Ultra-deep strike-slip breaking control grid-shaped reservoir geophysical characterization method and device, electronic equipment and medium

Abstract

The invention discloses an ultra-deep strike-slip breaking control grid-shaped reservoir geophysical representation method and device, electronic equipment and a medium. The method comprises the following steps: establishing a forward modeling model based on a grid-shaped reservoir body; grid-shaped reservoir body forward modeling simulation analysis is conducted on the forward modeling model; according to the result of the forward modeling analysis, constructing a deep learning sample library of the grid-shaped reservoir body; a U-net network model is established and trained, and grid-shaped reservoir body characterization is carried out through the trained network model. According to the method, on the basis of a conventional seismic attribute initial sample library, a self-adaptive deep learning sample library for reservoir body manual intervention is constructed according to manual interpretation results and innovatively researched and developed structure gradient attributes, instantaneous amplitude ratios and similarity transverse change rate attributes, and a fault control reservoir body is predicted and described by applying deep learning.

Description

Technical Field

[0001] This invention relates to the field of integrated seismic geology and engineering, and more specifically, to a geophysical characterization method, apparatus, electronic equipment, and medium for ultra-deep strike-slip fault-controlled grid-like reservoirs. Background Technology

[0002] Ultra-deep strike-slip fault-controlled reservoirs possess immense exploration potential. However, the fault-reservoir relationship is complex and difficult to characterize. Firstly, fault-controlled reservoirs are highly heterogeneous, making it difficult to establish their spatial configuration. Secondly, the relationship between the heterogeneous structure within fault-controlled reservoirs and their seismic reflection characteristics remains unclear.

[0003] There is a lack of research both domestically and internationally on the technical challenges of characterizing ultra-deep strike-slip fault-controlled grate reservoirs. Many articles focus only on a single technical aspect (such as the study of imaging algorithms) without conducting systematic research based on geological and seismic response characteristics to develop a comprehensive technology for the geophysical characterization of ultra-deep carbonate strike-slip fault-controlled grate reservoirs.

[0004] Therefore, it is necessary to develop a geophysical characterization method, device, electronic equipment, and medium for ultra-deep strike-slip fault-controlled grid-like reservoirs.

[0005] The information disclosed in the background section of this invention is intended only to enhance the understanding of the general background of this invention, and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] This invention proposes a geophysical characterization method, device, electronic equipment, and medium for ultra-deep strike-slip fault-controlled grid-like reservoirs. Based on the initial sample library of conventional seismic attributes, it constructs an adaptive deep learning sample library for artificial intervention of reservoirs by incorporating the results of artificial interpretation, as well as innovatively developed structural gradient attributes, instantaneous amplitude ratios, and similarity lateral change rate attributes. Deep learning is then applied to predict and characterize fault-controlled reservoirs.

[0007] In a first aspect, embodiments of this disclosure provide a geophysical characterization method for ultra-deep strike-slip fault-controlled grid-like reservoirs, including:

[0008] Establish a forward model based on gridded reservoirs;

[0009] Forward modeling analysis of the grid reservoir was performed on the aforementioned forward model;

[0010] Based on the results of forward modeling analysis, a deep learning sample library for graticular reservoirs is constructed.

[0011] A U-net network model is established and trained, and the trained network model is used to represent the raster reservoir.

[0012] As a specific implementation of this disclosure, establishing a forward model based on a grid-like reservoir includes:

[0013] Based on the actual seismic profile and combined with reservoir geology, a geological model was designed, and the reservoir characteristics of the target well section were determined from the seismic profile and actual drilling leakage.

[0014] The actual velocities of various reservoirs are calculated using well logging and well logging data, and the forward model is established based on the length-to-width ratio of the geological bodies obtained from actual well logging and well logging.

[0015] As one specific implementation of this disclosure, the reservoir characteristics include the number of reservoirs, beaded response characteristics, and reflection characteristics.

[0016] As a specific implementation of this disclosure, Kirchihoff migration imaging is simulated using the wave equation for the forward model to obtain forward simulation analysis results.

[0017] As a specific implementation of this disclosure, the forward simulation analysis results include:

[0018] Different types of reservoirs have different amplitude characteristics. The reservoir has a low velocity and a large absolute value of the corresponding trough amplitude. That is, the probability of encountering good reservoirs is high for beads with large amplitude values. The probability of venting and leakage is high in the parts with large amplitude values ​​inside the beads. Therefore, different beads in the same area are classified and sorted by the lateral change of amplitude, with the large amplitude energy as the target, and the parts with large amplitude inside the beads are identified as the target points for well trajectory design.

[0019] As a specific implementation of this disclosure, the deep learning sample library for the grid-like reservoir is constructed based on the results of forward simulation analysis, including:

[0020] When constructing the sample library, manually interpreted data is added as labels, and forward modeling analysis results are also added to the sample library. Attribute research is carried out for small- and medium-scale fault zones to expand the sample library established by conventional earthquake attributes. Then, sample library data augmentation processing is carried out to ensure the training quality of the deep learning network.

[0021] As a specific implementation of this disclosure, establishing and training a U-net network model includes:

[0022] The data in the deep learning sample library is divided into a training set and a validation set;

[0023] The U-net network model is established, trained using the training set, and validated using the validation set.

[0024] Secondly, this disclosure also provides a geophysical characterization device for ultra-deep strike-slip fault-controlled grid-like reservoirs, comprising:

[0025] The forward model building module establishes a forward model based on a grid-like reservoir.

[0026] The forward modeling simulation analysis module performs forward modeling simulation analysis on the gridded reservoir for the forward model.

[0027] The sample library construction module constructs a deep learning sample library for a raster reservoir based on the results of forward simulation analysis.

[0028] The representation module establishes and trains the U-net network model, and uses the trained network model to represent the raster storage collective.

[0029] As a specific implementation of this disclosure, establishing a forward model based on a grid-like reservoir includes:

[0030] Based on the actual seismic profile and combined with reservoir geology, a geological model was designed, and the reservoir characteristics of the target well section were determined from the seismic profile and actual drilling leakage.

[0031] The actual velocities of various reservoirs are calculated using well logging and well logging data, and the forward model is established based on the length-to-width ratio of the geological bodies obtained from actual well logging and well logging.

[0032] As one specific implementation of this disclosure, the reservoir characteristics include the number of reservoirs, beaded response characteristics, and reflection characteristics.

[0033] As a specific implementation of this disclosure, Kirchihoff migration imaging is simulated using the wave equation for the forward model to obtain forward simulation analysis results.

[0034] As a specific implementation of this disclosure, the forward simulation analysis results include:

[0035] Different types of reservoirs have different amplitude characteristics. The reservoir has a low velocity and a large absolute value of the corresponding trough amplitude. That is, the probability of encountering good reservoirs is high for beads with large amplitude values. The probability of venting and leakage is high in the parts with large amplitude values ​​inside the beads. Therefore, different beads in the same area are classified and sorted by the lateral change of amplitude, with the large amplitude energy as the target, and the parts with large amplitude inside the beads are identified as the target points for well trajectory design.

[0036] As a specific implementation of this disclosure, the deep learning sample library for the grid-like reservoir is constructed based on the results of forward simulation analysis, including:

[0037] When constructing the sample library, manually interpreted data is added as labels, and forward modeling analysis results are also added to the sample library. Attribute research is carried out for small- and medium-scale fault zones to expand the sample library established by conventional earthquake attributes. Then, sample library data augmentation processing is carried out to ensure the training quality of the deep learning network.

[0038] As a specific implementation of this disclosure, establishing and training a U-net network model includes:

[0039] The data in the deep learning sample library is divided into a training set and a validation set;

[0040] The U-net network model is established, trained using the training set, and validated using the validation set.

[0041] Thirdly, embodiments of this disclosure also provide an electronic device, the electronic device comprising:

[0042] Memory, which stores executable instructions;

[0043] A processor that executes the executable instructions in the memory to implement the geophysical characterization method for ultra-deep strike-slip fault-controlled grate reservoirs.

[0044] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned geophysical characterization method for ultra-deep strike-slip fault-controlled grate reservoirs.

[0045] Its beneficial effects are as follows:

[0046] This invention first establishes a geological model of a grid-like reservoir that conforms to the regional geological conditions. Based on this model, forward modeling analysis is conducted to analyze the vertical and horizontal heterogeneous geophysical response characteristics of the grid-like reservoir. The relative positions of venting and leakage in the beaded reflections are identified, and qualitative well trajectory design criteria for fault-controlled reservoirs are established. Building upon an initial sample library of conventional seismic attributes, and incorporating manually interpreted results, as well as innovatively developed structural gradient attributes, instantaneous amplitude ratios, and similarity lateral change rate attributes, an adaptive deep learning sample library for reservoir artificial intervention is constructed. Deep learning is then applied to predict and characterize fault-controlled reservoirs. Practice shows that this method effectively improves the accuracy of reservoir characterization and representation.

[0047] The methods and apparatus of the present invention have other features and advantages that will be apparent from or will be set forth in detail in the accompanying drawings and following detailed description, which together serve to explain the particular principles of the invention. Attached Figure Description

[0048] The above and other objects, features and advantages of the present invention will become more apparent from the more detailed description of exemplary embodiments of the invention in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same parts.

[0049] Figure 1 A flowchart illustrating the steps of a geophysical characterization method for ultra-deep strike-slip fault-controlled grid reservoirs according to an embodiment of the present invention is shown.

[0050] Figure 2a , Figure 2b , Figure 2c The actual seismic profile, forward modeling results, and schematic diagram of the forward model of well Shunbei 47X are shown respectively, according to an embodiment of the present invention.

[0051] Figure 3 A schematic diagram of the grid structure of the Shunbei 47X inclined shaft section according to an embodiment of the present invention is shown.

[0052] Figure 4 A schematic diagram of a forward model of the relationship between the location of the storage failure according to an embodiment of the present invention is shown.

[0053] Figure 5 A schematic diagram of the forward modeling of the transverse nonhomogeneous wave equation is shown according to an embodiment of the present invention.

[0054] Figure 6 A schematic diagram of the forward modeling of the longitudinal nonhomogeneous wave equation is shown according to an embodiment of the present invention.

[0055] Figure 7 A schematic diagram illustrating the construction of a multi-type training sample library according to an embodiment of the present invention is shown.

[0056] Figure 8 The diagram shows a seismic profile, a well logging interpretation result diagram, and a schematic diagram of the lateral variation rate of similarity according to an embodiment of the present invention.

[0057] Figure 9 A schematic diagram showing a comparison of training accuracy and validation accuracy during iterative training according to an embodiment of the present invention is illustrated.

[0058] Figure 10a , Figure 10b , Figure 10c , Figure 10d The diagrams show a seismic profile, coherence attributes, original deep learning prediction results, and corrected deep learning prediction results according to an embodiment of the present invention.

[0059] Figure 11A schematic diagram of a seismic profile, instantaneous energy, and deep learning-based “grid” internal reservoir distribution according to an embodiment of the present invention is shown.

[0060] Figure 12 A block diagram of a geophysical characterization device for an ultra-deep strike-slip fault-controlled grid reservoir according to an embodiment of the present invention is shown.

[0061] Explanation of reference numerals in the attached figures:

[0062] 201. Forward model establishment module; 202. Forward simulation analysis module; 203. Sample library construction module; 204. Characterization module. Detailed Implementation

[0063] Preferred embodiments of the invention will now be described in more detail. While preferred embodiments of the invention are described below, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0064] To facilitate understanding of the solutions and effects of the embodiments of the present invention, six specific application examples are given below. Those skilled in the art should understand that these examples are merely for the purpose of understanding the present invention, and any specific details therein are not intended to limit the present invention in any way.

[0065] Example 1

[0066] Figure 1 A flowchart illustrating the steps of a geophysical characterization method for ultra-deep strike-slip fault-controlled grid reservoirs according to an embodiment of the present invention is shown.

[0067] like Figure 1 As shown, the geophysical characterization method for ultra-deep strike-slip fault-controlled grid-like reservoirs includes:

[0068] Step 101: Establish a forward model based on a grid-like reservoir;

[0069] Step 102: Perform forward modeling analysis of the grid reservoir for the forward modeling model;

[0070] Step 103: Based on the results of forward simulation analysis, construct a deep learning sample library for the grid-like reservoir.

[0071] Step 104: Establish and train the U-net network model, and use the trained network model to represent the grid-like storage collective.

[0072] In one example, building a forward model based on a grid-like reservoir includes:

[0073] Based on the actual seismic profile and combined with reservoir geology, a geological model was designed, and the reservoir characteristics of the target well section were determined from the seismic profile and actual drilling leakage.

[0074] The actual velocities of various reservoirs are calculated using well logging and well logging data, and a forward model is established based on the length and width ratio of the geological bodies obtained from actual well logging and well logging.

[0075] In one example, reservoir characteristics include the number of reservoirs, beaded response characteristics, and reflection characteristics.

[0076] In one example, Kirchihoff migration imaging is simulated using the wave equation for a forward model, and the forward simulation analysis results are obtained.

[0077] In one example, the results of the forward simulation analysis include:

[0078] Different types of reservoirs have different amplitude characteristics. The reservoir has a low velocity and a large absolute value of the corresponding trough amplitude. That is, the probability of encountering good reservoirs is high for beads with large amplitude values. The probability of venting and leakage is high in the parts with large amplitude values ​​inside the beads. Therefore, different beads in the same area are classified and sorted by the lateral change of amplitude, with the large amplitude energy as the target, and the parts with large amplitude inside the beads are identified as the target points for well trajectory design.

[0079] In one example, based on the results of forward simulation analysis, the deep learning sample library for constructing a grid-like reservoir includes:

[0080] When constructing the sample library, manually interpreted data is added as labels, and forward modeling analysis results are also added to the sample library. Attribute research is carried out for small- and medium-scale fault zones to expand the sample library established by conventional earthquake attributes. Then, sample library data augmentation processing is carried out to ensure the training quality of the deep learning network.

[0081] In one example, building and training the U-net network model includes:

[0082] The data in the deep learning sample library is divided into a training set and a validation set;

[0083] A U-net network model is established, trained using a training set, and validated using a validation set.

[0084] Specifically, this invention first establishes a geological model of a grid-like reservoir that conforms to the regional geological conditions, conducts forward modeling analysis based on the grid-like reservoir geological model, analyzes the vertical and horizontal heterogeneous geophysical response characteristics of the grid-like reservoir, identifies the relative position of venting and leakage in the beaded reflection, and establishes a design criterion for the qualitative well trajectory of fault-controlled reservoirs.

[0085] Based on actual seismic profiles and reservoir geological understanding, a geological model is designed to determine the reservoir characteristics of the target well section from the seismic profiles and actual drilling losses. Using the actual velocities of various reservoirs calculated from well logging and well logging data as a basis, a forward model is established based on the length-to-width ratio of geological bodies obtained from actual well logging and well logging, and the setup parameters of the observation system are determined.

[0086] For a single well, the forward modeling model was simulated using the wave equation to simulate Kirchihoff migration imaging. The forward modeling results showed that the forward modeling results based on the "grid structure" model had a high similarity to the well-passing seismic profile, indicating that the grid structure model was very reasonable and that the migration imaging effect was good.

[0087] Forward modeling shows that reservoirs of the same size and profile, despite different velocities, exhibit similar waveform characteristics. This indicates that, without prior data such as drilling data, simply judging which "bead" (reservoir) will experience significant venting or leakage based on waveform alone will lead to errors. However, different types of reservoirs exhibit different amplitude characteristics. Lower reservoir velocities correspond to larger absolute values ​​of trough amplitude. Therefore, beads with larger amplitude values ​​have a higher probability of encountering good reservoirs, and areas within beads with larger amplitude values ​​have a higher probability of venting or leakage. Different beads within the same region can be classified and ranked by lateral amplitude variations, prioritizing those with higher amplitude energy. Simultaneously, areas within beads with larger amplitudes can be selected as target points for well trajectory design.

[0088] Through forward modeling of the transverse heterogeneity of grid-like reservoirs, it can be found that the seismic amplitude value corresponding to the venting and leakage point (high-quality reservoir) is relatively large. The "beads" can be selected by amplitude attributes, and the target point can be determined based on the amplitude extreme value inside the bead. In addition, the longitudinal heterogeneity forward modeling shows that the venting and leakage point (high-quality reservoir) is a high-probability event in the zero phase or lower part. Therefore, when designing the trajectory, the amplitude extreme value should be considered while diagonally crossing the zero phase or slightly lower part.

[0089] In the testing of deep learning models, model selection and construction are crucial, and training data is equally important. While modifying the model architecture to improve prediction accuracy, it's also essential to enhance the quality of the input data. When constructing the dataset, thorough data cleaning and labeling are vital; a high-quality dataset often improves both the quality of model training and prediction accuracy. The primary data sources for this invention's sample database are synthetic seismic data (forward modeling), attribute analysis, and manual interpretation.

[0090] When constructing the sample library, supplementing it with manually interpreted data as labels, and also adding the results of forward modeling of grid-like reservoirs to the sample library, can improve the current fine prediction effect of fault-controlled grid-like reservoirs based on deep learning.

[0091] For small- and medium-scale fault-controlled reservoirs and the "grid-like" structures within these reservoirs, conventional attributes often fail to meet the resolution requirements for reservoir prediction, particularly for the three seismic facies types. This study aims to improve the accuracy of reservoir characterization based on deep learning network models by developing attributes for small- and medium-scale fault zones and expanding the sample library established using conventional seismic attributes.

[0092] The higher-order anisotropic gradient method for structural energy utilizes the difference in amplitude between beaded and non-beaded seismic data to characterize the beaded profile, and uses higher-order gradients to quantitatively describe this amplitude difference. Simultaneously, considering the directional differences in the beaded morphology exhibited in 3D seismic data, higher-order gradients in the line and trace directions are introduced separately. These directional difference higher-order gradients are then fused and displayed in a three-dimensional manner, providing a more detailed description of the reservoir's geometric profile and distribution characteristics in 3D space. The beaded enhancement features are significant, and the fault linearity characteristics are better demonstrated.

[0093] Currently, commonly used methods for estimating seismic trace similarity mainly rely on algorithms based on cross-correlation and similarity. The main drawback of these algorithms is the averaging effect that occurs when combining correlation coefficients or similarity parameters calculated in different directions. The more traces calculated, the stronger this averaging effect becomes, severely reducing spatial resolution and the continuity of anomalies. For predicting the "grid-like" structure within fault-controlled reservoirs, the seismic scale is often limited to one or two traces, making it difficult to characterize using conventional coherence or similarity attributes.

[0094] The frequency-domain dip similarity lateral change rate detection technology mainly uses Hilbert transform to convert time-domain seismic attributes to the frequency domain, followed by dip scanning and similarity lateral change calculation to obtain the differences between seismic traces. This invention aims to improve the identification and characterization capabilities of the "grid-like" structure within reservoirs by employing similarity lateral change rate detection parameters with "small windows and small facets," thereby enhancing the prediction accuracy of reservoirs within reservoir groups.

[0095] Data augmentation (sample library enhancement) refers to methods that generate new training samples by making small changes to existing training samples. The purpose of data augmentation is twofold: to increase the amount of data and to improve its quality. After data augmentation, the diversity of training samples increases, and the model's dependence on certain attributes decreases accordingly. This effectively prevents the network from learning irrelevant features, thereby enhancing the model's generalization ability. Since the data labels currently involved in "raster" reservoirs are mostly derived from 2D seismic data, it is necessary to perform sample library data augmentation to ensure the training quality of deep learning networks.

[0096] Fault-controlled reservoirs, as the main reservoirs, have highly complex internal structures, and the reflection characteristics shown in seismic profiles are mostly a combined response of several reservoir types. Forward modeling using the wave equation revealed that the seismic response of the steep main fault is linear with weak reflections and the presence of in-phase axis faulting; secondary faults and small faults mainly exhibit chaotic weak reflections; cavities mainly show beaded or chaotic strong reflections. It was noted that fractures and fracture-cavities are associated, and the reflection anomalies in the seismic profiles are a combined response of faults and fracture-cavities at different scales.

[0097] Given the complexity of reservoir structures and types, fracture training data and labels generated from forward modeling synthetic records are used. The training process uses the training data to train the U-net network model, while validation data (not involved in the training) is used for validation. Data input must be standardized to avoid interference from amplitude errors.

[0098] Example 2

[0099] The present invention also provides a geophysical characterization device for ultra-deep strike-slip fault-controlled grid-like reservoirs, comprising:

[0100] The forward model building module establishes a forward model based on a grid-like reservoir.

[0101] The forward modeling simulation analysis module performs forward modeling simulation analysis of the gridded reservoir model.

[0102] The sample library construction module constructs a deep learning sample library for a raster reservoir based on the results of forward simulation analysis.

[0103] The representation module establishes and trains the U-net network model, and uses the trained network model to represent the raster storage collective.

[0104] In one example, building a forward model based on a grid-like reservoir includes:

[0105] Based on the actual seismic profile and combined with reservoir geology, a geological model was designed, and the reservoir characteristics of the target well section were determined from the seismic profile and actual drilling leakage.

[0106] The actual velocities of various reservoirs are calculated using well logging and well logging data, and a forward model is established based on the length and width ratio of the geological bodies obtained from actual well logging and well logging.

[0107] In one example, reservoir characteristics include the number of reservoirs, beaded response characteristics, and reflection characteristics.

[0108] In one example, Kirchihoff migration imaging is simulated using the wave equation for a forward model, and the forward simulation analysis results are obtained.

[0109] In one example, the results of the forward simulation analysis include:

[0110] Different types of reservoirs have different amplitude characteristics. The reservoir has a low velocity and a large absolute value of the corresponding trough amplitude. That is, the probability of encountering good reservoirs is high for beads with large amplitude values. The probability of venting and leakage is high in the parts with large amplitude values ​​inside the beads. Therefore, different beads in the same area are classified and sorted by the lateral change of amplitude, with the large amplitude energy as the target, and the parts with large amplitude inside the beads are identified as the target points for well trajectory design.

[0111] In one example, based on the results of forward simulation analysis, the deep learning sample library for constructing a grid-like reservoir includes:

[0112] When constructing the sample library, manually interpreted data is added as labels, and forward modeling analysis results are also added to the sample library. Attribute research is carried out for small- and medium-scale fault zones to expand the sample library established by conventional earthquake attributes. Then, sample library data augmentation processing is carried out to ensure the training quality of the deep learning network.

[0113] In one example, building and training the U-net network model includes:

[0114] The data in the deep learning sample library is divided into a training set and a validation set;

[0115] A U-net network model is established, trained using a training set, and validated using a validation set.

[0116] Specifically, this invention first establishes a geological model of a grid-like reservoir that conforms to the regional geological conditions, conducts forward modeling analysis based on the grid-like reservoir geological model, analyzes the vertical and horizontal heterogeneous geophysical response characteristics of the grid-like reservoir, identifies the relative position of venting and leakage in the beaded reflection, and establishes a design criterion for the qualitative well trajectory of fault-controlled reservoirs.

[0117] Based on actual seismic profiles and reservoir geological understanding, a geological model is designed to determine the reservoir characteristics of the target well section from the seismic profiles and actual drilling losses. Using the actual velocities of various reservoirs calculated from well logging and well logging data as a basis, a forward model is established based on the length-to-width ratio of geological bodies obtained from actual well logging and well logging, and the setup parameters of the observation system are determined.

[0118] For a single well, the forward modeling model was simulated using the wave equation to simulate Kirchihoff migration imaging. The forward modeling results showed that the forward modeling results based on the "grid structure" model had a high similarity to the well-passing seismic profile, indicating that the grid structure model was very reasonable and that the migration imaging effect was good.

[0119] Forward modeling shows that reservoirs of the same size and profile, despite different velocities, exhibit similar waveform characteristics. This indicates that, without prior data such as drilling data, simply judging which "bead" (reservoir) will experience significant venting or leakage based on waveform alone will lead to errors. However, different types of reservoirs exhibit different amplitude characteristics. Lower reservoir velocities correspond to larger absolute values ​​of trough amplitude. Therefore, beads with larger amplitude values ​​have a higher probability of encountering good reservoirs, and areas within beads with larger amplitude values ​​have a higher probability of venting or leakage. Different beads within the same region can be classified and ranked by lateral amplitude variations, prioritizing those with higher amplitude energy. Simultaneously, areas within beads with larger amplitudes can be selected as target points for well trajectory design.

[0120] Through forward modeling of the transverse heterogeneity of grid-like reservoirs, it can be found that the seismic amplitude value corresponding to the venting and leakage point (high-quality reservoir) is relatively large. The "beads" can be selected by amplitude attributes, and the target point can be determined based on the amplitude extreme value inside the bead. In addition, the longitudinal heterogeneity forward modeling shows that the venting and leakage point (high-quality reservoir) is a high-probability event in the zero phase or lower part. Therefore, when designing the trajectory, the amplitude extreme value should be considered while diagonally crossing the zero phase or slightly lower part.

[0121] In the testing of deep learning models, model selection and construction are crucial, and training data is equally important. While modifying the model architecture to improve prediction accuracy, it's also essential to enhance the quality of the input data. When constructing the dataset, thorough data cleaning and labeling are vital; a high-quality dataset often improves both the quality of model training and prediction accuracy. The primary data sources for this invention's sample database are synthetic seismic data (forward modeling), attribute analysis, and manual interpretation.

[0122] When constructing the sample library, supplementing it with manually interpreted data as labels, and also adding the results of forward modeling of grid-like reservoirs to the sample library, can improve the current fine prediction effect of fault-controlled grid-like reservoirs based on deep learning.

[0123] For small- and medium-scale fault-controlled reservoirs and the "grid-like" structures within these reservoirs, conventional attributes often fail to meet the resolution requirements for reservoir prediction, particularly for the three seismic facies types. This study aims to improve the accuracy of reservoir characterization based on deep learning network models by developing attributes for small- and medium-scale fault zones and expanding the sample library established using conventional seismic attributes.

[0124] The higher-order anisotropic gradient method for structural energy utilizes the difference in amplitude between beaded and non-beaded seismic data to characterize the beaded profile, and uses higher-order gradients to quantitatively describe this amplitude difference. Simultaneously, considering the directional differences in the beaded morphology exhibited in 3D seismic data, higher-order gradients in the line and trace directions are introduced separately. These directional difference higher-order gradients are then fused and displayed in a three-dimensional manner, providing a more detailed description of the reservoir's geometric profile and distribution characteristics in 3D space. The beaded enhancement features are significant, and the fault linearity characteristics are better demonstrated.

[0125] Currently, commonly used methods for estimating seismic trace similarity mainly rely on algorithms based on cross-correlation and similarity. The main drawback of these algorithms is the averaging effect that occurs when combining correlation coefficients or similarity parameters calculated in different directions. The more traces calculated, the stronger this averaging effect becomes, severely reducing spatial resolution and the continuity of anomalies. For predicting the "grid-like" structure within fault-controlled reservoirs, the seismic scale is often limited to one or two traces, making it difficult to characterize using conventional coherence or similarity attributes.

[0126] The frequency-domain dip similarity lateral change rate detection technology mainly uses Hilbert transform to convert time-domain seismic attributes to the frequency domain, followed by dip scanning and similarity lateral change calculation to obtain the differences between seismic traces. This invention aims to improve the identification and characterization capabilities of the "grid-like" structure within reservoirs by employing similarity lateral change rate detection parameters with "small windows and small facets," thereby enhancing the prediction accuracy of reservoirs within reservoir groups.

[0127] Data augmentation (sample library enhancement) refers to methods that generate new training samples by making small changes to existing training samples. The purpose of data augmentation is twofold: to increase the amount of data and to improve its quality. After data augmentation, the diversity of training samples increases, and the model's dependence on certain attributes decreases accordingly. This effectively prevents the network from learning irrelevant features, thereby enhancing the model's generalization ability. Since the data labels currently involved in "raster" reservoirs are mostly derived from 2D seismic data, it is necessary to perform sample library data augmentation to ensure the training quality of deep learning networks.

[0128] Fault-controlled reservoirs, as the main reservoirs, have highly complex internal structures, and the reflection characteristics shown in seismic profiles are mostly a combined response of several reservoir types. Forward modeling using the wave equation revealed that the seismic response of the steep main fault is linear with weak reflections and the presence of in-phase axis faulting; secondary faults and small faults mainly exhibit chaotic weak reflections; cavities mainly show beaded or chaotic strong reflections. It was noted that fractures and fracture-cavities are associated, and the reflection anomalies in the seismic profiles are a combined response of faults and fracture-cavities at different scales.

[0129] Given the complexity of reservoir structures and types, fracture training data and labels generated from forward modeling synthetic records are used. The training process uses the training data to train the U-net network model, while validation data (not involved in the training) is used for validation. Data input must be standardized to avoid interference from amplitude errors.

[0130] Example 3

[0131] The Ordovician strike-slip fault-controlled reservoirs in the Shunbei area possess immense exploration potential. However, the fault-reservoir relationship is complex and challenging to characterize. Firstly, fault-controlled reservoirs exhibit strong heterogeneity, making it difficult to establish their spatial configuration. Secondly, the relationship between the internal heterogeneous structure of fault-controlled reservoirs and seismic reflection characteristics remains unclear. To address these challenges, this invention establishes a grid-like reservoir geological model conforming to the regional geological conditions. Forward modeling analysis was conducted based on this model, analyzing the vertical and horizontal heterogeneous geophysical response characteristics of the grid-like reservoirs. The relative positions of venting and leakage in beaded reflections were identified, and qualitative well trajectory design criteria for fault-controlled reservoirs were established. Building upon an initial sample library of conventional seismic attributes, an adaptive deep learning sample library for reservoir artificial intervention was constructed, incorporating manually interpreted results, as well as innovatively developed structural gradient attributes, instantaneous amplitude ratios, and similarity lateral change rate attributes. Deep learning was then applied to predict and characterize fault-controlled reservoirs. Practice has shown that this method effectively improves the accuracy of reservoir characterization and representation.

[0132] Figure 2a , Figure 2b , Figure 2c The actual seismic profile, forward modeling results, and schematic diagram of the forward model of well Shunbei 47X are shown respectively, according to an embodiment of the present invention.

[0133] Taking Shunbei 47X well as an example, such as Figure 2a , Figure 2b , Figure 2cAs shown, based on the actual seismic profile and reservoir geology, the 47X geological model designed indicates the presence of multiple reservoirs in the Shunbei 47X well section, as revealed by the seismic profile and actual drilling leakage data. The seismic profile exhibits two distinct beaded response characteristics, multiple linear blank reflections, and subtle chaotic reflections. Therefore, the model design considers multiple faults, adding cavities and small fracture clusters superimposed on several main faults. This study uses the actual velocities of various reservoirs calculated from well logging and well logging data as a basis. A forward model is established based on the aspect ratio of geological bodies obtained from actual well logging and well logging. The observation system settings and parameters are: shot distance 50m, array length 7500m, Ricker wavelet 22Hz, and calculation grid 5m. A single-well forward model is then used, simulating Kirchihoff migration imaging using the wave equation. The forward simulation results show that the model established based on the "grid structure" exhibits high similarity to the well seismic profile, indicating that the grid structure model is very reasonable and that the migration imaging effect is good.

[0134] Figure 3 A schematic diagram of the grid structure of the Shunbei 47X inclined shaft section according to an embodiment of the present invention is shown.

[0135] To analyze the impact of different fault-reservoir location relationships in gridded reservoirs on the beaded imaging effect, a section of the Shunbei 47X deviated well is selected, such as... Figure 3 As shown in the geological model diagram, the Shunbei 47X well section contains multiple sets of Class I and III reservoirs. Each set of reservoirs exhibits strong heterogeneity, displaying a distinct "grid-like" structure, with even "grid-like" structures nested within each other. Multiple leakage points are also present. Shunbei 47X contains four types of reservoirs and multiple leakage points, indicating the development of large-scale reservoirs. The forward modeling parameters are identical to those of the single-well forward modeling parameters in the observation system. By controlling variables, the target layer is used as a template to study the beaded imaging characteristics under different fault-reservoir location relationships.

[0136] Figure 4 A schematic diagram of a forward model of the relationship between the location of the storage failure according to an embodiment of the present invention is shown.

[0137] from Figure 3 and Figure 4It can be seen that the beaded response on the left side of Shunbei 47X is composed of multiple reservoirs. The difference between Model I and Model II lies in the position of the grid-like reservoirs relative to the fault zone. In Model I, Class I reservoirs are mainly distributed within the main fault, while in Model II, Class I reservoirs are mainly distributed outside the main fault. Model III changes the arrangement of the reservoirs on the left side from an inclined arrangement to a parallel arrangement to test the influence of the grid-like reservoir arrangement angle on the beaded imaging effect. Models IV, V, and VI add small faults with a length of less than 10m to the previous models. Model V treats the reservoirs as part of the main fault zone and has a velocity abrupt change interface, while Model VI shows a uniform velocity change in the reservoirs without a velocity abrupt change interface. Model IV is used to test whether the dense development of small faults will affect beaded imaging. Model V is used to test the beaded imaging situation when the reservoirs are considered as part of the main fault and there is a velocity abrupt change interface. Model VI, based on Model V, shows a uniform velocity change in the reservoirs and tests whether a beaded response can be generated without a significant velocity abrupt change interface.

[0138] Forward modeling results show that the positional relationship between the fault zone and the reservoir has little impact on the beaded characteristics of the grid-like reservoir. Different reservoirs spaced less than 20m apart cannot form independent beads, but instead exhibit elongated beaded characteristics. Small reservoirs with a diameter less than 10m show no obvious seismic response. Abrupt or gradual contact between the reservoir and the fault zone, and the presence or absence of a velocity abrupt change interface, have little impact on seismic response characteristics. High-yield wells in the Shunbei No. 4 belt mostly exhibit strong-amplitude beaded morphology. Statistical analysis of forward modeling results from multiple wells using refined geological models shows that strong-amplitude beads represent the comprehensive seismic response of multiple different types of reservoirs. The better the reservoir's physical properties and the lower the velocity, the stronger the corresponding bead amplitude energy. Its waveform characteristics are mainly influenced by the scale and combination characteristics of the low-velocity, low-density reservoirs, while being less affected by the relatively high-velocity fault zone background.

[0139] Figure 5 A schematic diagram of the forward modeling of the transverse nonhomogeneous wave equation is shown according to an embodiment of the present invention.

[0140] To investigate the impact of the lateral heterogeneity of the "grid-like" structure on seismic reflection characteristics, Figure 5Based on the Shunbei 47X well model shown, an equivalent model was established to simplify the "grid-like" structure. In this equivalent model, yellow corresponds to reservoir types I and II (corresponding to drilling venting and leakage locations), and blue-green corresponds to reservoir type III (without venting or leakage). By changing the velocity values ​​of different equivalent models, the reservoir type was altered, and forward modeling analysis was performed to determine its amplitude and waveform characteristics. The forward modeling results show that reservoirs of the same size and contour, despite different assigned velocities, exhibit similar waveform characteristics. This indicates that without prior drilling data, simply judging which "bead" of venting or leakage occurs based on waveform alone will result in a larger error. However, different types of reservoirs have different amplitude characteristics. Type I and II reservoirs have low velocities and corresponding trough amplitudes with large absolute values. Therefore, the probability of encountering good reservoirs is higher when drilling beads with larger amplitude values. The probability of venting and leakage is higher when drilling into areas with larger amplitude values ​​inside the beads. Different beads in the same area can be classified and sorted by the lateral changes in amplitude, and the beads with larger amplitude energy can be selected as the preferred targets. At the same time, the areas with larger amplitudes inside the beads can be selected as the target points for well trajectory design.

[0141] Figure 6 A schematic diagram of the forward modeling of the longitudinal nonhomogeneous wave equation is shown according to an embodiment of the present invention.

[0142] To further analyze the location of the venting and leakage points on the seismic waveform, multiple forward modeling models were designed, altering the location of Class I reservoirs (corresponding to venting and leakage locations), as shown in the attached figure. Figure 6 Three models of different scales—75×75m, 50×50m, and 25×25m—were set up. For each reservoir group, the leakage point was placed in the middle, lower, and upper parts, respectively. A homogeneous model was also set up for comparison. Taking the first 75×75m model as an example, the Type I reservoir (corresponding to the leakage point) is in the middle of the reservoir group. From the forward modeling results of the wave equation, it can be seen that the leakage point is at the zero phase between red and black. When the Type I reservoir is in the lower part of the reservoir group, the leakage point is located slightly above the peak value, consistent with the convolution forward modeling results. When the Type I reservoir is in the upper part of the reservoir group, the leakage point is located slightly below the trough value. As the reservoir scale decreases, due to the interference of the top and bottom interfaces, the top and bottom no longer correspond to the troughs and peaks. Its heterogeneous internal structure tends to be more homogeneous than the homogeneous model, and the influence of the thin internal layers on the seismic response decreases. Therefore, the location of the leakage point in the small-scale reservoir cannot be accurately determined.

[0143] Through forward modeling of the transverse heterogeneity of grid-like reservoirs, it can be found that the seismic amplitude value corresponding to the venting and leakage point (high-quality reservoir) is relatively large. The "beads" can be selected by amplitude attributes, and the target point can be determined based on the amplitude extreme value inside the bead. In addition, the longitudinal heterogeneity forward modeling shows that the venting and leakage point (high-quality reservoir) is a high-probability event in the zero phase or lower part. Therefore, when designing the trajectory, the amplitude extreme value should be considered while diagonally crossing the zero phase or slightly lower part.

[0144] In the testing of deep learning models, model selection and construction are crucial, and training data is equally important. While attempting to improve prediction accuracy by changing the model architecture, it's also essential to enhance the quality of the input data. When constructing the dataset, thorough data cleaning and labeling are vital; a high-quality dataset often improves both the quality of model training and prediction accuracy. For fault-controlled reservoirs in the Shunbei area, the primary data sources for this invention's sample database are synthetic seismic data (forward modeling), attribute analysis, and manual interpretation.

[0145] Figure 7 A schematic diagram illustrating the construction of a multi-type training sample library according to an embodiment of the present invention is shown.

[0146] Traditional fault interpretation and reservoir prediction rely heavily on manual labor. While time-consuming, labor-intensive, and inefficient, the interpretation results incorporate the geological insights of professional technicians and are highly reliable. Considering the rich achievements in manual interpretation of fault-controlled reservoirs in the Shunbei area, supplementing the sample database with such data as labels, along with the results of forward modeling of gridded reservoirs, can improve the current accuracy of deep learning-based predictions for fault-controlled gridded reservoirs. Figure 7 As shown.

[0147] For small- and medium-scale fault-controlled reservoirs and the "grid-like" structures within these reservoirs, conventional attributes often fail to meet the resolution requirements for reservoir prediction, particularly for the three seismic facies types. This study aims to improve the accuracy of reservoir characterization based on deep learning network models by developing attributes for small- and medium-scale fault zones and expanding the sample library established using conventional seismic attributes.

[0148] The higher-order anisotropic gradient method for structural energy utilizes the difference in amplitude between beaded and non-beaded seismic data to characterize the beaded profile, and uses higher-order gradients to quantitatively describe this amplitude difference. Simultaneously, considering the directional differences in the beaded morphology exhibited in 3D seismic data, higher-order gradients in the line and trace directions are introduced separately. These directional difference higher-order gradients are then fused and displayed in a three-dimensional manner, providing a more detailed description of the reservoir's geometric profile and distribution characteristics in 3D space. The beaded enhancement features are significant, and the fault linearity characteristics are better demonstrated.

[0149] Currently, commonly used methods for estimating seismic trace similarity mainly rely on algorithms based on cross-correlation and similarity. The main drawback of these algorithms is the averaging effect that occurs when combining correlation coefficients or similarity parameters calculated in different directions. The more traces calculated, the stronger this averaging effect becomes, severely reducing spatial resolution and the continuity of anomalies. For predicting the "grid-like" structure within fault-controlled reservoirs, the seismic scale is often limited to one or two traces, making it difficult to characterize using conventional coherence or similarity attributes.

[0150] Figure 8 The diagram shows a seismic profile, a well logging interpretation result diagram, and a schematic diagram of the lateral variation rate of similarity according to an embodiment of the present invention.

[0151] The frequency-domain dip similarity lateral change rate detection technology mainly uses Hilbert transform to convert time-domain seismic attributes to the frequency domain, followed by dip scanning and similarity lateral change calculation to obtain the differences between seismic traces. This invention aims to improve the identification and characterization capabilities of the "grid-like" structure within reservoirs by employing similarity lateral change rate detection parameters with "small windows and small facets," thereby improving the prediction accuracy of reservoirs within reservoir groups. Figure 8 As shown, this demonstrates the actual application effect of the Shunbei 4-6H well.

[0152] Data augmentation (sample library enhancement) refers to methods that generate new training samples by making small changes to existing training samples. The goals of data augmentation are twofold: to increase the amount of data and to improve its quality. After data augmentation, the diversity of training samples increases, and the model's dependence on certain attributes decreases, effectively preventing the network from learning irrelevant features and thus enhancing the model's generalization ability. Since the data labels currently used in "raster" data storage systems are mostly derived from 2D seismic data, data augmentation is necessary to ensure the training quality of deep learning networks.

[0153] Figure 9 A schematic diagram showing a comparison of training accuracy and validation accuracy during iterative training according to an embodiment of the present invention is illustrated.

[0154] The deep Ordovician reservoirs in the Shunbei area are mainly controlled by ultra-deep strike-slip faults and secondary faults associated with fault activity. Fault-controlled reservoirs, as the primary reservoirs, have highly complex internal structures, and the reflection characteristics shown in seismic profiles are largely a combination of responses from several reservoir types. Forward modeling using wave equations revealed that the seismic response of the steep main fault is linear with weak reflections and the presence of in-phase axis faulting; secondary and small faults exhibit mainly chaotic weak reflections; and cavities mainly show beaded or chaotic strong reflections. It was noted that fractures and fracture-cavities are associated, and the reflection anomalies in the seismic profiles are a combined response of faults and fracture-cavities at different scales. Given the complexity of the reservoir structure and types, a total of 220 training data points and labels were generated using forward modeling synthetic records, each with a dimension size of 128×128×128. The training process used 200 training data points to train the U-net network model, and 20 modified validation data points (not used in training) were used for validation. Data input was standardized to avoid interference from amplitude errors. After 100 rounds of iterative training, the U-net model achieved a prediction accuracy of 97.8% on the fracture training data and 95.9% on the fracture validation data. Figure 9 As shown.

[0155] Figure 10a , Figure 10b , Figure 10c , Figure 10d The diagrams show a seismic profile, coherence attributes, original deep learning prediction results, and corrected deep learning prediction results according to an embodiment of the present invention.

[0156] from Figure 10a , Figure 10b , Figure 10c , Figure 10d It can be seen that the corrected deep learning fracture prediction effect has the highest resolution, rich spatial extension details of fracture characterization, good integrity and source of main fractures, and secondary and adjacent fractures are also completely predicted; the overall vertical development law of associated small fractures and interlayer cracks shows the typical characteristics of fracture-controlled reservoirs.

[0157] The deep learning-based reservoir "grid" structure identification also uses the same U-net network structure model, but with different training data. The training data is used to remove features of disordered weak reflection, beaded strong reflection, and beaded disordered reflection. Combined with "grid" sensitive attributes based on forward modeling, including higher-order structural gradients and similarity lateral variation rates, training data for small-to-medium scale fault-controlled fracture-vuggy reservoirs in Shunbei is created.

[0158] Figure 11 A schematic diagram of a seismic profile, instantaneous energy, and deep learning-based “grid” internal reservoir distribution according to an embodiment of the present invention is shown.

[0159] Using the Shunbei 10X well area as the test area, the seismic sensitivity attributes of conventional fault-controlled reservoirs were first calculated, including tensor, instantaneous energy, disorder, coherence, structural gradient, and similarity lateral variation rate. These attributes were then used to establish and train a deep neural network model to predict the "grid-like" structure within the Shunbei 10X target point, i.e., the distribution of high-quality reservoirs. The process mainly consisted of three parts: First, multiple attributes were extracted based on the reservoir's characteristics in the seismic data. For each attribute, a small amount of master survey line and time slice data was preprocessed, and a large amount of data was automatically generated as training samples for the network using data augmentation methods. Then, corresponding models were established and trained and tested on two types of samples with different input attributes. Finally, to comprehensively consider all attributes and obtain more comprehensive and accurate prediction results, ensemble learning methods were used to fuse multiple models and obtain optimized prediction results for the "grid-like" structure within the reservoir, such as... Figure 11 As shown in the figure, the comparison between the reservoir prediction by the deep learning method and the conventional seismic attribute prediction at the Shunbei 10X well point is shown. The prediction results have a high degree of consistency with the drilling, indicating the reliability of the method.

[0160] Example 4

[0161] Figure 12 A block diagram of a geophysical characterization device for an ultra-deep strike-slip fault-controlled grid reservoir according to an embodiment of the present invention is shown.

[0162] like Figure 12 As shown, this geophysical characterization device for ultra-deep strike-slip fault-controlled grid-like reservoirs includes:

[0163] Forward model establishment module 201 establishes a forward model based on a grid-like reservoir;

[0164] Forward simulation analysis module 202 performs forward simulation analysis of the grid reservoir for the forward model;

[0165] The sample library construction module 203 constructs a deep learning sample library for the raster reservoir based on the results of forward simulation analysis.

[0166] The representation module 204 establishes and trains the U-net network model, and uses the trained network model to represent the raster storage collective.

[0167] In one example, building a forward model based on a grid-like reservoir includes:

[0168] Based on the actual seismic profile and combined with reservoir geology, a geological model was designed, and the reservoir characteristics of the target well section were determined from the seismic profile and actual drilling leakage.

[0169] The actual velocities of various reservoirs are calculated using well logging and well logging data, and a forward model is established based on the length and width ratio of the geological bodies obtained from actual well logging and well logging.

[0170] In one example, reservoir characteristics include the number of reservoirs, beaded response characteristics, and reflection characteristics.

[0171] In one example, Kirchihoff migration imaging is simulated using the wave equation for a forward model, and the forward simulation analysis results are obtained.

[0172] In one example, the results of the forward simulation analysis include:

[0173] Different types of reservoirs have different amplitude characteristics. The reservoir has a low velocity and a large absolute value of the corresponding trough amplitude. That is, the probability of encountering good reservoirs is high for beads with large amplitude values. The probability of venting and leakage is high in the parts with large amplitude values ​​inside the beads. Therefore, different beads in the same area are classified and sorted by the lateral change of amplitude, with the large amplitude energy as the target, and the parts with large amplitude inside the beads are identified as the target points for well trajectory design.

[0174] In one example, based on the results of forward simulation analysis, the deep learning sample library for constructing a grid-like reservoir includes:

[0175] When constructing the sample library, manually interpreted data is added as labels, and forward modeling analysis results are also added to the sample library. Attribute research is carried out for small- and medium-scale fault zones to expand the sample library established by conventional earthquake attributes. Then, sample library data augmentation processing is carried out to ensure the training quality of the deep learning network.

[0176] In one example, building and training the U-net network model includes:

[0177] The data in the deep learning sample library is divided into a training set and a validation set;

[0178] A U-net network model is established, trained using a training set, and validated using a validation set.

[0179] Example 5

[0180] This disclosure provides an electronic device comprising: a memory storing executable instructions; and a processor executing the executable instructions in the memory to implement the aforementioned geophysical characterization method for ultra-deep strike-slip fault-controlled grate reservoirs.

[0181] An electronic device according to an embodiment of the present disclosure includes a memory and a processor.

[0182] This memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0183] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory.

[0184] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.

[0185] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0186] Example 6

[0187] This disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the described geophysical characterization method for ultra-deep strike-slip fault-controlled grate reservoirs.

[0188] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When these non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the methods described in the foregoing embodiments of the present disclosure are performed.

[0189] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0190] Those skilled in the art should understand that the above description of the embodiments of the present invention is only intended to illustrate the beneficial effects of the embodiments of the present invention, and is not intended to limit the embodiments of the present invention to any of the examples given.

[0191] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A geophysical characterization method for ultra-deep strike-slip fault-controlled grid-like reservoirs, characterized in that, include: Establish a forward model based on gridded reservoirs; Forward modeling analysis of the grid reservoir was performed on the aforementioned forward model; Based on the results of forward modeling analysis, a deep learning sample library for graticular reservoirs is constructed. A U-net network model is established and trained, and the trained network model is used to represent the raster reservoir.

2. The geophysical characterization method for ultra-deep strike-slip fault-controlled grid-like reservoirs according to claim 1, wherein, Establishing a forward model based on a grid-like reservoir includes: Based on the actual seismic profile and combined with reservoir geology, a geological model was designed, and the reservoir characteristics of the target well section were determined from the seismic profile and actual drilling leakage. The actual velocities of various reservoirs are calculated using well logging and well logging data, and the forward model is established based on the length-to-width ratio of the geological bodies obtained from actual well logging and well logging.

3. The geophysical characterization method for ultra-deep strike-slip fault-controlled grid-like reservoirs according to claim 2, wherein, The reservoir characteristics include the number of reservoirs, beaded response characteristics, and reflection characteristics.

4. The geophysical characterization method for ultra-deep strike-slip fault-controlled grid-like reservoirs according to claim 1, wherein, For the aforementioned forward model, Kirchihoff migration imaging was simulated using the wave equation to obtain forward modeling simulation analysis results.

5. The geophysical characterization method for ultra-deep strike-slip fault-controlled grid-like reservoirs according to claim 4, wherein, The forward simulation analysis results include: Different types of reservoirs have different amplitude characteristics. The reservoir has a low velocity and a large absolute value of the corresponding trough amplitude. That is, the probability of encountering good reservoirs is high for beads with large amplitude values. The probability of venting and leakage is high in the parts with large amplitude values ​​inside the beads. Therefore, different beads in the same area are classified and sorted by the lateral change of amplitude, with the large amplitude energy as the target, and the parts with large amplitude inside the beads are identified as the target points for well trajectory design.

6. The geophysical characterization method for ultra-deep strike-slip fault-controlled grid-like reservoirs according to claim 1, wherein, Based on the results of forward simulation analysis, the deep learning sample library for the raster reservoir is constructed as follows: When constructing the sample library, manually interpreted data is added as labels, and forward modeling analysis results are also added to the sample library. Attribute research is carried out for small- and medium-scale fault zones to expand the sample library established by conventional earthquake attributes. Then, sample library data augmentation processing is carried out to ensure the training quality of the deep learning network.

7. The geophysical characterization method for ultra-deep strike-slip fault-controlled grid-like reservoirs according to claim 1, wherein, Building and training the U-net network model includes: The data in the deep learning sample library is divided into a training set and a validation set; The U-net network model is established, trained using the training set, and validated using the validation set.

8. A geophysical characterization device for ultra-deep strike-slip fault-controlled grid-like reservoirs, characterized in that, include: The forward model building module establishes a forward model based on a grid-like reservoir. The forward modeling simulation analysis module performs forward modeling simulation analysis on the gridded reservoir for the forward model. The sample library construction module constructs a deep learning sample library for a raster reservoir based on the results of forward simulation analysis. The representation module establishes and trains the U-net network model, and uses the trained network model to represent the raster storage collective.

9. An electronic device, characterized in that, The electronic device includes: Memory, which stores executable instructions; A processor that executes the executable instructions in the memory to implement the geophysical characterization method for ultra-deep strike-slip fault-controlled grate reservoirs according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the geophysical characterization method for ultra-deep strike-slip fault-controlled grate reservoirs as described in any one of claims 1-7.

Images