Self-supervised learning and well-constrained seismic large model construction method and device
Patent Information
- Application Number
- CN202510343240.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-21
- Publication Date
- 2026-09-22
AI Technical Summary
[0006]然而，值得注意的是，当前大模型的研究与应用主要集中在NLP与计算机视觉领域，对于油气勘探等特定行业的应用探索尚显不足
[0038]本发明实施例提供的上述技术方案的有益效果至少包括：
Smart Images
Figure CN122797657A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas exploration and development technology, and in particular to a method and apparatus for constructing a large seismic model with self-supervised learning and well constraints. Background Technology
[0002] Since the emergence of the Transformer model architecture in 2017, the field of Natural Language Processing (NLP) has witnessed an unprecedented leap forward. This innovation has not only spurred a host of pre-trained models based on self-supervised learning principles, such as BERT, GPT, and T5, but has also led the trend of using massive amounts of text data for pre-training to capture deep-level language structure and semantic information. These models have accumulated rich language representation capabilities through the pre-training stage, demonstrating powerful generalization potential.
[0003] Subsequently, for diverse NLP tasks, only a small amount of labeled data is needed to fine-tune the pre-trained model, quickly adapting it to specific needs and greatly improving the flexibility and efficiency of model deployment. This "pre-training-fine-tuning" paradigm marks the official entry of the NLP field into a new era of industrial implementation, effectively alleviating the pain points of fragmentation, high cost, and long cycle in the traditional AI development process.
[0004] Previously, each AI application scenario often required building a model from scratch, undergoing a lengthy and uncertain iterative optimization process. If the results did not meet expectations, the entire process had to be restarted. Now, pre-trained models, as a powerful foundational platform, significantly reduce repetitive work and development costs through their cross-scenario general knowledge transfer capabilities. At the same time, their high initial accuracy and robustness greatly reduce the need for labeled data during the fine-tuning process, drastically shortening the development cycle to within a few days or even hours.
[0005] The rise of this industrialized development model has completely revolutionized the research and development paradigm of AI applications, enabling developers to quickly build and deploy high-performance AI solutions based on standardized pre-trained models. This has improved efficiency, lowered the technical threshold, and promoted the popularization and application of AI technology.
[0006] However, it is worth noting that current research and applications of large-scale models are mainly concentrated in the fields of NLP and computer vision, with insufficient exploration of applications in specific industries such as oil and gas exploration. Particularly in the crucial area of seismic data processing, although seismic information comprehensively covers the exploration area, well data, as a valuable resource for verifying subsurface structures and reservoir information, is limited in quantity, posing a challenge to its direct integration into large-scale model construction.
[0007] Therefore, there is an urgent need to develop a self-supervised learning method for constructing large-scale earthquake models that integrates high accuracy, strong generalization ability, and good interpretability, in order to overcome existing limitations and promote the innovation and development of earthquake data processing technology. Summary of the Invention
[0008] This invention provides a method and apparatus for constructing a large seismic model based on self-supervised learning and well constraints. By deeply integrating the self-supervised learning mechanism with well constraints, a new paradigm for constructing large seismic models is realized. Well information is directly incorporated into the loss function as a key constraint, thereby enhancing the prediction accuracy and robustness of the large seismic model in complex geological environments.
[0009] In a first aspect, embodiments of the present invention provide a method for constructing a large seismic model using self-supervised learning and well constraints, comprising:
[0010] Seismic profile data are extracted from the seismic dataset and used as samples to construct a sample set;
[0011] The training sample set is obtained by combining the sample occlusion ratio and occlusion size;
[0012] The selected coding network is trained using a training sample set through self-supervised learning to obtain an initial large-scale earthquake model.
[0013] Establish well-pass seismic trace label data, and iteratively optimize the initial large seismic model based on the loss function to obtain a large seismic model, which is used to extract feature information from the input seismic data.
[0014] Optionally, the establishment of well-through seismic trace label data includes:
[0015] Along the main seismic survey line, seismic data with a set number of traces are expanded outwards from the well trajectory as the center to obtain well-passing seismic trace label data.
[0016] Along the direction of the seismic connection line, seismic data with a set number of traces are expanded to both sides of the well trajectory as the center to obtain the well-passing seismic trace label data.
[0017] Starting from the set azimuth angle, the azimuth angles are traversed on the plane at set azimuth angle intervals. Along the direction of the currently traversed azimuth angle, the seismic data of the set number of traces are expanded to both sides with the well trajectory as the center, thus obtaining the well-passing seismic trace label data.
[0018] Optionally, the seismic dataset includes pre-stack seismic data volumes and post-stack seismic data volumes.
[0019] Optionally, the step of performing self-supervised learning training on the selected encoding network using the training sample set includes:
[0020] Feature analysis is performed on samples in the training sample set by selecting an encoding network, and self-supervised learning training is carried out based on the feature analysis results.
[0021] Optionally, if the sample is post-stack seismic profile data, perform feature analysis on the sample, including:
[0022] The analysis focuses on the strength of the amplitude, its horizontal and vertical variations, the waveform's shape, and its horizontal continuity and discontinuity; correspondingly,
[0023] If the sample is pre-stack seismic profile data, perform feature analysis on the sample, including:
[0024] Partial overlay of pre-stack data was performed, and based on the obtained partially overlaid seismic data, the strength of amplitude, the information on amplitude variation in the horizontal and vertical directions, the shape of the waveform, and the horizontal continuity and discontinuity of the waveform were analyzed.
[0025] Optionally, the training sample set is obtained by combining the sample occlusion ratio and occlusion size, including:
[0026] Different training sample sets are obtained by combining different sample occlusion ratios and occlusion sizes; correspondingly,
[0027] The step of using a training sample set to perform self-supervised learning training on a selected encoding network includes:
[0028] The selected encoding network is trained using each training sample set in turn through self-supervised learning.
[0029] Optionally, the encoding network includes multiple cascaded convolutional modules, each containing a convolutional layer, a normalization layer, and an activation layer, and each convolutional module is used to extract feature information from the input seismic data.
[0030] Secondly, embodiments of the present invention provide a large-scale seismic model, which is constructed using any of the self-supervised learning and well-constrained large-scale seismic model construction methods described above.
[0031] Thirdly, embodiments of the present invention provide a self-supervised learning and well-constrained large seismic model construction apparatus, comprising:
[0032] The sample set construction module is used to extract seismic profile data from the seismic dataset as samples and construct a sample set.
[0033] The training sample set construction module is used to combine samples according to their occlusion ratio and occlusion size to obtain the training sample set;
[0034] The initial earthquake large-scale model training module is used to perform self-supervised learning training on a selected coding network using a training sample set to obtain the initial earthquake large-scale model.
[0035] The earthquake large model training module is used to establish well-pass seismic trace label data, and iteratively optimize the initial earthquake large model based on the loss function to obtain the earthquake large model, which is used to extract feature information from the input earthquake data.
[0036] Fourthly, embodiments of the present invention provide a computer storage medium storing computer-executable instructions, which, when executed by a processor, implement any of the above-mentioned self-supervised learning and well-constrained large seismic model construction methods.
[0037] Fifthly, embodiments of this disclosure provide a server, 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 any of the above-described self-supervised learning and well-constrained large seismic model construction methods.
[0038] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following:
[0039] (1) The self-supervised learning and well-constrained large-scale seismic model construction method provided in this embodiment of the invention extracts seismic profile data from the seismic dataset, constructs a training sample set, and performs self-supervised learning training on the encoding network to obtain an initial large-scale seismic model. The initial large-scale seismic model has strong predictive ability for seismic data in conventional directions. Based on this, well-passing seismic trace label data is established, and well-passing seismic traces are directly integrated into the loss function as key constraints, which enhances the prediction accuracy and robustness of the large-scale seismic model in complex geological environments. Its predictive ability in any direction is strong. Therefore, the final large-scale seismic model has high computational accuracy, good accuracy, strong generalization ability, and strong interpretability. This method realizes a new paradigm for the construction of large-scale seismic models by deeply integrating the self-supervised learning mechanism and well constraints.
[0040] (2) The self-supervised learning and well-constrained seismic large model construction method provided in this embodiment of the invention not only makes full use of the existing rich seismic exploration data resources, but also greatly reduces the burden of manual annotation, effectively reducing manpower costs and time consumption.
[0041] Other features and advantages of the invention will be set forth in the description which follows, 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, claims, and drawings.
[0042] 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
[0043] 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:
[0044] Figure 1This is a flowchart of the self-supervised learning and well-constrained large seismic model construction method in an embodiment of the present invention;
[0045] Figure 2 This is a flowchart illustrating the self-supervised learning task for constructing a large-scale earthquake model in an embodiment of the present invention;
[0046] Figure 3 This is a schematic diagram illustrating the extraction of seismic trace tag data through wells in an embodiment of the present invention;
[0047] Figure 4 This is a flowchart illustrating the specific implementation of the self-supervised learning and well-constrained large seismic model construction method in this embodiment of the invention.
[0048] Figure 5 This is a structural diagram of the earthquake large model construction device based on self-supervised learning and well constraints in an embodiment of the present invention. Detailed Implementation
[0049] 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.
[0050] It should be understood that the terminology used in this invention is merely for describing particular embodiments and is not intended to limit the invention. Furthermore, with respect to numerical ranges in this invention, it should be understood that each intermediate value between the upper and lower limits of the range is also specifically disclosed. Every smaller range between any stated value or intermediate value within a stated range, and any other stated value or intermediate value within said range, is also included in this invention. The upper and lower limits of these smaller ranges may be independently included or excluded from the range.
[0051] Unless otherwise stated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. While only preferred methods and materials have been described herein, any methods and materials similar or equivalent to those described herein may be used in the implementation or testing of this invention. All references to this specification are incorporated by way of citation to disclose and describe methods and / or materials associated with those references. In the event of any conflict with any incorporated reference, the content of this specification shall prevail.
[0052] Example
[0053] This invention provides a method for constructing a large seismic model using self-supervised learning and well constraints, the process of which is described in detail below. Figure 1 As shown, it includes the following steps:
[0054] Step S11: Extract seismic profile data from the seismic dataset as samples to construct a sample set.
[0055] We collect and organize pre-stack and post-stack seismic data volumes generated during seismic exploration to construct a seismic dataset.
[0056] Furthermore, pre-stack seismic data volumes, mainly referring to CRP gather data, may also include corresponding partial stacking data obtained according to different stacking methods.
[0057] The collected seismic data were optimized. For pre-stack data, high-frequency noise, random noise, and linear oblique interference noise were removed, and layer flattening was performed. For post-stack data, high-frequency noise and random noise were removed.
[0058] Noise can be eliminated by filtering data using low-pass filtering, median filtering, and high-pass filtering; after converting the signal to the frequency domain using Fourier transform, the frequency components can be adjusted to achieve the purpose of noise reduction.
[0059] Seismic profile data can be extracted from seismic datasets using various extraction methods. For example, random profiles can be extracted along different inline, crossline, and arbitrary tangent directions of the seismic data.
[0060] Step S12: Combine the samples according to the occlusion ratio and occlusion size to obtain the training sample set.
[0061] The masking ratio refers to the proportion of data in the original profile data that is masked. In other words, it's the percentage of masked samples out of the total number of samples in the original sample set. Different masking ratios can be set according to actual needs, such as 25%, 50%, 75%, etc.
[0062] Masking size refers to the physical extent of the data being masked. Different masking sizes, such as 8x8, 16x16, and 32x32, can be used to simulate different ranges of missing data.
[0063] The masked data can be filled with 0 values or Gaussian noise values.
[0064] Step S13: Use the training sample set to perform self-supervised learning training on the selected coding network to obtain the initial large earthquake model.
[0065] The coding network consists of multiple cascaded convolutional modules, each containing a convolutional layer, a normalization layer, and an activation layer. Each convolutional module is used to extract feature information from the input seismic data.
[0066] Feature analysis is performed on samples in the training sample set by selecting an encoding network, and self-supervised learning training is carried out based on the results of the feature analysis.
[0067] If the sample is post-stack seismic profile data, the feature analysis of the sample can include analyzing the strength of the amplitude, the information on the horizontal and vertical changes of the amplitude, the shape of the waveform, the horizontal continuity and discontinuity of the waveform, to obtain the structural characteristics of the target layer and to make reservoir predictions.
[0068] If the sample is pre-stack seismic profile data, the feature analysis of the sample may include partially stacking the pre-stack data, and based on the obtained partially stacked seismic data, analyzing the strength of the amplitude, the information on the horizontal and vertical changes of the amplitude, the shape of the waveform, and the horizontal continuity and discontinuity of the waveform.
[0069] The relationship between amplitude and offset (incident angle) in pre-stack data is processed. The amplitude at the same location varies with different offsets (incident angles). Based on this variation, Vp, Vs, and ρ, obtained from the simplified Zoeppritz equation, can be used to calculate various elastic parameters. Based on rock physics analysis, elastic parameters sensitive to reservoir fluids in this area are selected. Simultaneously, combined with geological understanding of structure and hydrocarbon accumulation, the location of high-quality reservoirs is analyzed.
[0070] In some embodiments, different training sample sets can be obtained by combining different sample occlusion ratios and occlusion sizes; and each training sample set can be used sequentially to perform self-supervised learning training on the selected encoding network.
[0071] By changing the masking ratio, different degrees of data loss can be simulated, thereby assessing the tolerance for data loss.
[0072] By changing the masking size, we can better understand the impact of missing data at different ranges on subsequent processing and analysis.
[0073] In some embodiments, a masking value can be set, which refers to the specific value of the data to be masked. Different values of data can be selected from the original profile data for masking, depending on actual needs. This can simulate different types of data loss scenarios, thereby evaluating the processing effect on different signal types.
[0074] During the construction of the training sample set, it can be combined and adjusted according to actual needs to generate a series of sample sets under different conditions. These sample sets can be used to test and evaluate the performance of algorithms and models based on profile data, and can also provide a reference for subsequent data processing and analysis, such as... Figure 2 As shown.
[0075] Step S14: Establish well-through seismic trace label data, iteratively optimize the initial large seismic model based on the loss function to obtain the large seismic model, which is used to extract feature information from the input seismic data.
[0076] Establish well-through seismic trace label data, including at least one of the following methods:
[0077] (1) Along the main seismic survey line, seismic data with a set number of traces are expanded to both sides of the well trajectory as the center to obtain well-passing seismic trace label data;
[0078] (2) Along the direction of the seismic connection line, the seismic data of a set number of traces are expanded to both sides of the well trajectory as the center to obtain the well-passing seismic trace label data.
[0079] (3) Starting from the set azimuth angle, traverse the azimuth angle on the plane at the set azimuth angle intervals. Along the direction of the currently traversed azimuth angle, expand the seismic data of the set number of traces to both sides with the well trajectory as the center to obtain the well-passing seismic trace label data.
[0080] For example, starting from 0 azimuth, five well-crossing seismic traces are extracted every 30 degrees azimuth on the plane, centered on the well trajectory, to expand the well-crossing seismic trace label data. See also Figure 3 As shown.
[0081] Different expansion methods can be tried, such as expanding outwards along the actual well trajectory by 2, 4, or 6 channels, and by spacing them at 15-degree, 30-degree, or 45-degree angles on the plane. During the label extraction process, labels can be combined and adjusted according to actual needs to generate a series of labels under different conditions. These labels can be used to test and evaluate the performance of algorithms and models based on profile data, and can also provide a reference for subsequent data processing and analysis.
[0082] Using well-pass seismic trace label data as a quality control condition, the parameters in the self-supervised learning large seismic model are iterated using a loss function.
[0083] The quality control of the well-pass seismic trace label data is carried out in multiple directions, including the quality control of the main survey line and the connecting line, as well as the oblique quality control. The quality control is more stringent, and its loss function is more conducive to improving the accuracy of the large seismic model.
[0084] Based on the different samples constructed earlier, a pre-trained basic network architecture is used to input the constructed samples into the network for encoding. The network learns to extract useful features and can demonstrate good performance even on unlabeled data, thereby constructing a large-scale earthquake model.
[0085] The self-supervised learning and well-constrained large-scale seismic model construction method provided in this invention extracts seismic profile data from a seismic dataset to construct a training sample set. The encoding network is then trained using self-supervised learning to obtain an initial large-scale seismic model, which exhibits strong predictive ability for seismic data in conventional directions. Based on this, well-passing seismic trace label data is established, and these traces are directly incorporated into the loss function as key constraints. This enhances the prediction accuracy and robustness of the large-scale seismic model in complex geological environments, demonstrating strong predictive ability in any direction. Therefore, the final large-scale seismic model boasts high computational accuracy, good precision, strong generalization ability, and strong interpretability. This method, through the deep integration of self-supervised learning mechanisms and well-constrained conditions, realizes a new paradigm for large-scale seismic model construction.
[0086] The self-supervised learning and well-constrained large seismic model construction method provided in this invention not only makes full use of the abundant existing seismic exploration data resources, but also greatly reduces the burden of manual annotation, effectively reducing labor costs and time consumption.
[0087] The following examples illustrate the effectiveness of the self-supervised learning and well-constrained large-scale seismic model construction method provided in this invention:
[0088] This method was applied to the prediction of acoustic impedance properties in the xxx oilfield. The overall prediction results showed an error of less than 3% compared to the expert interpreter, meeting operational application requirements. Furthermore, the intelligent identification significantly improved the timeliness of the predictions. Without compromising or even improving interpretation quality, the required sample size was reduced by more than 15 times, and production efficiency was increased by more than 40 times.
[0089] In summary, the self-supervised learning and well-constrained large-scale seismic model construction method provided in this embodiment of the invention involves seismic data collection, seismic data preprocessing, construction of a self-supervised learning task based on the processed seismic data, and construction of a large-scale seismic model based on the constructed self-supervised learning task. Figure 4 As shown in the figure, this method makes full use of a large amount of unlabeled seismic data and performs self-learning on the seismic data. The whole method improves the development efficiency of intelligent models in the field of seismology, and the models are more accurate and robust.
[0090] Based on the inventive concept of this invention, embodiments of this invention also provide a self-supervised learning and well-constrained large-scale seismic model construction device, the structure of which is as follows: Figure 5 As shown, it includes:
[0091] The sample set construction module 51 is used to extract seismic profile data from the seismic dataset as samples to construct a sample set;
[0092] The training sample set construction module 52 is used to combine samples according to the occlusion ratio and occlusion size to obtain the training sample set;
[0093] The initial earthquake large model training module 53 is used to perform self-supervised learning training on the selected coding network using the training sample set to obtain the initial earthquake large model.
[0094] The earthquake large model training module 54 is used to establish well-pass seismic trace label data, iteratively optimize the initial earthquake large model based on the loss function, and obtain the earthquake large model, which is used to extract feature information from the input earthquake data.
[0095] In some embodiments, the earthquake large model training module 54, which establishes well-pass seismic trace label data, is used for:
[0096] Along the main seismic survey line, seismic data of a set number of traces are expanded outwards from the well trajectory as the center to obtain well-passing seismic trace label data; along the seismic tie line, seismic data of a set number of traces are expanded outwards from the well trajectory as the center to obtain well-passing seismic trace label data; starting from a set azimuth angle, the azimuth angles are traversed at set azimuth angle intervals on the plane, and along the direction of the currently traversed azimuth angle, seismic data of a set number of traces are expanded outwards from the well trajectory as the center to obtain well-passing seismic trace label data.
[0097] In some embodiments, the initial large-scale earthquake model training module 53, which utilizes a training sample set to perform self-supervised learning training on a selected coding network, is used for:
[0098] Feature analysis is performed on samples in the training sample set by selecting an encoding network, and self-supervised learning training is carried out based on the feature analysis results.
[0099] In some embodiments, if the sample is post-stack seismic profile data, the initial large-scale seismic model training module 53 performs feature analysis on the sample for the following purposes:
[0100] The analysis focuses on the strength of the amplitude, its horizontal and vertical variations, the waveform's shape, and its horizontal continuity and discontinuity; correspondingly,
[0101] If the sample is pre-stack seismic profile data, perform feature analysis on the sample, including:
[0102] Partial overlay of pre-stack data was performed, and based on the obtained partially overlaid seismic data, the strength of amplitude, the information on amplitude variation in the horizontal and vertical directions, the shape of the waveform, and the horizontal continuity and discontinuity of the waveform were analyzed.
[0103] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.
[0104] Based on the inventive concept of this invention, embodiments of this invention also provide a computer storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-mentioned method for constructing large seismic models with self-supervised learning and well constraints.
[0105] Based on the inventive concept of this invention, this embodiment of the invention also provides a server, 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-mentioned self-supervised learning and well-constrained large seismic model construction method.
[0106] Unless otherwise specifically stated, terms such as processing, calculation, operation, determination, display, etc., may refer to the actions and / or processes of one or more processing or computing systems or similar devices that represent the manipulation and conversion of data representing physical (e.g., electronic) quantities within the registers or memory of the processing system into other data similarly representing physical quantities within the memory, registers, or other such information storage, transmission, or display devices of the processing system. Information and signals can be represented using any of a variety of different techniques and methods. For example, data, instructions, commands, information, signals, bits, symbols, and chips mentioned throughout the above description can be represented by voltage, current, electromagnetic waves, magnetic fields or particles, light fields or particles, or any combination thereof.
[0107] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process may be rearranged without departing from the scope of this disclosure. The appended method claims provide elements of various steps in an exemplary order and are not intended to limit the scope to the specific order or hierarchy described.
[0108] In the detailed description above, various features are combined together in a single embodiment to simplify this disclosure. This approach to disclosure should not be construed as reflecting an intention that embodiments of the claimed subject matter require more features than those stated in each claim. Rather, as reflected in the appended claims, the invention is presented with fewer features than all of the features in a single disclosed embodiment. Therefore, the appended claims are hereby clearly incorporated into the detailed description, wherein each claim stands alone as a preferred embodiment of the invention.
[0109] Those skilled in the art will also understand that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments herein can be implemented as electronic hardware, computer software, or a combination thereof. To clearly illustrate the interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps described above are generally described in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Those skilled in the art can implement the described functionality in alternative ways for each specific application; however, such implementation decisions should not be construed as departing from the scope of this disclosure.
[0110] The steps of the methods or algorithms described in conjunction with the embodiments herein can be directly embodied in hardware, software modules executed by a processor, or a combination thereof. The software modules can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. The ASIC can reside in a user terminal. Alternatively, the processor and storage medium can exist as discrete components in the user terminal.
[0111] For software implementation, the techniques described in this application can be implemented using modules (e.g., procedures, functions, etc.) that perform the functions described in this application. This software code can be stored in memory units and executed by a processor. The memory units can be implemented within the processor or outside the processor; in the latter case, they are communicatively coupled to the processor via various means, as is well known in the art.
[0112] The foregoing description includes examples of one or more embodiments. It is certainly impossible to describe all possible combinations of components or methods in order to describe the above embodiments, but those skilled in the art will recognize that further combinations and arrangements of the various embodiments are possible. Therefore, the embodiments described herein are intended to cover all such changes, modifications, and variations that fall within the scope of the appended claims. Furthermore, the term "comprising" as used in the specification or claims is interpreted in a manner similar to the term "including," as interpreted when used as a conjunction in the claims. Additionally, the use of any term "or" in the specification of the claims is intended to mean "non-exclusive or."
Claims
1. A method for constructing a large seismic model using self-supervised learning and well constraints, characterized in that, include: Seismic profile data are extracted from the seismic dataset and used as samples to construct a sample set; The training sample set is obtained by combining the sample occlusion ratio and occlusion size; The selected coding network is trained using a training sample set through self-supervised learning to obtain an initial large-scale earthquake model. Establish well-pass seismic trace label data, and iteratively optimize the initial large seismic model based on the loss function to obtain a large seismic model, which is used to extract feature information from the input seismic data.
2. The method according to claim 1, characterized in that, The established well-pass seismic trace label data includes: Along the main seismic survey line, seismic data with a set number of traces are expanded outwards from the well trajectory as the center to obtain well-passing seismic trace label data. Along the direction of the seismic connection line, seismic data with a set number of traces are expanded to both sides of the well trajectory as the center to obtain the well-passing seismic trace label data. Starting from the set azimuth angle, the azimuth angles are traversed on the plane at set azimuth angle intervals. Along the direction of the currently traversed azimuth angle, the seismic data of the set number of traces are expanded to both sides with the well trajectory as the center, thus obtaining the well-passing seismic trace label data.
3. The method according to claim 1, characterized in that, The earthquake dataset includes pre-stack earthquake data volumes and post-stack earthquake data volumes.
4. The method according to claim 3, characterized in that, The step of using a training sample set to perform self-supervised learning training on a selected encoding network includes: Feature analysis is performed on samples in the training sample set by selecting an encoding network, and self-supervised learning training is carried out based on the feature analysis results.
5. The method according to claim 4, characterized in that, If the sample is post-stack seismic profile data, perform feature analysis on the sample, including: The analysis focuses on the strength of the amplitude, its horizontal and vertical variations, the waveform's shape, and its horizontal continuity and discontinuity; correspondingly, If the sample is pre-stack seismic profile data, perform feature analysis on the sample, including: Partial overlay of pre-stack data was performed, and based on the obtained partially overlaid seismic data, the strength of amplitude, the information on amplitude variation in the horizontal and vertical directions, the shape of the waveform, and the horizontal continuity and discontinuity of the waveform were analyzed.
6. The method according to claim 1, characterized in that, The training sample set is obtained by combining the sample occlusion ratio and occlusion size, including: Different training sample sets are obtained by combining different sample occlusion ratios and occlusion sizes; correspondingly, The step of using a training sample set to perform self-supervised learning training on a selected encoding network includes: The selected encoding network is trained using each training sample set in turn through self-supervised learning.
7. The method according to claim 1, characterized in that, The coding network includes multiple cascaded convolutional modules, each containing a convolutional layer, a normalization layer, and an activation layer. Each convolutional module is used to extract feature information from the input seismic data.
8. A large-scale earthquake model, characterized in that, The large-scale seismic model is constructed using the self-supervised learning and well-constrained large-scale seismic model construction method described in any one of claims 1 to 7.
9. A device for constructing a large seismic model using self-supervised learning and well constraints, characterized in that, include: The sample set construction module is used to extract seismic profile data from the seismic dataset as samples and construct a sample set. The training sample set construction module is used to combine samples according to their occlusion ratio and occlusion size to obtain the training sample set; The initial earthquake large-scale model training module is used to perform self-supervised learning training on a selected coding network using a training sample set to obtain the initial earthquake large-scale model. The earthquake large model training module is used to establish well-pass seismic trace label data, and iteratively optimize the initial earthquake large model based on the loss function to obtain the earthquake large model, which is used to extract feature information from the input earthquake data.
10. A computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions, which, when executed by a processor, implement the self-supervised learning and well-constrained large seismic model construction method according to any one of claims 1 to 7.
11. A server, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the self-supervised learning and well-constrained large seismic model construction method according to any one of claims 1 to 7.