Data cross-modal association method and device in oil and gas field database, equipment and medium

Abstract

The invention relates to the technical field of oil and gas field databases, and discloses a data cross-modal association method and device in an oil and gas field database, equipment and a medium, and the method comprises the steps: receiving an original seismic waveform time sequence inputted in advance, and generating a time-frequency spectrogram through Fourier transform; receiving a pre-input rock core scanning image set, extracting local binary pattern texture features from the rock core scanning image set, and obtaining a spatially continuous three-dimensional rock core digital model based on the local binary pattern texture features; injecting the time-frequency spectrogram and the three-dimensional rock core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector; constructing a comparative learning framework of the spectrogram and the logging text, and generating a time-space aligned semantic feature matrix based on the comparative learning framework; and based on the cross-modal unified semantic vector and the semantic feature matrix, obtaining a dynamic association map of the logging and the seismic waveform.

Description

Technical Field

[0001] This application relates to the technical field of oil and gas field databases, and in particular to a method, device, equipment and medium for cross-modal association of data in an oil and gas field database. Background Art

[0002] In the process of oil and gas field exploration and development, seismic waveform data, core image data and well logging data are three types of key multi-modal data sources. The seismic waveform time series can generate a time-frequency spectrum diagram through Fourier transform, reflecting the time-frequency characteristics of the formation structure; the core scan image can construct a three-dimensional digital core model through texture feature extraction methods such as local binary pattern (LBP) to characterize the microscopic structure of the reservoir rock; while the well logging data records the formation physical property parameters at different depths in text form. How to achieve effective association between these multi-modal data is the core technical challenge to improve the accuracy of oil and gas reservoir interpretation.

[0003] In the prior art, most of them adopt independent feature extraction strategies (such as Fourier transform of seismic data, LBP feature extraction of core images, and statistical analysis methods of well logging data), resulting in a lack of unity in the semantic representation space of different modal data and making it difficult to establish an accurate cross-modal mapping relationship. Summary of the Invention

[0004] One or more embodiments of this specification provide a method, device, equipment and medium for cross-modal association of data in an oil and gas field database to solve the technical problems raised in the background art.

[0005] One or more embodiments of this specification adopt the following technical solutions: A method for cross-modal association of data in an oil and gas field database provided by one or more embodiments of this specification, the method includes: Receiving a pre-input original seismic waveform time series and generating a time-frequency spectrum diagram through Fourier transform; Receiving a pre-input set of core scan images, extracting local binary pattern texture features from the set of core scan images, and obtaining a spatially continuous three-dimensional core digital model based on the local binary pattern texture features; Injecting the time-frequency spectrum diagram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector; Constructing a contrastive learning framework for the spectrum diagram and well logging text, and generating a spatio-temporally aligned semantic feature matrix based on the contrastive learning framework; Based on the cross-modal unified semantic vector and the semantic feature matrix, obtaining a dynamic association map of well logging and seismic waveforms.

[0006] It should be noted that through the above content, the embodiments of this specification have the following beneficial effects: Cross-modal semantic unity enhancement: By injecting spectrograms during earthquakes and 3D core digital models into a hierarchical semantic framework, multi-modal semantic space mapping of seismic waveforms, core images, and logging texts is achieved. This method breaks through the limitations of traditional independent feature extraction strategies, establishes deep semantic associations of cross-modal data in a unified semantic vector space, and effectively bridges the semantic gap between the micro time-frequency characteristics of seismic waveforms and the macrostructure of core textures.

[0007] Space-time alignment accuracy optimization: A contrastive learning framework is used to jointly model spectrograms and logging texts. By dynamically perceiving the non-linear relationship between the time dimension of seismic waveforms and the depth space dimension of logging, space-time consistency alignment of cross-modal features is achieved. This mechanism significantly improves the feature misalignment problem caused by the separation of space-time dimensions in traditional methods, providing a high-precision semantic feature matrix for subsequent dynamic associations.

[0008] Innovation of dynamic association mechanism: By fusing unified semantic vectors and space-time alignment feature matrices, a dynamically updatable association map is constructed. This map can adapt to the real-time update of logging data during the drilling process, overcoming the defect that traditional static matching mechanisms cannot reflect the dynamic changes of oil and gas reservoirs, and providing real-time data support for the adjustment of development plans.

[0009] Breakthrough in 3D modeling efficiency: In the stage of constructing the core digital model, through an optimization algorithm for the spatial continuity of local binary pattern texture features, while ensuring the accuracy of microstructural characterization, the processing complexity of large-scale core scan image sets is significantly reduced. This method effectively solves the efficiency bottleneck problem caused by feature redundancy in the traditional 3D modeling process.

[0010] Enhanced engineering application value: Through the deep association and dynamic visualization expression of multi-modal data, the abnormal areas of seismic waveforms, the pore structure characteristics of cores, and the physical properties of logging are cross-verified, significantly improving the accuracy of reservoir fluid identification and the rationality of well trajectory design, providing core technical support for the intelligent exploration and development of oil and gas fields.

[0011] Furthermore, receiving the pre-input original seismic waveform time series and generating a spectrogram through Fourier transform includes: Receiving the original seismic waveform time series, using the short-time Fourier transform algorithm, setting a specified window, and generating a spectrogram matrix containing time, frequency, and amplitude; Obtaining a time-depth conversion model established based on logging acoustic travel time data: Inputting the spectrogram matrix into the time-depth conversion model to generate a spectrogram of depth and time.

[0012] It should be noted that through the above content, the embodiments of this specification have the following beneficial effects: Improvement of time-frequency feature resolution: By combining the short-time Fourier transform algorithm with a configurable window function, while maintaining the frequency resolution, it accurately captures the local time-varying characteristics of seismic waveforms, and the generated three-dimensional time-frequency spectrogram matrix (time-frequency-amplitude) realizes the fine-grained characterization of microstratigraphic structure features, providing a more spatio-temporally resolved basic feature expression for subsequent cross-modal associations.

[0013] Optimization of depth domain alignment accuracy: Based on the acoustic travel time data of well logging, a time-depth conversion model is constructed to map the traditional time-domain seismic spectrum to the depth coordinate system, establishing a unified depth reference for seismic waveforms and well logging data. This conversion mechanism effectively solves the problem of cross-modal data space misalignment caused by the non-linearity of the time-depth relationship, and significantly improves the matching accuracy between seismic features and the true depth position of the formation.

[0014] Enhancement of multi-dimensional feature fusion: The generated depth-time dual-domain time-frequency spectrogram synchronously retains the time evolution law of the original waveform and the depth distribution characteristics after conversion, forming a composite feature expression with both time sensitivity and spatial interpretability. This multi-dimensional feature fusion mechanism provides a unified spatial anchor point for the cross-scale association of seismic data with core and well logging data.

[0015] Intelligentization of data processing flow: Through the automatic parameter mapping of the time-depth conversion model, it replaces the experience-dependent method of manually calibrating the seismic time-depth relationship in the past. While reducing subjective errors, it realizes the batch depth calibration of large-scale seismic data, and significantly improves the engineering implementation efficiency of the data preprocessing link.

[0016] Furthermore, receiving the pre-input core scan images, extracting local binary pattern texture features from the core scan images, and obtaining a spatially continuous three-dimensional core digital model based on the local binary pattern texture features includes: Receiving a set of core scan images; Calculating the LBP value for each pixel point of each core scan image to generate an LBP encoded map; Statistical LBP histogram features of the LBP encoded map to characterize the local texture distribution of the core; Based on the LBP histogram similarity of adjacent slices, establishing a spatial correspondence relationship between slices; Stacking two-dimensional slices into a three-dimensional voxel grid based on the spatial correspondence relationship to obtain the three-dimensional core digital model.

[0017] It should be noted that through the above content, the embodiments of this specification have the following beneficial effects: Enhanced Microstructure Characterization Ability: By extracting the pixel-by-pixel texture features of the image encoded by Local Binary Pattern (LBP), the spatial heterogeneity of micro-geological features such as pore structure and mineral particle distribution in core scanning images is completely retained. The LBP histogram statistical mechanism effectively captures the texture pattern differences at different lithological interfaces, significantly improving the characterization accuracy of the digital model for the complex microstructure of reservoir rocks.

[0018] Guaranteed Three-dimensional Spatial Continuity: Based on the spatial correspondence established by the similarity of LBP histograms of adjacent slices, the limitation of the traditional two-dimensional image stacking method that relies on manual alignment is broken through. This method automatically constructs the topological connection between slices through the statistical correlation of texture features, ensuring the structural continuity of the three-dimensional voxel grid in the vertical and horizontal directions, and truly restoring the sedimentary sequence characteristics of underground rock layers.

[0019] Optimized Multi-scale Feature Fusion: The generation process of the three-dimensional voxel grid synchronously integrates the micro-texture information encoded by pixel-level LBP and the macro-statistical characteristics of slice-level histograms, forming a multi-scale fusion core digital representation system. This fusion mechanism enables the model to not only reflect the pore structure details at the millimeter level but also characterize the spatial distribution law of lithological units at the meter level.

[0020] Automated Upgrade of the Modeling Process: The full-process algorithmic processing from LBP feature extraction to three-dimensional reconstruction replaces the empirical operation mode that relies on manual annotation of slice correspondence in traditional methods. This automated process not only avoids the introduction of subjective errors but also greatly improves the processing efficiency of large-scale core scanning image sets, providing the ability to generate standardized data for the construction of the oilfield digital twin system.

[0021] Furthermore, the step of injecting the time-frequency spectrogram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector includes: Defining a template as the hierarchical semantic framework, the hierarchical semantic framework includes a cross-attention mechanism between the time-frequency spectrogram and the three-dimensional core digital model; Based on the cross-attention mechanism between the time-frequency spectrogram and the three-dimensional core digital model, obtaining interaction features; Inputting the interaction features into a fully connected layer to reduce the dimension to a cross-modal unified semantic vector.

[0022] It should be noted that through the above content, the embodiments of this specification have the following beneficial effects: Deep Optimization of Cross-modal Feature Interaction: By constructing a cross-attention mechanism between the time-frequency spectrum diagram and the 3D core model, the dynamic interactive perception of seismic waveform time-frequency features and core micro-texture features is realized. This mechanism breaks through the limitation of simply splicing multi-modal features in traditional methods, establishes implicit association rules between seismic wave propagation characteristics and rock physical properties in the feature space, and significantly improves the fineness of cross-modal semantic alignment.

[0023] Enhancement of Hierarchical Semantic Representation: The hierarchical semantic framework performs multi-scale fusion of the macroscopic stratigraphic interface features of seismic data and the microscopic pore structure features of the core model through a hierarchical feature abstraction mechanism. This hierarchical and progressive semantic integration method not only retains the unique feature expressions of each modality but also constructs a cross-scale geological semantic association network, enhancing the geological interpretability of the unified semantic vector.

[0024] Improvement of the Compactness of the Semantic Vector Space: A fully connected layer is used to perform directional dimensionality reduction on high-dimensional interaction features. While retaining the cross-modal key semantic information, it eliminates seismic time-frequency noise and core texture redundant features. The generated compact semantic vector not only reduces the subsequent computational complexity but also strengthens the class separability of different geological units through orthogonalization processing of the feature space.

[0025] Breakthrough in the Interpretability of Geological Laws: The construction process of the cross-attention weight matrix essentially reveals the physical association pattern between seismic wave attenuation characteristics and physical property parameters such as rock permeability. This interpretable feature interaction mechanism provides a new theoretical path for reservoir parameter inversion, breaking through the application limitations of traditional black-box models in oil and gas geological research.

[0026] Furthermore, generating a spatio-temporally aligned semantic feature matrix based on the contrast learning framework includes: Mutually correcting the spectrum diagram and well logging text based on the attention mechanism of the contrast learning framework to obtain fused features; Compressing the fused features into a spatio-temporally aligned semantic feature matrix.

[0027] It should be noted that the embodiments of this specification have the following beneficial effects through the above content: Breakthrough in Cross-modal Dynamic Correction Ability: Through the cross-attention mechanism of the contrast learning framework, two-way feature correction of the spectrum diagram and well logging text is realized. This mechanism enables the time-evolution features of seismic waveforms to correct the depth interpretation deviation of well logging text, while the spatial constraint of well logging parameters feeds back the calibration of the physical meaning of the spectrum diagram, forming a self-optimizing feature interaction system with spatio-temporal consistency.

[0028] Spatiotemporal Coupling Precision Leap: During the feature fusion process, through the dynamic allocation strategy of attention weights, the non-linear mapping relationship between the propagation time of seismic signals and the spatial depth of well logs is automatically captured. This technology breaks through the limitations of traditional linear interpolation methods and significantly improves the spatiotemporal matching precision of key geological elements such as the identification of the top and bottom interfaces of oil and gas reservoirs and the location of formation pinch-out points.

[0029] Semantic Feature Matrix Compactification Innovation: Using a feature compression algorithm to perform orthogonal dimensionality reduction on high-dimensional fusion features, while eliminating redundant information between modalities, the correlative expression of seismic waveform anomaly patterns and well log parameter mutation features is retained. The generated semantic matrix has both low-dimensionality and high information entropy characteristics, providing a lightweight and highly discriminative feature base for subsequent dynamic correlation map construction.

[0030] Enhanced Geological Interpretation Credibility: The attention correction trajectory formed during the contrast learning process can reversely analyze the physical interaction mechanism between seismic attributes and well log responses. This interpretable correction mechanism provides a transparent decision-making basis for reservoir hydrocarbon-bearing property identification and effectively overcomes the "black box" risk existing in traditional end-to-end models.

[0031] Furthermore, obtaining the dynamic correlation map of well logs and seismic waveforms based on the cross-modal unified semantic vector and the semantic feature matrix includes: Horizontally splicing the cross-modal unified semantic vector and the semantic feature matrix to generate a fusion feature matrix; Defining well log nodes and seismic nodes in the fusion feature matrix and determining the similarity between well log nodes and seismic nodes; Based on the similarity, obtaining the dynamic correlation map of well logs and seismic waveforms.

[0032] It should be noted that the embodiments of this specification have the following beneficial effects through the above content: Enhanced Multimodal Feature Complementarity: Through the horizontal splicing of the cross-modal unified semantic vector and the spatiotemporal alignment semantic matrix, triple information fusion of seismic waveform time-frequency features, core microstructure features, and well log physical property parameters is achieved. This fusion mechanism breaks through the representation limitations of single-modal features, enables cross-verification between the propagation law of seismic signals and rock physical properties, and significantly improves the joint discrimination ability of reservoir sensitive features.

[0033] Breakthrough in Dynamic Correlation Topology Self-Adaptability: Based on the node connection mechanism of similarity calculation, it can automatically sense the real-time correlation changes between well log parameter changes and seismic waveform features. This technology breaks through the rigid constraints of traditional fixed-threshold correlation rules, forms an intelligent correlation network that can be dynamically reconstructed with the increment of drilling data, and effectively adapts to the dynamic evolution requirements of geological cognition during the development of oil and gas reservoirs.

[0034] Geological anomaly body correlation precision leap: Define logging-seismic heterogeneous nodes in the fusion feature space, and accurately capture the hidden correlation patterns between abnormal sections of logging curves and seismic waveform distortion areas through high-dimensional feature similarity measurement. This technology significantly improves the correlation accuracy of key geological interpretation tasks such as fault identification and gas-bearing anomaly detection, and solves the technical pain point that traditional methods are insensitive to weak signal correlation.

[0035] Upgrade of engineering decision-making support visualization: Through the visual expression of the connection strength and topological structure of nodes in the dynamic correlation map, the spatial matching relationship between seismic reflection interfaces and logging interpretation conclusions is intuitively presented. This interactive map expression provides a transparent intelligent support interface for engineering decisions such as real-time adjustment of well trajectories and delineation of reservoir sweet spots.

[0036] Furthermore, the dynamic correlation map of logging and seismic waveforms obtained based on the similarity includes: Screen node pairs corresponding to the similarity greater than a preset threshold, where the node pairs are associated logging nodes and seismic nodes: Generate the dynamic correlation map based on the node pairs.

[0037] It should be noted that through the above content, the embodiments of this specification have the following beneficial effects: Innovation of dynamic adaptive correlation mechanism: Through the intelligent screening mechanism of preset thresholds, the dynamic optimization of the correlation relationship between logging-seismic nodes is realized. This technology breaks through the rigid constraints of traditional fixed correlation rules, and can automatically adjust the correlation strength threshold according to the actual geological scenario to ensure that the topological structure of the map evolves adaptively with data quality and geological complexity.

[0038] Improvement of correlation noise suppression ability: Based on the node pair screening strategy of similarity threshold, false correlation signals caused by data acquisition noise or cross-modal semantic deviation are effectively filtered. This mechanism significantly improves the credibility of the geological significance of node connections in the correlation map, and avoids misjudging random perturbations of seismic waveforms as effective reservoir responses.

[0039] Enhanced focus on key geological targets: The threshold screening process essentially forms a directional enhancement of strong correlation features, making the map automatically focus on the highly correlated regions between abnormal sections of logging parameters and seismic waveform anomaly areas. This focusing effect greatly improves the recognition sensitivity of key geological targets such as the boundaries of hidden oil and gas reservoirs and small faults.

[0040] Breakthrough in engineering implementation interpretability: The preset threshold mechanism provides a transparent decision-making basis for generating the correlation map. Geological engineers can intuitively control the balance between the sparsity and reliability of the correlation network by adjusting the threshold. This interpretable interaction method significantly enhances the trust and acceptance of technical personnel in the results generated by artificial intelligence.

[0041] A data cross-modal association device in an oil and gas field database provided by one or more embodiments of this specification includes: A spectrogram generation unit that receives a pre-input original seismic waveform time series and generates a time-frequency spectrogram through Fourier transform; A core digital model determination unit that receives a pre-input core scan image set, extracts local binary pattern texture features from the core scan image set, and obtains a spatially continuous three-dimensional core digital model based on the local binary pattern texture features; A semantic vector determination unit that injects the time-frequency spectrogram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector; A feature matrix generation unit that constructs a contrastive learning framework for the spectrogram and well logging text, and generates a spatio-temporally aligned semantic feature matrix based on the contrastive learning framework; An association graph determination unit that obtains a dynamic association graph of well logging and seismic waveforms based on the cross-modal unified semantic vector and the semantic feature matrix.

[0042] A data cross-modal association device in an oil and gas field database provided by one or more embodiments of this specification includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to: Receive a pre-input original seismic waveform time series and generate a time-frequency spectrogram through Fourier transform; Receive a pre-input core scan image set, extract local binary pattern texture features from the core scan image set, and obtain a spatially continuous three-dimensional core digital model based on the local binary pattern texture features; Inject the time-frequency spectrogram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector; Construct a contrastive learning framework for the spectrogram and well logging text, and generate a spatio-temporally aligned semantic feature matrix based on the contrastive learning framework; Obtain a dynamic association graph of well logging and seismic waveforms based on the cross-modal unified semantic vector and the semantic feature matrix.

[0043] A non-volatile computer storage medium provided by one or more embodiments of this specification stores computer-executable instructions, and when the computer-executable instructions are executed by a computer, they can implement: Receive a pre-input original seismic waveform time series and generate a time-frequency spectrogram through Fourier transform; Receive a pre - input core scan image set, extract local binary pattern texture features from the core scan image set, and obtain a spatially continuous three - dimensional core digital model based on the local binary pattern texture features; Inject the time - frequency spectrogram and the three - dimensional core digital model into a pre - generated hierarchical semantic framework to obtain a cross - modal unified semantic vector; Construct a contrastive learning framework for the spectrogram and well - logging text, and generate a spatio - temporal alignment semantic feature matrix based on the contrastive learning framework; Based on the cross - modal unified semantic vector and the semantic feature matrix, obtain a dynamic correlation map of well - logging and seismic waveforms.

[0044] The above - mentioned at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects: Improvement of cross - modal semantic unity: By injecting the seismic time - frequency spectrogram and the three - dimensional core digital model into a hierarchical semantic framework, a multi - modal semantic space mapping of seismic waveforms, core images, and well - logging text is realized. This method breaks through the limitations of traditional independent feature extraction strategies, establishes deep semantic associations of cross - modal data in a unified semantic vector space, and effectively bridges the semantic gap between the micro - time - frequency characteristics of seismic waveforms and the macro - structure of core textures.

[0045] Optimization of spatio - temporal alignment accuracy: Adopt a contrastive learning framework to jointly model the spectrogram and well - logging text, and realize the spatio - temporal consistency alignment of cross - modal features by dynamically perceiving the non - linear relationship between the time dimension of seismic waveforms and the depth space dimension of well - logging. This mechanism significantly improves the feature misalignment problem caused by the spatio - temporal dimension splitting in traditional methods, and provides a high - precision semantic feature matrix for subsequent dynamic correlation.

[0046] Innovation of dynamic correlation mechanism: By fusing the unified semantic vector and the spatio - temporal alignment feature matrix, a dynamically updatable correlation map is constructed. This map can adapt to the real - time update of well - logging data during the drilling process, overcomes the defect that traditional static matching mechanisms cannot reflect the dynamic changes of oil and gas reservoirs, and provides real - time data support for the adjustment of development plans.

[0047] Breakthrough in three - dimensional modeling efficiency: In the stage of constructing the core digital model, through the spatial continuity optimization algorithm of local binary pattern texture features, while ensuring the accuracy of micro - structure representation, the processing complexity of a large - scale core scan image set is significantly reduced. This method effectively solves the efficiency bottleneck problem caused by feature redundancy in the traditional three - dimensional modeling process.

[0048] Enhanced engineering application value: Through the deep association and dynamic visualization of multi-modal data, the cross-verification of seismic waveform anomaly areas, core pore structure characteristics, and logging physical property parameters is achieved, significantly improving the accuracy of reservoir fluid identification and the rationality of well trajectory design, providing core technical support for the intelligent exploration and development of oil and gas fields. Brief Description of the Drawings

[0049] To more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in this specification. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings: Figure 1 It is a schematic flowchart of a method for cross-modal association of data in an oil and gas field database provided by one or more embodiments of this specification; Figure 2 It is a schematic structural diagram of a device for cross-modal association of data in an oil and gas field database provided by one or more embodiments of this specification; Figure 3 It is a schematic structural diagram of a device for cross-modal association of data in an oil and gas field database provided by one or more embodiments of this specification. Detailed Embodiments

[0050] The embodiments of this specification provide a method, device, device, and medium for cross-modal association of data in an oil and gas field database.

[0051] To enable those skilled in the art to better understand the technical solutions in this specification, the following will clearly and completely describe the technical solutions in the embodiments of this specification with reference to the drawings in the embodiments of this specification. Obviously, the described embodiments are only some embodiments of this specification, rather than all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this specification.

[0052] Figure 1 It is a schematic flowchart of a method for cross-modal association of data in an oil and gas field database provided by one or more embodiments of this specification. This process can be executed by a cross-modal data association system. Some input parameters or intermediate results in the process allow manual intervention and adjustment to help improve accuracy.

[0053] The method process steps of the embodiments of this specification are as follows: S101, Receive the pre-input original seismic waveform time series and generate a spectrogram through Fourier transform.

[0054] In the embodiments of this specification, the following specific implementation solutions can be adopted: Seismic data preprocessing: Denoise and normalize the input seismic waveform time series to eliminate the high-frequency interference signals introduced by the acquisition equipment.

[0055] Sliding window framing: Use an overlapping Hamming window to frame and cut the continuous seismic signal to ensure the local stationarity of time-frequency analysis.

[0056] Short-time Fourier transform: Perform spectrum calculation frame by frame to generate a three-dimensional time-frequency spectrum matrix containing time, frequency, and amplitude.

[0057] Time-depth domain conversion: Integrate the sonic logging slowness data and map the time axis to the true formation depth coordinates through a depth calibration network.

[0058] S102. Receive a pre-input core scan image set, extract local binary pattern texture features from the core scan image set, and obtain a spatially continuous three-dimensional core digital model based on the local binary pattern texture features.

[0059] In the embodiments of this specification, the following specific implementation solutions can be adopted: Core image enhancement: Use the histogram equalization algorithm to eliminate the brightness difference in the scan image and improve the visual recognition of the pore structure.

[0060] Multi-scale LBP extraction: Calculate the rotation-invariant local binary pattern layer by layer within the range of 3×3 to 15×15 pixels to capture rock texture features at different granularities.

[0061] Interlayer topological association: Analyze the similarity of the LBP histogram sequences of adjacent slices through the dynamic time warping algorithm to construct a three-dimensional spatial adjacency graph.

[0062] Voxel model reconstruction: Optimize the interlayer connection relationship based on the Markov random field and use the marching cubes algorithm to generate a continuous three-dimensional grid model.

[0063] S103. Inject the time-frequency spectrum diagram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector.

[0064] In the embodiments of this specification, the following specific implementation solutions can be adopted: Semantic framework initialization: Construct a three-level fusion architecture including underlying signal features, middle-layer physical properties, and high-layer geological semantics.

[0065] Bidirectional attention injection: Deploy a multi-head cross-attention module in the middle layer to establish an association mapping between the seismic frequency band energy and the core porosity.

[0066] Gated Feature Aggregation: Integrate semantic expressions at different levels through residual connections, and adopt an adaptive weight allocation strategy to balance the modal contribution degrees.

[0067] Semantic Space Compression: Use a variational autoencoder to perform non-linear dimensionality reduction on the fused features, and generate low-dimensional dense cross-modal semantic vectors.

[0068] S104. Construct a contrastive learning framework for spectrogram and logging text, and generate a spatio-temporal aligned semantic feature matrix based on the contrastive learning framework.

[0069] In the embodiments of this specification, the following specific implementation schemes can be adopted: Dual-Modal Encoder Construction: Design parallel convolutional neural network and Transformer architectures to extract the time-frequency features of the spectrogram and the semantics of the logging text respectively.

[0070] Deformable Attention Alignment: Deploy an attention module with adjustable receptive fields in the feature space to capture the non-linear relationship between the propagation delay of seismic waves and the depth offset of logging.

[0071] Contrastive Loss Optimization: Adopt a hard example mining strategy to construct positive and negative sample pairs, and drive the alignment of the feature space through a cosine similarity loss function.

[0072] Matrix Orthogonalization Processing: Perform principal component analysis on the optimized feature matrix to eliminate redundant dimensions and improve the linear separability of cross-modal features.

[0073] S105. Based on the cross-modal unified semantic vector and the semantic feature matrix, obtain the dynamic association map of logging and seismic waveforms.

[0074] In the embodiments of this specification, the following specific implementation schemes can be adopted: Multi-Modal Feature Concatenation: Concatenate the semantic vector and the feature matrix along the channel dimension, construct a joint representation space and perform whitening processing.

[0075] Heterogeneous Node Definition: Divide logging nodes according to the sampling interval of logging depth, and define seismic nodes according to the spatial distribution of seismic trace gathers.

[0076] Dynamic Similarity Calculation: Adopt an adaptive kernel function to calculate the association strength between nodes in real time, and realize the incremental update of the association weight through a sliding window mechanism.

[0077] Knowledge Graph Visualization: Generate a dynamic topology map based on the force-directed layout algorithm, and provide functions for querying node attributes and backtracking association paths.

[0078] It should be noted that through the above content, the embodiments of this specification have the following beneficial effects: Cross-modal semantic unity enhancement: By injecting the spectrogram during an earthquake and the 3D core digital model into a hierarchical semantic framework, a multi-modal semantic space mapping of seismic waveforms, core images, and logging texts is achieved. This method breaks through the limitations of traditional independent feature extraction strategies, establishes deep semantic associations for cross-modal data in a unified semantic vector space, and effectively bridges the semantic gap between the micro time-frequency characteristics of seismic waveforms and the macro structure of core textures.

[0079] Space-time alignment accuracy optimization: A contrastive learning framework is used to jointly model the spectrogram and logging text. By dynamically perceiving the non-linear relationship between the time dimension of seismic waveforms and the depth space dimension of logging, the space-time consistency alignment of cross-modal features is achieved. This mechanism significantly improves the feature misalignment problem caused by the separation of space-time dimensions in traditional methods, providing a high-precision semantic feature matrix for subsequent dynamic associations.

[0080] Innovation of dynamic association mechanism: By fusing the unified semantic vector and the space-time alignment feature matrix, a dynamically updatable association map is constructed. This map can adaptively update the logging data in real time during the drilling process, overcoming the defect that traditional static matching mechanisms cannot reflect the dynamic changes of oil and gas reservoirs, and providing real-time data support for the adjustment of development plans.

[0081] Breakthrough in 3D modeling efficiency: In the stage of constructing the core digital model, through an optimization algorithm for the spatial continuity of local binary pattern texture features, while ensuring the accuracy of microstructural characterization, the processing complexity of a large-scale core scan image set is significantly reduced. This method effectively solves the efficiency bottleneck problem caused by feature redundancy in the traditional 3D modeling process.

[0082] Enhanced engineering application value: Through the deep association and dynamic visualization expression of multi-modal data, the abnormal areas of seismic waveforms, the pore structure characteristics of cores, and the physical properties of logging are cross-verified, significantly improving the accuracy of reservoir fluid identification and the rationality of well trajectory design, providing core technical support for the intelligent exploration and development of oil and gas fields.

[0083] Furthermore, when receiving the pre-input original seismic waveform time series and generating a spectrogram through Fourier transform, the original seismic waveform time series can be received, the short-time Fourier transform algorithm can be used, a specified window can be set, and a spectrogram matrix including time, frequency, and amplitude can be generated; obtain a time-depth conversion model established based on logging acoustic travel time data: input the spectrogram matrix into the time-depth conversion model to generate a spectrogram of depth and time.

[0084] In the embodiments of this specification, the following specific implementation schemes can be adopted: Step 1: Generation of spectrogram matrix Seismic data preprocessing: Perform baseline correction and outlier removal on the input raw seismic waveform time series to eliminate the DC offset and burst interference introduced by the acquisition system; Sliding window framing: Use a Hamming window function with adjustable length to perform overlapping framing on the continuous waveform to balance the time resolution and the effect of spectrum leakage suppression; Short-time Fourier transform execution: Calculate the spectral distribution of the seismic signal in each time window frame by frame to generate a three-dimensional matrix containing time stamps, frequency components, and amplitude intensities; Time-frequency map normalization: Perform logarithmic compression and normalization on the spectral amplitude values to enhance the visual recognition of weak reflection signals.

[0085] Step 2: Generation of time-depth domain spectrogram Loading of time-depth conversion model: Call the pre-trained depth conversion network, which establishes a time-depth mapping relationship based on well logging acoustic travel time data; Data format adaptation: Convert the time axis data of the time-frequency spectrogram matrix into the tensor format required by the model input, while retaining the frequency and amplitude dimension information; Prediction of depth coordinates: Obtain the formation depth value corresponding to each time stamp through inference of the time-depth conversion model, and establish a time-depth mapping lookup table; Synthesis of dual-domain spectrogram: Align and fuse the original time axis and the predicted depth axis coordinates to generate a two-dimensional spectral profile diagram that simultaneously annotates time and depth.

[0086] It should be noted that through the above content, the embodiments of this specification have the following beneficial effects: Improvement of time-frequency feature resolution: Through the short-time Fourier transform algorithm combined with a configurable window function, while maintaining the frequency resolution, accurately capture the local time-varying characteristics of the seismic waveform. The generated three-dimensional time-frequency spectrogram matrix (time-frequency-amplitude) realizes the fine-grained characterization of the microscopic formation structure features, providing a more spatio-temporally resolved basic feature expression for subsequent cross-modal association.

[0087] Optimization of depth domain alignment accuracy: Based on well logging acoustic travel time data, construct a time-depth conversion model, map the traditional time-domain seismic spectrum to the depth coordinate system, and establish a unified depth reference for seismic waveforms and well logging data. This conversion mechanism effectively solves the problem of cross-modal data space misalignment caused by the non-linearity of the time-depth relationship, and significantly improves the matching accuracy between seismic features and the true depth position of the formation.

[0088] Enhancement of multi-dimensional feature fusion: The generated depth-time dual-domain spectrogram synchronously retains the time evolution law of the original waveform and the depth distribution characteristics after conversion, forming a composite feature expression with both time sensitivity and spatial interpretability. This multi-dimensional feature fusion mechanism provides a unified spatial anchor point for the cross-scale association of seismic data with core and well logging data.

[0089] Intelligent data processing flow: Through the automated parameter mapping of the time-to-depth conversion model, the traditional experience-dependent method of manually calibrating the time-to-depth relationship of earthquakes is replaced. While reducing subjective errors, batch depth calibration of large-scale seismic data is realized, which significantly improves the engineering implementation efficiency of the data preprocessing link.

[0090] Furthermore, when receiving the pre-input core scan image, extracting the local binary pattern texture features from the core scan image, and obtaining the spatially continuous three-dimensional core digital model based on the local binary pattern texture features, the core scan image set can be first received; the LBP value is calculated for each pixel point of each core scan image to generate an LBP coding image; the LBP histogram features of the LBP coding image are counted to characterize the local texture distribution of the core; based on the similarity of the LBP histograms of adjacent slices, a spatial correspondence relationship between slices is established; based on the spatial correspondence relationship, the two-dimensional slices are stacked into a three-dimensional voxel grid to obtain the three-dimensional core digital model.

[0091] In the examples of this specification, the following specific implementation schemes can be used: Step 1: Core image preprocessing Image quality verification: Perform resolution detection and distortion correction on the input core scanning image set, and remove abnormal slices with scanning artifacts or blur and distortion; Lighting equalization processing: Adaptive histogram equalization algorithm is used to eliminate the brightness difference between adjacent slices to ensure the consistency of texture feature extraction; Image segmentation and cutting: Divide large-scale core scans into overlapping sub-regions to improve the accuracy of local texture feature extraction.

[0092] Step 2: LBP coding map generation Multi-scale LBP operator configuration: Define rotation-invariant LBP operators with different neighborhood radii in the range of 3×3 to 15×15 pixels; Pixel-by-pixel feature calculation: Perform circular neighborhood sampling on each pixel, compare the grayscale values ​​of the central pixel and the neighboring pixels to generate binary code; Coding map optimization: Isolated coding noise is eliminated through morphological closing operations, retaining effective texture patterns reflecting pore structures and mineral boundaries.

[0093] Step 3: Statistical modeling of texture features Local histogram construction: Divide the LBP coding map into grid units and count the distribution frequencies of different coding modes in each unit; Feature vector fusion: local histograms are aggregated along the radial and circumferential dimensions of the core slice to generate a global feature vector that characterizes the texture distribution of the entire slice; Feature dimensionality reduction processing: The principal component analysis technique is used to compress the feature dimensions and retain the key texture patterns for differentiating different lithologies.

[0094] Step Four: Establishment of interlayer spatial relationships Dynamic similarity matching: The similarity of adjacent slice histogram sequences is calculated through the dynamic time warping algorithm to capture the texture gradual change law caused by sedimentary cycles. Misalignment compensation and calibration: Based on cross-correlation peak detection, the spatial offset of slices caused by the inclination of core sampling is automatically corrected. Adjacency matrix generation: A weighted graph structure representing the three-dimensional topological connection relationship between slices is constructed, and the weight value is positively correlated with the histogram similarity.

[0095] Step Five: Three-dimensional model reconstruction Voxel space initialization: The resolution and spatial range of the three-dimensional voxel grid are defined according to the core diameter and slice spacing. Interlayer interpolation optimization: The Markov random field model is used to optimize the voxel filling strategy between adjacent slices to eliminate staircase artifacts. Surface smoothing processing: The anisotropic diffusion filtering algorithm is used to enhance the continuity of the pore-skeleton boundary to generate a high-fidelity three-dimensional core model.

[0096] It should be noted that the embodiments of this specification have the following beneficial effects through the above content: Enhanced microstructure characterization ability: Through the extraction of pixel-by-pixel texture features of the local binary pattern (LBP) encoded map, the spatial heterogeneity of microscopic geological features such as pore structures and mineral particle distributions in core scan images is completely retained. The LBP histogram statistical mechanism effectively captures the texture pattern differences at different lithology interfaces, significantly improving the characterization accuracy of the digital model for the complex microscopic structure of reservoir rocks.

[0097] Guaranteed three-dimensional spatial continuity: Based on the spatial correspondence relationship established by the similarity of adjacent slice LBP histograms, the limitation of the traditional two-dimensional image stacking method relying on manual alignment is broken through. This method automatically constructs the topological connection between slices through the statistical correlation of texture features, ensuring the structural continuity of the three-dimensional voxel grid in the vertical and horizontal directions and truly restoring the sedimentary sequence characteristics of underground rock layers.

[0098] Multi-scale feature fusion and optimization: The generation process of the three-dimensional voxel grid synchronously integrates the microscopic texture information encoded by pixel-level LBP and the macroscopic statistical characteristics of slice-level histograms, forming a multi-scale fusion core digital representation system. This fusion mechanism enables the model to not only reflect the millimeter-level pore structure details but also characterize the spatial distribution law of meter-level lithology units.

[0099] Automated Upgrade of Modeling Process: The full-process algorithmic processing from LBP feature extraction to 3D reconstruction replaces the empirical operation mode that relies on manual annotation of the corresponding relationship of slices in traditional methods. This automated process not only avoids the introduction of subjective errors but also greatly improves the processing efficiency of large-scale core scan image sets, providing the ability to generate standardized data for the construction of the oilfield digital twin system.

[0100] Further, when injecting the spectrogram and the 3D core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector, a template can be defined as the hierarchical semantic framework, and the hierarchical semantic framework includes a cross-attention mechanism between the spectrogram and the 3D core digital model; based on the cross-attention mechanism between the spectrogram and the 3D core digital model, interaction features are obtained; the interaction features are input into a fully connected layer to reduce the dimension to a cross-modal unified semantic vector.

[0101] In the embodiments of this specification, the following specific implementation schemes can be adopted: Step 1: Construction of Hierarchical Semantic Framework Design of multi-level feature abstraction architecture: Construct a three-level hierarchical processing architecture including underlying signal features, middle-level physical properties, and high-level geological semantics; Alignment of modal feature spaces: Perform band energy normalization on the spectrogram matrix and perform isotropic voxel resampling on the 3D core model to unify the spatial resolution benchmark; Configuration of cross-attention module: Deploy a multi-head cross-attention mechanism in the middle layer of the framework and set two-way association channels for the seismic band dimension and the core spatial dimension.

[0102] Step 2: Extraction of Cross-modal Interaction Features Dynamic calculation of attention weights: Establish an association weight matrix between the band energy of the spectrogram and the pore distribution of the core through a query-key-value matching mechanism; Bidirectional projection fusion of features: Project the core texture feature vector into the seismic frequency domain space, and simultaneously map the time-frequency features into the 3D space of the core to form bidirectional feature enhancement; Residual feature compensation: Integrate the original modal features and attention-weighted features through skip connections to retain key detail information in the cross-modal interaction process.

[0103] Step 3: Generation of Unified Semantic Vector Feature channel concatenation: Concatenate the time-frequency-core interaction features and the high-level semantic features output by the hierarchical framework along the feature dimension; Adaptive feature compression: Adopt a bottleneck fully connected layer structure to gradually reduce the dimension to screen cross-modal common features and suppress modal-specific noise; Semantic space regularization: Apply an orthogonal constraint loss to the output vector to ensure the distinguishability of semantic vectors of different geological units in the low-dimensional space.

[0104] It should be noted that the embodiments of this specification have the following beneficial effects through the above content: Deep optimization of cross-modal feature interaction: By constructing a cross-attention mechanism between the time-frequency spectrogram and the three-dimensional core model, dynamic interactive perception of seismic waveform time-frequency features and core micro-texture features is achieved. This mechanism breaks through the limitations of traditional methods that simply splice multi-modal features, establishes implicit association rules between seismic wave propagation characteristics and rock physical properties in the feature space, and significantly improves the fineness of cross-modal semantic alignment.

[0105] Enhancement of hierarchical semantic representation: The hierarchical semantic framework fuses the macroscopic stratigraphic interface features of seismic data and the microscopic pore structure features of the core model through a hierarchical feature abstraction mechanism. This hierarchical and progressive semantic integration method not only retains the unique feature expressions of each modality but also constructs a cross-scale geological semantic association network, enhancing the geological interpretability of the unified semantic vector.

[0106] Improvement of semantic vector space compactness: Use a fully connected layer to perform directional dimensionality reduction on high-dimensional interaction features, eliminating seismic time-frequency noise and core texture redundant features while retaining cross-modal key semantic information. The generated compact semantic vector not only reduces the subsequent computational complexity but also strengthens the class separability of different geological units through orthogonalization processing of the feature space.

[0107] Breakthrough in geological law interpretability: The construction process of the cross-attention weight matrix essentially reveals the physical association pattern between seismic wave attenuation characteristics and physical property parameters such as rock permeability. This interpretable feature interaction mechanism provides a new theoretical path for reservoir parameter inversion, breaking through the application limitations of traditional black-box models in oil and gas geological research.

[0108] Furthermore, when generating a spatio-temporally aligned semantic feature matrix based on the contrast learning framework, the spectrogram and well log text can be mutually corrected based on the attention mechanism of the contrast learning framework to obtain a fused feature; and the fused feature is compressed into a spatio-temporally aligned semantic feature matrix.

[0109] In the embodiments of this specification, the following specific implementation schemes can be adopted: Step 1: Cross-modal two-way correction Construction of a dual-modal encoder: Design a convolutional neural network to extract the time-frequency features of the spectrogram, and simultaneously use the Transformer architecture to encode the semantic information of the well log text; Deployable Deformable Attention Mechanism: Deploy an attention module with adjustable receptive fields in the feature space to capture the non-linear mapping relationship between the propagation time delay of seismic waves and the offset of logging depth; Feature Cross-Correction Execution: Through the cross-attention gating mechanism, use logging parameter semantics to correct the frequency band energy distribution of the spectrogram, and at the same time enhance the keyword weights of logging texts with time-frequency anomaly features; Residual Compensation Fusion: Perform weighted summation of the original modal features and the attention-corrected features to retain the effective information increment in the cross-modal interaction process.

[0110] Step 2: Semantic Matrix Compression and Alignment Feature Orthogonalization Processing: Perform Gram-Schmidt orthogonalization on the fused features to eliminate redundant information dimensions between the spectrum-text modalities; Spatio-Temporal Constraint Compression: Adopt spatial pyramid pooling technology to compress the feature dimensions, and simultaneously impose time continuity loss and depth consistency loss; Matrix Optimization and Calibration: Extract spatio-temporal shared basis vectors through non-negative matrix factorization technology, and reconstruct a low-rank semantic matrix with spatio-temporal alignment characteristics; Dynamic Update Mechanism: Deploy a sliding window strategy to achieve incremental update of matrix parameters, adapting to the temporal changes of real-time data streams during the drilling process.

[0111] It should be noted that the embodiments of this specification have the following beneficial effects through the above content: Breakthrough in Cross-Modal Dynamic Correction Ability: Through the cross-attention mechanism of the contrastive learning framework, achieve bidirectional feature correction between the spectrogram and logging texts. This mechanism enables the time evolution features of seismic waveforms to correct the depth interpretation deviation of logging texts, and at the same time the spatial constraints of logging parameters feed back to calibrate the physical meaning of the spectrogram, forming a self-optimizing feature interaction system with spatio-temporal consistency.

[0112] Jump in Spatio-Temporal Coupling Accuracy: During the feature fusion process, through the dynamic allocation strategy of attention weights, automatically capture the non-linear mapping relationship between the propagation time of seismic signals and the logging depth space. This technology breaks through the limitations of traditional linear interpolation methods and significantly improves the spatio-temporal matching accuracy of key geological elements such as the identification of the top and bottom interfaces of oil and gas reservoirs and the location of formation pinch-out points.

[0113] Innovation in Compactification of Semantic Feature Matrix: Use feature compression algorithms to perform orthogonal dimensionality reduction on high-dimensional fused features, eliminating redundant information between modalities while retaining the correlative expression of seismic waveform anomaly patterns and logging parameter mutation features. The generated semantic matrix has both low-dimensionality and high information entropy characteristics, providing a lightweight and highly discriminative feature basis for subsequent dynamic correlation map construction.

[0114] Enhanced Geological Interpretation Credibility: The attention correction trajectory formed during the contrastive learning process can reversely analyze the physical interaction mechanism between seismic attributes and logging responses. This interpretable correction mechanism provides a transparent decision-making basis for identifying hydrocarbon-bearing properties of reservoirs, effectively overcoming the "black box" risk of traditional end-to-end models.

[0115] Further, when obtaining the dynamic correlation map of logging and seismic waveforms based on the cross-modal unified semantic vector and the semantic feature matrix, the cross-modal unified semantic vector and the semantic feature matrix can be horizontally concatenated to generate a fused feature matrix; logging nodes and seismic nodes are defined in the fused feature matrix, and the similarity between the logging nodes and the seismic nodes is determined; the dynamic correlation map of logging and seismic waveforms is obtained based on the similarity.

[0116] In the embodiments of this specification, the following specific implementation schemes can be adopted: Step 1: Multimodal Feature Fusion Feature Dimension Alignment: Perform zero-padding operation on the cross-modal unified semantic vector to make its channel dimension consistent with the column width of the semantic feature matrix; Horizontal Concatenation Execution: Concatenate the semantic vectors along the column direction of the feature matrix, and use bilinear interpolation to eliminate scale differences to generate a joint representation matrix; Matrix Whitening Processing: Perform ZCA whitening operation on the fused matrix to eliminate the linear correlation between features of different modalities.

[0117] Step 2: Heterogeneous Node Definition Logging Node Division: Cut the row vector of the fused matrix according to the logging depth sampling interval, and each depth unit corresponds to a logging node, carrying physical properties such as porosity and permeability; Seismic Node Extraction: Cut the column vector of the matrix along the spatial distribution of the seismic trace gather, and each trace gather node is associated with seismic attributes such as time-frequency energy and reflection intensity; Node Attribute Enhancement: Aggregate the features of adjacent nodes through the graph attention mechanism to strengthen the discriminative expression of nodes in reservoir fluid identification.

[0118] Step 3: Dynamic Similarity Calculation Multi-scale Kernel Function Configuration: Design a hybrid metric function combining Euclidean distance and cosine similarity to balance local details and global distribution characteristics; Adaptive Threshold Setting: Dynamically adjust the similarity discrimination threshold based on the node feature distribution density to achieve strong correlation focusing in sparse regions and noise suppression in dense regions; Incremental Update Mechanism: Deploy a sliding time window to track the latest logging data and trigger real-time iterative update of the association weights.

[0119] Step 4: Correlation Map Generation Topological structure optimization: The force-directed layout algorithm is adopted to arrange the spatial positions of nodes, so that strongly associated nodes are clustered and weakly associated nodes are dispersed; Dynamic visualization rendering: Integrate WebGL technology to realize the interactive display of the atlas, support the pop-up query of node attributes and the animation tracing of associated paths; Abnormal association warning: Set an association intensity mutation detection module to trigger a sweet spot warning signal when the similarity between well logging-seismic nodes changes suddenly.

[0120] It should be noted that the embodiments of this specification have the following beneficial effects through the above content: Enhanced multi-modal feature complementarity: Through the horizontal splicing of cross-modal unified semantic vectors and spatio-temporal aligned semantic matrices, triple information fusion of seismic waveform time-frequency features, core microstructure features, and well logging physical property parameters is achieved. This fusion mechanism breaks through the representation limitations of single-modal features, enables cross-verification of seismic signal propagation laws and rock physical properties, and significantly improves the joint discrimination ability of reservoir sensitive features.

[0121] Breakthrough in dynamic association topology self-adaptability: Based on the node connection mechanism of similarity calculation, it can automatically sense the real-time correlation changes between well logging parameters and seismic waveform features. This technology breaks through the rigid constraints of traditional fixed-threshold association rules, forms an intelligent association network that can be dynamically reconstructed with the increment of drilling data, and effectively adapts to the dynamic evolution requirements of geological cognition during the development of oil and gas reservoirs.

[0122] Jump in the association accuracy of geological anomalies: Define well logging-seismic heterogeneous nodes in the fusion feature space, and accurately capture the hidden association patterns between abnormal segments of well logging curves and seismic waveform distortion areas through high-dimensional feature similarity measurement. This technology significantly improves the association accuracy of key geological interpretation tasks such as fault identification and gas-bearing anomaly detection, and solves the technical pain point that traditional methods are insensitive to weak signal associations.

[0123] Upgrade of engineering decision-making support visualization: The dynamic association atlas visually presents the spatial matching relationship between seismic reflection interfaces and well logging interpretation conclusions through the visual expression of node connection strength and topological structure. This interactive atlas expression method provides a transparent intelligent support interface for engineering decisions such as real-time adjustment of well trajectories and delineation of reservoir sweet spots.

[0124] Furthermore, when obtaining the dynamic association atlas of well logging and seismic waveforms based on the similarity, filter the node pairs corresponding to the similarity greater than the preset threshold, and the node pairs are associated well logging nodes and seismic nodes: Generate the dynamic association atlas based on the node pairs.

[0125] In the embodiments of this specification, the following specific implementation schemes can be adopted: Step 1: Intelligent threshold screening Dynamic threshold initialization: Based on the density distribution of the node feature space, the kernel density estimation algorithm is used to automatically calculate the initial similarity threshold; Multi-scale threshold optimization: Region subsets are divided for different geological units, and local optimal thresholds are calculated respectively to adapt to the reservoir heterogeneity characteristics; Noise suppression processing: Isolated low-similarity pseudo-associations are eliminated through morphological opening operations, and continuous high-confidence node connections are retained; Key area focusing: Threshold downsampling is implemented for sensitive areas such as sweet spots and fault zones to enhance the ability to capture weak association signals.

[0126] Step 2: Heterogeneous node processing Node attribute inheritance: Physical property parameters such as porosity and oil saturation are inherited for the selected logging nodes, and attributes such as time-frequency energy and coherence volume are associated with seismic nodes; Composite node construction: Virtual composite nodes are created for nodes with multiple associations to represent complex geological phenomena at the well-seismic intersection; Node index optimization: A spatial index based on GeoHash and an inverted index of feature vectors are established to support multi-dimensional fast retrieval.

[0127] Step 3: Dynamic graph construction Graph database initialization: The Neo4j graph database is used to construct a node-relationship storage structure, and multi-dimensional edge attributes such as "association strength" and "spatial distance" are defined; Incremental update mechanism: The Kafka message queue is deployed to capture new node pairs in real time, triggering local reconstruction of the graph topology; Dynamic weight assignment: The association weight decay coefficient is set according to the drilling timeliness to ensure that the graph reflects the latest geological understanding state; Topology optimization execution: The spectral clustering algorithm is used to partition subgraphs in high-density association regions to improve the interpretability of the graph structure.

[0128] Step 4: Visualization and interaction Multi-level rendering: The LOD technology is integrated to achieve multi-scale visualization of the overall view of the entire work area and local details, supporting mouse wheel zooming and drag-and-drop roaming; Dynamic focus tracking: A line-of-sight focus rendering optimization algorithm is deployed to highlight and enhance the display of associated nodes around the current drilling trajectory; Intelligent drilling function: Support for right-clicking on any node to trigger the backtracking of the association path, and visually display the geological evidence chain of well-seismic association; Abnormal association warning: When an unclosed association loop or an edge with a strength mutation is detected, a flashing alarm and log record are automatically triggered.

[0129] Step 5: Graph maintenance and update Version management: Adopt a Git-like mechanism to record the evolution process of the atlas, supporting comparative analysis and backtracking of any historical version; Drift detection module: Deploy a concept drift detection algorithm to trigger threshold recalibration when the node feature distribution changes significantly; Self-learning optimization: Automatically adjust the topological layout parameters through a reinforcement learning mechanism to continuously improve the cognitive friendliness of the atlas visualization.

[0130] It should be noted that the embodiments of this specification have the following beneficial effects through the above content: Innovation of dynamic adaptive association mechanism: Through an intelligent screening mechanism with preset thresholds, realize the dynamic optimization of the association relationship between logging-seismic nodes. This technology breaks through the rigid constraints of traditional fixed association rules and can automatically adjust the association strength threshold according to the actual geological scenario to ensure the adaptive evolution of the atlas topological structure with data quality and geological complexity.

[0131] Improvement of association noise suppression ability: Based on the screening strategy of node pairs with similarity thresholds, effectively filter false association signals caused by data acquisition noise or cross-modal semantic deviation. This mechanism significantly improves the credibility of the geological significance of node connections in the association atlas and avoids misjudging random perturbations of seismic waveforms as effective reservoir responses.

[0132] Enhancement of key geological target focusing: The threshold screening process essentially forms a directional enhancement of strong association features, making the atlas automatically focus on the highly correlated regions of logging parameter mutation segments and seismic waveform anomaly areas. This focusing effect greatly improves the recognition sensitivity of key geological targets such as the boundaries of subtle oil and gas reservoirs and small faults.

[0133] Breakthrough in engineering implementation interpretability: The preset threshold mechanism provides a transparent decision-making basis for the generation of the association atlas. Geological engineers can intuitively control the balance between the sparsity and reliability of the association network through threshold adjustment. This interpretable interaction method significantly enhances the trust and acceptance of technicians in the results generated by artificial intelligence.

[0134] Figure 2 The structural schematic diagram of a data cross-modal association device in an oil and gas field database provided for one or more embodiments of this specification includes: a spectrogram generation unit 201, a core digital model determination unit 202, a semantic vector determination unit 203, a feature matrix generation unit 204, and an association atlas determination unit 205.

[0135] The spectrogram generation unit 201 receives the pre-input original seismic waveform time series and generates a time-frequency spectrogram through Fourier transform; The core digital model determination unit 202 receives a pre-input core scan image set, extracts local binary pattern texture features from the core scan image set, and obtains a spatially continuous three-dimensional core digital model based on the local binary pattern texture features; The semantic vector determination unit 203 injects the time-frequency spectrogram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector; The feature matrix generation unit 204 constructs a contrastive learning framework for the spectrogram and well logging text, and generates a spatio-temporally aligned semantic feature matrix based on the contrastive learning framework; The association map determination unit 205 obtains a dynamic association map of well logging and seismic waveforms based on the cross-modal unified semantic vector and the semantic feature matrix.

[0136] Figure 3 The structural schematic diagram of a data cross-modal association device in an oil and gas field database provided by one or more embodiments of this specification includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: Receive a pre-input original seismic waveform time series, and generate a time-frequency spectrogram through Fourier transform; Receive a pre-input core scan image set, extract local binary pattern texture features from the core scan image set, and obtain a spatially continuous three-dimensional core digital model based on the local binary pattern texture features; Inject the time-frequency spectrogram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector; Construct a contrastive learning framework for the spectrogram and well logging text, and generate a spatio-temporally aligned semantic feature matrix based on the contrastive learning framework; Obtain a dynamic association map of well logging and seismic waveforms based on the cross-modal unified semantic vector and the semantic feature matrix.

[0137] A non-volatile computer storage medium provided by one or more embodiments of this specification stores computer-executable instructions, and when the computer-executable instructions are executed by a computer, they can: Receive a pre-input original seismic waveform time series, and generate a time-frequency spectrogram through Fourier transform; Receive a pre-input set of core scan images, extract local binary pattern texture features from the core scan image set, and obtain a spatially continuous three-dimensional core digital model based on the local binary pattern texture features; Inject the time-frequency spectrogram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector; Construct a contrastive learning framework for the spectrogram and well logging text, and generate a spatio-temporally aligned semantic feature matrix based on the contrastive learning framework; Based on the cross-modal unified semantic vector and the semantic feature matrix, obtain a dynamic correlation map of well logging and seismic waveforms.

[0138] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the embodiments of devices, equipment, and non-volatile computer storage media, since they are basically similar to the method embodiments, the description is relatively simple, and reference can be made to the partial description of the method embodiments for the relevant parts.

[0139] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and reference can be made to the partial description of the method embodiments for the relevant parts.

[0140] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0141] In the embodiments provided in this application, it should be understood that the disclosed devices / network devices and methods can be implemented in other ways. For example, the device / network device embodiments described above are only illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0142] The unit described as a separation component may or may not be physically separated. The component shown as a unit may or may not be a physical unit, that is, it may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0143] In addition, in each embodiment of the present application, each functional unit may be integrated in a processing unit, may exist separately as individual physical units, or two or more units may be integrated in one unit. The above units may be implemented in the form of hardware or in the form of software.

[0144] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods of the present application can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0145] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for cross-modal association of data in an oil and gas field database, characterized in that Including: Receiving a pre-input original seismic waveform time series and generating a time-frequency spectrogram through Fourier transform; Receiving a pre-input core scan image set, extracting local binary pattern texture features from the core scan image set, and obtaining a spatially continuous three-dimensional core digital model based on the local binary pattern texture features; Injecting the time-frequency spectrogram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector; Constructing a contrastive learning framework for the spectrogram and well logging text, and generating a spatio-temporally aligned semantic feature matrix based on the contrastive learning framework; Based on the cross-modal unified semantic vector and the semantic feature matrix, obtaining a dynamic correlation map of well logging and seismic waveforms.

2. The method according to claim 1, wherein The receiving the pre-input original seismic waveform time series and generating a time-frequency spectrogram through Fourier transform includes: Receiving the original seismic waveform time series, using the short-time Fourier transform algorithm, setting a specified window, and generating a time-frequency spectrogram matrix including time, frequency, and amplitude; Obtaining a time-depth conversion model established based on well logging acoustic travel time data: Inputting the time-frequency spectrogram matrix into the time-depth conversion model to generate a time-frequency spectrogram of depth and time.

3. The method according to claim 1, wherein The receiving the pre-input core scan image, extracting local binary pattern texture features from the core scan image, and obtaining a spatially continuous three-dimensional core digital model based on the local binary pattern texture features includes: Receiving a core scan image set; Calculating the LBP value for each pixel point of each core scan image to generate an LBP encoded map; Statistical LBP histogram features of the LBP encoded map to characterize the local texture distribution of the core; Based on the LBP histogram similarity of adjacent slices, establishing a spatial correspondence relationship between slices; Stacking two-dimensional slices into a three-dimensional voxel grid based on the spatial correspondence relationship to obtain the three-dimensional core digital model.

4. The method according to claim 1, characterized in that The injecting the time-frequency spectrogram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector includes: Defining a template as a hierarchical semantic framework, where the hierarchical semantic framework includes a cross-attention mechanism between the time-frequency spectrogram and the three-dimensional core digital model; Based on the cross-attention mechanism between the time-frequency spectrogram and the three-dimensional core digital model, obtaining interaction features; Inputting the interaction features into a fully connected layer to reduce the dimension to a cross-modal unified semantic vector.

5. The method according to claim 1, characterized in that, The generating a spatio-temporally aligned semantic feature matrix based on the contrastive learning framework includes: Mutually correcting the spectrogram and well logging text based on the attention mechanism of the contrastive learning framework to obtain a fused feature; Compressing the fused feature into a spatio-temporally aligned semantic feature matrix.

6. The method according to claim 1, characterized in that, The obtaining a dynamic correlation map of well logging and seismic waveforms based on the cross-modal unified semantic vector and the semantic feature matrix includes: Horizontally splicing the cross-modal unified semantic vector and the semantic feature matrix to generate a fused feature matrix; Defining well logging nodes and seismic nodes in the fused feature matrix and determining the similarity between the well logging nodes and the seismic nodes; Based on the similarity, obtaining a dynamic correlation map of well logging and seismic waveforms.

7. The method according to claim 6, characterized in that, The dynamic association map of well logging and seismic waveforms obtained based on the similarity includes: Filtering node pairs corresponding to the similarity greater than a preset threshold, where the node pairs are associated well logging nodes and seismic nodes: Generating the dynamic association map based on the node pairs.

8. A data cross-modal association device in an oil and gas field database, characterized in that, Including: A spectrogram generation unit that receives a pre-input original seismic waveform time series and generates a time-frequency spectrogram through Fourier transform; A core digital model determination unit that receives a pre-input set of core scan images, extracts local binary pattern texture features from the core scan image set, and obtains a spatially continuous three-dimensional core digital model based on the local binary pattern texture features; A semantic vector determination unit that injects the time-frequency spectrogram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector; A feature matrix generation unit that constructs a contrastive learning framework for the spectrogram and well logging text, and generates a spatio-temporally aligned semantic feature matrix based on the contrastive learning framework; An association map determination unit that obtains a dynamic association map of well logging and seismic waveforms based on the cross-modal unified semantic vector and the semantic feature matrix.

9. An apparatus for cross-modal association of data in an oil and gas field database, characterized in that, Including: At least one processor; And, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to: Receive a pre-input original seismic waveform time series and generate a time-frequency spectrogram through Fourier transform; Receive a pre-input set of core scan images, extract local binary pattern texture features from the core scan image set, and obtain a spatially continuous three-dimensional core digital model based on the local binary pattern texture features; Inject the time-frequency spectrogram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector; Construct a contrastive learning framework for the spectrogram and well logging text, and generate a spatio-temporally aligned semantic feature matrix based on the contrastive learning framework; Obtain a dynamic association map of well logging and seismic waveforms based on the cross-modal unified semantic vector and the semantic feature matrix.

10. A non-volatile computer storage medium, characterized in that, Stores computer-executable instructions that, when executed by a computer, can implement: Receive a pre-input original seismic waveform time series and generate a time-frequency spectrogram through Fourier transform; Receive a pre-input set of core scan images, extract local binary pattern texture features from the core scan image set, and obtain a spatially continuous three-dimensional core digital model based on the local binary pattern texture features; Inject the time-frequency spectrogram and the three-dimensional core digital model into a pre-generated hierarchical semantic framework to obtain a cross-modal unified semantic vector; Construct a contrastive learning framework for the spectrogram and well logging text, and generate a spatio-temporally aligned semantic feature matrix based on the contrastive learning framework; Obtain a dynamic association map of well logging and seismic waveforms based on the cross-modal unified semantic vector and the semantic feature matrix.

Images