A downhole geological structure imaging system based on azimuth gamma rays

Abstract

This invention relates to a downhole geological structure imaging system based on azimuth gamma rays, belonging to the field of downhole logging and geological exploration technology. It utilizes a multi-azimuth detector array and multi-modal data acquisition and fusion, leveraging the Transformer architecture and GAN, along with self-supervised learning to enhance scarce or low-quality data, and combines a multi-head cross-attention mechanism to accurately identify complex formations. The system supports multiple tasks such as lithological classification, fault prediction, and abnormal formation alerts, achieving real-time, high-resolution imaging and safety warnings, and is suitable for oil and gas drilling and underground engineering scenarios.

Description

Technical Field

[0001] This invention belongs to the field of downhole logging and geological exploration technology, specifically relating to a downhole geological structure imaging system based on azimuth gamma rays. Background Technology

[0002] With the increasing demands of oil and gas field exploration and development, accurately acquiring formation information and identifying key structures such as faults and fractures in complex downhole environments has become a crucial research direction in current drilling and logging technologies. Azimuth gamma-ray technology, capable of acquiring radioactive signals within a 360° circumferential range, has been widely applied in measurement while drilling and formation evaluation. Through azimuth gamma-ray detection, the directionality of gamma-ray counts can be analyzed in conjunction with sensor attitude information, resulting in more refined depictions of geological structures.

[0003] Chinese invention patent CN106285632B discloses an azimuth gamma measurement device. It acquires gamma count values ​​and probe attitude by using a focused gamma sensor, a three-axis accelerometer, and a three-axis fluxgate. The collected gamma data is then normalized to the same coordinate system in different directions, realizing continuous measurement of azimuth gamma and multi-attitude adaptation. This device can effectively determine the properties and changes of complex strata and improve the drilling success rate.

[0004] Chinese invention patent CN109581522B discloses a probe-type azimuth gamma probe for a drilling rig. By setting a gamma ray probe composed of a NaI crystal and a photomultiplier tube in the probe and combining it with three gravity acceleration sensors, the direction of gamma detection is determined, realizing the directional measurement of drill collar gamma data. This overcomes the problems of complex structure, high cost and low accuracy of traditional methods, and has certain practical value.

[0005] The above design achieves directional gamma measurement through an azimuth gamma probe and attitude sensor, meeting the basic requirements for downhole formation identification. However, it still has certain limitations, such as significant deficiencies in multimodal data fusion, real-time high-resolution imaging, and dynamic prediction of complex formations. It struggles to adequately address the comprehensive analysis needs of various geological anomalies (faults, fractures, lithological transitions, etc.). Furthermore, current azimuth gamma logging systems still have room for improvement in areas such as data completion, noise suppression, and remote transmission visualization. Upgrading these aspects could significantly improve downhole exploration efficiency and substantially reduce trial-and-error costs and potential risks caused by incomplete information during drilling. Summary of the Invention

[0006] In response to the aforementioned existing technologies, a downhole geological structure imaging system based on azimuth gamma rays is proposed. By integrating advanced spatiotemporal modeling, multimodal data fusion, and gamma completion algorithms, it aims to achieve more accurate and dynamic intelligent identification and imaging of downhole formation structures, providing an efficient and reliable solution for geological exploration and oil and gas development.

[0007] To address the aforementioned problems, this invention provides a downhole geological structure imaging system based on azimuth gamma rays.

[0008] In summary, this application has the following beneficial effects:

[0009] 1. Through a ring-shaped multi-directional gamma-ray detector array, all-round monitoring of the downhole formation can be achieved, and changes in radioactivity intensity from any direction can be detected in a timely manner. The dynamic signal adjustment unit can adjust the detector sensitivity in real time according to environmental factors such as wellbore eccentricity and drilling fluid influence, further reducing the interference of noise on the detection results.

[0010] 2. Utilizing multimodal devices such as resistivity, acoustic wave, and temperature and pressure sensors, the system acquires key formation information such as conductivity, density, and porosity while monitoring formation radioactivity characteristics. The data synchronization interface ensures high temporal and spatial consistency between gamma-ray data and other modal data, providing a solid foundation for subsequent multimodal fusion and imaging. This enhances the system's ability to perceive complex formation environments and effectively reduces data loss or timing discrepancies during real-time acquisition and transmission, making the detection results more reliable. For downhole environments with high noise levels or unstable acquisition conditions, this invention introduces preprocessing and denoising mechanisms, significantly improving the signal-to-noise ratio of the original gamma-ray signal and multimodal data. By analyzing historical data and real-time acquired signals, it automatically identifies and corrects gamma-ray attenuation or abnormal fluctuations caused by drilling fluid absorption, wellbore eccentricity, etc. Combined with advanced algorithm modules such as GAN and self-supervised learning, it can further improve data reliability under conditions of scarce or low-quality data, ensuring the accuracy and consistency of data in subsequent modeling and fusion stages, and providing stable support for acquiring high-quality signals even in harsh downhole environments.

[0011] 3. The Transformer-based spatiotemporal modeling unit can perform multiple tasks such as lithological classification, structural analysis, and fault prediction of downhole formations, improving the system's functional integration. The highly automated deep learning mechanism reduces reliance on manual intervention. When downhole sensing data has insufficient sampling or low resolution, GAN can generate high-resolution gamma-ray feature maps to achieve data augmentation and missing data completion. Self-supervised learning utilizes pseudo-label generation and comparative learning, significantly reducing the need for manually labeled data, enabling the model to continuously optimize even in large-scale unlabeled or weakly labeled data scenarios. It adapts to various sudden acquisition interferences and can establish deep correlations between gamma rays and multimodal data such as resistivity, acoustic waves, temperature, and pressure. It can identify subtle stratigraphic features that are originally difficult to detect. Through more interactions and comparisons between modes, it reduces missed or false detections, making stratigraphic imaging more realistic and the details clearer. The system maintains high accuracy and robustness in multimodal and multi-task scenarios. It is particularly sensitive to the identification of key geological structures such as faults, fissures, and lithological transition zones. It reduces the dependence on large-scale, high-quality manual annotation and improves feasibility when the original data is insufficient or there is severe noise interference.

[0012] 4. By considering non-uniform sampling or dynamic sampling intervals in temporal location encoding, the model becomes more adaptable to changes in actual downhole acquisition frequency. Information such as wellbore dip angle and eccentricity is incorporated into spatial location encoding, helping the system to perform more detailed modeling of data acquired by the detector at different azimuth angles and depths. In the continuous time-space domain, the system can predict the next moment or azimuth based on the signal at the current azimuth or time, allowing for early judgment of fault activity or lithological boundary movement. Combined with GAN data completion and self-supervised strategies, this significantly improves prediction accuracy and real-time performance in dynamic scenarios, more completely capturing the dynamic process of formation changes with time and azimuth, providing a scientific basis for real-time downhole decision-making, and responding promptly to potential sudden formation activity (such as microseismic events or fluid mutations) to prevent safety hazards in downhole operations.

[0013] 5. The multi-task support module incorporates an "abnormal formation alarm" function. Once abnormal signals such as faults, fractures, or pressure surges are detected, the system can issue alarms in real time. Multimodal cross-validation improves the reliability of alarms and reduces false alarms caused by relying solely on single-modal data. Combined with 2D / 3D imaging and a visualization interface, geological engineers can quickly interpret and make decisions. It supports historical data overlay analysis, which helps to track downhole environmental changes over long periods, forming predictive and planning production strategies, improving the safety factor and real-time monitoring efficiency of downhole operations, and effectively avoiding construction delays or accidents caused by formation anomalies. The visualization display method meets the needs of modern operations, allowing the team to intuitively grasp the overall situation downhole. Attached Figure Description

[0014] Figure 1 This is a system architecture diagram for this application; Detailed Implementation

[0015] To make the objectives, technical solutions, and beneficial effects of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, so as to facilitate understanding by those skilled in the art.

[0016] Example 1

[0017] like Figure 1 As shown, this embodiment focuses on how to build and operate a downhole geological structure imaging system based on azimuth gamma rays, covering core units such as gamma ray detection module, multimodal data acquisition module, signal processing and correction module, intelligent modeling and multimodal fusion module, and data transmission and display module.

[0018] The system comprises several hardware and software modules, forming an integrated solution from downhole data acquisition to surface visualization and analysis. It enables real-time acquisition of gamma-ray and other modal sensor data downhole, and efficient processing, fusion, and visualization imaging. The entire process includes:

[0019] The gamma-ray detection module is used to acquire the azimuth gamma-ray intensity and energy spectrum data of the downhole formation, forming the raw gamma-ray signal input.

[0020] A ring-shaped multi-directional gamma-ray detector array is used to distribute the detectors in a 360° orientation; each detector is responsible for collecting gamma-ray signals in a specific direction.

[0021] Equipped with a dynamic signal adjustment unit, the detector sensitivity is adjusted in real time according to different environments (such as drilling fluid properties, wellbore eccentricity, etc.) to maintain high accuracy in signal acquisition.

[0022] In addition to gamma rays, the multimodal data acquisition module incorporates multimodal sensors such as resistivity and acoustic logging to collect information on the conductivity, formation density, and porosity of the downhole environment; it also monitors environmental parameters such as temperature and pressure.

[0023] The data synchronization interface processes timestamps uniformly during the acquisition process to ensure that the time dimension of gamma-ray data is aligned with that of other modal data, providing a foundation for subsequent multimodal fusion.

[0024] The signal processing and correction module preprocesses the raw acquired data, such as normalization and noise reduction, to improve the signal-to-noise ratio of the gamma-ray signal.

[0025] By applying a dynamic correction mechanism, errors caused by downhole environmental factors such as wellbore eccentricity and drilling fluid absorption are reduced by analyzing historical data and current real-time signals.

[0026] The intelligent modeling and multimodal fusion module adopts a core algorithm framework based on Transformer to generate high-resolution imaging maps of downhole geological structures and realize multi-task functions, such as lithology classification, geological structure analysis, and fault prediction.

[0027] It includes a multimodal feature fusion unit, a dynamic weight allocation unit, and a spatiotemporal modeling unit, which respectively perform fusion, weighted scheduling, and in-depth modeling of different modal data at the time-space level.

[0028] The data transmission and display module uses high-speed wireless communication or fiber optic technology to transmit massive amounts of downhole acquisition and analysis results back to the surface in real time.

[0029] It provides an intuitive visualization interface that can display two-dimensional / three-dimensional geological structure maps, lithological distribution maps, fault location maps, and various real-time signal trend maps; it also supports overlaying historical data on the same layer for comparative analysis.

[0030] Working principle

[0031] A ring-shaped multi-azimuth detector array is deployed on the downhole instrument, with each detector corresponding to a different azimuth angle, to collect real-time data on the intensity and energy spectrum of gamma rays in the formation.

[0032] The detector's sensitivity is dynamically adjusted based on environmental disturbances to minimize the interference of noise on the results.

[0033] Simultaneously with gamma-ray detection, resistivity, acoustic logging, temperature, and pressure sensors work together to acquire multiple modal signals that reflect formation characteristics.

[0034] Through the data synchronization interface, data from different time segments are aligned in a unified manner, enabling synchronization of azimuth gamma rays with other modal signals in both time and space dimensions.

[0035] In the preprocessing unit, the system performs normalization, filtering, and noise reduction on the acquired gamma-ray data and other modal data to remove random noise from the external environment and the electrical equipment itself.

[0036] The dynamic correction mechanism corrects for gamma ray attenuation or abnormal fluctuations caused by factors such as wellbore eccentricity and changes in drilling fluid density, based on historical and current data, making subsequent modeling and analysis more reliable.

[0037] The system utilizes the Transformer encoder-decoder architecture to perform spatiotemporal modeling on the input gamma-ray data (and other modal data):

[0038] Data input: The gamma-ray signals from each azimuth detector are combined into an input sequence, supplemented with timestamp information, to form a spatiotemporal dataset;

[0039] Autoregressive modeling: using the Transformer decoder part to predict the next signal based on the signal at the current time or location;

[0040] Attention mechanism: The multi-head attention module automatically learns which azimuth points and time periods are more characteristic of stratigraphic changes, thereby deeply exploring key details such as lithological transitions and fault fractures;

[0041] Location coding: Add location coding to the azimuth and depth information of each detector to help the model better understand the differences in the spatial and temporal distribution of the strata;

[0042] Multimodal fusion

[0043] Feature-level fusion: Mapping sensor data such as resistivity and acoustic waves to the same feature space as gamma rays, and forming a multimodal fusion vector by splicing or weighting.

[0044] Decision-level fusion: For certain tasks (such as lithological classification or fault location prediction), each mode independently generates prediction results, and then voting or weighted averaging are performed to finally generate the optimal prediction.

[0045] Dynamic weight allocation: Based on the quality of downhole signals (e.g., weak resistivity signals or severe gamma ray interference at certain times), the system will automatically adjust the weight of different modal characteristics in the final result;

[0046] Spatiotemporal modeling unit: Combining temporal and spatial location coding, it identifies the dynamic change trends of strata and outputs more spatiotemporally resolved stratigraphic structure prediction results.

[0047] Multitasking support

[0048] Lithological classification: Based on gamma-ray energy spectrum and multimodal fusion characteristics, the lithological types of strata are identified.

[0049] Geological structure analysis: Automatically identifies the dip angle of strata, fractures, and fault locations.

[0050] Fault prediction: By fusing historical data with current location signals, the activity trend and possible location of faults can be predicted.

[0051] For ease of explanation, the following formulas use general symbols, and the specific parameters can be adjusted according to the actual application scenario.

[0052] Transformer encoder-decoder structure

[0053] Attention Mechanism

[0054]

[0055] Multi-head attention

[0056] Extending the above single-head attention mechanism to h parallel attention heads can be represented as:

[0057]

[0058] Autoregressive Modeling

[0059] In the decoder section, an autoregressive strategy can be used to predict signals at subsequent time steps. For time step t, the next signal is predicted. It can be represented as:

[0060]

[0061] Positional Encoding

[0062] To enable the Transformer model to perceive the temporal and spatial location information of the input sequence, the following sine and cosine functions are often used for location encoding. Let the location index be pos and the dimension index be i, then a common temporal location encoding can be defined as:

[0063]

[0064] If information such as azimuth angle α and well depth δ is introduced into the spatial location coding, pos can be replaced with a composite location index related to α and δ, and then sine and cosine mapping can be performed in the same way:

[0065]

[0066] By combining temporal location encoding with spatial location encoding, a more complete spatiotemporal distribution embedding can be constructed for modeling dynamic features of downhole formations.

[0067] Multimodal feature fusion

[0068]

[0069] Feature-level fusion

[0070] The multimodal vectors mapped above are concatenated or weighted to obtain the fusion vector F. t

[0071]

[0072] Decision-level integration

[0073]

[0074] Dynamic weight allocation

[0075] To address potential quality fluctuations in gamma-ray and multimodal data, signal quality scoring (SQ) can be used. i (t) Dynamically adjust the weights of each modality during the fusion process. Let i represent the modality index and t be the time step. Define the weight update function as follows:

[0076]

[0077] Spatiotemporal modeling unit

[0078] Combining the positional encoding, attention mechanism, autoregressive prediction, and multimodal fusion discussed above, the fused feature sequence F t The input is fed into the encoder-decoder structure of the Transformer:

[0079] The encoder performs deep processing on the input sequence and outputs a spatiotemporal context vector Z. t .

[0080] In autoregressive mode, the decoder combines historical information Z≤t and attention allocation to predict the next stratigraphic signal or geological structure information.

[0081]

[0082] This results in more spatiotemporal resolution predictions of stratigraphic structure, such as lithological classification, fault location, or other real-time monitoring indicators.

[0083] Imaging output and visualization

[0084] After completing the fusion modeling of gamma-ray and other modal data, the system dynamically generates two-dimensional or three-dimensional geological structure maps to help geologists intuitively understand the actual situation downhole.

[0085] If data loss or severe interference occurs during the acquisition process, AI algorithms can be used to complete or interpolate the missing data, thereby improving the integrity and accuracy of the image.

[0086] With the help of high-speed data transmission units, massive amounts of stratigraphic information and imaging results are sent to the ground operation platform in real time, enabling linkage between the well and the surface. The visualization interface provides geological engineers with multi-level and multi-angle graphical displays, including lithological distribution maps, fault location maps, etc.

[0087] With a 360° azimuth gamma-ray detector array, the system can monitor the radioactive characteristics of the strata in an all-round and continuous manner, and improve accuracy with the help of a dynamic sensitivity adjustment unit.

[0088] The multimodal data synchronous acquisition and dynamic correction mechanism significantly reduces environmental interference and improves signal reliability.

[0089] By fusing resistivity, acoustic waves, temperature, and pressure data with gamma-ray data in a multimodal manner, the system gains more comprehensive data support when dealing with complex geological formations.

[0090] It enables multiple functions such as lithology classification, geological structure analysis, and fault prediction on the same platform, providing a "one-stop" solution for downhole decision-making and geological assessment.

[0091] With the support of high-speed wireless or fiber optic communication technology, massive amounts of downhole data can be acquired and analyzed in real time on the ground; the visualization interface can not only display two-dimensional / three-dimensional geological maps, but also overlay and compare historical data to assist geological engineers in making scientific judgments.

[0092] The modular system architecture described in this embodiment can be easily upgraded or integrated with more algorithms and hardware units (such as new sensors and enhanced AI models), leaving room for future technological development.

[0093] In summary, Example 1, through a detailed description of the gamma-ray detection module, multimodal data acquisition module, signal processing and correction module, intelligent modeling and multimodal fusion module, and data transmission and display module, demonstrates a complete technical chain from omnidirectional downhole data acquisition to high-resolution surface geological imaging. This system not only significantly enhances the monitoring, diagnosis, and prediction capabilities of downhole formations but also possesses high flexibility and multi-task compatibility, making it extremely valuable for applications in fields such as oil drilling, geological exploration, and underground engineering.

[0094] Example 2:

[0095] like Figure 1 As shown, this embodiment, based on the aforementioned Embodiment 1 (i.e., the overall system structure and basic function implementation), additionally introduces advanced algorithm modules such as Generative Adversarial Networks (GANs) and self-supervised learning, further enhancing the depth and breadth of intelligent modeling and multimodal fusion. GANs can generate high-resolution gamma-ray feature maps, compensating for insufficient sampling in cases of scarce data; simultaneously, the self-supervised learning module significantly reduces the need for manual annotation through methods such as pseudo-label generation, improving the model's accuracy and robustness under low-quality or incomplete data.

[0096] Adversarial learning modules (GANs) are designed to address the common problems of insufficient or low-resolution data in downhole operations.

[0097] By having the generator and discriminator of the adversarial generative network compete against each other, the realism and resolution of the synthesized gamma-ray feature data are continuously "improved".

[0098] Complementing the basic models such as the spatiotemporal Transformer and attention mechanism in Example 1, it provides more stable and richer input data for subsequent multimodal fusion and multi-task support.

[0099] The self-supervised learning module addresses challenges that may arise in downhole environments, such as insufficient manually labeled data and difficulty in obtaining real labels.

[0100] By using pseudo-label generation and self-supervised objective optimization, the model can automatically learn stratigraphic features and correlation patterns without requiring large-scale manual annotation.

[0101] By combining the self-attention mechanism in the Transformer framework, more hidden spatiotemporal features can be discovered, making the system more resilient and adaptable when dealing with complex situations such as low signal-to-noise ratio and formation anomalies.

[0102] Compared to Example 1, Example 2 focuses on a detailed discussion of how the two new modules described above can be embedded into an existing system, and their impact on data quality improvement and model robustness.

[0103] How the Adversarial Learning Module (GAN) works

[0104] Introducing a GAN typically involves two sub-networks:

[0105] Generator: Based on the initial gamma-ray data or multimodal fusion vector as input, it generates a "fake" high-resolution gamma-ray feature map through a multi-layer convolutional / deconvolutional network or Transformer encoder.

[0106] Discriminator: Trained in opposition to the generator, it is responsible for determining whether the input gamma-ray feature map is genuine data from actual collection or fabricated by the generator.

[0107] By scoring the "realism" of the generator's output using a discriminator, the generator is forced to continuously optimize itself and output more realistic and higher resolution gamma-ray feature maps.

[0108] Specific procedures

[0109] Select some low-resolution or incomplete gamma-ray data acquired from downhole as input to the generator; multimodal data (such as resistivity and sonic logging) can be regarded as condition vectors to help the generator better understand the formation background.

[0110] The generator outputs a gamma-ray feature map that is "completed" or "enhanced" in the spatiotemporal dimension. This feature map surpasses traditional interpolation algorithms in terms of spatial resolution and detail fidelity.

[0111] The discriminator's role is to compare the generator's output with the actual acquired "real" gamma-ray images and learn to distinguish between the genuine and the fake.

[0112] If the discriminator has a high recognition rate, it forces the generator to further improve the realism of the forged image; if the discriminator has a low recognition rate, it indicates that the generator's performance has been significantly improved.

[0113] Loss function and training

[0114] Adversarial loss is typically used, but perceptual loss or reconstruction loss can also be added to constrain the feature layer.

[0115] In the iterations of Mini-batch Stochastic Gradient Descent (MSD) or other optimizers, the generator and discriminator continuously improve each other until training converges, resulting in high-quality gamma-ray feature maps that can fill in missing data and enhance resolution.

[0116] Adversarial Loss

[0117] Let the generator be denoted as G, the discriminator as D, and the noisy or conditional input distribution (such as sparse gamma-ray data, multimodal features, etc.) as z. Assume the true distribution is p. data Adopting adversarial objectives:

[0118] Discriminator loss:

[0119]

[0120] Generator loss:

[0121]

[0122] Perceptual Loss

[0123] Perceptual loss is often used to measure the difference between the generated image (or feature map) and the real image in the high-level feature space. Let ϕ represent a pre-trained feature extraction network (such as a partial convolutional layer of VGG or ResNet), then:

[0124]

[0125] By comparing in a high-level semantic space, the model is able to better preserve details and structural features.

[0126] Reconstruction Loss

[0127] In data completion or super-resolution scenarios, reconstruction errors between the original and actual resolution data are typically introduced, for example:

[0128]

[0129] Alternatively, mean square error can be used:

[0130]

[0131] In scenarios with low-quality or missing data, reconstruction loss can ensure that the generator's output has sufficient pixel-level consistency with the real gamma-ray image, reducing overfitting or blurring issues.

[0132] Synthetic generator loss

[0133] To simultaneously consider adversarial loss, perceptual loss, and reconstruction loss, they can be weighted and summed to obtain the final generator objective function:

[0134]

[0135] Mini-batch SGD and training process

[0136] In iterations of optimizers such as Mini-batch Stochastic Gradient Descent (SGD) or Adam, the discriminator and generator are updated alternately:

[0137] Discriminator Update:

[0138]

[0139] Generator update:

[0140]

[0141] Where θ D ,θ G η represents the parameters of the discriminator and generator, respectively, and η is the learning rate. Through the interaction between the two, the realism and resolution of the generated image are continuously improved until the loss converges.

[0142] Synergy with Transformer architecture

[0143] The high-resolution gamma-ray feature map generated by GAN will be fed back into the Transformer spatiotemporal modeling unit in Example 1 for further analysis and fusion.

[0144] When performing multimodal feature fusion or attention allocation, the "data augmentation" effect of GAN can supplement the originally scarce or noisy data, thereby allowing the Transformer's self-attention mechanism to capture richer stratigraphic details.

[0145] GAN data augmentation

[0146] The trained generator G will produce a completed or enhanced high-resolution gamma-ray feature map, denoted as G. .

[0147] Will Treated as a new data sample, compared with the original multimodal data r t ,s t Together with the data, these are input into the Transformer spatiotemporal modeling unit, thereby capturing richer stratigraphic details in self-attention, location encoding, and autoregressive prediction.

[0148] Its role in multimodal feature fusion It can be stitched together with other modal features (feature-level fusion) or output predictions independently (decision-level fusion). Under the dynamic weight allocation mechanism, if the GAN output quality is high, its weight in the final stratigraphic structure imaging or fault prediction task will be appropriately increased; otherwise, its influence will be automatically reduced to ensure the robustness of the overall system.

[0149] Integrated pipeline

[0150] Step 1: Train the GAN to obtain the ability to generate high-quality gamma-ray images;

[0151] Step 2: The GAN output and the actual collected data are fed into the Transformer module to perform multi-head attention, position encoding and spatiotemporal modeling;

[0152] Step 3: Under the multimodal fusion and self-attention mechanism, the system performs multi-task analysis on the downhole formation (such as lithological classification, fault prediction, anomaly alarm, etc.) to improve accuracy and interpretability.

[0153] Through the above loss function and training process, this invention can successfully generate reliable high-resolution feature maps in environments with missing or noisy gamma-ray data, and closely integrate them with the Transformer spatiotemporal modeling unit, making the intelligent analysis and prediction of complex geological structures more accurate, real-time and robust.

[0154] High-resolution imaging: Significantly improves the detail and visualization of geological formations, enabling better identification of minute structures such as fractures and faults.

[0155] Addressing scarce data: Reduce reliance on high-density sampling to maintain system performance even under conditions of tight drilling time, sensor failure, and complex downhole terrain.

[0156] Sharing with other modalities: The fake data output by the generator can be fused with resistivity and acoustic data at the same time, so that multimodal learning can obtain more dimensions and clearer features when modeling, further improving the ability to identify and predict complex strata.

[0157] Self-supervised learning module

[0158] When applied to the massive amounts of data collected downhole, the effective labels are often very limited; in addition, the labeling process is time-consuming and requires the deep involvement of professional geologists.

[0159] There is a large amount of low-quality data (very noisy, incomplete, etc.), and using it directly will weaken the model's performance.

[0160] In unsupervised or weakly supervised scenarios, the model can use its own structure to discover "learnable" clues in the data; by completing self-prediction tasks or generating pseudo-labels, it can gradually approach the learning effect of real annotations.

[0161] Compared with traditional supervised learning, self-supervised learning requires little or no manually labeled data, greatly improving the feasibility of large-scale data processing in downhole scenarios.

[0162] Based on a small amount of existing real labels or historical statistical information, preliminary "label inference" is first performed on a large batch of unlabeled gamma-ray / multimodal data. For example, the system can use the spatiotemporal features learned in Transformer to automatically assign the same or similar labels to similar signal segments.

[0163] The generated pseudo-labels are treated as "soft labels". The model prediction results are compared with the pseudo-labels to continuously update the model parameters by minimizing a certain loss function (such as cross-entropy or mean squared error).

[0164] Through iterative iteration, the pseudo-labels continuously "improve" with the feedback from the model, while the model gradually "corrects" itself under the guidance of the pseudo-labels, forming a positive cycle.

[0165] In some cases, a contrastive learning approach can be used to perturb the same batch of data (such as adding noise, randomly rotating the orientation, etc.) to train the model to distinguish between "same source" and "different source" data, thereby learning a more robust feature representation.

[0166] This process complements GAN in data augmentation: GAN provides high-quality synthesized or completed data, and the self-supervised module then generates more transformed samples from this data, making the model more tolerant to noise or interference.

[0167] The multimodal fusion unit mentioned in Example 1 can work in conjunction with self-supervised learning: when a certain modality lacks human labels, the corresponding pseudo-labels can be inferred by using the existing "weak supervision information" of other modalities.

[0168] The dynamic weight allocation mechanism (described in Example 1) can determine the weight allocation in the fusion stage based on the credibility of the pseudo-labels of each modality; if the credibility of the pseudo-labels of a certain modality is enhanced, the system will automatically increase its contribution to the final decision.

[0169] Self-supervised strategies can be pre-trained or trained in parallel on a large amount of unlabeled data, significantly reducing the need for long-term labeling by professional geological experts. Even in the face of harsh downhole environments, severe noise, or incomplete data collection by some sensors, the system can continuously optimize through pseudo-labels and self-learning mechanisms, maintaining a high tolerance and recognition accuracy for anomalies. In actual exploration, as new data sources continuously enter, the system can continuously "learn-correct-relearn" through self-supervised learning strategies to achieve long-term evolutionary intelligent geological imaging.

[0170] This embodiment completely adopts the hardware and core software architecture of Embodiment 1: the basic functions such as the ring-shaped multi-directional gamma ray detector, multi-modal sensor integration, signal processing and correction process, and Transformer-based intelligent modeling remain unchanged.

[0171] The "Intelligent Modeling and Multimodal Fusion Module" has added two new sub-modules: adversarial learning and self-supervised learning, as an upgrade to the existing algorithm framework.

[0172] Example 1 provides a robust spatiotemporal modeling mechanism, offering powerful "comparison data" and "structured embeddings" for GANs and self-supervised learning.

[0173] The high-resolution gamma-ray feature map generated by GAN can not only improve the support for multiple tasks such as fault prediction and lithology classification, but also be deeply coupled with the dynamic weight allocation mechanism in Example 1.

[0174] The self-supervised learning pseudo-label generation and contrastive learning mechanism can produce high learning efficiency on originally sparse or noisy datasets, realizing "adaptive evolution" of intelligent stratigraphic imaging.

[0175] This embodiment is particularly suitable for scenarios where the downhole environment is dynamic and changeable, and where there are gaps or unstable quality in the sampling data. It can continuously improve data availability and model inference performance at each stage.

[0176] With the increase in the types of downhole sensors, GANs and self-supervised modules can also be extended to more modal data, thereby further improving the overall imaging accuracy and geological analysis level.

[0177] Adversarial learning modules (GANs) enable the system to acquire "high-fidelity, high-resolution" gamma-ray feature maps even when data is scarce or resolution is insufficient, significantly improving the details of stratigraphic imaging and the ability to detect anomalous structures.

[0178] The self-supervised learning module utilizes pseudo-labels and contrastive learning mechanisms, which greatly alleviates the pressure of manual annotation and enables the system to have strong self-iteration capabilities and robustness to environmental noise.

[0179] Compared to Example 1, Example 2, through the introduction of these two major modules, provides a leapfrog "intelligent upgrade" for the downhole geological structure imaging system, truly realizing high-precision stratum identification and dynamic prediction under complex geological environments and conditions with a large amount of unlabeled data.

[0180] In summary, while inheriting the complete functionality of Example 1, Example 2 focuses on the deep integration and application of GAN and self-supervised learning, providing an effective solution to the challenges of scarce data and noise in downhole geological imaging, and further promoting the application value of this system in actual oil and gas field drilling, geological exploration and other fields.

[0181] Example 3:

[0182] like Figure 1 As shown, based on the first two embodiments, the system already possesses gamma-ray detection, multimodal fusion (including advanced methods such as GAN and self-supervised learning), and spatiotemporal Transformer modeling capabilities. However, for more complex downhole environments, we need to further enhance the following three aspects:

[0183] Multi-head cross-attention mechanism

[0184] By conducting in-depth correlation analysis of gamma-ray signals with other modes (resistivity, sound waves, temperature, pressure, etc.), we can identify implicit features that exist between different modes but have not yet been explicitly captured.

[0185] More refined spatiotemporal location coding

[0186] Further refine the temporal and spatial location coding to ensure that, under the spatiotemporal Transformer architecture, the details of the strata's temporal changes and distribution characteristics in different orientations can be fully captured.

[0187] Abnormal formation alarm module

[0188] In addition to multi-task support, a new real-time early warning function has been added to perform online analysis of gamma-ray and multi-modal data collected downhole. Once formation anomalies such as faults and fractures are detected, an alarm will be issued to assist downhole operations and safety decisions.

[0189] Overall, Embodiment 3 will "embed and upgrade" the technical modules of the two aforementioned embodiments, and enable formation anomalies to be detected and alerted more accurately and at a faster speed through multi-head cross-attention and location coding mechanisms.

[0190] Multi-head cross-attention is similar to the "self-attention" mentioned earlier, but it specifically emphasizes the interaction and matching between different modalities or different data sources.

[0191] Traditional multi-head attention often learns structural associations within the same modality or sequence, while multi-head "cross" attention can create richer couplings between gamma-ray features and resistivity features, acoustic features, and even data completed by GAN.

[0192] Specific procedures

[0193] For input feature alignment, basic preprocessing (normalization, embedding, position encoding, etc.) is performed on gamma-ray data and other modal data to ensure that they can interact with each other in the same depth dimension.

[0194] Cross-attention calculation: In a certain Attention layer of the Transformer, the input comes from gamma rays and other modalities respectively, forming three sets of vectors: Query, Key, and Value.

[0195] Multiple attention heads will "observe" the interaction between these two major feature sources from different angles, and output attention distributions in different dimensions;

[0196] Ultimately, the outputs of all attention heads are concatenated, weighted, or linearly transformed in the channel or feature dimensions to generate a high-dimensional fused feature.

[0197] Latent feature mining benefits from multimodal cross attention, which allows weakly relevant information that was previously difficult to detect (for example, when a gamma-ray signal suddenly increases in a certain direction, and the acoustic data at the corresponding location also shows anomalies in certain frequency bands) to be better captured and given higher feature weights.

[0198] This helps identify subtle formation changes, such as lithological transitions, potential fault boundaries, and may even provide real-time indications of complex factors such as drilling fluid disturbances.

[0199] Compared to self-attention, cross-attention is more targeted in terms of the source of query, key, and value, and can establish deep-level correlations between gamma-ray data and other modalities (resistivity, sound waves, etc.).

[0200] Cross-Attention Calculation

[0201] Suppose we have two sets of inputs:

[0202]

[0203] In a cross-attention layer, it is usually set that:

[0204]

[0205] Single-head cross attention

[0206] The calculation of single-head cross-attention is similar to that of self-attention, except that the input sources are distinguished into different modalities:

[0207]

[0208] Multi-Head Cross-Attention

[0209] If h parallel cross-attention heads are used, it can be expanded as follows:

[0210]

[0211] After concatenating the outputs of all attention heads along either the channel or feature dimensions, and then passing the W... O Linear transformation ultimately generates high-dimensional fused features.

[0212] Hidden Feature Mining

[0213] Through the aforementioned cross-attention mechanism, the system can dynamically calculate the correlation weights between "gamma rays" and "other modes," and extract those weak correlations that were originally difficult to detect. For example:

[0214] Weak correlation enhancement

[0215] When a gamma-ray signal x in a certain direction i When (Query) suddenly increases at a certain moment, and the sound wave or resistivity in the corresponding Y (Key & Value) also shows subtle changes in a specific frequency band, the results are compared by QK. ⊤ The dot product result, the attention module will be used for these (x) i ,y j Assigning higher weights to these indicates a potential coupling relationship in terms of stratigraphic features.

[0216] Focus from multiple angles

[0217] The multi-head mechanism ensures that each attention head can "observe" the data from different angles:

[0218] For example, head 1 focuses on time-series patterns (short-term fluctuations), head 2 focuses on spatial proximity or filtering of noise signals, and head 3 focuses on high-frequency characteristics, etc.

[0219] During the final fusion, the outputs of these different attention heads are spliced ​​together and linearly transformed to form a more complete high-dimensional cross-feature.

[0220] Formation anomaly identification

[0221] By capturing the multi-head cross-attention distribution of gamma rays and other modal signals, anomalies can be detected promptly in terms of location or time.

[0222] If the attention weight remains high in certain areas, and it is not just a single mode that is abnormal, it suggests that there may be more significant stratigraphic changes such as faults, fractures, or wellbore collapses in this area.

[0223] In practical applications, these implicit features not only help to accurately characterize lithological transitions and potential fault boundaries, but also enable continuous tracking and subsequent model analysis to predict and warn of downhole drilling fluid disturbances and sudden geological conditions, helping engineers to accurately adjust drilling plans or take timely protective measures.

[0224] Achieving a deep coupling of multimodal features such as gamma rays, resistivity, and acoustic waves allows for a more nuanced expression of the system between "feature-level fusion" and "decision-level fusion".

[0225] Enhanced sensitivity to anomalous formations: When faced with anomalies such as faults, fractures, or fluid activity, only one modal signal may show abnormalities. Through multi-head cross-attention, not only can this abnormality be captured, but it can also be correlated and analyzed to see if other modalities also show corresponding phenomena, thereby improving the confidence of anomaly detection.

[0226] Adaptation to GAN-generated data: The high-resolution gamma-ray feature map generated by GAN mentioned in Example 2 can also be compared and integrated with actual multimodal data through a cross-attention mechanism, thereby avoiding the negative impact of "fake data" on the overall judgment.

[0227] Spatiotemporal modeling based on Transformer:

[0228] Temporal Positional Encoding: In Examples 1 and 2, we have already discussed using timestamp information to characterize the dynamic changes of gamma-ray signals.

[0229] In Example 3, the time position encoding not only includes encoding different time steps using traditional sine / cosine functions, but also introduces the concept of dynamic time intervals:

[0230] For certain non-uniform or "skipped" sampling (such as downhole equipment pausing, accelerating, or changing position), the system can adaptively adjust the encoding method to make the model more accurately reflect the real time distribution of data acquisition.

[0231] This more flexible time coding strategy can further reduce the model's dependence on a "fixed sampling frequency" and maintain good temporal consistency modeling even when there are temporary fluctuations in the downhole environment.

[0232] Spatial Positional Encoding is similar to temporal encoding, but it also incorporates some extensions:

[0233] Considering that the azimuth layout of downhole detectors is not always uniformly distributed at 360° (there may be a certain degree of tilt or offset in some cases), this embodiment can additionally record the actual azimuth coordinates of the detector, well diameter, well wall roughness, and other information, and convert them into learnable embedding vectors.

[0234] Meanwhile, in multi-head cross-attention, these spatial location codes will also be used to guide attention matching between different modalities, enabling high-dimensional fusion features to accurately reflect the real geographical location and depth distribution downhole.

[0235] Construction of spatial location coordinates

[0236] Suppose that for the j-th detector or acquisition location, we have recorded the following spatial information:

[0237]

[0238] This information can be mapped using some learnable or fixed function to obtain a scalar or low-dimensional vector pos.j ,For example:

[0239]

[0240] Where f(.) is a function that can be normalized or nonlinearly transformed for wellbore roughness, etc., or it can be directly concatenated.

[0241] {α j ,δ j ,ρ j Encoding is performed in a high-dimensional space.

[0242] Spatial location coding formula

[0243] Similar to time-location encoding, for the location index pos j Embedding is performed using sine and cosine functions. Let the position encoding vector have dimension d. model The index i ∈ {0, 1, 2, ..., d} model -1}. Then the spatial location encoding of the j-th acquisition point can be defined as:

[0244]

[0245] If we consider α j ,δ j ,ρ j Each has a different effect on positional encoding, and they can be substituted or weighted separately during embedding, for example:

[0246]

[0247] Then, map the sine and cosine functions to d as described above. model 3D space.

[0248] Applications of multi-head cross attention

[0249] In the multimodal cross-attention layer, the spatial location encoding vector PE is... space (pos j After being concatenated or added to the original feature vectors (such as gamma intensity, resistivity, sound waves, etc.) collected by the detector, it can be used as input for Query, Key, and Value, thereby achieving more accurate orientation matching.

[0250] Precise orientation matching

[0251] Different detectors have significantly different spatial location codes due to variations in tilt angle, eccentricity, or formation roughness.

[0252] Cross-attention takes these positional encodings into account when calculating similarity, ensuring that the mapping between multimodalities is more physically meaningful.

[0253] Guiding implicit associations

[0254] When the azimuth angles are similar or the wellbore roughness is similar, the corresponding pos j It might be closer;

[0255] Attention distribution will classify such locations as “potentially relevant” or “neighborhood” areas, making subtle stratigraphic changes (such as the extension of fractures) easier to capture.

[0256] Through the above spatial location coding, this embodiment can flexibly adapt to detector arrangement methods that are not strictly uniformly distributed, while fully integrating multi-source geological environment information such as well diameter, roughness, and dip angle, effectively improving the performance of multi-head cross-attention in downhole formation structure identification and dynamic analysis.

[0257] Richer and more flexible spatiotemporal coding allows Transformer to accurately capture subtle changes in the distribution of gamma-ray signals over time and in terms of orientation.

[0258] In conjunction with GANs and self-supervised learning: When the completed data generated by GANs or obtained through self-supervised learning is integrated into the spatiotemporal modeling unit, these location codes can better constrain the consistency between the completed data and the real stratigraphic environment, thereby preventing overfitting problems such as "false faults" or "displaced lithology".

[0259] Abnormal formation alarm function in the multi-task support module

[0260] The aforementioned embodiments already include multi-tasking capabilities, such as lithology classification, geological structure analysis, and fault prediction. However, in actual downhole operations, timely detection and early warning of faults, fractures, or other abnormal formations are often crucial and can directly affect operational safety and construction efficiency.

[0261] Therefore, the abnormal formation alarm module aims to achieve "real-time detection and rapid alarm" by integrating the modeling results of gamma rays and multimodal signals.

[0262] Real-time detection process

[0263] Signal monitoring: The system continuously monitors the multimodal fusion features and the predicted values ​​of the spatiotemporal Transformer at each time step or at each well depth interval.

[0264] Triggering mechanism: When the gamma ray intensity or other modal signal in a certain direction exceeds a threshold (e.g., a significant abrupt change, spike, or continuous fluctuation), the cross-attention unit will simultaneously evaluate:

[0265] Is it merely single-modal random noise? Do other modal data also show corresponding anomalies? Have similar situations in historical data led to significant changes in the stratigraphic structure? Alarm generation: If the anomaly is supported by high confidence through multimodal cross-validation or historical model review analysis, the system will send an alarm to the ground platform in real time, including: anomaly category judgment (e.g., suspected fault, fracture, well collapse, etc.); depth and azimuth of the anomalous strata; and properties of the surrounding strata (e.g., possible lithology and porosity characteristics).

[0266] Decision support: Once an early warning is issued, surface geological engineers can respond based on the severity level of the warning and relevant information, such as adjusting the drilling path, strengthening wellbore support, and investigating potential risks.

[0267] At each time step t or at each well depth δ interval, the system will perform multimodal fusion of features F. t and the predicted value output by the Transformer Continuous monitoring is performed. We can define a multimodal monitoring vector: M t =[F t , ,…],in:

[0268] F t Including features derived from the fusion of gamma rays, resistivity, and sound waves;

[0269] The predicted value of the spatiotemporal Transformer (or multi-task prediction, such as fault risk scoring, etc.);

[0270] Other contextual information (such as wellbore inclination angle, real-time pressure, etc.) can be added as needed.

[0271] Triggering Mechanism

[0272] Anomaly detection threshold

[0273] To determine whether formation anomalies have occurred, a threshold τ can be set for each modal characteristic or predicted value. i Let M t =[mt,1,m t ,2 ,…,m t ,K ], where m t Let i represent the value of the i-th mode (or index) at time t. Then, the condition for "preliminary anomaly detection" is defined as follows:

[0274] m t ,i>τ i or |m t ,i − |>η i

[0275] where represents the mean (or median) of this feature in historical data, and η i represents the threshold set based on the standard deviation or other robust metrics. If this condition is satisfied for the duration of Δt time steps or well intervals for this feature, it is regarded as "suspected anomaly".

[0276] Multi-modal cross-validation

[0277] Once a suspected anomaly occurs, the cross-attention unit will conduct a synchronous evaluation to determine whether anomalies occur simultaneously in the same orientation or adjacent depths among different modalities. We can define the cross-attention distribution A t (from the multi-head cross-attention calculation):

[0278]

[0279] Q i corresponds to the gamma-ray modality, and K j , V j correspond to other modalities (resistivity, acoustic wave, etc.);

[0280] If A t (i, j) is greater than a certain threshold λ, it indicates a strong correlation between modality i and modality j at this position / moment. When the suspected anomaly mainly comes from a single modality and A t (i, j) is low, it may be just random noise; if significant anomalies occur simultaneously in multiple modalities and the attention score is high, it may be a real formation change.

[0281] Historical comparison

[0282] Before making a final judgment, the current moment M t will also be compared with the similar patterns of M t′ in historical data (t′ < t). For example:

[0283]

[0284] If it is found that a similar feature pattern has occurred in history and has been confirmed as an actual situation such as "severe fault" or "wellbore collapse", the credibility of this anomaly detection is increased;

[0285] If the similarity is very low or the corresponding situation has been determined as "equipment noise", the alarm level can be lowered.

[0286] Alarm Generation

[0287] If the above multimodal verification and historical comparison results show a high anomaly confidence level, an alarm can be sent in real time. Let the anomaly confidence level evaluation function be:

[0288]

[0289]

[0290] Decision Support

[0291] Once an alarm is issued, the surface geological engineer will make a corresponding decision based on the following criteria:

[0292] Adjust the drilling path: if the risk of fault or fracture is high;

[0293] Strengthen wellbore support: such as when Caving or wellbore roughness is excessive;

[0294] Investigate potential risks: If the alarm level is not high, but historical data shows that such anomalies may develop into serious failures;

[0295] Ignore or mark: If on-site inspection and background data confirm that the alarm is a "false alarm", it can be marked in the human-computer interaction system.

[0296] This process can be summarized as follows:

[0297]

[0298] The method employs "real-time perception" of multimodal fusion features and Transformer prediction output, combined with cross-attention and historical comparison for "trigger judgment," ultimately generating alarms and assisting decision-making upon anomaly confirmation. This approach can promptly detect significant risks such as formation faults, fractures, or wellbore collapse, significantly enhancing the safety and efficiency of downhole logging and geological exploration.

[0299] The model enhanced by GAN and self-supervised learning, as described in Example 2, enables the abnormal stratum alarm module to maintain a high detection accuracy even in scenarios with missing or noisy data.

[0300] The multi-head cross-attention mechanism allows the alarm system to quickly screen for "false anomalies" (noise in one modality while other modalities are normal), reducing false alarms; it also allows real anomalies to be corroborated by more modalities, thereby improving the confidence of alarms.

[0301] By leveraging multi-head cross-attention, the system can more fully combine multimodal features, capturing more key information related to formation fractures, lithological transitions, and pressure anomalies in the hard-to-observe "hidden layers." Temporal and spatial location encoding makes the spatiotemporal Transformer more "realistic" when processing data, especially in cases of non-uniform sampling or eccentric detector deployment, it can still maintain a high level of spatiotemporal modeling accuracy.

[0302] This addresses the real-time and security requirements of multi-task support in the previous embodiments, provides a rapid alarm channel for actual downhole operations, significantly reduces operational risks, and provides a clearer and more actionable basis for subsequent decision-making.

[0303] The GAN and self-supervised learning results introduced in Example 2 are not separated at this stage. Instead, through the cross-attention mechanism and further refinement of location coding, the quality and accuracy of formation anomaly alarms are improved simultaneously, making the system more flexible and practical in real oil and gas fields or geological exploration scenarios.

[0304] Example 3, building upon the foundational functionalities, GAN data augmentation, and self-supervised learning achieved in the first two examples, further highlights the significant value of multi-head cross-attention, refined spatiotemporal location coding, and anomalous formation alerts for downhole geological structure imaging systems. Through this series of improvements:

[0305] The system can establish deeper implicit correlations between multimodal signals and generate more accurate and robust high-dimensional fusion features;

[0306] By leveraging the spatiotemporal Transformer to perform high-level modeling of temporal and spatial distribution, we can promptly detect and provide early warnings of anomalous locations in complex strata.

[0307] The significant improvements in real-time performance, automation, and collaboration provide geologists and downhole work teams with powerful "predictive" capabilities, greatly reducing construction risks and uncertainties.

[0308] Thus, Example 3, together with the previous two examples, forms a complete multimodal, multi-stage, and highly intelligent geological structure imaging system, which has extremely important technical support and practical value for oil and gas field drilling, geological disaster early warning, and underground engineering construction.

[0309] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A downhole geological structure imaging system based on azimuth gamma rays, characterized in that: Includes the following modules: The gamma-ray detection module is used to collect azimuth gamma-ray intensity and energy spectrum data of the formation in the well. The multimodal data acquisition module is used to acquire downhole multimodal data other than gamma rays; The signal processing and correction module is used for preprocessing and dynamic correction of gamma-ray data and multimodal data; The intelligent modeling and multimodal fusion module, based on the Transformer architecture and multimodal data fusion algorithm, is used to generate high-resolution imaging maps of downhole geological structures and to perform multiple tasks, including: Data input: The gamma-ray intensity data of each azimuth detector constitutes the input sequence; the input sequence is divided by timestamps to form a spatiotemporal dataset; Intelligent Modeling: Processing Gamma-Ray Data Using the Transformer Architecture Autoregressive modeling: predicting signal changes at the next location or time point based on the current detector signal; Attention mechanism: By using multi-head attention to capture the correlation between gamma-ray intensity in terms of location and time, the heterogeneity of the formation can be analyzed; The azimuth and depth positions of each gamma-ray detection point are represented by a location code. The multimodal feature fusion unit employs feature-level fusion and decision-level fusion, utilizing deep features extracted from modal data such as gamma rays, resistivity, and acoustic waves to generate unified multimodal fusion features; The dynamic weight allocation unit dynamically adjusts the weights of different modal features based on signal quality and task requirements using a modal sensing mechanism. The spatiotemporal modeling unit, based on the spatiotemporal Transformer architecture, captures the spatiotemporal correlation of formation signals through temporal and spatial location encoding, and is used for the identification and prediction of dynamic changes in formations. Multitasking support module: Lithological classification: Predicting stratigraphic lithology using gamma-ray energy dispersive spectroscopy data; Geological structure analysis: Identify the dip angle of strata, fractures, and fault locations; Fault prediction: By combining historical data and current azimuth signals, the location of faults or anomalous strata is inferred by predicting future signal trends; Imaging output: By comparing the predicted results with the actual data, two-dimensional or three-dimensional geological structure maps are dynamically generated. It supports using AI algorithms to complete missing data and improve the integrity of imaging; The data transmission and display module is used to transmit the collected and analyzed data to the ground in real time and display it visually.

2. The downhole geological structure imaging system based on azimuth gamma rays according to claim 1, characterized in that: The gamma-ray detection module includes: a ring-shaped multi-directional gamma-ray detector array, distributed in a 360° orientation, with each detector responsible for acquiring gamma-ray signals in a specific direction; and a dynamic signal adjustment unit that adjusts the detector sensitivity in real time to adapt to different downhole environments.

3. The downhole geological structure imaging system based on azimuth gamma rays according to claim 1, characterized in that: The multimodal data acquisition module includes: a resistivity sensor for acquiring formation conductivity information; an acoustic logging sensor for acquiring formation density and porosity; temperature and pressure sensors for monitoring downhole environmental parameters; and a data synchronization interface for synchronizing and integrating gamma-ray data with multimodal data.

4. The downhole geological structure imaging system based on azimuth gamma rays according to claim 1, characterized in that: The signal processing and correction module includes: a preprocessing unit, which normalizes and denoises the gamma-ray data to improve the signal-to-noise ratio; and a dynamic correction mechanism, which automatically corrects signal errors caused by environmental changes such as drilling fluid absorption and wellbore eccentricity by analyzing historical data and real-time signals.

5. The downhole geological structure imaging system based on azimuth gamma rays according to claim 1, characterized in that: The data transmission and display module includes: a data transmission unit that uses high-speed wireless communication or fiber optic technology to support low-latency transmission of large data volumes; and a visualization interface for displaying two-dimensional or three-dimensional geological structure maps, lithological distribution maps, fault location maps, and real-time signal trend maps, and supports historical data overlay analysis.

6. The downhole geological structure imaging system based on azimuth gamma rays according to claim 1, characterized in that: The intelligent modeling and multimodal fusion module also includes an adversarial learning module, which generates high-resolution gamma-ray feature maps through generative adversarial networks, thereby enhancing the utilization of scarce data.

7. The downhole geological structure imaging system based on azimuth gamma rays according to claim 1, characterized in that: The intelligent modeling and multimodal fusion module includes a self-supervised learning module, which reduces the reliance on manually labeled data and improves the model's robustness to low-quality data through pseudo-label generation and self-supervised objective optimization.

8. The downhole geological structure imaging system based on azimuth gamma rays according to claim 1, characterized in that: The intelligent modeling and multimodal fusion module captures the implicit correlation between gamma rays and other modal data through a multi-head cross-attention mechanism, generating high-dimensional fusion features for the identification of characteristics in complex strata.

9. The downhole geological structure imaging system based on azimuth gamma rays according to claim 1, characterized in that: The spatiotemporal modeling unit is based on the Transformer architecture and uses temporal position encoding to capture the dynamic characteristics of gamma-ray signals changing over time, and spatial position encoding to characterize the spatial distribution characteristics of gamma-ray signals in different directions.

10. A downhole geological structure imaging system based on azimuth gamma rays according to claim 1, characterized in that: The multi-task support module further includes an abnormal formation alarm module, which detects abnormal formation signals such as faults or fractures in real time and generates alarms to assist downhole decision-making.

Images