A high-resolution reservoir prediction method under seismic full-wave information mining

Abstract

The present application relates to the technical field of reservoir prediction, and particularly relates to a high-resolution reservoir prediction method under seismic full-wave information mining. The method comprises the following steps: acquiring seismic full-wave information data; performing frequency-domain joint analysis on the seismic full-wave information data to generate a multi-scale full-wave feature tensor; performing seismic full-wave signal multi-scale decoupling on the multi-scale full-wave feature tensor to generate seismic multi-wave field separation data; performing implicit reservoir feature mapping on the seismic multi-wave field separation data to generate initial reservoir feature mapping data; performing nonlinear correlation analysis on the initial reservoir feature mapping data to generate high-order reservoir feature data; and performing inversion linkage optimization on the high-order reservoir feature data and the seismic multi-wave field separation data to generate reservoir dynamic response optimization data. Through multi-wave field information extraction, high-resolution reservoir detail display, dynamic reservoir change prediction and confidence evaluation, the present application improves the accuracy and resolution of reservoir prediction.

Description

Technical Field

[0001] This invention relates to the field of reservoir prediction technology, and in particular to a high-resolution reservoir prediction method based on seismic full-wave information mining. Background Technology

[0002] Initially, seismic exploration relied primarily on the traditional seismic reflection method, inferring geological structures by analyzing the reflection of seismic waves at subsurface interfaces. However, this method has limited effectiveness in complex geological environments, particularly posing significant challenges in predicting deep or complex reservoirs. With advancements in technology, especially in computer technology and data processing capabilities, seismic full-wave information mining technology has gradually emerged. This technology analyzes the full waveform information of seismic waves, including reflected, transmitted, and scattered waves, to obtain richer subsurface information. In particular, holographic analysis of the wavefield allows for more accurate identification of key reservoir characteristics such as physical properties, porosity, and water saturation, thereby improving the resolution of reservoir prediction. In recent years, with the development of deep learning and big data technologies, the combination of seismic full-wave information mining and machine learning has significantly improved the accuracy and efficiency of reservoir prediction. However, current traditional seismic reservoir prediction methods often rely solely on traditional seismic reflection wave information, neglecting other wave fields in the full wave information (such as transmitted waves and scattered waves). They also suffer from low resolution, making it difficult to identify details and accurately characterize reservoir features, thus resulting in low accuracy and resolution in reservoir prediction. Summary of the Invention

[0003] Therefore, it is necessary to provide a high-resolution reservoir prediction method based on seismic full-wave information mining to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a high-resolution reservoir prediction method based on seismic full-wave information mining is provided, the method comprising the following steps:

[0005] Step S1: Acquire seismic full-wave information data; perform frequency domain joint analysis on the seismic full-wave information data to generate multi-scale full-wave feature tensors; perform multi-scale decoupling of the seismic full-wave signal on the multi-scale full-wave feature tensors to generate seismic multi-wave field separation data.

[0006] Step S2: Perform implicit reservoir feature mapping on the seismic multi-wavelength separation data to generate initial reservoir feature mapping data; perform nonlinear correlation analysis on the initial reservoir feature mapping data to generate high-order reservoir feature data; perform inversion linkage optimization on the high-order reservoir feature data and the seismic multi-wavelength separation data to generate optimized reservoir dynamic response data; perform reservoir detail enhancement on the optimized reservoir dynamic response data to generate a high-resolution reservoir detail display map;

[0007] Step S3: Perform dynamic reservoir feature change analysis on the high-resolution reservoir detail map to generate dynamic reservoir feature change data; predict reservoir feature changes on the high-resolution reservoir detail map based on the dynamic reservoir feature change data to generate a reservoir feature change prediction map.

[0008] Step S4: Classify the reservoirs in the reservoir feature change prediction map to generate a reservoir classification map; calculate the confidence level of the reservoir classification map to generate reservoir prediction confidence data; compare the reservoir prediction confidence data with the preset confidence threshold and output the results until a high-resolution reservoir prediction distribution map is generated.

[0009] This invention, by acquiring full-wave information and performing joint frequency domain analysis, can extract rich subsurface information from multiple scales and wavefields, overcoming the limitations of traditional methods that rely solely on a single waveform signal. Multi-scale decoupling and wavefield separation techniques further enhance signal resolution, allowing for independent processing of features from different wavefields, providing clearer foundational data for subsequent high-precision reservoir prediction. Implicit reservoir feature mapping and nonlinear correlation analysis reveal deep-seated reservoir characteristic relationships, generating more accurate high-order reservoir feature data. The inversion-linked optimization method establishes close connections between multi-wavefield data, generating more realistic reservoir dynamic responses. Reservoir detail enhancement techniques improve reservoir structure resolution, ensuring high-precision detail display and thus improving the accuracy of reservoir prediction. Dynamic reservoir feature change analysis can monitor reservoir change trends in real time and identify potential change patterns. Based on this, the reservoir feature change prediction map provides a forward-looking prediction of future reservoir evolution, helping to make more accurate development decisions and effectively improving the dynamic management capabilities of reservoirs. Through reservoir classification and confidence calculation, precise regional division and risk assessment of reservoirs can be performed, further improving the accuracy of reservoir prediction. Confidence assessment ensures the reliability and credibility of the prediction results, avoiding errors caused by uncertainty. The resulting high-resolution reservoir prediction distribution map provides a reliable basis for oil and gas exploration and development. Therefore, this invention improves the accuracy and resolution of reservoir prediction through multi-wave field information extraction, high-resolution reservoir detail display, dynamic reservoir change prediction, and confidence assessment.

[0010] Preferably, step S1 includes the following steps:

[0011] Step S11: Acquire full-wave seismic information data using a multi-point distributed seismic signal acquisition array;

[0012] Step S12: Perform data preprocessing on the seismic full-wave information data to generate standard seismic full-wave information data. The data preprocessing includes data cleaning, missing value imputation, and data standardization.

[0013] Step S13: Perform time-domain analysis on the full-wave signal to generate full-wave signal time-domain data; perform fast Fourier transform on the full-wave signal time-domain data to generate full-wave signal frequency-domain data; perform joint frequency-domain analysis based on the full-wave signal time-domain data and full-wave signal frequency-domain data to generate multi-scale full-wave feature tensors.

[0014] Step S14: Based on the preset seismic wave propagation characteristic coefficients, perform multi-scale decoupling of the seismic full-wave signal on the multi-scale full-wave characteristic tensor to generate seismic multi-wave field separation data.

[0015] This invention preprocesses seismic full-wave information, including data cleaning, missing value imputation, and standardization, to remove noise and invalid data, ensuring that the final data is more representative and usable, providing high-quality foundational data for subsequent analysis. The combination of time-domain and frequency-domain analysis comprehensively captures the characteristics of seismic signals from different perspectives (time and frequency domains). The multi-scale full-wave feature tensor generated through joint frequency-domain analysis helps to understand the propagation characteristics of seismic waves at multiple scales, improving the depth and accuracy of signal analysis. Multi-scale decoupling based on seismic wave propagation characteristic coefficients effectively separates different wavefields and extracts the characteristics of various seismic waves. This is crucial for accurately identifying and analyzing different types of seismic waves (such as body waves and surface waves), contributing to a better understanding of seismic wave propagation in subsurface media. The time-domain, frequency-domain, and multi-scale analyses in steps S13 and S14 provide a multi-dimensional perspective for seismic signal processing, enhancing data resolution capabilities and helping to extract meaningful information in complex seismic environments. Through multi-angle analysis and decoupling techniques, the analytical capability and accuracy of seismic signals are effectively improved, possessing significant practical application value for earthquake monitoring, early warning, and disaster assessment.

[0016] Preferably, step S14 includes the following steps:

[0017] Step S141: Perform seismic wave propagation velocity analysis on the multi-scale full-wave characteristic tensor to generate seismic wave propagation velocity data; classify the seismic wave types based on the seismic wave propagation velocity data on the multi-scale full-wave characteristic tensor to generate seismic wave type classification data, which includes P-wave data, S-wave data and surface wave data.

[0018] Step S142: Perform variational mode decomposition on the P-wave data, S-wave data, and surface wave data to generate P-wave mode decomposition data, S-wave mode decomposition data, and surface wave mode decomposition data; perform wavefield separation on the P-wave data, S-wave data, and surface wave data using the P-wave mode decomposition data, S-wave mode decomposition data, and surface wave mode decomposition data to generate seismic full-wave wavefield separation data;

[0019] Step S143: Perform wavefield feature coupling analysis on the seismic full-wave wavefield separation data to generate wavefield feature coupling data; use the wavefield feature coupling data to decouple the seismic full-wave signal at multiple scales using the multi-scale full-wave feature tensor to generate seismic multi-wavefield separation data.

[0020] This invention, through seismic wave propagation velocity analysis and seismic wave type classification, can accurately distinguish between P-waves, S-waves, and surface waves. This not only improves the structured processing level of seismic wave data but also provides strong support for subsequent wavefield separation and decoupling, helping to more clearly identify the type and propagation path of seismic waves. Processing P-wave, S-wave, and surface wave data using variational mode decomposition (VMD) technology effectively extracts the independent characteristics of each wave type, avoiding mutual interference between different wavefields, thus achieving more accurate wavefield separation. This step facilitates in-depth analysis of the characteristics of each wavefield, especially in complex seismic wave propagation environments, accurately revealing the contributions of different waves. Feature coupling analysis of the separated wavefield data helps reveal the potential relationships and interactions between different wavefields, thereby providing a more comprehensive understanding of seismic wave behavior and its propagation patterns in the subsurface medium. This analysis plays a crucial role in predicting seismic wave propagation trends and analyzing the coupling effects between wavefields. The multi-scale decoupling in step S143 not only helps separate different wavefields in seismic waves but also further extracts valuable seismic wave characteristic data. Multi-scale decoupling improves signal resolution, providing more detailed support for real-time monitoring, analysis, and early warning of seismic signals. The combined processing of these multiple steps generates high-quality seismic signal data, enhancing the accuracy of earthquake monitoring and expanding its application potential in earthquake early warning, geological exploration, and structural safety assessment. By combining seismic wave classification, wavefield separation, and characteristic coupling analysis, the multi-scale decoupling process of seismic wave signals is optimized, enhancing the understanding of seismic wave propagation characteristics and providing a reliable data foundation for subsequent earthquake disaster monitoring and prediction.

[0021] Preferably, step S2 includes the following steps:

[0022] Step S21: Extract implicit reservoir features from the seismic multi-wave field separation data to obtain seismic implicit reservoir feature data; perform spatial distribution mapping on the seismic multi-wave field separation data based on the seismic implicit reservoir feature data to generate initial reservoir feature mapping data.

[0023] Step S22: Decouple the initial reservoir feature mapping data by physical properties to generate reservoir physical property data; perform nonlinear correlation analysis on the reservoir physical property data to generate high-order reservoir feature data; perform inversion linkage optimization on the high-order reservoir feature data and seismic multi-wave field separation data to generate reservoir dynamic response optimization data.

[0024] Step S23: Decompose the reservoir dynamic response optimization data into frequency components to generate a multi-frequency reservoir feature map; enhance the reservoir details of the multi-frequency reservoir feature map to generate a high-resolution reservoir detail display map.

[0025] This invention, through implicit reservoir feature extraction, can deeply explore the propagation characteristics of seismic waves in the subsurface medium, and the generated implicit reservoir feature data provides a more accurate foundation for reservoir feature analysis. By spatially mapping these features, they are applied to the spatial analysis of reservoirs, helping to determine the distribution of reservoirs over a wide area and providing clear spatial basis for reservoir exploration and development. Decoupling of reservoir physical properties allows for the separate analysis of the influence between different properties (such as porosity and elastic modulus), avoiding confusion between different properties. This decoupling analysis can more clearly demonstrate the physical characteristics of the reservoir, providing an accurate interpretation of the relationship between seismic wave signals and reservoir properties. Nonlinear correlation analysis further enhances the understanding of complex physical properties, making reservoir characteristic analysis more accurate. Through inversion-linked optimization, combining high-order reservoir feature data with seismic wave data can effectively optimize the dynamic response model of the reservoir. This helps to better understand the dynamic behavior of the reservoir, such as oil and gas flow or water permeability, thereby improving the efficiency of oil and gas field development and water resource utilization. Frequency component decomposition allows for the separate analysis of different frequency characteristics of a reservoir, revealing reservoir details and behaviors corresponding to different frequency components. Reservoir detail enhancement highlights key reservoir features, particularly in complex reservoir structures, helping engineers more clearly identify potential resource-rich areas or anomalies. The generated high-resolution reservoir detail maps provide detailed reservoir information, resulting in better visualization of reservoir spatial and physical properties. Through these high-resolution maps, geologists and engineers can more accurately assess, make decisions, and optimize reservoirs, providing crucial data support for oil and gas exploration, production processes, and resource assessment. In-depth analysis and optimization of seismic multifield separation data improves the accuracy of reservoir feature extraction, and by optimizing reservoir dynamic response and detail enhancement, more efficient and accurate reservoir assessment methods are provided. These analyses are of significant value for improving the analytical capabilities of seismic data and its applications in oil and gas exploration and other fields.

[0026] Preferably, spatial distribution mapping of seismic multi-wave field separation data based on seismic implicit reservoir characteristic data includes:

[0027] Seismic acquisition points were identified from the seismic multi-wave field separation data to obtain seismic acquisition point information data; geographic coordinates were then identified from the seismic acquisition point information data to obtain geographic spatial coordinate data.

[0028] Seismic multi-wave field separation data and geospatial coordinate data are spatially mapped to a grid to generate seismic multi-wave field grid-mapped data; the propagation range of the seismic multi-wave field grid-mapped data is evolved based on the implicit reservoir characteristics of seismic waves to generate seismic propagation evolution path data.

[0029] Based on earthquake propagation evolution path data, the grid cells of the earthquake multi-wave field grid mapping data are refined to generate seismic wave spatial propagation area data; local implicit mode extraction is performed on the seismic wave implicit reservoir characteristic data to obtain seismic wave local implicit mode data.

[0030] Based on the local implicit model data of seismic waves, the reservoir characteristic values ​​of the seismic wave spatial propagation area data are calculated by grid cell to obtain the reservoir characteristic values ​​of the grid cells; based on the reservoir characteristic values ​​of the grid cells, the spatial frame joint mapping of the seismic multi-wavefield grid mapping data is performed to generate the initial reservoir characteristic mapping data.

[0031] This invention ensures that all data is accurately mapped to actual geographical locations by confirming seismic acquisition points and obtaining their geographic coordinates. This not only provides accurate foundational information for subsequent data analysis and spatial mapping but also guarantees the reliability and consistency of seismic data. By mesh mapping seismic multi-wave field separation data with geospatial coordinate data, seismic data can be spatially divided into multiple grid cells. This refined spatial grid mapping helps provide higher-resolution seismic data, allowing for a more detailed representation and analysis of reservoir spatial distribution characteristics within local areas. By analyzing the propagation range evolution of the mesh-mapped data based on implicit reservoir characteristic data of seismic waves, the propagation path and evolution process of seismic waves can be revealed. This not only improves the understanding of seismic wave propagation mechanisms but also helps predict the propagation trend and impact range of seismic waves under different conditions. Local implicit pattern extraction can identify the unique propagation characteristics of seismic waves in different regions, thereby helping to deepen the understanding of the local behavior of seismic waves. Combining local implicit pattern data with the calculation of reservoir characteristic values ​​of grid cells can generate more accurate reservoir physical property data, providing a scientific basis for subsequent reservoir evaluation and optimization. Based on reservoir characteristic values ​​of grid cells, a joint spatial frame mapping can be performed to tightly integrate seismic multi-wave field grid data with reservoir characteristic values, generating initial reservoir characteristic mapping data. This joint mapping can better reflect the relationship between seismic waves and reservoir characteristics, providing a more intuitive and accurate model for the spatial distribution and physical property analysis of reservoirs. Through the above series of spatial distribution mapping and feature extraction processes, detailed and high-precision reservoir characteristic mapping data can be obtained. This helps to more accurately assess the spatial distribution, physical properties, and dynamic response of reservoirs, thereby optimizing reservoir exploration and development strategies and improving the utilization efficiency of oil and gas resources. By spatially mapping the implicit reservoir characteristic data of seismic waves, high-resolution and high-precision reservoir characteristic mapping data are provided, which not only improves the understanding of the seismic wave propagation process but also enhances the accuracy of reservoir assessment and optimization, and has important practical application value for oil and gas exploration, resource assessment, and development.

[0032] Preferably, the inversion and linkage optimization of high-order reservoir characteristic data and seismic multi-wavelength separation data includes:

[0033] Frequency domain analysis was performed on the seismic multi-wave field separation data to generate seismic full-wave frequency domain data; nonlinear wave equations were constructed on the seismic full-wave data based on the seismic full-wave frequency domain data to generate reservoir medium wave equations.

[0034] Finite difference calculations are performed based on the reservoir medium wave equation to obtain reservoir medium wave difference data; numerical simulation of the reservoir medium wave equation is then performed using the reservoir medium wave difference data to generate simulated seismic wave field data.

[0035] By simulating seismic wavefield data, high-order reservoir characteristic data are inverted and optimized to generate optimized dynamic response data for the reservoir.

[0036] This invention, through frequency domain analysis of seismic multi-wave field separation data, converts time-domain signals into frequency-domain data, revealing the propagation characteristics of different frequency components in the reservoir. This process facilitates refined analysis of reservoir wave characteristics, providing a necessary data foundation for subsequent wave equation construction and reservoir optimization. By constructing nonlinear wave equations for the reservoir medium based on full-wave frequency domain seismic data, the propagation behavior of seismic waves in the reservoir can be accurately described. Nonlinear wave equations can simulate the complex interaction between subsurface media and seismic waves; considering nonlinear effects, the simulation results are more realistic and accurate. After finite-difference calculation of the reservoir medium wave equations, wave difference data can be obtained. This data is further used for numerical simulation to generate simulated seismic wavefield data. This method can accurately simulate the propagation behavior of seismic waves in complex reservoir environments, providing scientific simulation results for subsequent reservoir response analysis and optimization. By inverting and optimizing high-order reservoir characteristic data using simulated seismic wavefield data, the dynamic response of the reservoir can be obtained more accurately. The inversion optimization process links reservoir properties, structure, and dynamic behavior. By matching simulated wavefield data with higher-order reservoir characteristics, it provides a more accurate reservoir response model. The generated optimized reservoir dynamic response data helps to accurately assess the reservoir's performance under different seismic wave actions, especially its response to wave action and its impact on hydrocarbon flow. The optimized response data not only improves the prediction accuracy of reservoir dynamic characteristics but also provides more precise decision support for oil and gas exploration and reservoir development. Through inversion-linked optimization, the dynamic response of the reservoir can be accurately obtained, further improving the efficiency of oil and gas field development. Optimization of reservoir dynamic response provides a more precise basis for oil and gas development planning and implementation, especially in the process of oil and gas reservoir exploitation, helping to predict and adjust production plans and improve resource utilization.

[0037] Preferably, step S23 includes the following steps:

[0038] Step S231: Perform wavelet packet decomposition on the reservoir dynamic response optimization data to generate low-frequency component data and high-frequency component data; perform geological labeling on the low-frequency component data and high-frequency component data to generate low-frequency component labeled data and high-frequency component labeled data.

[0039] Step S232: Reconstruct the frequency band features of the reservoir dynamic response optimization data to generate multi-frequency time-domain images; perform spatial feature mapping on the multi-frequency time-domain images to generate multi-frequency reservoir feature maps;

[0040] Step S233: Enhance the texture features of the multi-frequency reservoir feature map based on the high-frequency component annotation data to generate local texture feature enhancement data of the map;

[0041] Step S234: Based on the low-frequency component annotation data, perform trend fitting on the multi-frequency reservoir feature map to generate map trend fitting data; optimize the overall trend of the multi-frequency reservoir feature map using the map trend fitting data to generate map overall trend optimized data.

[0042] Step S235: Optimize and integrate the multi-frequency reservoir feature map based on the local texture feature enhancement data and the overall trend optimization data of the map to obtain a high-resolution reservoir detail map.

[0043] This invention effectively decomposes reservoir dynamic response optimization data into low-frequency and high-frequency components through wavelet packet decomposition, capturing the overall trend and detailed features of the reservoir response, respectively. The low-frequency components primarily reflect the large-scale behavior of the reservoir, revealing its macroscopic dynamic response; the high-frequency components capture the high-frequency changes in reservoir details, providing more precise clues for microstructural analysis. Geological labeling allows for a more accurate interpretation of these frequency components, helping to establish a direct correlation between physical phenomena and reservoir characteristics. Frequency band feature reconstruction of the reservoir dynamic response optimization data generates multi-frequency time-domain images, providing a detailed representation of the reservoir's time-domain characteristics across multiple frequency bands. This image clearly demonstrates the reservoir's time response characteristics at different frequencies, aiding in a better understanding of the reservoir's dynamic behavior. Through spatial feature mapping, the generated multi-frequency reservoir feature map provides crucial spatial information for subsequent analysis and processing, facilitating a global understanding of the reservoir's various characteristics. The labeled high-frequency components enhance the texture features of the map, thereby improving the local details of the reservoir map. Texture enhancement helps reveal microscopic features and heterogeneity in reservoirs, such as pore distribution and fracture structure. This process effectively improves the resolution of reservoir features, making the analysis of complex reservoirs more accurate. Labeled low-frequency components facilitate trend fitting of multi-frequency reservoir feature maps, leading to overall trend optimization. This process helps identify regular changes in reservoirs at large scales, providing a foundation for predicting overall reservoir behavior. Overall trend optimization improves the understanding of long-term reservoir evolution characteristics by removing noise and enhancing signal reliability. By combining local texture enhancement data and overall trend optimization data, multi-frequency reservoir feature maps are optimized and integrated to generate high-resolution reservoir detail maps. These optimized and integrated high-resolution images reveal rich reservoir details, including microstructure and macroscopic variations, with extremely high accuracy and clarity.

[0044] Preferably, step S233 includes the following steps:

[0045] Based on the high-frequency component annotation data, high-frequency texture regions are located in the multi-frequency reservoir feature map to generate high-frequency texture region data; feature regions are separated from the high-frequency texture region data to generate high-frequency texture separated region data.

[0046] The gray-level co-occurrence matrix is ​​calculated on the high-frequency texture separation region data to obtain the texture information data of the high-frequency texture region; local contrast enhancement is performed on the texture information data of the high-frequency texture region to generate texture enhancement data of the high-frequency texture region; texture component matching and alignment are performed on the high-frequency texture region data based on the texture enhancement data of the high-frequency texture region to generate high-frequency texture alignment region data.

[0047] By using high-frequency texture alignment region data, the multi-frequency reservoir feature map is updated with map details to generate local texture feature enhancement data.

[0048] This invention locates high-frequency texture regions in multi-frequency reservoir feature maps based on high-frequency component annotation data. This accurately identifies detailed parts of the reservoir map, especially high-frequency texture information, which is crucial for revealing subtle changes and details within the reservoir. Feature region separation of the high-frequency texture region data further refines the high-frequency texture information, allowing different texture regions to be analyzed independently, improving the accuracy and targeting of subsequent processing. Calculating the gray-level co-occurrence matrix of the separated high-frequency texture regions helps extract the spatial arrangement patterns of the textures. This process reveals texture features in the reservoir map, such as roughness, uniformity, and contrast, which are essential for revealing the reservoir's microstructure. The obtained high-frequency texture region texture information data provides a reliable foundation for subsequent local contrast enhancement and texture enhancement. Local contrast enhancement of the high-frequency texture region texture information data highlights the details and features of the texture in the reservoir map, making the reservoir's microstructure clearer. Local contrast enhancement helps amplify texture features, making important details in the reservoir more prominent and facilitating the identification of potential reservoir heterogeneity regions. Texture component matching and alignment are performed based on high-frequency texture enhancement data to ensure effective alignment and fusion of texture information from different frequencies within the same map. This process helps eliminate deviations between different frequency information, ensuring the overall consistency and accuracy of the map. Detailed updates are then performed on the multi-frequency reservoir feature map using high-frequency texture alignment region data, further improving the map's clarity and resolution, making the reservoir's texture features more accurately reflect its actual physical properties and structural characteristics. The resulting enhanced local texture feature data provides a more refined and clearer display of reservoir details, especially in high-frequency texture regions, further improving the map's operability and visualization. The enhanced map effectively reveals the reservoir's microscopic heterogeneity, providing more accurate data support for reservoir analysis and evaluation.

[0049] Preferably, step S3 includes the following steps:

[0050] Step S31: Perform dynamic reservoir feature change analysis on the high-resolution reservoir detail display map to generate dynamic reservoir feature change data; divide the dynamic reservoir feature change data into a dataset to generate a model training set and a model test set;

[0051] Step S32: Use a convolutional neural network algorithm to train the model on the training set to generate a pre-model for predicting reservoir characteristic changes; use a model test set to optimize and iterate the pre-model for predicting reservoir characteristic changes to generate a model for predicting reservoir characteristic changes.

[0052] Step S33: Import the dynamic reservoir characteristic change data into the reservoir characteristic change prediction model to predict the reservoir characteristic change, thereby generating reservoir characteristic change prediction data.

[0053] Step S34: Dynamically update the high-resolution reservoir detail map using reservoir feature change prediction data to generate a reservoir feature change prediction map.

[0054] This invention analyzes dynamic reservoir characteristic changes using high-resolution reservoir detail maps, enabling the identification and extraction of temporal trends in reservoir characteristics such as pressure, porosity, and fluid distribution. This process helps acquire time-series data on reservoir changes, providing data support for subsequent predictive modeling. By partitioning the dynamic reservoir characteristic change data into training and test sets, the prediction model can be effectively trained and its performance validated, ensuring the model's generalization ability and prediction accuracy. Training the model on the training set using a CNN algorithm automatically learns complex patterns of reservoir characteristic changes from the data. Especially when dealing with high-dimensional data, CNN effectively captures local spatial and temporal features, improving prediction accuracy. Iterative optimization of the model using the test set continuously improves its predictive ability, reducing overfitting or underfitting, ensuring that the generated reservoir characteristic change prediction model accurately reflects the actual changes in the reservoir. After model training and optimization, dynamic reservoir characteristic change data is imported into the reservoir characteristic change prediction model for prediction. This allows for accurate simulation of the evolution trend of reservoir characteristics, such as predicting changes in porosity and fluid flow in the reservoir over future time periods. This predictive capability is crucial for anticipating reservoir changes. Based on the reservoir characteristic change prediction data, a reservoir characteristic change prediction map is generated by dynamically updating a high-resolution reservoir detail map, clearly showing the reservoir's state changes at future moments. This process helps decision-makers assess the optimal timing for reservoir development, optimize production strategies, and predict extraction risks. Accurate reservoir characteristic change prediction not only improves the accuracy of reservoir assessment but also provides data support for subsequent reservoir management and extraction decisions. For example, predicting changes in reservoir pressure, temperature, and fluid flow can effectively help engineers adjust extraction plans or design new reservoir development schemes. The prediction map provides a visual way to help geologists and engineers intuitively understand the dynamic changes of the reservoir, improving the efficiency and scientific basis of decision-making and reducing uncertainty in the development process. By dynamically predicting changes in reservoir characteristics, potential reservoir problems (such as decreased fluid permeability or abnormal pressure) can be identified in advance, thereby enabling preventive measures to be taken, risks to be reduced, and production plans to be optimized.

[0055] Preferably, step S4 includes the following steps:

[0056] Step S41: Assess the reservoir potential of the reservoir characteristic change prediction map and generate reservoir potential assessment data; classify the reservoirs based on the reservoir potential assessment data and generate a reservoir classification map.

[0057] Step S42: Perform actual reservoir observations on the reservoir classification map to generate actual reservoir observation result data; calculate the confidence level of the actual reservoir observation result data and the reservoir classification map to generate reservoir prediction confidence level data;

[0058] Step S43: Compare the reservoir prediction confidence data with the preset confidence threshold. When the reservoir prediction confidence data is greater than or equal to the preset confidence threshold, output the map based on the reservoir classification map to generate a high-resolution reservoir prediction distribution map. When the reservoir prediction confidence data is less than the preset confidence threshold, return to step S3 for feature re-extraction until the reservoir prediction confidence data is greater than or equal to the preset confidence threshold.

[0059] This invention quantifies reservoir potential by analyzing reservoir characteristic change prediction maps, providing a basis for subsequent development decisions. Accurate reservoir potential assessment identifies highly productive oil or gas areas, maximizing resource utilization. Based on reservoir potential assessment data, reservoirs are classified into different categories (e.g., high-yield, low-yield, or high-development-difficulty areas). This allows engineers to develop targeted development plans and optimize extraction schedules. Actual reservoir observations provide field data, which is compared with the reservoir classification maps to verify the accuracy of predictions. The introduction of actual reservoir observation data makes reservoir classification and prediction more closely reflect reality, reducing prediction errors. Combined with confidence level calculations, comparing actual observation data with reservoir classification maps allows for the calculation of reservoir prediction confidence. This process helps identify which reservoir areas have more reliable predictions, providing a scientific basis for decision-making. By comparing with preset confidence thresholds, reservoir areas with high confidence levels can be selected, ensuring the accuracy and reliability of predictions. If the reservoir prediction confidence data exceeds a set threshold, a high-resolution reservoir prediction distribution map is generated, providing clear direction for adjusting the exploitation strategy. When the confidence of the prediction result is insufficient, it automatically returns to step S3 for feature re-extraction and optimization until the confidence requirement is met. This feedback mechanism ensures the reliability and scientific validity of the final output reservoir prediction distribution map. Through confidence calculation and feedback adjustment, step S4 achieves adaptive optimization, thereby improving the accuracy of the prediction results. The generation of reservoir classification maps and high-resolution reservoir prediction distribution maps helps decision-makers more accurately identify the reservoir characteristics and development potential of different areas. Before exploitation, high-resolution reservoir prediction maps allow for early prediction of dynamic changes in the reservoir, helping to avoid risks. Reservoir classification maps and reservoir prediction distribution maps provide visual support for exploitation plans. Based on these maps, engineers can develop more precise exploitation plans, such as prioritizing the development of high-yield areas and adjusting development strategies to address the challenges of low-yield or difficult-to-develop areas. This process reduces errors from human intervention, making the entire reservoir assessment and development process more scientific and refined. Through progressive optimization of confidence levels and a feedback mechanism, the extraction and prediction of reservoir characteristics can be continuously improved, reducing losses caused by erroneous predictions during development. The final high-resolution reservoir prediction distribution map facilitates more accurate resource allocation, avoiding resource waste or exploitation of uneconomical areas. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the steps involved in a high-resolution reservoir prediction method based on seismic full-wave information mining.

[0061] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0062] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S3.

[0063] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0064] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0065] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0066] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0067] To achieve the above objectives, please refer to Figures 1 to 3 A high-resolution reservoir prediction method based on seismic full-wave information mining, the method comprising the following steps:

[0068] Step S1: Acquire seismic full-wave information data; perform frequency domain joint analysis on the seismic full-wave information data to generate multi-scale full-wave feature tensors; perform multi-scale decoupling of the seismic full-wave signal on the multi-scale full-wave feature tensors to generate seismic multi-wave field separation data.

[0069] Step S2: Perform implicit reservoir feature mapping on the seismic multi-wavelength separation data to generate initial reservoir feature mapping data; perform nonlinear correlation analysis on the initial reservoir feature mapping data to generate high-order reservoir feature data; perform inversion linkage optimization on the high-order reservoir feature data and the seismic multi-wavelength separation data to generate optimized reservoir dynamic response data; perform reservoir detail enhancement on the optimized reservoir dynamic response data to generate a high-resolution reservoir detail display map;

[0070] Step S3: Perform dynamic reservoir feature change analysis on the high-resolution reservoir detail map to generate dynamic reservoir feature change data; predict reservoir feature changes on the high-resolution reservoir detail map based on the dynamic reservoir feature change data to generate a reservoir feature change prediction map.

[0071] Step S4: Classify the reservoirs in the reservoir feature change prediction map to generate a reservoir classification map; calculate the confidence level of the reservoir classification map to generate reservoir prediction confidence data; compare the reservoir prediction confidence data with the preset confidence threshold and output the results until a high-resolution reservoir prediction distribution map is generated.

[0072] This invention, by acquiring full-wave information and performing joint frequency domain analysis, can extract rich subsurface information from multiple scales and wavefields, overcoming the limitations of traditional methods that rely solely on a single waveform signal. Multi-scale decoupling and wavefield separation techniques further enhance signal resolution, allowing for independent processing of features from different wavefields, providing clearer foundational data for subsequent high-precision reservoir prediction. Implicit reservoir feature mapping and nonlinear correlation analysis reveal deep-seated reservoir characteristic relationships, generating more accurate high-order reservoir feature data. The inversion-linked optimization method establishes close connections between multi-wavefield data, generating more realistic reservoir dynamic responses. Reservoir detail enhancement techniques improve reservoir structure resolution, ensuring high-precision detail display and thus improving the accuracy of reservoir prediction. Dynamic reservoir feature change analysis can monitor reservoir change trends in real time and identify potential change patterns. Based on this, the reservoir feature change prediction map provides a forward-looking prediction of future reservoir evolution, helping to make more accurate development decisions and effectively improving the dynamic management capabilities of reservoirs. Through reservoir classification and confidence calculation, precise regional division and risk assessment of reservoirs can be performed, further improving the accuracy of reservoir prediction. Confidence assessment ensures the reliability and credibility of the prediction results, avoiding errors caused by uncertainty. The resulting high-resolution reservoir prediction distribution map provides a reliable basis for oil and gas exploration and development. Therefore, this invention improves the accuracy and resolution of reservoir prediction through multi-wave field information extraction, high-resolution reservoir detail display, dynamic reservoir change prediction, and confidence assessment.

[0073] In this embodiment of the invention, reference Figure 1The above is a flowchart illustrating the steps of a high-resolution reservoir prediction method based on seismic full-wave information mining according to the present invention. In this example, the high-resolution reservoir prediction method based on seismic full-wave information mining includes the following steps:

[0074] Step S1: Acquire seismic full-wave information data; perform frequency domain joint analysis on the seismic full-wave information data to generate multi-scale full-wave feature tensors; perform multi-scale decoupling of the seismic full-wave signal on the multi-scale full-wave feature tensors to generate seismic multi-wave field separation data.

[0075] In this embodiment of the invention, full-wave information is acquired using seismic detection equipment (such as seismographs). These devices typically acquire signals including different types of seismic waves such as P-waves, S-waves, and surface waves. The acquired raw seismic data is stored in a database, and preprocessing processes include denoising, time synchronization, and calibration to ensure data quality. Spectral analysis techniques, such as Fast Fourier Transform (FFT) or wavelet transform, are used to convert the time-domain seismic wave data into the frequency domain. Each type of seismic wave has different frequency characteristics, and frequency domain analysis helps to separate waveform features in different frequency bands. Wavelet transform or multi-scale Fourier analysis methods are used to process the signal to obtain wave characteristics at different frequency and time scales. Each scale represents a different resolution of the signal, which helps to capture the propagation characteristics of seismic waves at different time and spatial scales. The data obtained from multi-scale analysis are integrated into a multi-dimensional feature tensor, where each element represents the wave intensity or other characteristics (such as amplitude, phase, etc.) at different frequencies and time scales. Decoupling algorithms (such as Principal Component Analysis (PCA) and Independent Component Analysis (ICA) are employed to process multi-scale full-wave characteristic tensors. The goal is to decompose complex seismic full-wave information into multiple independent signal components (multi-wavefield separation). Each component corresponds to a specific wavefield (e.g., P-wave, S-wave, Rayleigh wave, etc.). Through the decoupling process, different types of wavefield signals are separated, and each wavefield signal is reconstructed for subsequent analysis or prediction. Multi-wavefield separated data can more accurately reflect the characteristics of different wavefields, facilitating subsequent seismic wavefield analysis, source location, and seismic model construction.

[0076] Step S2: Perform implicit reservoir feature mapping on the seismic multi-wavelength separation data to generate initial reservoir feature mapping data; perform nonlinear correlation analysis on the initial reservoir feature mapping data to generate high-order reservoir feature data; perform inversion linkage optimization on the high-order reservoir feature data and the seismic multi-wavelength separation data to generate optimized reservoir dynamic response data; perform reservoir detail enhancement on the optimized reservoir dynamic response data to generate a high-resolution reservoir detail display map;

[0077] In this embodiment of the invention, multiple wavefield signals extracted from seismic multi-wavefield separation data are used as input. These wavefield signals contain preliminary information about the reservoir, including the propagation characteristics of seismic waves, reflected waves, and refracted waves. Implicit mapping methods (such as deep learning algorithms based on neural networks, support vector machines, etc.) are used to map the multi-wavefield separation data to the reservoir's feature space. Implicit mapping can uncover the complex relationship between seismic data and reservoir physical properties (such as porosity, permeability, etc.). Through this process, initial reservoir feature mapping data is generated, which reflects the coarse properties and hierarchical structure of the reservoir. The initial reservoir feature mapping data is analyzed using nonlinear models (such as deep neural networks, convolutional neural networks, random forests, etc.). At this point, nonlinear analysis helps extract complex patterns in the data that cannot be revealed by linear methods, especially the influence of seismic wave propagation paths, reflection intensities, etc., on reservoir properties. The relationship between different reservoir characteristics (such as rock type, reservoir geometry, etc.) and seismic wave characteristics is modeled to reveal nonlinear relationships. For example, there is a complex nonlinear mapping relationship between reflection intensities and porosity or permeability in different regions of the reservoir. This process generates high-order reservoir characteristic data, containing more refined reservoir properties such as interlayer variations and fracture characteristics. Based on this high-order reservoir characteristic data, an inversion model is constructed. Using seismic inversion techniques (such as least squares inversion and Bayesian inversion), the seismic data and reservoir characteristics are linked and optimized during the inversion process, iteratively adjusting model parameters. During the inversion process, the high-order reservoir characteristic data is linked with seismic multi-wavefield separation data. By optimizing the model, the error between the wavefield data and the reservoir response is minimized. This linked optimization accurately reflects the dynamic response of the reservoir, especially its response characteristics under different seismic events. Finally, through inversion and optimization, optimized dynamic response data of the reservoir is obtained. This data accurately reflects the dynamic behavior of the reservoir under different conditions, such as stress changes and fluid permeability. Data augmentation techniques (such as super-resolution reconstruction and image processing algorithms) are used to enhance the details of the optimized dynamic response data. These techniques can improve the resolution of reservoir characteristics by supplementing missing information and amplifying subtle changes in the reservoir. Detailed optimization algorithms are applied to enhance the representation of important reservoir features, such as fracture networks, pore structure, and permeability distribution. This process involves image filtering, denoising, and edge detection techniques to highlight the details within the reservoir. Finally, the enhanced reservoir details are presented as high-resolution visualizations, which can be used for further analysis and decision support, helping geologists and engineers to more accurately understand the reservoir structure and properties.

[0078] Step S3: Perform dynamic reservoir feature change analysis on the high-resolution reservoir detail map to generate dynamic reservoir feature change data; predict reservoir feature changes on the high-resolution reservoir detail map based on the dynamic reservoir feature change data to generate a reservoir feature change prediction map.

[0079] In this embodiment of the invention, a set of high-resolution reservoir detail images is acquired. These images typically display detailed reservoir features, including the spatial distribution of attributes such as porosity, permeability, and fracture networks. If reservoir detail images at multiple time points are available, a time-series dataset can be constructed, with each image reflecting the current state of the reservoir. If data at multiple time points is unavailable, hypothetical time-series data can be generated through simulation or other means. Image analysis techniques, such as image differentiation, edge detection, and image registration (e.g., phase-correlation-based image registration methods), are used to detect feature changes in the reservoir detail images. For example, by comparing changes in images at different time points, dynamic changes in features such as porosity, fractures, and permeability in the reservoir can be extracted. The detected changes are quantified and recorded to generate dynamic reservoir feature change data. This data typically includes the magnitude, direction, and rate of reservoir attribute changes in the time series, and also involves local regional trends (such as fracture propagation or changes in porosity). Based on dynamic reservoir characteristic change data, appropriate prediction methods are selected, such as time series prediction based on machine learning, convolutional neural networks (CNN) or long short-term memory networks (LSTM) in deep learning, or mathematical models based on traditional numerical methods such as the finite difference method. The model is trained using historical data and dynamic reservoir characteristic change data. The goal of this model is to learn the patterns of reservoir characteristic changes over time and accurately predict future changes in reservoir characteristics. The prediction model is optimized through cross-validation, error evaluation, and other methods to ensure that it can effectively capture the dynamic evolution of reservoir characteristics. The trained prediction model is applied to high-resolution reservoir detail maps, predicting changes in reservoir characteristics at future time points based on the current reservoir state (image data) and historical change data. This process can output predicted changes in porosity, permeability, fracture propagation, etc. Reservoir characteristic change prediction maps are generated based on the model prediction results. These prediction maps will show the state of the reservoir at future time points or under different environmental conditions, reflecting the changing trends in different areas (such as areas of fracture propagation, areas of increased or decreased porosity, etc.). The prediction maps are visualized to facilitate further analysis and decision-making by geologists, engineers, and other expert teams. The changes in the reservoir can be displayed through color changes, 3D graphics, or dynamic charts.

[0080] Step S4: Classify the reservoirs in the reservoir feature change prediction map to generate a reservoir classification map; calculate the confidence level of the reservoir classification map to generate reservoir prediction confidence data; compare the reservoir prediction confidence data with the preset confidence threshold and output the results until a high-resolution reservoir prediction distribution map is generated.

[0081] In this embodiment of the invention, a specific classification algorithm is selected based on data (such as changes in porosity, permeability, and fracture density) from a reservoir feature change prediction map. Commonly used classification methods include Support Vector Machine (SVM), Random Forest (RF), Decision Tree, k-means clustering, and deep learning (such as Convolutional Neural Networks (CNN)). The reservoir feature change prediction map is converted into a data format for the specific input classification algorithm, such as feature vectors or image pixels (which may be two-dimensional or three-dimensional data depending on the prediction map). The classification model is trained using known reservoir data (such as historical data and actual exploration results), and the model learns the relationship between reservoir feature changes and categories. For example, the reservoir is divided into different categories, such as "high-permeability zone," "low-permeability zone," and "fracture zone." The trained classification model is applied to classify the reservoir feature change prediction map to obtain reservoir category information for each region or pixel. These categories can represent the distribution of different reservoir attributes, such as porosity, permeability, and fracture density. Based on the classification results, a reservoir classification map is generated. This atlas uses color or different regions to represent different reservoir categories. It can be a two-dimensional image or a three-dimensional model, showing the spatial distribution of different reservoir categories. Confidence typically represents the reliability of the classification result, i.e., the probability that the model predicts the reservoir category. For example, the output probability of the classifier can be used to define the confidence of each predicted region. For deep learning models, the predicted probability of each pixel is usually obtained (e.g., the probability of belonging to a certain category). For each predicted region or pixel, its confidence of belonging to a certain category is calculated. For example, maximum a posteriori probability (MAP) or Bayesian inference methods are used to calculate the confidence score for each reservoir category. The calculated confidence values ​​are assigned to each region in the reservoir classification atlas to obtain reservoir prediction confidence data. This data provides a confidence score for each reservoir region or pixel, representing the reliability of the classification result. Based on the confidence calculation results, a confidence map is generated. This map can be a two-dimensional image, where the value of each pixel represents the confidence of that region. Colors can represent high confidence levels; for example, dark colors represent high confidence, and light colors represent low confidence. Based on practical application requirements, a confidence threshold is set. This threshold is used to filter reliable areas for reservoir prediction. The threshold can be determined through historical data or expert experience, typically set to 0.7, 0.8, or 0.9, indicating that the prediction result is considered reliable only when the confidence level exceeds this value. The confidence value of each region or pixel is compared with the preset threshold. If the confidence level is higher than the threshold, the reservoir prediction for that region is considered reliable; if it is lower than the threshold, the prediction result is considered unreliable or requires further confirmation. Based on the comparison results of the confidence level and the threshold, a high-resolution reservoir prediction distribution map is generated. The map will display all regions with confidence levels exceeding the threshold; these regions represent reliable reservoir prediction results. High-confidence regions are highlighted using different colors or different transparency indicators.This map can help geological engineers, exploration teams, or development decision-makers to visually see reliable reservoir distributions and development areas.

[0082] Preferably, step S1 includes the following steps:

[0083] Step S11: Acquire full-wave seismic information data using a multi-point distributed seismic signal acquisition array;

[0084] Step S12: Perform data preprocessing on the seismic full-wave information data to generate standard seismic full-wave information data. The data preprocessing includes data cleaning, missing value imputation, and data standardization.

[0085] Step S13: Perform time-domain analysis on the full-wave signal to generate full-wave signal time-domain data; perform fast Fourier transform on the full-wave signal time-domain data to generate full-wave signal frequency-domain data; perform joint frequency-domain analysis based on the full-wave signal time-domain data and full-wave signal frequency-domain data to generate multi-scale full-wave feature tensors.

[0086] Step S14: Based on the preset seismic wave propagation characteristic coefficients, perform multi-scale decoupling of the seismic full-wave signal on the multi-scale full-wave characteristic tensor to generate seismic multi-wave field separation data.

[0087] In this embodiment of the invention, a multi-point distributed seismic signal acquisition array is deployed within the target area. Each acquisition point should be equipped with a highly sensitive seismic sensor capable of capturing full-wave information generated during seismic activity in real time. The acquired data includes information such as the amplitude, frequency, and phase of seismic waves, covering different types of seismic waves from P-waves and S-waves to surface waves, forming a full-wave information dataset. The multi-point array should cover different geographical areas to ensure comprehensive capture of different seismic wave propagation modes. Noise data is removed, and outliers caused by sensor malfunctions or data transmission errors are corrected. Interpolation algorithms (such as linear interpolation and spline interpolation) are used to fill in missing data in the seismic signal due to equipment failure or transmission problems. Data acquired from different seismic stations are standardized, for example, by processing the data according to a standardized range (e.g., mean 0, variance 1) to reduce differences caused by different equipment or environments and ensure data comparability. The temporal variation trend of the full-wave signal is analyzed, and irrelevant frequency components are removed using time-domain filtering techniques, retaining the main seismic wave information to generate time-domain data. The time-domain data is converted into frequency-domain data to reveal the frequency components in the signal. The FFT algorithm effectively extracts the amplitude and phase of each frequency component in a signal. By jointly analyzing time-domain and frequency-domain data, multi-scale information is extracted based on waveform variation features. Wavelet transform or other multi-scale methods are used to analyze signal characteristics at different time and frequency scales, generating a multi-scale full-wave feature tensor. This feature tensor is a key data structure for seismic signal identification, classification, and further analysis. Using seismological theory or experimental data, propagation characteristic coefficients for different types of seismic waves are determined, including wave velocity, propagation path, and attenuation coefficient. Based on these seismic wave propagation characteristic coefficients, decoupling algorithms (such as principal component analysis or matrix factorization) are used to decompose the multi-scale full-wave feature tensor, separating various seismic wavefields (such as P-waves, S-waves, and surface waves) from the original signal. The decoupled results provide independent signals for each wavefield, reflecting the characteristics of different types of seismic waves in the time and frequency domains. The resulting multi-wavefield seismic data is used for subsequent seismic wave analysis and disaster prediction.

[0088] Preferably, step S14 includes the following steps:

[0089] Step S141: Perform seismic wave propagation velocity analysis on the multi-scale full-wave characteristic tensor to generate seismic wave propagation velocity data; classify the seismic wave types based on the seismic wave propagation velocity data on the multi-scale full-wave characteristic tensor to generate seismic wave type classification data, which includes P-wave data, S-wave data and surface wave data.

[0090] Step S142: Perform variational mode decomposition on the P-wave data, S-wave data, and surface wave data to generate P-wave mode decomposition data, S-wave mode decomposition data, and surface wave mode decomposition data; perform wavefield separation on the P-wave data, S-wave data, and surface wave data using the P-wave mode decomposition data, S-wave mode decomposition data, and surface wave mode decomposition data to generate seismic full-wave wavefield separation data;

[0091] Step S143: Perform wavefield feature coupling analysis on the seismic full-wave wavefield separation data to generate wavefield feature coupling data; use the wavefield feature coupling data to decouple the seismic full-wave signal at multiple scales using the multi-scale full-wave feature tensor to generate seismic multi-wavefield separation data.

[0092] In this embodiment of the invention, the propagation velocities of seismic waves of different types are obtained based on the propagation characteristics of seismic waves, using existing seismological theories or experimental data. The specific propagation velocity can be estimated by the time delay of seismic wave propagation in the Earth's crust. Seismic wave propagation velocity data is generated by calculating the propagation velocities of signals in different frequency bands within the multi-scale full-wave feature tensor. This data helps distinguish different wave types, especially when there are significant differences in wave velocity (e.g., P-waves, S-waves, and surface waves). Based on the seismic wave propagation velocity data, classification algorithms (e.g., support vector machines or K-means clustering) are applied to classify the multi-scale full-wave feature tensor into different wave types. According to the propagation velocity of seismic waves in the medium, the signals can be divided into three main types: P-waves (P-waves), S-waves (S-waves), and surface waves, generating seismic wave type classification data. This data contains the signals corresponding to each type of seismic wave, with P-wave, S-wave, and surface wave data independently labeled for easy subsequent processing. For P-wave, S-wave, and surface wave data, variational mode decomposition (VMD) is used for mode decomposition. Virtual Mode Decomposition (VMD) can decompose nonlinear, non-stationary signals into several intrinsic mode functions (IMFs), which represent the main oscillatory components of the signal. VMD is performed on P-wave, S-wave, and surface wave signals to obtain P-wave mode decomposition data, S-wave mode decomposition data, and surface wave mode decomposition data, respectively. These data capture multiple frequency and time-domain features of different waveforms. Based on the mode decomposition data, multi-channel wavefield separation techniques (such as Independent Component Analysis (ICA) or Non-negative Matrix Factorization (NMF)) are used to further separate the wavefields of P-wave, S-wave, and surface wave signals. The goal of this process is to extract the independent contribution of each wave type and remove redundant or interfering signals. This method allows for clearer separation of each wavefield in the seismic wave, generating the final seismic full-wave wavefield separation data. Coupled analysis is then performed on the seismic full-wave wavefield separation data to discover the interrelationships or common features between different wavefields. By employing methods such as cross-correlation analysis and mutual information analysis, the coupling characteristics among P-waves, S-waves, and surface waves are identified, generating wavefield characteristic coupling data. This data contains coupling information of different wavefields at multiple scales, helping to reveal the interactions between wavefields and their overall impact on seismic wave propagation. Using the generated wavefield characteristic coupling data, combined with previous seismic wave type classification data and mode decomposition data, further multi-scale decoupling of the multi-scale full-wave characteristic tensor is performed through decoupling algorithms (such as frequency domain decoupling or wavelet packet decoupling). The goal of this process is to separate multiple independent wavefield data from complex seismic wave signals, accurately reflecting the propagation characteristics of each type of seismic wave at different scales. Through this decoupling step, final seismic multi-wavefield separation data is generated, containing independent signals of all seismic wave types, which can be used for further seismic analysis and prediction.

[0093] As an example of the present invention, reference is made to Figure 2 As shown, in this example, step S2 includes:

[0094] Step S21: Extract implicit reservoir features from the seismic multi-wave field separation data to obtain seismic implicit reservoir feature data; perform spatial distribution mapping on the seismic multi-wave field separation data based on the seismic implicit reservoir feature data to generate initial reservoir feature mapping data.

[0095] Step S22: Decouple the initial reservoir feature mapping data by physical properties to generate reservoir physical property data; perform nonlinear correlation analysis on the reservoir physical property data to generate high-order reservoir feature data; perform inversion linkage optimization on the high-order reservoir feature data and seismic multi-wave field separation data to generate reservoir dynamic response optimization data.

[0096] Step S23: Decompose the reservoir dynamic response optimization data into frequency components to generate a multi-frequency reservoir feature map; enhance the reservoir details of the multi-frequency reservoir feature map to generate a high-resolution reservoir detail display map.

[0097] In this embodiment of the invention, implicit features related to reservoir characteristics are extracted from seismic multi-wave field separation data using deep learning or traditional seismic processing algorithms (such as reflection coefficient method, time-frequency analysis, etc.). These features include reflection intensity, reflection time, amplitude variation, phase information, etc. Automatic feature extraction is performed using deep neural networks (such as convolutional neural networks CNN) or other machine learning models to identify implicit reservoir features from seismic signals, such as lithology, porosity, and aquifers, generating seismic wave implicit reservoir feature data. This is a preliminary identification of the reservoir, containing reservoir information reflected during seismic wave reflection and propagation. Based on the seismic wave implicit reservoir feature data, the spatial distribution characteristics of the seismic signal are used for mapping, and the feature data is mapped to three-dimensional space using a Geographic Information System (GIS) or spatial interpolation algorithms. Methods such as Kriging interpolation or Inverse Distance Weighted (IDW) can be used to spatially map the extracted implicit reservoir features, generating initial reservoir feature mapping data. This data displays the preliminary characteristic distribution of the reservoir in three-dimensional space, including attributes such as layer thickness and porosity. Physical property decoupling is then performed on the initial reservoir feature mapping data. Typically, reservoir characteristics are decoupled using physical property parameters such as wave velocity, density, porosity, and permeability. Physical modeling methods or machine learning methods (such as Principal Component Analysis (PCA)) can be used to extract information related to physical properties from seismic wave data. Common decoupling methods include using wave velocity and density inversion algorithms to identify lithological variations, fluid states, porosity, etc., in the reservoir, generating reservoir physical property data that displays the reservoir's lithological characteristics, porosity, permeability, and other physical properties. Nonlinear correlation analysis is then performed on the reservoir physical property data to explore the complex relationships between different physical properties. For example, the nonlinear relationship between porosity and permeability, or the correlation between lithology and fluid saturation, can be analyzed using methods such as nonlinear regression analysis, support vector machines (SVM), and neural networks (e.g., deep autoencoders) to identify higher-order nonlinear characteristics between different physical properties, generating higher-order reservoir characteristic data. This data can reflect deeper physical phenomena within the reservoir and reveal its complexity. Inversion and optimization can be performed by combining higher-order reservoir characteristic data with seismic multi-wavelength separation data. Seismic inversion techniques can be used to jointly optimize reservoir characteristic data with seismic wave reflection data, improving the accuracy of the reservoir model. Inversion methods such as waveform inversion, least squares inversion, and full waveform inversion can be used to optimize the dynamic response of the reservoir, generating optimized reservoir dynamic response data. This data can accurately predict the reservoir's response under different pressure and fluid conditions, helping to optimize oil and gas exploration and production strategies. Frequency component decomposition can be performed on the optimized reservoir dynamic response data. Methods such as Fourier transform, wavelet transform, or EMD (Empirical Mode Decomposition) can be used to decompose the signal into different frequency components to help identify the reservoir's response characteristics at different frequencies.The decomposed data reveals the high-frequency and low-frequency response characteristics of the reservoir, aiding in the analysis of different reservoir properties, such as physical elasticity and fluid state. A multi-frequency reservoir feature map is generated, displaying the characteristic distribution of the reservoir at different frequencies and providing more comprehensive reservoir information. Based on the multi-frequency reservoir feature map, reservoir detail enhancement is performed. Image processing techniques such as super-resolution reconstruction and detail enhancement filtering can be used to enhance low-resolution details in the map. High-resolution reconstruction algorithms are employed to elevate the reservoir's microstructure and details to a higher spatial resolution, thereby better showcasing the reservoir's detailed features and generating a high-resolution reservoir detail map. This map displays the reservoir's microstructural features, helping to further analyze reservoir performance, such as recoverability, permeability, and production potential.

[0098] Preferably, spatial distribution mapping of seismic multi-wave field separation data based on seismic implicit reservoir characteristic data includes:

[0099] Seismic acquisition points were identified from the seismic multi-wave field separation data to obtain seismic acquisition point information data; geographic coordinates were then identified from the seismic acquisition point information data to obtain geographic spatial coordinate data.

[0100] Seismic multi-wave field separation data and geospatial coordinate data are spatially mapped to a grid to generate seismic multi-wave field grid-mapped data; the propagation range of the seismic multi-wave field grid-mapped data is evolved based on the implicit reservoir characteristics of seismic waves to generate seismic propagation evolution path data.

[0101] Based on earthquake propagation evolution path data, the grid cells of the earthquake multi-wave field grid mapping data are refined to generate seismic wave spatial propagation area data; local implicit mode extraction is performed on the seismic wave implicit reservoir characteristic data to obtain seismic wave local implicit mode data.

[0102] Based on the local implicit model data of seismic waves, the reservoir characteristic values ​​of the seismic wave spatial propagation area data are calculated by grid cell to obtain the reservoir characteristic values ​​of the grid cells; based on the reservoir characteristic values ​​of the grid cells, the spatial frame joint mapping of the seismic multi-wavefield grid mapping data is performed to generate the initial reservoir characteristic mapping data.

[0103] In this embodiment of the invention, the location of each acquisition point is confirmed based on the acquisition device from which the seismic signal originates, such as a seismograph or vibration meter. The GPS or other positioning system of the seismic signal acquisition device can be used to record the specific location of each acquisition point. Seismic data processing software is used to process the seismic signal to determine the coordinates of the acquisition points and extract the acquisition point information data. The result is seismic acquisition point information data, including the acquisition point number, acquisition time, and acquisition signal type. Using the GPS information recorded in the seismic acquisition point information data, the accurate geographic coordinates (longitude, latitude, elevation, etc.) of each acquisition point are confirmed. Geographic Information System (GIS) software can be used to interface the acquisition point data with a map coordinate system to obtain the geospatial coordinate data of each acquisition point. This data includes the precise location of each acquisition point, providing a basis for subsequent spatial mapping. The seismic multi-wavelength separation data is combined with the geospatial coordinate data, and based on a geographic coordinate system (such as a UTM coordinate system or a latitude and longitude coordinate system), the data is mapped to a predetermined grid system. Spatial interpolation algorithms (such as Kriging interpolation or inverse distance weighting) are used to map seismic data onto a uniform grid, ensuring that each seismic data point corresponds to a grid cell in space. The generated seismic multi-wavefield grid mapping data can accurately represent the multi-wavefield characteristics at different locations during seismic wave propagation. Based on implicit reservoir characteristic data of seismic waves (such as wave velocity, reflectivity, and absorption characteristics), the propagation range evolution of the seismic multi-wavefield grid mapping data is analyzed. Wave equations or physics-based seismic wave propagation models are used to simulate the propagation process of seismic waves under different geological conditions, generating seismic propagation evolution path data. This data describes the propagation path and evolution process of seismic waves from the source to each acquisition point. Based on the seismic propagation evolution path data, the seismic multi-wavefield grid mapping data is refined. Through sensitivity analysis of propagation path and wave velocity changes, more refined grid division is performed on areas affecting propagation. Grid cells can be locally densified based on propagation velocity, wave type, and reflection characteristics, thereby improving the resolution of the seismic wave propagation area. The result is the generation of seismic wave spatial propagation region data, which demonstrates the different regions of seismic wave propagation and reflects the complexity of propagation through refined grid cells. Local implicit modes are extracted from the seismic wave implicit reservoir characteristic data using deep learning or mode decomposition techniques (such as variational mode decomposition, VMD). Mode decomposition is performed on seismic data from different frequency bands or time windows to extract implicit modes related to seismic wave propagation, reflection, and reservoir characteristics, generating local implicit mode data that describes the wave patterns and characteristics of the reservoir in different local regions. Reservoir characteristic values ​​are calculated for each refined grid cell of the seismic wave spatial propagation region data using this local implicit mode data.By analyzing the implicit patterns and wave propagation characteristics within different grid cells, characteristic values ​​related to reservoir properties (such as porosity, permeability, and lithology) are calculated, generating grid cell reservoir characteristic values. This helps to further describe the physical properties and reflection characteristics of the reservoir. Using the grid cell reservoir characteristic values ​​and seismic multi-wavefield grid mapping data, a joint mapping is performed to generate a complete reservoir spatial framework. Spatial mapping algorithms (such as seismic data inversion and interactive mapping) are employed to combine the reservoir characteristic values ​​with seismic signal data, forming a preliminary reservoir model and generating initial reservoir characteristic mapping data. This data represents the spatial distribution characteristics of the reservoir and contains joint information on seismic wave propagation and reservoir characteristics.

[0104] Preferably, the inversion and linkage optimization of high-order reservoir characteristic data and seismic multi-wavelength separation data includes:

[0105] Frequency domain analysis was performed on the seismic multi-wave field separation data to generate seismic full-wave frequency domain data; nonlinear wave equations were constructed on the seismic full-wave data based on the seismic full-wave frequency domain data to generate reservoir medium wave equations.

[0106] Finite difference calculations are performed based on the reservoir medium wave equation to obtain reservoir medium wave difference data; numerical simulation of the reservoir medium wave equation is then performed using the reservoir medium wave difference data to generate simulated seismic wave field data.

[0107] By simulating seismic wavefield data, high-order reservoir characteristic data are inverted and optimized to generate optimized dynamic response data for the reservoir.

[0108] In this embodiment of the invention, seismic multi-wave field separation data is converted from the time domain to the frequency domain using Fourier transform or wavelet transform to extract the frequency components of seismic waves. Frequency analysis methods (such as Fast Fourier Transform, FFT) are used to divide the seismic data into multiple frequency bands, enabling clearer analysis of waveform characteristics at different frequencies and generating full-wave frequency domain data. This data contains seismic wave information with different frequency components, aiding in a deeper understanding of the propagation characteristics of seismic waves in reservoir media. Based on the full-wave frequency domain data, nonlinear wave equations describing the propagation of seismic waves in reservoir media are constructed. These equations consider factors such as the nonlinear effects of waves, the complexity of the medium, and changes in wave velocity. Nonlinear wave equations can be analogous to conventional wave equations (such as the Helmholtz equation and wave propagation equation) and incorporate nonlinear terms (such as reflection and refraction terms dependent on medium properties). These equations can more accurately describe the propagation of seismic waves in complex reservoirs, thus influencing the wave field inversion process. Numerical calculations are performed using the reservoir medium wave equations via the finite difference method (FDM). The finite difference method can discretize continuous wave equations, transforming them into solvable difference equations. By using a gridded reservoir model (e.g., orthogonal or unstructured grids), the equations are discretized in the spatiotemporal domain, and the wave value at each grid point is calculated, generating reservoir wave difference data. This data represents the numerical simulation results of seismic wave propagation in the reservoir medium, reflecting the wave propagation characteristics, reflection, refraction, and other phenomena. Based on the reservoir wave difference data, the reservoir medium wave equations are numerically solved and simulated. Explicit or implicit numerical methods can be used, such as the Lax-Friedrichs method or gradient descent. During the numerical simulation, the propagation of seismic waves in the reservoir is simulated by incorporating the reservoir's physical properties (such as elastic modulus, density, porosity, etc.), considering the reservoir's non-homogeneity and anisotropy, generating simulated seismic wavefield data. This data reflects the entire process of seismic wave propagation in the reservoir medium, including wave reflection, refraction, and interactions, helping to better understand the dynamic response of the reservoir. High-order reservoir characteristic data are inverted and optimized using simulated seismic wavefield data. During the inversion process, optimization algorithms such as least squares, genetic algorithms, and backpropagation can be employed to minimize the error between the simulated and actual seismic wavefields. Based on the inversion results, reservoir physical parameters, such as porosity, permeability, and wave velocity, are optimized to more accurately describe the actual dynamic response of the reservoir, generating optimized reservoir dynamic response data. This data details the dynamic changes of the reservoir under different wavefield conditions, optimizes the reservoir characteristic model, and provides support for oil and gas exploration and development.

[0109] Preferably, step S23 includes the following steps:

[0110] Step S231: Perform wavelet packet decomposition on the reservoir dynamic response optimization data to generate low-frequency component data and high-frequency component data; perform geological labeling on the low-frequency component data and high-frequency component data to generate low-frequency component labeled data and high-frequency component labeled data.

[0111] Step S232: Reconstruct the frequency band features of the reservoir dynamic response optimization data to generate multi-frequency time-domain images; perform spatial feature mapping on the multi-frequency time-domain images to generate multi-frequency reservoir feature maps;

[0112] Step S233: Enhance the texture features of the multi-frequency reservoir feature map based on the high-frequency component annotation data to generate local texture feature enhancement data of the map;

[0113] Step S234: Based on the low-frequency component annotation data, perform trend fitting on the multi-frequency reservoir feature map to generate map trend fitting data; optimize the overall trend of the multi-frequency reservoir feature map using the map trend fitting data to generate map overall trend optimized data.

[0114] Step S235: Optimize and integrate the multi-frequency reservoir feature map based on the local texture feature enhancement data and the overall trend optimization data of the map to obtain a high-resolution reservoir detail map.

[0115] In this embodiment of the invention, reservoir dynamic response optimization data is decomposed into multiple frequency bands, particularly low-frequency and high-frequency components, using wavelet packet decomposition (WPD). Wavelet packet decomposition can capture different frequency information of the signal more precisely, making it suitable for analyzing non-stationary signals, such as reservoir dynamic response signals. During the decomposition process, appropriate wavelet basis functions (such as Daubechies wavelets) can be selected, and the number of decomposition layers can be set to balance frequency and temporal resolution based on the characteristics of the reservoir data. Low-frequency and high-frequency component data are labeled according to the geological characteristics of the reservoir (such as rock layer type, porosity, and permeability). Manual labeling or automatic labeling methods based on machine learning, such as support vector machines (SVM) or convolutional neural networks (CNN), can be used. The labeled low-frequency and high-frequency component data will contain geological information about the reservoir, such as rock layer boundaries and porosity variations, which is helpful for subsequent feature analysis and map construction. Based on the low-frequency and high-frequency component data after wavelet packet decomposition, the features of each frequency band are recombined using inverse wavelet packet transform or frequency band reconstruction methods to generate multi-frequency time-domain images. These images can display the temporal characteristics of reservoir dynamic response at different frequency bands, reflecting the temporal changes of the reservoir. Spatial feature mapping is then performed on the multi-frequency time-domain images. The spatial mapping process typically involves aligning the time-domain data with geographic spatial coordinates (e.g., based on the geological coordinate system of the exploration area) and using spatial interpolation methods (such as Kriging interpolation, inverse distance weighted interpolation, etc.) to generate spatial distribution data. The generated multi-frequency reservoir feature map will contain the distribution of reservoir features in different frequency bands and can reveal the spatial heterogeneity of the reservoir. Based on the high-frequency component labeled data, image processing methods are used to enhance the texture features of the multi-frequency reservoir feature map. High-frequency components usually contain detailed and rapidly changing information, which helps to capture subtle changes in the reservoir. Methods such as Gabor filtering and Laplace pyramids are used for texture enhancement to highlight small-scale structural features inside the reservoir, such as fractures and pores. Enhanced data can help identify and highlight the heterogeneity and complexity of reservoirs at local scales. Based on low-frequency component annotation data, trend fitting techniques are used to analyze multi-frequency reservoir feature maps. Low-frequency components typically reflect the overall trend and large-scale characteristics of the reservoir, and therefore can be used to fit the long-term trend or variation patterns of the reservoir. Trend modeling is performed using methods such as multinomial regression, curve fitting, and locally weighted regression to obtain map trend fitting data, i.e., the large-scale variation patterns of the reservoir. Based on the map trend fitting data, the overall trend of the multi-frequency reservoir feature maps is optimized to remove unnecessary noise and fluctuations, highlighting the global trend and macrostructure of the reservoir, generating optimized overall trend data. This data demonstrates the basic structure and trend of the reservoir, contributing to a deeper understanding of the main characteristics and geological background of the reservoir.This study combines enhanced local texture features and optimized overall trend data from reservoir feature maps to optimize and integrate multi-frequency reservoir feature maps. By fusing local details and global trends, the complex characteristics of the reservoirs can be presented more comprehensively. Optimization and integration are achieved using weighted averaging and fusion algorithms, enabling the maps to simultaneously display details and macroscopic structures while preserving the true physical properties of the reservoirs. The optimized and integrated data generates a high-resolution reservoir detail map, which displays both the details and overall trends of the reservoirs, exhibiting high spatial resolution and accurately revealing their complex structures.

[0116] Preferably, step S233 includes the following steps:

[0117] Based on the high-frequency component annotation data, high-frequency texture regions are located in the multi-frequency reservoir feature map to generate high-frequency texture region data; feature regions are separated from the high-frequency texture region data to generate high-frequency texture separated region data.

[0118] The gray-level co-occurrence matrix is ​​calculated on the high-frequency texture separation region data to obtain the texture information data of the high-frequency texture region; local contrast enhancement is performed on the texture information data of the high-frequency texture region to generate texture enhancement data of the high-frequency texture region; texture component matching and alignment are performed on the high-frequency texture region data based on the texture enhancement data of the high-frequency texture region to generate high-frequency texture alignment region data.

[0119] By using high-frequency texture alignment region data, the multi-frequency reservoir feature map is updated with map details to generate local texture feature enhancement data.

[0120] In this embodiment of the invention, information from high-frequency component annotation data is combined with a multi-frequency reservoir feature map, and texture regions are located using image processing techniques. High-frequency components typically reflect details and rapidly changing structures, mainly concentrated in areas such as small fractures and pores in the reservoir. Algorithms such as edge detection and texture analysis (e.g., Canny edge detection, Laplacian pyramid, or Sobel operator) are used to analyze the multi-frequency map and identify high-frequency texture regions. These regions represent high-frequency variations or complex details in the reservoir. By locating these high-frequency regions, high-frequency texture region data is generated, which contains detailed information about the reservoir under high-frequency components. Based on the high-frequency texture region data, complex texture regions in the image are further separated. The separation process can extract different texture feature regions through threshold segmentation, region growing algorithms, etc. Pixel clustering (e.g., K-means clustering) or morphological segmentation methods (e.g., morphological opening and closing operations) are used to segment the high-frequency texture regions into several feature regions to more finely analyze the reservoir characteristics of different regions, generating high-frequency texture separation region data, which contains spatial information and feature identifiers of different texture feature regions. Gray-Level Co-occurrence Matrix (GLCM) calculations are performed on the high-frequency texture separation region data. GLCM is a method for describing image texture, revealing the spatial relationships and texture features between pixels in an image. GLCMs at different directions and distances are calculated, and texture features such as contrast, uniformity, energy, and entropy are extracted. These features help to gain a deeper understanding of the texture information in high-frequency regions of the reservoir. High-frequency texture region texture information data is obtained, containing detailed texture features of the reservoir's high-frequency texture regions, such as roughness and contrast. Based on the high-frequency texture region texture information data, local contrast enhancement is performed. Contrast enhancement can enhance details in the image and highlight subtle structural changes in the reservoir. Local contrast adjustment algorithms (such as local histogram equalization or the CLAHE algorithm) are used to enhance the high-frequency texture region texture information data to improve the discernibility of reservoir texture details, generating high-frequency texture region texture enhancement data. This data is characterized by clearer and easier-to-analyze high-frequency texture regions of the reservoir, facilitating subsequent image processing and analysis. Based on high-frequency texture region enhancement data, texture regions at different frequency bands are matched and aligned. The purpose of texture component matching and alignment is to ensure that the reservoir texture features can be accurately aligned at different image scales. Image registration techniques (such as feature point-based registration algorithms, optical flow methods, or mutual information methods) are used to accurately align the texture regions in the high-frequency texture region data, ensuring that all texture features correspond in the same spatial coordinates, generating high-frequency texture aligned region data. This data can accurately map the reservoir texture features and guarantee the consistency of texture regions at different scales.By utilizing high-frequency texture alignment region data, the spectral details of multi-frequency reservoir feature maps are updated. Updating the spectral details enhances local details in the image while preserving global trends, improving the overall resolution and accuracy of the map. Texture enhancement algorithms (such as hierarchical weighted fusion or multi-scale fusion methods) are employed to fuse the updated high-frequency texture data with the original spectral data, thereby optimizing the detail rendering of the map and generating enhanced local texture feature data. This data reveals enhanced details of the reservoir's fine structures, providing a more accurate reservoir characterization.

[0121] As an example of the present invention, reference is made to Figure 3 As shown, step S3 in this example includes:

[0122] Step S31: Perform dynamic reservoir feature change analysis on the high-resolution reservoir detail display map to generate dynamic reservoir feature change data; divide the dynamic reservoir feature change data into a dataset to generate a model training set and a model test set;

[0123] Step S32: Use a convolutional neural network algorithm to train the model on the training set to generate a pre-model for predicting reservoir characteristic changes; use a model test set to optimize and iterate the pre-model for predicting reservoir characteristic changes to generate a model for predicting reservoir characteristic changes.

[0124] Step S33: Import the dynamic reservoir characteristic change data into the reservoir characteristic change prediction model to predict the reservoir characteristic change, thereby generating reservoir characteristic change prediction data.

[0125] Step S34: Dynamically update the high-resolution reservoir detail map using reservoir feature change prediction data to generate a reservoir feature change prediction map.

[0126] In this embodiment of the invention, high-resolution reservoir detail maps (obtained from previous steps) are used to provide the fine structure and features of the reservoir. Image processing and dynamic analysis algorithms (such as time series analysis, wavelet transform, edge detection, etc.) are employed to analyze feature changes in the reservoir images. By comparing reservoir detail maps at different times or states, dynamic change information of the reservoir at each moment is extracted, generating dynamic reservoir feature change data. This data reflects the feature changes of the reservoir at different time points or states, including the dynamic evolution of features such as porosity, fractures, and permeability. The dynamic reservoir feature change data is divided into two datasets: a model training set and a model test set. The training set is used to train the prediction model, and the test set is used to evaluate the model's performance and generalization ability. The dataset is randomly divided into training and test sets using common data partitioning methods (such as the 80 / 20 or 70 / 30 rule). This ensures that the test and training sets are representative in terms of data distribution. Two datasets are obtained: a model training set (for model training) and a model test set (for model validation and optimization). Using the data in the model training set, a convolutional neural network (CNN) is selected as the main deep learning algorithm. CNNs can extract multi-level feature information from image data, making them particularly suitable for predicting spatial and local structural changes in reservoir features. A specific CNN architecture for reservoir change prediction is constructed, typically including multiple convolutional layers, pooling layers, and fully connected layers. Backpropagation training is performed using a training set to adjust network weights and minimize the loss function (e.g., mean squared error). This yields a preliminary reservoir characteristic change prediction model, capable of predicting reservoir characteristic change trends to a certain extent. The trained pre-model is evaluated using data from a test set, and the model is optimized based on the test set results. Model hyperparameters (e.g., kernel size, learning rate, number of layers) are tuned using methods such as cross-validation and grid search, and advanced optimization algorithms (e.g., Adam, RMSProp) are used for iterative model training to improve the model's accuracy and stability. Through multiple iterations of optimization, a final reservoir characteristic change prediction model is generated, capable of accurately predicting the trend of reservoir characteristics changing over time or in a given state. Dynamic reservoir characteristic change data (obtained from step S31) is imported into the reservoir characteristic change prediction model to predict characteristic changes. Using the trained and optimized reservoir characteristic change prediction model, forward inference is performed on new dynamic reservoir characteristic change data to predict changes in reservoir characteristics at future times or under different environments, generating reservoir characteristic change prediction data. This data reflects the changes in reservoir characteristics at the predicted time or state, including features such as porosity changes and fracture evolution. The previous high-resolution reservoir detail map is updated using the reservoir characteristic change prediction data (obtained from step S33). This step combines the predicted reservoir characteristic change data with the original map to obtain a new image representation.Image fusion techniques or pixel-level update methods (such as image deformation, color mapping, etc.) are used to overlay the reservoir feature change prediction data onto the original reservoir display map to obtain an updated reservoir detail map, generating a reservoir feature change prediction map. This is a dynamically updated image that shows the evolution of reservoir characteristics at different time points or under different environmental conditions, providing a more accurate prediction of reservoir change trends.

[0127] Preferably, step S4 includes the following steps:

[0128] Step S41: Assess the reservoir potential of the reservoir characteristic change prediction map and generate reservoir potential assessment data; classify the reservoirs based on the reservoir potential assessment data and generate a reservoir classification map.

[0129] Step S42: Perform actual reservoir observations on the reservoir classification map to generate actual reservoir observation result data; calculate the confidence level of the actual reservoir observation result data and the reservoir classification map to generate reservoir prediction confidence level data;

[0130] Step S43: Compare the reservoir prediction confidence data with the preset confidence threshold. When the reservoir prediction confidence data is greater than or equal to the preset confidence threshold, output the map based on the reservoir classification map to generate a high-resolution reservoir prediction distribution map. When the reservoir prediction confidence data is less than the preset confidence threshold, return to step S3 for feature re-extraction until the reservoir prediction confidence data is greater than or equal to the preset confidence threshold.

[0131] In this embodiment of the invention, a reservoir characteristic change prediction map generated in step S34 is used, which shows the characteristic changes of the reservoir at different time points or under different conditions. Reservoir evaluation algorithms (such as comprehensive analysis of reservoir evaluation index, permeability, and porosity) are applied to assess the reservoir's potential. The reservoir potential assessment model can be based on geological, petrological, and physical models, combined with historical data, to generate reservoir potential assessment data. This data describes the reservoir's development potential, including recoverable reserves, production capacity, and reservoir quality. The reservoir characteristic change prediction map is then classified using the reservoir potential assessment data. Reservoir classification is based on reservoir attributes such as permeability, porosity, and saturation, dividing the reservoir into different types. Clustering algorithms (such as K-means, hierarchical clustering, etc.) or supervised learning algorithms (such as decision trees, random forests, etc.) are used for reservoir classification. Reservoirs can be classified according to different potential assessment results (such as high potential, medium potential, low potential), generating a reservoir classification map. This map shows the distribution of different reservoir types, providing support for subsequent analysis and development. Actual reservoir observations are conducted using reservoir classification maps. These observations include drilling data, field survey data, and production data, providing practical feedback on reservoir characteristics and performance. Actual reservoir data is acquired using practical observation methods (such as seismic exploration, downhole measurements, and production testing), and correlation analysis is performed based on the reservoir classification maps to generate actual reservoir observation results data. This data includes actually observed reservoir characteristics and performance data, such as actual porosity, permeability, pressure, and production capacity. Confidence scores are calculated using the actual reservoir observation results data and the reservoir classification maps to assess the accuracy and reliability of the reservoir classification. Statistical methods (such as Bayesian inference and Monte Carlo simulation) or confidence scoring models in machine learning (such as the probability output of decision trees and the confidence score of support vector machines) are applied to calculate the confidence score. This step combines the actual observation data and model predictions to generate a confidence value, producing reservoir prediction confidence data, which represents the confidence level of the reservoir classification and prediction. The reservoir prediction confidence data is compared with a preset confidence threshold. The confidence threshold is a standard set according to project requirements, representing the degree of confidence in the reservoir prediction. A simple conditional judgment (such as an if statement) is used to compare the reservoir prediction confidence data with the threshold. If the reservoir prediction confidence data is greater than or equal to the preset threshold, the process continues to the next step; otherwise, it returns to step S3 for feature re-extraction. If the confidence data meets the threshold requirements, a high-resolution reservoir prediction distribution map is generated based on the reservoir classification map. This map shows the high-resolution prediction of reservoir characteristics in spatial distribution, presenting information such as reservoir exploitability and development potential. Data visualization techniques (such as heatmaps and contour maps) are used to combine the reservoir classification map with the spatial information and prediction data for high-resolution rendering, generating the high-resolution reservoir prediction distribution map, providing an intuitive reference for subsequent development decisions.If the reservoir prediction confidence data does not reach the preset threshold, the process returns to step S3 for feature re-extraction. By adjusting the feature extraction method or using more high-resolution observation data for re-extraction, the model is improved to increase prediction accuracy. Based on the adjusted features and optimized prediction results, the confidence score is calculated again until the preset confidence threshold is met.

[0132] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0133] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A high-resolution reservoir prediction method based on seismic full-wave information mining, characterized in that, Includes the following steps: Step S1: Acquire seismic full-wave information data; perform frequency domain joint analysis on the seismic full-wave information data to generate multi-scale full-wave feature tensors; perform multi-scale decoupling of the seismic full-wave signal on the multi-scale full-wave feature tensors to generate seismic multi-wave field separation data. Step S2: Perform implicit reservoir feature mapping on the seismic multi-wavelength separation data to generate initial reservoir feature mapping data; perform nonlinear correlation analysis on the initial reservoir feature mapping data to generate high-order reservoir feature data; perform inversion linkage optimization on the high-order reservoir feature data and the seismic multi-wavelength separation data to generate optimized reservoir dynamic response data; perform reservoir detail enhancement on the optimized reservoir dynamic response data to generate a high-resolution reservoir detail display map; Step S3: Perform dynamic reservoir feature change analysis on the high-resolution reservoir detail map to generate dynamic reservoir feature change data; predict reservoir feature changes on the high-resolution reservoir detail map based on the dynamic reservoir feature change data to generate a reservoir feature change prediction map. Step S4: Classify the reservoirs in the reservoir feature change prediction map to generate a reservoir classification map; calculate the confidence level of the reservoir classification map to generate reservoir prediction confidence data; compare the reservoir prediction confidence data with the preset confidence threshold and output the results until a high-resolution reservoir prediction distribution map is generated.

2. The high-resolution reservoir prediction method based on seismic full-wave information mining according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Acquire full-wave seismic information data using a multi-point distributed seismic signal acquisition array; Step S12: Perform data preprocessing on the seismic full-wave information data to generate standard seismic full-wave information data. The data preprocessing includes data cleaning, missing value imputation, and data standardization. Step S13: Perform time-domain analysis on the full-wave signal to generate full-wave signal time-domain data; perform fast Fourier transform on the full-wave signal time-domain data to generate full-wave signal frequency-domain data; perform joint frequency-domain analysis based on the full-wave signal time-domain data and full-wave signal frequency-domain data to generate multi-scale full-wave feature tensors. Step S14: Based on the preset seismic wave propagation characteristic coefficients, perform multi-scale decoupling of the seismic full-wave signal on the multi-scale full-wave characteristic tensor to generate seismic multi-wave field separation data.

3. The high-resolution reservoir prediction method based on seismic full-wave information mining according to claim 2, characterized in that, Step S14 includes the following steps: Step S141: Perform seismic wave propagation velocity analysis on the multi-scale full-wave characteristic tensor to generate seismic wave propagation velocity data; classify the seismic wave types based on the seismic wave propagation velocity data on the multi-scale full-wave characteristic tensor to generate seismic wave type classification data, which includes P-wave data, S-wave data and surface wave data. Step S142: Perform variational mode decomposition on the P-wave data, S-wave data, and surface wave data to generate P-wave mode decomposition data, S-wave mode decomposition data, and surface wave mode decomposition data; perform wavefield separation on the P-wave data, S-wave data, and surface wave data using the P-wave mode decomposition data, S-wave mode decomposition data, and surface wave mode decomposition data to generate seismic full-wave wavefield separation data; Step S143: Perform wavefield feature coupling analysis on the seismic full-wave wavefield separation data to generate wavefield feature coupling data; use the wavefield feature coupling data to decouple the seismic full-wave signal at multiple scales using the multi-scale full-wave feature tensor to generate seismic multi-wavefield separation data.

4. The high-resolution reservoir prediction method based on seismic full-wave information mining according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Extract implicit reservoir features from the seismic multi-wave field separation data to obtain seismic implicit reservoir feature data; perform spatial distribution mapping on the seismic multi-wave field separation data based on the seismic implicit reservoir feature data to generate initial reservoir feature mapping data. Step S22: Decouple the initial reservoir feature mapping data by physical properties to generate reservoir physical property data; perform nonlinear correlation analysis on the reservoir physical property data to generate high-order reservoir feature data; perform inversion linkage optimization on the high-order reservoir feature data and seismic multi-wave field separation data to generate reservoir dynamic response optimization data. Step S23: Decompose the reservoir dynamic response optimization data into frequency components to generate a multi-frequency reservoir feature map; enhance the reservoir details of the multi-frequency reservoir feature map to generate a high-resolution reservoir detail display map.

5. The high-resolution reservoir prediction method based on seismic full-wave information mining according to claim 4, characterized in that, Spatial distribution mapping of seismic multi-wave field separation data based on seismic implicit reservoir characteristic data includes: Seismic acquisition points were identified from the multi-wavelength separation data to obtain seismic acquisition point information data; geographic coordinates were then identified from the seismic acquisition point information data to obtain geographic spatial coordinate data. Seismic multi-wave field separation data and geospatial coordinate data are spatially mapped to a grid to generate seismic multi-wave field grid-mapped data; the propagation range of the seismic multi-wave field grid-mapped data is evolved based on the implicit reservoir characteristics of seismic waves to generate seismic propagation evolution path data. Based on earthquake propagation evolution path data, the grid cells of the earthquake multi-wave field grid mapping data are refined to generate seismic wave spatial propagation area data; local implicit mode extraction is performed on the seismic wave implicit reservoir characteristic data to obtain seismic wave local implicit mode data. Based on the local implicit model data of seismic waves, the reservoir characteristic values ​​of the seismic wave spatial propagation area data are calculated by grid cell to obtain the reservoir characteristic values ​​of the grid cells; based on the reservoir characteristic values ​​of the grid cells, the spatial frame joint mapping of the seismic multi-wavefield grid mapping data is performed to generate the initial reservoir characteristic mapping data.

6. The high-resolution reservoir prediction method based on seismic full-wave information mining according to claim 4, characterized in that, The inversion and linkage optimization of high-order reservoir characteristic data and seismic multi-wavelength separation data includes: Frequency domain analysis was performed on the seismic multi-wave field separation data to generate seismic full-wave frequency domain data; nonlinear wave equations were constructed on the seismic full-wave data based on the seismic full-wave frequency domain data to generate reservoir medium wave equations. Finite difference calculations are performed based on the reservoir medium wave equation to obtain reservoir medium wave difference data; numerical simulation of the reservoir medium wave equation is then performed using the reservoir medium wave difference data to generate simulated seismic wave field data. By simulating seismic wavefield data, high-order reservoir characteristic data are inverted and optimized to generate optimized dynamic response data for the reservoir.

7. The high-resolution reservoir prediction method based on seismic full-wave information mining according to claim 4, characterized in that, Step S23 includes the following steps: Step S231: Perform wavelet packet decomposition on the reservoir dynamic response optimization data to generate low-frequency component data and high-frequency component data; perform geological labeling on the low-frequency component data and high-frequency component data to generate low-frequency component labeled data and high-frequency component labeled data. Step S232: Reconstruct the frequency band features of the reservoir dynamic response optimization data to generate multi-frequency time-domain images; perform spatial feature mapping on the multi-frequency time-domain images to generate multi-frequency reservoir feature maps; Step S233: Enhance the texture features of the multi-frequency reservoir feature map based on the high-frequency component annotation data to generate local texture feature enhancement data of the map; Step S234: Based on the low-frequency component annotation data, perform trend fitting on the multi-frequency reservoir feature map to generate map trend fitting data; optimize the overall trend of the multi-frequency reservoir feature map using the map trend fitting data to generate map overall trend optimized data. Step S235: Optimize and integrate the multi-frequency reservoir feature map based on the local texture feature enhancement data and the overall trend optimization data of the map to obtain a high-resolution reservoir detail map.

8. The high-resolution reservoir prediction method based on seismic full-wave information mining according to claim 7, characterized in that, Step S233 includes the following steps: Based on the high-frequency component annotation data, high-frequency texture regions are located in the multi-frequency reservoir feature map to generate high-frequency texture region data; feature regions are separated from the high-frequency texture region data to generate high-frequency texture separated region data. The gray-level co-occurrence matrix is ​​calculated on the high-frequency texture separation region data to obtain the texture information data of the high-frequency texture region; local contrast enhancement is performed on the texture information data of the high-frequency texture region to generate texture enhancement data of the high-frequency texture region; texture component matching and alignment are performed on the high-frequency texture region data based on the texture enhancement data of the high-frequency texture region to generate high-frequency texture alignment region data. By using high-frequency texture alignment region data, the multi-frequency reservoir feature map is updated with map details to generate local texture feature enhancement data.

9. The high-resolution reservoir prediction method based on seismic full-wave information mining according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Perform dynamic reservoir feature change analysis on the high-resolution reservoir detail display map to generate dynamic reservoir feature change data; divide the dynamic reservoir feature change data into a dataset to generate a model training set and a model test set; Step S32: Use a convolutional neural network algorithm to train the model on the training set to generate a pre-model for predicting reservoir characteristic changes; use a model test set to optimize and iterate the pre-model for predicting reservoir characteristic changes to generate a model for predicting reservoir characteristic changes. Step S33: Import the dynamic reservoir characteristic change data into the reservoir characteristic change prediction model to predict the reservoir characteristic change, thereby generating reservoir characteristic change prediction data. Step S34: Dynamically update the high-resolution reservoir detail map using reservoir feature change prediction data to generate a reservoir feature change prediction map.

10. The high-resolution reservoir prediction method based on seismic full-wave information mining according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Assess the reservoir potential of the reservoir characteristic change prediction map and generate reservoir potential assessment data; classify the reservoirs in the reservoir characteristic change prediction map based on the reservoir potential assessment data and generate a reservoir classification map. Step S42: Perform actual reservoir observations on the reservoir classification map to generate actual reservoir observation result data; calculate the confidence level of the actual reservoir observation result data and the reservoir classification map to generate reservoir prediction confidence level data; Step S43: Compare the reservoir prediction confidence data with the preset confidence threshold. When the reservoir prediction confidence data is greater than or equal to the preset confidence threshold, output the map based on the reservoir classification map to generate a high-resolution reservoir prediction distribution map. When the reservoir prediction confidence data is less than the preset confidence threshold, return to step S3 for feature re-extraction until the reservoir prediction confidence data is greater than or equal to the preset confidence threshold.

Images