthetic seismic data volume and a river channel data volume annotated in the synthetic seismic data volume;
[0017] The convolutional neural network model is trained using samples from the training sample set. The synthetic seismic data volume is input into the convolutional neural network model, and the river channel data volume in the seismic data volume is output to determine the model training parameters in the convolutional neural network model.
[0018] Optionally, obtaining the training sample set may specifically include:
[0019] The randomly generated single-channel reflection coefficients are horizontally expanded to obtain a three-dimensional horizontal layered reflection coefficient model within a preset sampling point range;
[0020] Numerical simulations were performed based on the principle of river channel deposition to generate a three-dimensional reflection coefficient model with a random number of river channels.
[0021] Based on the three-dimensional horizontal layered reflection coefficient model and the three-dimensional reflection coefficient model with a random number of channels, a three-dimensional horizontal layered reflection coefficient model including channels is obtained.
[0022] After performing folding deformation processing and fracture deformation processing on the three-dimensional reflection coefficient model containing a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels, respectively, they are convolved with preset seismic wavelets to obtain the synthetic seismic data volume and the channel data volume marked in the synthetic seismic data volume.
[0023] Optionally, the step of performing folding deformation processing and fault deformation processing on the three-dimensional reflection coefficient model containing a random number of channels and the three-dimensional horizontally layered reflection coefficient model containing channels, respectively, and then convolving them with a preset seismic wavelet to obtain the synthetic seismic data volume and the channel data volume marked in the synthetic seismic data volume, may include:
[0024] The three-dimensional reflection coefficient model with a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels are subjected to folding deformation processing to obtain a three-dimensional folded deformation model with a random number of channels and a three-dimensional folded deformation model containing channels.
[0025] Fracture deformation was performed on the three-dimensional fold deformation model with a random number of channels and the three-dimensional fold deformation model containing channels, respectively, to obtain the three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels.
[0026] The three-dimensional fracture deformation model with a random number of river channels and the three-dimensional fracture deformation model containing river channels are respectively convolved with seismic wavelets of preset bandwidth and dominant frequency to obtain the synthetic seismic data volume and the river channel data volume marked in the synthetic seismic data volume.
[0027] or,
[0028] The three-dimensional reflection coefficient model with a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels are subjected to fracture deformation to obtain the three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels.
[0029] Folding deformation processing was performed on the three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels, respectively, to obtain the three-dimensional folding deformation model with a random number of channels and the three-dimensional folding deformation model containing channels.
[0030] The three-dimensional fold deformation model containing a random number of river channels and the three-dimensional fold deformation model containing river channels are respectively convolved with seismic wavelets of preset bandwidth and dominant frequency to obtain the synthetic seismic data volume and the river channel data volume marked in the synthetic seismic data volume.
[0031] Optionally, before obtaining the synthetic seismic data volume, the process may further include:
[0032] Random noise with a preset signal-to-noise ratio is added to the synthesized seismic data volume.
[0033] Thirdly, embodiments of the present invention provide an application of river data identified by the river identification method described in the first aspect in oil and gas exploration and development.
[0034] Fourthly, embodiments of the present invention provide a river identification device, which may include:
[0035] The preprocessing module is used to preprocess the three-dimensional post-stack seismic data based on the maximum amplitude value of the acquired three-dimensional post-stack seismic data;
[0036] The identification module is used to input the preprocessed three-dimensional post-stack seismic data into a pre-trained convolutional neural network model to identify river data in the three-dimensional post-stack seismic data.
[0037] The recovery module is used to recover the river data in the identified three-dimensional post-stack seismic data based on the maximum amplitude value, so as to obtain the river identification result in the three-dimensional post-stack seismic data.
[0038] Fifthly, embodiments of the present invention provide a deep learning model training apparatus, which may include:
[0039] The acquisition module is used to acquire a training sample set, wherein each sample in the training sample set includes a synthetic seismic data volume and a river channel data volume annotated in the synthetic seismic data volume;
[0040] The training module is used to train the convolutional neural network model with samples from the training sample set, input the synthetic seismic data volume into the convolutional neural network model, and output the river channel data volume in the seismic data volume to determine the model training parameters in the convolutional neural network model.
[0041] In a sixth aspect, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the river identification method as described in the first aspect, or the deep learning model training method as described in the second aspect.
[0042] In a seventh aspect, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the river identification method as described in the first aspect, or the deep learning model training method as described in the second aspect.
[0043] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following:
[0044] This invention provides a river channel identification method, a deep learning model training method, an apparatus, and related equipment. The river channel identification method includes: preprocessing the acquired three-dimensional post-stack seismic data based on the maximum amplitude value; inputting the preprocessed three-dimensional post-stack seismic data into a pre-trained convolutional neural network model to identify river channel data within the three-dimensional post-stack seismic data; and recovering the identified river channel data based on the maximum amplitude value to obtain the river channel identification result from the three-dimensional post-stack seismic data. This method can accurately and quantitatively identify complex river systems, especially small rivers, in three-dimensional seismic data. Compared with traditional river channel identification methods, both identification accuracy and efficiency are significantly improved. It can provide efficient technical tools to support paleogeomorphological and sedimentary environment analysis, oil and gas enrichment area selection, oil and gas development plan preparation, and well location and well trajectory design in river-related research areas.
[0045] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.
[0046] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0047] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0048] Figure 1 This is a flowchart illustrating the deep learning model training method provided in Embodiment 1 of the present invention;
[0049] Figure 2 This is a schematic diagram of the sample preparation process provided in Embodiment 1 of the present invention;
[0050] Figure 3 This is one example of the synthetic seismic data volume provided in Embodiment 1 of the present invention;
[0051] Figure 4 for Figure 3 Example of a river channel data volume annotated in the middle;
[0052] Figure 5 This is a second example of the synthetic seismic data volume provided in Embodiment 1 of the present invention;
[0053] Figure 6 for Figure 5 Example of a river channel data volume annotated in the middle;
[0054] Figure 7 This is a schematic diagram of the deep learning model training device provided in Embodiment 1 of the present invention;
[0055] Figure 8 This is a flowchart illustrating the river identification method provided in Embodiment 2 of the present invention;
[0056] Figure 9 This is a schematic diagram showing the intelligent complex river identification results provided in Embodiment 2 of the present invention in a cross-section.
[0057] Figure 10 This is a schematic diagram showing the intelligent complex river identification results provided in Embodiment 2 of the present invention on a slice along the layer.
[0058] Figure 11 This is a schematic diagram of the river identification device provided in Embodiment 2 of the present invention. Detailed Implementation
[0059] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
[0060] In recent years, artificial intelligence has become a core driving force for a new round of technological revolution and industrial transformation. Countries around the world are scrambling to establish a foothold, making it a fiercely contested arena for showcasing innovative strength. Currently, major global oil and gas companies and oilfield service companies are actively engaged in the research and development of intelligent technologies to win market competition and seize technological advantages. Addressing the existing problems in river identification, research on three-dimensional intelligent river identification methods is being conducted to overcome the problems and shortcomings of traditional methods. This is of great significance both in enhancing the core competitiveness of listed companies and in improving the quality and efficiency of actual production.
[0061] Example 1
[0062] Embodiment 1 of the present invention provides a method for training a deep learning model, referring to... Figure 1 As shown, the method may include the following steps:
[0063] Step S11: Obtain the training sample set. Each sample in the training sample set includes a synthetic seismic data volume and a river channel data volume labeled in the synthetic seismic data volume.
[0064] It should be noted that the inventors' invention involves manually labeling existing seismic data obtained through actual exploration to mark river channels. Existing manual labeling methods are not only time-consuming and labor-intensive, but their accuracy also depends entirely on the skill level of the cartographer. Furthermore, when labeling river channel data, only qualitative analysis is possible; quantitative analysis is not possible for small rivers or complex river systems, making effective identification and labeling impossible. Therefore, the inventors of this application innovatively propose a method based on simulated sample data generation. That is, each sample in the training sample set mentioned above in this embodiment is synthesized through simulation and is not actual seismic data obtained through exploration. The process of simulating the creation of training samples not only reduces workload and difficulty but also enables quantitative simulation. This means that the simulation process fully considers the river channel's own elements, such as channel width, thickness, and curvature, while also incorporating complex factors such as internal amplitude changes, channel intersections, and tectonic movements, to obtain accurate model training parameters for the model during training.
[0065] Step S12: Train the convolutional neural network model using samples from the training sample set, input the synthetic seismic data volume into the convolutional neural network model, and output the river channel data volume from the seismic data volume to determine the model training parameters in the convolutional neural network model.
[0066] In this embodiment of the invention, when creating the training samples, a deep learning training set that directly contains expert knowledge and accurate river information is generated based on geological principles. Then, a three-dimensional intelligent river identification network is designed and implemented based on deep learning theory. The generated deep learning training set is then used to train the three-dimensional intelligent river identification network, enabling it to learn complex river features in three-dimensional post-stack seismic data in a high-dimensional nonlinear space. This achieves intelligent river identification and provides a powerful, intelligent, low-cost, high-efficiency, and high-precision technical support for the identification of complex river systems in post-stack three-dimensional data volumes.
[0067] In an optional embodiment, the inventors base their work on the fact that a reflection coefficient is generated when a subsurface impedance difference interface exists. Based on the convolution principle, this reflection coefficient is convolved with a seismic wavelet to generate a synthetic seismic record. (See also...) Figure 2 As shown, obtaining the training sample set in step S11 above specifically includes:
[0068] Step S21: Horizontally expand the randomly generated single-channel reflection coefficient to obtain a three-dimensional horizontal layered reflection coefficient model within the preset sampling point range.
[0069] This embodiment of the invention uses a sample data pair consisting of a synthetic seismic data volume and a river channel data volume labeled within the synthetic seismic data volume as an example for illustration. In this step, when creating the above sample data pair, the single-channel reflection coefficient is horizontally expanded to generate a three-dimensional horizontally layered reflection coefficient model with a size of 96×96×96.
[0070] Step S22: Perform numerical simulation based on the principle of river channel deposition to generate a three-dimensional reflection coefficient model with a random number of river channels.
[0071] In this step, within the above sampling range of 96×96×96, a three-dimensional reflection coefficient model is generated with a random number of channels ranging from 0 to 100.
[0072] Step S23: Based on the three-dimensional horizontal layered reflection coefficient model and the three-dimensional reflection coefficient model with a random number of channels, obtain the three-dimensional horizontal layered reflection coefficient model containing the channels.
[0073] In this step, the three-dimensional horizontal layered reflectance coefficient model is added to the three-dimensional reflectance coefficient model containing a random number of river channels, resulting in a three-dimensional horizontal layered reflectance coefficient model that includes the river channels. In this step, the models are added together; that is, the corresponding values for each sample point in the three-dimensional model are added together.
[0074] Step S24: After performing folding deformation processing and fracture deformation processing on the three-dimensional reflection coefficient model with a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels, respectively, they are convolved with preset seismic wavelets to obtain the synthetic seismic data volume and the channel data volume marked in the synthetic seismic data volume.
[0075] It should be noted that the order in which the folding deformation processing and fracture deformation processing are performed on the above model in this step is not important. Folding deformation processing can be performed first, followed by fracture deformation processing, or vice versa; this embodiment of the invention does not impose a specific limitation on this. In this embodiment, the preset seismic wavelet can be a seismic wavelet with a bandwidth of 80Hz and a dominant frequency that varies randomly between 15 and 40Hz. Convolution with the seismic wavelet yields the synthetic seismic data volume and the river channel data volume annotated within the synthetic seismic data volume.
[0076] In step S12 above, during model training, it can be based on Figure 2 The method described in the text generates 1000 pairs of synthetic seismic data volumes and accurately labeled river data volumes. For each pair of synthetic seismic data volumes and accurately labeled river data volumes that closely approximate the actual data, the maximum value of the synthetic seismic data volume that closely approximates the actual data in the sample pair is first calculated. Then, the synthetic seismic data volume and the accurately labeled river data volume are divided by the maximum value to obtain a normalized sample pair. All the normalized sample pairs are randomly sorted to obtain the deep learning training set.
[0077] In this embodiment of the invention, the above data is normalized in order to reduce the training workload and improve training efficiency during model training.
[0078] In one specific embodiment, step S24 above may specifically include:
[0079] Folding deformation processing was performed on the three-dimensional reflection coefficient model with a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels, respectively, to obtain the three-dimensional folded deformation model with a random number of channels and the three-dimensional folded deformation model containing channels.
[0080] Fracture deformation was performed on a three-dimensional fold deformation model with a random number of channels and a three-dimensional fold deformation model containing channels, respectively, to obtain a three-dimensional fracture deformation model with a random number of channels and a three-dimensional fracture deformation model containing channels.
[0081] The three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels are respectively convolved with seismic wavelets with preset bandwidth and dominant frequency to obtain synthetic seismic data volume and channel data volume labeled in synthetic seismic data volume.
[0082] or,
[0083] Fracture deformation was performed on the three-dimensional reflection coefficient model with a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels, respectively, to obtain the three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels.
[0084] Folding deformation processing was performed on the three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels, respectively, to obtain the three-dimensional folding deformation model with a random number of channels and the three-dimensional folding deformation model containing channels.
[0085] The three-dimensional fold deformation model with a random number of channels and the three-dimensional fold deformation model containing channels are respectively convolved with seismic wavelets of preset bandwidth and dominant frequency to obtain synthetic seismic data volume and channel data volume labeled in the synthetic seismic data volume.
[0086] Reference Figure 3 One example of a synthetic seismic data volume shown, and Figure 4 The river data volume marked in the middle, Figure 3 and Figure 4 The vertical coordinate represents the longitudinal survey line number, the horizontal coordinate represents the connecting survey line number, the vertical coordinate represents time in seconds (s), and the color scale represents the amplitude. For example... Figure 3 and Figure 4 As shown, the accurately labeled river data volume in the sample pair generated by this invention is consistent with the changes in the synthetic seismic data volume, and the river location is accurately labeled.
[0087] Reference Figure 5 Example 2 of synthetic seismic data volumes and Figure 6 The river channel data volume is labeled in the figure, and this data is used to train the samples in the training sample set. Figure 5 and Figure 6 The vertical coordinate represents the longitudinal survey line number, the horizontal coordinate represents the connecting survey line number, the vertical coordinate represents time in seconds (s), and the color scale represents the amplitude. For example... Figure 5 and Figure 6 As shown, the synthetic seismic data volume generated by this invention contains rich and diverse information on folds and fault structures, closely approximating actual data. Furthermore, the river channel locations in the accurately labeled river channel data volume are precisely labeled, providing accurate labeled samples for network training. The existence of a sufficient number of labeled data in the deep learning training set lays a solid big data foundation for the rapid and accurate implementation of deep learning-based intelligent river channel identification methods.
[0088] In another optional embodiment, before obtaining the synthetic seismic data volume, the method may further include adding random noise with a preset signal-to-noise ratio to the synthetic seismic data volume, so as to obtain a synthetic seismic data volume that closely approximates the actual data.
[0089] In another alternative embodiment, the inventors have made detailed feasibility and necessity designs for elements such as the number of layers, the scale of the convolutional layers, the number of channels, the number of filters, and the activation function of the three-dimensional convolutional network according to the requirements of the present invention. The convolutional neural network model described above in this invention is constructed according to the principles of deep learning, with the following deep learning layers built sequentially: An input layer is constructed, consisting of a single convolutional layer with a size of 3×3×3×nc×nf (nf convolutional kernels form a 3×3×3 convolutional filter, for example, nc has 1 channel and nf has 100 channels); an intermediate layer is constructed, consisting of several building blocks, each block being composed of a convolutional layer, a batch normalization layer, and a modified linear unit layer connected sequentially, with each convolutional layer having a size of 3×3×3×nc×nf (100 convolutional kernels form a 3×3×3 convolutional filter, nc has 100 channels); and an output layer is constructed, consisting of a single convolutional layer and a nonlinear unit layer, with the convolutional layer having a size of 3×3×3×nc×1 (1 convolutional kernel forms a 3×3×3 convolutional filter, nc has 100 channels), and the nonlinear unit layer employing a hyperbolic tangent function.
[0090] During training, the convolutional neural network model is trained using a training sample set (deep learning training set). Mini-batch stochastic gradient descent is used to optimize the mean squared error objective function. For example, the batch size is 4, the training period is 50, and the learning rate is 0.001.
[0091] Based on the same inventive concept, this invention also provides a deep learning model training device, referring to... Figure 7 As shown, the device may include an acquisition module 71 and a training module 72, and its working principle is as follows:
[0092] The acquisition module 71 is used to acquire a training sample set, wherein each sample in the training sample set includes a synthetic seismic data volume and a river channel data volume annotated in the synthetic seismic data volume;
[0093] The training module 72 is used to train the convolutional neural network model with samples from the training sample set, input the synthetic seismic data volume into the convolutional neural network model, and output the river channel data volume in the seismic data volume to determine the model training parameters in the convolutional neural network model.
[0094] In an optional embodiment, the acquisition module 71 is specifically used for:
[0095] The randomly generated single-channel reflection coefficients are horizontally expanded to obtain a three-dimensional horizontal layered reflection coefficient model within a preset sampling point range;
[0096] Numerical simulations were performed based on the principle of river channel deposition to generate a three-dimensional reflection coefficient model with a random number of river channels.
[0097] Based on the three-dimensional horizontal layered reflection coefficient model and the three-dimensional reflection coefficient model with a random number of channels, a three-dimensional horizontal layered reflection coefficient model including channels is obtained.
[0098] After performing folding deformation processing and fracture deformation processing on the three-dimensional reflection coefficient model containing a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels, respectively, they are convolved with preset seismic wavelets to obtain the synthetic seismic data volume and the channel data volume marked in the synthetic seismic data volume.
[0099] In one specific embodiment, the acquisition module 71 is specifically used for:
[0100] The three-dimensional reflection coefficient model with a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels are subjected to folding deformation processing to obtain a three-dimensional folded deformation model with a random number of channels and a three-dimensional folded deformation model containing channels.
[0101] Fracture deformation was performed on the three-dimensional fold deformation model with a random number of channels and the three-dimensional fold deformation model containing channels, respectively, to obtain the three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels.
[0102] The three-dimensional fracture deformation model with a random number of river channels and the three-dimensional fracture deformation model containing river channels are respectively convolved with seismic wavelets of preset bandwidth and dominant frequency to obtain the synthetic seismic data volume and the river channel data volume marked in the synthetic seismic data volume.
[0103] or,
[0104] The three-dimensional reflection coefficient model with a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels are subjected to fracture deformation to obtain the three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels.
[0105] Folding deformation processing was performed on the three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels, respectively, to obtain the three-dimensional fold deformation model with a random number of channels and the three-dimensional fold deformation model containing channels.
[0106] The three-dimensional fold deformation model containing a random number of river channels and the three-dimensional fold deformation model containing river channels are respectively convolved with seismic wavelets of preset bandwidth and dominant frequency to obtain the synthetic seismic data volume and the river channel data volume marked in the synthetic seismic data volume.
[0107] In another optional embodiment, the acquisition module 71 is further configured to add random noise with a preset signal-to-noise ratio to the synthetic seismic data volume.
[0108] Based on the same inventive concept, this embodiment of the invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described deep learning model training method.
[0109] Based on the same inventive concept, this embodiment of the invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described deep learning model training method.
[0110] The principles by which the above-mentioned devices, media, and related equipment in the embodiments of the present invention solve the problem are similar to those of the aforementioned methods. Therefore, their implementation can refer to the implementation of the aforementioned methods, and repeated details will not be repeated.
[0111] Example 2
[0112] Embodiment 2 of this invention provides a river channel identification method. This method, based on a pre-trained convolutional neural network model, can only identify complex river channels in 3D seismic data. (Refer to...) Figure 8 As shown, the method may include the following steps:
[0113] Step S81: Based on the maximum amplitude of the acquired three-dimensional post-stack seismic data, preprocess the three-dimensional post-stack seismic data.
[0114] Specifically, the preprocessing process involves dividing the amplitude of the 3D post-stack seismic data by its maximum amplitude, ensuring that the amplitude of the preprocessed 3D post-stack seismic data falls between -1 and 1. This yields the preprocessed 3D post-stack seismic data, while simultaneously recording the maximum amplitude.
[0115] Step S82: Input the preprocessed 3D post-stack seismic data into a pre-trained convolutional neural network model to identify river channel data in the 3D post-stack seismic data.
[0116] Step S83: Recover the river channel data in the identified three-dimensional post-stack seismic data based on the maximum amplitude value to obtain the river channel identification result in the three-dimensional post-stack seismic data.
[0117] It should be noted that the pre-trained convolutional neural network model in the embodiments of the present invention can be trained in the manner described in Embodiment 1, or it can be trained in other ways. The embodiments of the present invention do not specifically limit this.
[0118] In this embodiment of the invention, the maximum value of the actual three-dimensional post-stack seismic data of the river channel to be identified is calculated, and the actual three-dimensional post-stack seismic data of the river channel to be identified is divided by the maximum value to make the data range between -1.0 and 1.0, so as to obtain the preprocessed three-dimensional post-stack seismic data, and the maximum value is recorded. The preprocessed three-dimensional post-stack seismic data is input into the trained convolutional neural network model (three-dimensional intelligent river channel recognition network) to obtain intelligently identified complex river channel data. The intelligently identified complex river channel data is multiplied by the recorded maximum value to obtain the intelligent complex river channel recognition result after maximum value recovery.
[0119] Reference Figure 9 The diagram shown illustrates the intelligent complex river channel identification results in a cross-section. Figure 9 The horizontal coordinate in the graph represents the connecting survey line number, and the vertical coordinate represents time, in seconds (s). For example... Figure 9 As shown, all river channel reflections submerged in a highly reflective background on the cross-section were accurately and quickly identified.
[0120] Reference Figure 10 The diagram shown illustrates the intelligent complex river channel identification results displayed on a slice. Figure 10 The vertical coordinates represent the connecting line number, and the vertical coordinates represent the longitudinal survey line number. For example... Figure 10 As shown, both wide and very narrow river channels stacked together on the target layer were accurately identified. The intelligent river identification method in this embodiment of the invention has the ability to identify complex river systems. A 10GB data volume can be identified in just one hour, while manual interpretation would take several days and the accuracy could not be guaranteed. This fully demonstrates the effectiveness and value of the invention.
[0121] Furthermore, since this invention is based on deep learning algorithms and involves only simple 3D convolution operations performed on a GPU, it boasts high computational efficiency. The relevant functional modules of this invention have been integrated into an industrial software system, providing intelligent, low-cost, high-efficiency, and high-precision technical support for the identification of complex river systems in stacked 3D data volumes.
[0122] Furthermore, the three-dimensional intelligent river identification network of this invention can accurately and quantitatively identify complex river systems, especially small rivers, in three-dimensional seismic data. Compared with traditional river identification methods, both the identification accuracy and efficiency are significantly improved. It can provide efficient technical tools to support paleogeographic and sedimentary environment analysis, oil and gas enrichment area selection, oil and gas development plan preparation, well location and well trajectory design in river-related research areas.
[0123] In another alternative embodiment, the convolutional neural network model is pre-trained via the following steps:
[0124] Obtain a training sample set, where each sample includes a synthetic seismic data volume and a river channel data volume annotated within the synthetic seismic data volume;
[0125] The convolutional neural network model is trained using samples from the training sample set. The synthetic seismic data volume is input into the convolutional neural network model, and the river channel data volume in the seismic data volume is output to determine the model training parameters in the convolutional neural network model.
[0126] Based on the same inventive concept, this invention also provides a river identification device, referring to... Figure 11 As shown, the device may include a preprocessing module 111, an identification module 112, and a recovery module 113, and its working principle is as follows:
[0127] The preprocessing module 111 is used to preprocess the three-dimensional post-stack seismic data based on the maximum amplitude value of the acquired three-dimensional post-stack seismic data;
[0128] The identification module 112 is used to input the preprocessed three-dimensional post-stack seismic data into a pre-trained convolutional neural network model to identify the river data in the three-dimensional post-stack seismic data;
[0129] The recovery module 113 is used to recover the river data in the identified three-dimensional post-stack seismic data based on the maximum amplitude value, so as to obtain the river identification result in the three-dimensional post-stack seismic data.
[0130] In an optional embodiment, the preprocessing module 111 is specifically used to divide the amplitude of the three-dimensional post-stack seismic data by the maximum amplitude value, so that the amplitude of the preprocessed three-dimensional post-stack seismic data is in the range of -1 and 1.
[0131] In another optional embodiment, the convolutional neural network model in the recognition module 112 is pre-trained through the following steps:
[0132] Obtain a training sample set, wherein each sample in the training sample set includes a synthetic seismic data volume and a river channel data volume annotated in the synthetic seismic data volume;
[0133] The convolutional neural network model is trained using samples from the training sample set. The synthetic seismic data volume is input into the convolutional neural network model, and the river channel data volume in the seismic data volume is output to determine the model training parameters in the convolutional neural network model.
[0134] Based on the same inventive concept, this embodiment of the invention also provides an application of the river data identified by the above-mentioned river identification method in oil and gas exploration and development.
[0135] Specifically, it can provide efficient technical tools to support paleogeomorphological and sedimentary environment analysis in river-related research areas, selection of oil and gas enrichment areas, preparation of oil and gas development plans, and design of well locations and well trajectories. In basin and prospective scale analysis, understanding river channel characteristics helps in the quantitative analysis of geomorphology and accurate understanding of sedimentary processes, and helps in understanding the sedimentary environment and the main directions of sediment migration over time. In reservoir scale analysis, determining the volume of river sandstone and the properties of the fluids filling it is crucial. In the oil and gas field development stage, accurately depicting the macroscopic planar distribution characteristics of river channels, sedimentation, and their variation patterns can provide important basis for the selection of development target blocks, preparation of oil and gas development plans, and design of well locations and well trajectories.
[0136] Based on the same inventive concept, this embodiment of the invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described river identification method.
[0137] Based on the same inventive concept, this embodiment of the invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described river identification method.
[0138] The principles by which the above-mentioned devices, media, and related equipment in the embodiments of the present invention solve the problem are similar to those of the aforementioned methods. Therefore, their implementation can refer to the implementation of the aforementioned methods, and repeated details will not be repeated.
[0139] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.
[0140] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0141] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0142] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0143] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for river channel identification, characterized in that, include: Based on the maximum amplitude value of the acquired three-dimensional post-stack seismic data, the three-dimensional post-stack seismic data is preprocessed; wherein, the preprocessing involves dividing the amplitude of the three-dimensional post-stack seismic data by the maximum amplitude value, so that the amplitude of the preprocessed three-dimensional post-stack seismic data is between -1 and 1; The preprocessed 3D post-stack seismic data is input into a pre-trained convolutional neural network model to identify river channel data in the 3D post-stack seismic data. Based on the maximum amplitude, the river channel data in the identified three-dimensional post-stack seismic data is recovered to obtain the river channel identification result in the three-dimensional post-stack seismic data; The convolutional neural network model is pre-trained through the following steps: Obtain a training sample set, wherein each sample in the training sample set includes a synthetic seismic data volume and a river channel data volume annotated in the synthetic seismic data volume; The convolutional neural network model is trained using samples from the training sample set. The synthetic seismic data volume is input into the convolutional neural network model, and the river channel data volume in the seismic data volume is output to determine the model training parameters in the convolutional neural network model. Specifically, obtaining the training sample set includes: The randomly generated single-channel reflection coefficients are horizontally expanded to obtain a three-dimensional horizontal layered reflection coefficient model within a preset sampling point range; Numerical simulations were performed based on the principle of river channel deposition to generate a three-dimensional reflection coefficient model with a random number of river channels. Based on the three-dimensional horizontal layered reflection coefficient model and the three-dimensional reflection coefficient model with a random number of channels, a three-dimensional horizontal layered reflection coefficient model including channels is obtained. After performing folding deformation processing and fracture deformation processing on the three-dimensional reflection coefficient model containing a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels, respectively, they are convolved with preset seismic wavelets to obtain the synthetic seismic data volume and the channel data volume marked in the synthetic seismic data volume.
2. A deep learning model training method, characterized in that, The convolutional neural network model determined by the method is applied to the river identification method of claim 1, including: Obtain a training sample set, wherein each sample in the training sample set includes a synthetic seismic data volume and a river channel data volume annotated in the synthetic seismic data volume; The convolutional neural network model is trained using samples from the training sample set. The synthetic seismic data volume is input into the convolutional neural network model, and the river channel data volume in the seismic data volume is output to determine the model training parameters in the convolutional neural network model.
3. The method according to claim 2, characterized in that, The acquisition of the training sample set specifically includes: The randomly generated single-channel reflection coefficients are horizontally expanded to obtain a three-dimensional horizontal layered reflection coefficient model within a preset sampling point range; Numerical simulations were performed based on the principle of river channel deposition to generate a three-dimensional reflection coefficient model with a random number of river channels. Based on the three-dimensional horizontal layered reflection coefficient model and the three-dimensional reflection coefficient model with a random number of channels, a three-dimensional horizontal layered reflection coefficient model including channels is obtained. After performing folding deformation processing and fracture deformation processing on the three-dimensional reflection coefficient model containing a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels, respectively, they are convolved with preset seismic wavelets to obtain the synthetic seismic data volume and the channel data volume marked in the synthetic seismic data volume.
4. The method according to claim 3, characterized in that, The process involves performing folding and fracture deformation processing on the three-dimensional reflection coefficient model containing a random number of river channels and the three-dimensional horizontally layered reflection coefficient model containing river channels, respectively, followed by convolution with a preset seismic wavelet to obtain the synthetic seismic data volume and the river channel data volume labeled in the synthetic seismic data volume, including: The three-dimensional reflection coefficient model with a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels are subjected to folding deformation processing to obtain a three-dimensional folded deformation model with a random number of channels and a three-dimensional folded deformation model containing channels. Fracture deformation was performed on the three-dimensional fold deformation model with a random number of channels and the three-dimensional fold deformation model containing channels, respectively, to obtain the three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels. The three-dimensional fracture deformation model with a random number of river channels and the three-dimensional fracture deformation model containing river channels are respectively convolved with seismic wavelets of preset bandwidth and dominant frequency to obtain the synthetic seismic data volume and the river channel data volume marked in the synthetic seismic data volume. or, The three-dimensional reflection coefficient model with a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels are subjected to fracture deformation to obtain the three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels. Folding deformation processing was performed on the three-dimensional fracture deformation model with a random number of channels and the three-dimensional fracture deformation model containing channels, respectively, to obtain the three-dimensional folding deformation model with a random number of channels and the three-dimensional folding deformation model containing channels. The three-dimensional fold deformation model containing a random number of river channels and the three-dimensional fold deformation model containing river channels are respectively convolved with seismic wavelets of preset bandwidth and dominant frequency to obtain the synthetic seismic data volume and the river channel data volume marked in the synthetic seismic data volume.
5. The method according to claim 3, characterized in that, Before obtaining the synthetic seismic data volume, the process also includes: Random noise with a preset signal-to-noise ratio is added to the synthesized seismic data volume.
6. The application of river data identified by the river identification method according to claim 1 in oil and gas exploration and development.
7. A river channel identification device, characterized in that, include: The preprocessing module is used to preprocess the three-dimensional post-stack seismic data based on the maximum amplitude value of the acquired three-dimensional post-stack seismic data; wherein, the preprocessing involves dividing the amplitude of the three-dimensional post-stack seismic data by the maximum amplitude value, so that the amplitude of the preprocessed three-dimensional post-stack seismic data is between -1 and 1; The identification module is used to input the preprocessed three-dimensional post-stack seismic data into a pre-trained convolutional neural network model to identify river data in the three-dimensional post-stack seismic data. The recovery module is used to recover the river channel data in the identified three-dimensional post-stack seismic data based on the maximum amplitude value, so as to obtain the river channel identification result in the three-dimensional post-stack seismic data; The convolutional neural network model is pre-trained through the following steps: Obtain a training sample set, wherein each sample in the training sample set includes a synthetic seismic data volume and a river channel data volume annotated in the synthetic seismic data volume; The convolutional neural network model is trained using samples from the training sample set. The synthetic seismic data volume is input into the convolutional neural network model, and the river channel data volume in the seismic data volume is output to determine the model training parameters in the convolutional neural network model. Specifically, obtaining the training sample set includes: The randomly generated single-channel reflection coefficients are horizontally expanded to obtain a three-dimensional horizontal layered reflection coefficient model within a preset sampling point range; Numerical simulations were performed based on the principle of river channel deposition to generate a three-dimensional reflection coefficient model with a random number of river channels. Based on the three-dimensional horizontal layered reflection coefficient model and the three-dimensional reflection coefficient model with a random number of channels, a three-dimensional horizontal layered reflection coefficient model including channels is obtained. After performing folding deformation processing and fracture deformation processing on the three-dimensional reflection coefficient model containing a random number of channels and the three-dimensional horizontal layered reflection coefficient model containing channels, respectively, they are convolved with preset seismic wavelets to obtain the synthetic seismic data volume and the channel data volume marked in the synthetic seismic data volume.
8. A deep learning model training device, characterized in that, The convolutional neural network model determined by the device is applied to the river identification method of claim 1, including: The acquisition module is used to acquire a training sample set, wherein each sample in the training sample set includes a synthetic seismic data volume and a river channel data volume annotated in the synthetic seismic data volume; The training module is used to train the convolutional neural network model with samples from the training sample set, input the synthetic seismic data volume into the convolutional neural network model, and output the river channel data volume in the seismic data volume to determine the model training parameters in the convolutional neural network model.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the river identification method as described in claim 1, or the deep learning model training method as described in any one of claims 2 to 5.
10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the river identification method as described in claim 1, or the deep learning model training method as described in any one of claims 2 to 5.