Seismic horizon identification method and device and storage medium
By training a multi-objective seismic horizon identification network using a convolutional neural network and combining it with an adaptive adjustment operator and fault correction, the problem of low accuracy in seismic horizon identification was solved, achieving high-precision seismic horizon identification and accurate seismic interpretation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA NAT PETROLEUM CORP
- Filing Date
- 2022-11-07
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies have low accuracy in identifying seismic horizons, which affects the accuracy of seismic interpretation, especially near faults. Furthermore, artificial intelligence methods lack sufficient accuracy in identifying these horizons.
A multi-target horizon recognition network was trained using a convolutional neural network. By acquiring seismic data volumes and multi-target horizon recognition sample sets, manually interpreted seismic horizons and auxiliary seismic horizons were processed after annotation. The network was then fused using an adaptive adjustment operator to identify target seismic horizons and perform fault correction.
It improves the accuracy of seismic horizon identification, ensures the accuracy and efficiency of seismic interpretation, and is suitable for the detection and identification of complex seismic horizons.
Smart Images
Figure CN118053061B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of seismic interpretation technology, and in particular to a method, apparatus and storage medium for identifying seismic horizons. Background Technology
[0002] Seismic interpretation refers to the process of determining the geological significance of each layer (i.e., seismic horizon) in a seismic profile based on its waveform characteristics and geological patterns. Therefore, identifying seismic horizons is particularly important for seismic interpretation.
[0003] In related technologies, seismic horizons are typically identified by performing image edge detection on seismic profiles, obtaining the edge contours, and then filling these contours. However, these methods often suffer from low accuracy in identifying seismic horizons due to the low precision of the identified edge contours, thus hindering accurate seismic interpretation. Summary of the Invention
[0004] In view of this, this application provides a method, apparatus and storage medium for identifying seismic horizons, which can accurately identify seismic horizons and improve the accuracy of seismic interpretation.
[0005] Specifically, the following technical solutions are included:
[0006] In a first aspect, embodiments of this application provide a method for identifying seismic horizons, the method comprising:
[0007] Acquiring earthquake data and multi-target layer identification sample set Among them, the earthquake data body Used to generate sample seismic profiles W r The multi-target hierarchical identification sample set Including the sample seismic profile W r Manually interpreted seismic horizons obtained after annotation processing and the artificially interpreted seismic horizon At least four relevant auxiliary seismic horizons;
[0008] According to the sample seismic profile W r and multi-target layer identification sample set Train a convolutional neural network to obtain a multi-target hierarchical recognition network Net. L The sample seismic profile W r and multi-target layer identification sample set These correspond to the input and output of the convolutional neural network;
[0009] The seismic profile to be identified is substituted into the multi-target horizon identification network Net. L In the process, a transitional seismic horizon and at least four auxiliary seismic horizons associated with the transitional seismic horizon are obtained;
[0010] The transitional seismic horizon and at least four auxiliary seismic horizons associated with it are fused to determine the target seismic horizon.
[0011] In some embodiments, the method based on the sample seismic profile W r and multi-target layer identification sample set Train a convolutional neural network to obtain a multi-target hierarchical recognition network Net. L include:
[0012] Convolutional neural networks are constructed, wherein the convolutional neural network includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a branch network, a seventh convolutional layer, and an eighth convolutional layer connected in series, wherein the branch network includes at least five sub-networks connected in parallel, and each sub-network includes a first upsampling layer, a fourth convolutional layer, a second upsampling layer, a fifth convolutional layer, a third upsampling layer, and a sixth convolutional layer connected in series;
[0013] The sample seismic profile W r and multi-target layer identification sample set The input is substituted into the convolutional neural network for training to obtain the multi-target hierarchical recognition network Net. L .
[0014] In some embodiments, the sample seismic profile W r and multi-target layer identification sample set The input is substituted into the convolutional neural network for training to obtain the multi-target hierarchical recognition network Net. L include:
[0015] The sample seismic profile W r and multi-target layer identification sample set Substitute the data into the convolutional neural network for training to obtain the transition layer recognition network;
[0016] Obtain known seismic profiles and artificially interpreted seismic horizons obtained after annotating the known seismic profiles, and at least four auxiliary seismic horizons related to the artificially interpreted seismic horizons;
[0017] Substituting the known seismic profile into the transition horizon identification network yields the predicted seismic horizon and at least four auxiliary seismic horizons associated with the predicted seismic horizon.
[0018] The target loss value is calculated based on the predicted seismic horizon and at least four auxiliary seismic horizons associated with the predicted seismic horizon, as well as the artificially interpreted seismic horizon and at least four auxiliary seismic horizons associated with the artificially interpreted seismic horizon obtained after annotating the known seismic profile.
[0019] In response to the target loss value being less than or equal to a threshold, the transition layer identification network is used as the multi-target layer identification network Net. L .
[0020] In some embodiments, the transitional seismic horizon and at least four auxiliary seismic horizons associated with the transitional seismic horizon include a set of horizons along the survey line direction and a set of horizons perpendicular to the survey line direction. The process of fusing the transitional seismic horizon and the at least four auxiliary seismic horizons associated with the transitional seismic horizon to determine the target seismic horizon involves... include:
[0021] Obtain the adaptive adjustment operator α;
[0022] Based on the adaptive adjustment operator α, the set of horizons along the survey line direction, and the set of horizons perpendicular to the survey line direction, the target seismic horizon is calculated.
[0023] In some embodiments, the set of horizons along the survey line includes the target seismic horizons along the survey line. The set of horizons perpendicular to the survey line direction includes the target seismic horizons perpendicular to the survey line direction. The target seismic horizon is calculated based on the adaptive adjustment operator α, the set of horizons along the survey line direction, and the set of horizons perpendicular to the survey line direction. include:
[0024] Based on the adaptive adjustment operator α, the target seismic horizon along the survey line direction is... and the target seismic horizon perpendicular to the survey line direction The target seismic horizon is obtained by performing a weighted summation. The target seismic horizon The following formula is used for calculation:
[0025]
[0026] In some embodiments, the process of fusing the transitional seismic horizon and at least four auxiliary seismic horizons associated with the transitional seismic horizon to determine the target seismic horizon involves... Subsequently, the method further includes:
[0027] For the target seismic horizon Fault correction processing is performed to obtain continuous seismic horizons after fault correction.
[0028] In some embodiments, the acquisition of seismic data volume D and multi-target horizon identification sample set include:
[0029] Obtain seismic data volume D and original manually interpreted seismic horizons, wherein the original manually interpreted seismic horizons are the labeled horizons obtained after annotating the seismic profiles generated from the seismic data volume D;
[0030] Interpolation processing is performed on the original manually interpreted seismic horizons to obtain at least four original auxiliary seismic horizons. The original manually interpreted seismic horizons and the at least four original auxiliary seismic horizons are then used as a multi-objective sample set.
[0031] The seismic data volume D is augmented to obtain the augmented seismic data volume.
[0032] For the multi-target sample set and the augmented seismic data volume Augmentation processing is performed to obtain the seismic data volume. and the multi-target layer identification sample set
[0033] In some embodiments, the original artificially interpreted seismic horizon includes multiple center points p i =(x i ,y i ,t i The interpolation process performed on the original manually interpreted seismic horizons to obtain at least four original auxiliary seismic horizons includes:
[0034] Obtain the center point p i =(x i ,y i ,t i Centered on t i Time along survey line x i Direction and perpendicular to the survey line y i Earthquake traces in the direction The maximum and minimum values of , where i is the subscript index of the original manually interpreted seismic horizon, and the value of i is . This indicates the number of center points;
[0035] Based on the earthquake trace Find the maxima and minima to obtain the sequence of local maxima points. and local minimum point sequence Among them, u j This means that at t j Time along survey line x i Direction and perpendicular to the survey line y i Earthquake traces in the direction The maximum point of v k This means that at t k Time along survey line x i Direction and perpendicular to the survey line y i Earthquake traces in the direction The minimum point, The number of maxima. Let j be the number of local minima, and j take the value of 1. The value of k is
[0036] For the local maximum point sequence and local minimum point sequence After sorting, the maximum point u is obtained. j and the minimum point v k ;
[0037] The maximum point u j and the minimum point v k The corresponding center point p i The four sets and These are the four original auxiliary seismic horizons.
[0038] Secondly, embodiments of this application provide a seismic horizon identification device, the device comprising:
[0039] The acquisition module is used to acquire seismic data volumes. and multi-target layer identification sample set Among them, the earthquake data body Used to generate sample seismic profiles W r The multi-target hierarchical identification sample set Including the sample seismic profile W r Manually interpreted seismic horizons obtained after annotation processing and the artificially interpreted seismic horizon At least four relevant auxiliary seismic horizons;
[0040] Training module, used to train based on the sample seismic profile W r and multi-target layer identification sample set Train a convolutional neural network to obtain a multi-target hierarchical recognition network Net.L The sample seismic profile W r and multi-target layer identification sample set These correspond to the input and output of the convolutional neural network;
[0041] The identification module is used to input the seismic profile to be identified into the multi-target horizon identification network Net. L In the process, a transitional seismic horizon and at least four auxiliary seismic horizons associated with the transitional seismic horizon are obtained;
[0042] The fusion module is used to fuse the transitional seismic horizon and at least four auxiliary seismic horizons related to the transitional seismic horizon to determine the target seismic horizon.
[0043] Thirdly, embodiments of this application provide a non-volatile computer-readable storage medium that, when instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to perform the seismic horizon identification method as described in the first aspect above.
[0044] The seismic horizon identification method provided in this application embodiment identifies the seismic horizon by analyzing the seismic data volume. The generated sample seismic profile W r and multi-target layer identification sample set The included sample seismic profile W r Manually interpreted seismic horizons obtained after annotation processing and artificial interpretation of seismic horizon At least four relevant auxiliary seismic horizons are fed into a convolutional neural network for training, resulting in the multi-target horizon recognition network Net. L When it is necessary to identify seismic horizons in a seismic profile to be identified, the seismic profile to be identified can be substituted into the multi-target horizon identification network Net. L In this method, a transitional seismic horizon and at least four auxiliary seismic horizons associated with it are obtained. The target seismic horizon is then obtained by fusing these two horizons. Since the multi-target horizon identification sample set in this method includes manually interpreted seismic horizons obtained after annotating sample seismic profiles and at least four auxiliary seismic horizons associated with these manually interpreted horizons, the training samples are sufficiently rich, enabling the multi-target horizon identification network Net trained based on these samples to achieve high performance. L It has high identification accuracy, enabling accurate identification of seismic horizons and improving the accuracy of seismic interpretation. Attached Figure Description
[0045] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0046] Figure 1 A flowchart illustrating a method for identifying seismic horizons provided in an embodiment of this application;
[0047] Figure 2 This application provides a method for identifying seismic horizons, which involves obtaining seismic data volumes. and multi-target layer identification sample set Flowchart;
[0048] Figure 3 In a seismic horizon identification method provided in this application embodiment, the seismic profile W is used as the basis for identification. r and multi-target layer identification sample set Train a convolutional neural network to obtain a multi-target hierarchical recognition network Net. L Flowchart;
[0049] Figure 4 In a seismic horizon identification method provided in this application embodiment, a transitional seismic horizon and at least four auxiliary seismic horizons related to the transitional seismic horizon are fused to determine the target seismic horizon. Flowchart;
[0050] Figure 5 An image showing the effect of seismic horizon identification using a seismic horizon identification method provided in this application.
[0051] Figure 6 This is a schematic diagram of the structure of a seismic horizon identification device provided in an embodiment of this application. Detailed Implementation
[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0053] Seismic interpretation refers to the process of determining the geological significance of each layer (i.e., seismic horizon) in a seismic profile based on its waveform characteristics and geological regularities. Seismic interpretation is a crucial stage in seismic exploration engineering, and identifying seismic horizons is of paramount importance for its success.
[0054] Currently, there are two types of methods for identifying seismic horizons in related technologies:
[0055] One method involves identifying seismic horizons using phase axis tracing. Phase axis tracing is the process of identifying and tracing waves reflected from the same interface within a seismic profile. Since actual seismic horizons often correspond to phase axes on the seismic profile, continuous tracing of phase axes on the seismic profile is a prerequisite for obtaining valid seismic horizons. This method is time-consuming and labor-intensive, relying on subjective judgments made by seismic interpreters based on their knowledge and experience, and is influenced by prior human knowledge.
[0056] Another type of method is the artificial intelligence-based seismic horizon identification method, such as single seed point tracking, waveform similarity comparison, image edge detection, and multi-attribute feature fusion. This type of method has low accuracy in identifying seismic horizons, failing to precisely pinpoint them and further affecting the accuracy of seismic interpretation. Furthermore, these methods are heavily influenced by faults, leading to inaccurate identification of horizons near faults, reducing the accuracy of horizon identification and impacting the accuracy of seismic interpretation. In addition, frequent modification and editing of incorrectly identified points can also affect subsequent operations such as seismic interpretation and mapping.
[0057] To meet the requirements of seismic horizon identification, it is necessary to introduce efficient, accurate, and practical intelligent algorithms. With the development of artificial intelligence, especially the rise and development of research represented by pattern recognition and deep learning, more and more new methods are being applied to the field of seismic interpretation technology. To address the problems existing in related technologies, this application proposes a method for seismic horizon identification.
[0058] Figure 1 A flowchart illustrating a seismic horizon identification method provided in this application embodiment is available. Figure 1 The method includes the following steps.
[0059] Step 101, Obtain the seismic data volume and multi-target layer identification sample set Among them, earthquake data volume Used to generate sample seismic profiles W r Multi-target hierarchical identification sample set This includes manually interpreted seismic horizons obtained after annotating the sample seismic profile W. and artificial interpretation of seismic horizon At least four related auxiliary seismic horizons.
[0060] See Figure 2 This step includes the following sub-steps:
[0061] Step 1011: Obtain the seismic data volume D and the original manually interpreted seismic horizons.
[0062] Among them, the original manually interpreted seismic horizons are the labeled horizons obtained after annotating the seismic profiles generated from the seismic data volume D.
[0063] In order to conduct subsequent network training, it is necessary to obtain the original manually interpreted seismic horizons.
[0064] It should be noted that the original manually interpreted seismic horizons are the horizons marked on the seismic profile generated by the seismic data volume D by seismic interpreters using professional seismic interpretation software.
[0065] Step 1012: Interpolate the original manually interpreted seismic horizons to obtain at least four original auxiliary seismic horizons, and use the original manually interpreted seismic horizons and at least four original auxiliary seismic horizons as a multi-objective sample set.
[0066] Interpolation can effectively process invalid values in the seismic data volume D, yielding at least four original auxiliary seismic horizons. These four original auxiliary seismic horizons, which are related to the original manually interpreted seismic horizons, can be used to describe the original manually interpreted seismic horizons from multiple perspectives. They also provide rich samples for subsequent network training, facilitating feature extraction.
[0067] In some embodiments, the interpolation process may be performed using bilinear interpolation, which can improve the quality of image scaling.
[0068] In some embodiments, the original manually interpreted seismic horizon includes multiple center points p i =(x i ,y i ,t i ).
[0069] The specific implementation method for interpolating the original manually interpreted seismic horizons to obtain at least four original auxiliary seismic horizons can be as follows:
[0070] (1) Obtain the center point p i =(x i ,y i ,t i Centered on t i Time along survey line x i Direction and perpendicular to the survey line y i Earthquake traces in the direction The maximum and minimum values of .
[0071] Where i is the subscript index of the original manually interpreted seismic horizon, and the value of i is... This indicates the number of center points.
[0072] (2) Based on seismic traces Find the maxima and minima to obtain the sequence of local maxima points. and local minimum point sequence
[0073] Among them, u j This means that at t j Time along survey line x i Direction and perpendicular to the survey line y i Earthquake traces in the direction The maximum point of v k This means that at t k Time along survey line x i Direction and perpendicular to the survey line y i Earthquake traces in the direction The minimum point, The number of maxima. Let j be the number of local minima, and j take the value of 1. The value of k is
[0074] (3) For the sequence of local maxima and local minimum point sequence After sorting, the maximum point u is obtained. j and the minimum point v k .
[0075] It is understandable that the sorting process is performed according to chronological order, and the sorted sequence of maximum points is as follows: The sequence of minimum points is
[0076] (4) The maximum point u j and the minimum point v k The corresponding center point p i The four sets and These are the four original auxiliary seismic horizons.
[0077] Step 1013: Expand the seismic data volume D to obtain the expanded seismic data volume.
[0078] Expanding the seismic data volume D can increase the richness of the sample and further improve the data accuracy.
[0079] In some embodiments, due to the multi-target sample set after interpolation processing The dataset contains at least four original auxiliary seismic horizons, therefore, for the interpolated multi-target sample set... By matching data points, the seismic data volume D can be expanded, and the sampling interval can be modified from the initial 0.4-0.8ms to 0.1ms.
[0080] Step 1014, for the multi-target sample set and the expanded seismic data volume Augmentation processing is performed to obtain the seismic data volume. and multi-target layer identification sample set
[0081] Augmentation processing can be used to transform multi-target sample sets. and the expanded seismic data volume The richer data samples improve the performance of the multi-target hierarchical recognition network Net obtained during subsequent training. L The recognition accuracy.
[0082] In some embodiments, the multi-target sample set is processed according to the following formula. The middle represents all p of the layer. i Point augmentation processing is performed to obtain a multi-target layer recognition sample set.
[0083]
[0084] In the formula, η and μ represent the coordinates of the transformed point, x and y represent the coordinates of the point before the transformation, and a1,b1,c1,d1,e1,f1 and a2,b2,c2,d2,e2,f2 represent the transformation parameters.
[0085] In some embodiments, the augmented seismic data volume is processed according to the following formula. All seismic trace data points are augmented to obtain the seismic data volume.
[0086]
[0087] In the formula, m and n represent the coordinates of the transformed point, g and h represent the coordinates of the point before the transformation, and a1,b1,c1,d1,e1,f1 and a2,b2,c2,d2,e2,f2 represent the transformation parameters.
[0088] Understandably, in order to ensure the correspondence, for multi-target sample sets... and the expanded seismic data volume The transformation parameters for augmentation are consistent.
[0089] Step 102, based on the sample seismic profile W r and multi-target layer identification sample set Train a convolutional neural network to obtain a multi-target hierarchical recognition network Net. L Among them, the sample seismic profile W r and multi-target layer identification sample set These correspond to the input and output of a convolutional neural network.
[0090] By training a convolutional neural network, a multi-object recognition network Net is obtained. L This prepares for the subsequent identification of seismic horizons.
[0091] See Figure 3 This step may include the following sub-steps:
[0092] Step 1021: Construct a convolutional neural network.
[0093] The convolutional neural network includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a branch network, a seventh convolutional layer, and an eighth convolutional layer connected in series. The branch network includes at least five sub-networks connected in parallel. Each sub-network includes a first upsampling layer, a fourth convolutional layer, a second upsampling layer, a fifth convolutional layer, a third upsampling layer, and a sixth convolutional layer connected in series.
[0094] Based on actual sample requirements, and taking into account both data volume and processing speed, this convolutional neural network is constructed to facilitate subsequent seismic horizon identification.
[0095] It's important to note that convolutional layers can enhance image features and reduce noise, facilitating feature extraction. The more convolutional layers, the more detailed and complex the feature extraction; conversely, fewer convolutional layers result in simpler and more basic feature extraction. The appropriate number of convolutional layers needs to be determined based on the actual sample data. Pooling layers can simplify the output of convolutional layers, reduce feature dimensionality, and compress data. Pooling layers are typically placed after convolutional layers, and their number is determined based on the actual sample data.
[0096] In some embodiments, the first, second, third, fourth, fifth, sixth, seventh, and eighth convolutional layers each include three 3x3 convolutional kernels and perform convolutions of the same scale.
[0097] In some embodiments, the first pooling layer, the second pooling layer, and the third pooling layer all employ max pooling.
[0098] In some embodiments, the branch network includes at least five sub-networks connected in parallel and has a merge function that can merge the at least five parallel-connected sub-networks together. The merge function can be a concat function.
[0099] In some embodiments, the first upsampling layer, the second upsampling layer, and the third upsampling layer in the branch network are used to magnify the image, giving it a higher resolution.
[0100] In some embodiments, seismic wave group characteristics are used for horizon identification, and multi-target detection is used to constrain the scope of horizon identification.
[0101] Step 1022, transfer the sample seismic profile W r and multi-target layer identification sample set The data is fed into a convolutional neural network for training, resulting in the multi-object hierarchical recognition network Net. L .
[0102] By training the constructed convolutional neural network, a multi-target hierarchical recognition network Net is obtained. L This facilitates the identification of seismic horizons.
[0103] In some embodiments, seismic wave group features are used for seismic horizon identification, and multi-target detection is used to constrain the scope of seismic horizon identification.
[0104] It should be noted that using seismic wave group characteristics for seismic horizon identification and employing multi-target detection to constrain the scope of seismic horizon identification can improve the efficiency of seismic horizon identification. Extracting the geological features of seismic horizons by dynamically calculating their seismic attributes has the advantages of low storage space consumption and high identification accuracy.
[0105] This step can be achieved as follows:
[0106] (1) The sample seismic profile W r and multi-target layer identification sample set The data is then fed into a convolutional neural network for training to obtain a transition layer recognition network.
[0107] Through training, a transitional layer recognition network that needs to be verified is obtained, thus preparing for verification.
[0108] (2) Obtain known seismic profiles and artificially interpreted seismic horizons obtained after annotating the known seismic profiles, and at least four auxiliary seismic horizons related to the artificially interpreted seismic horizons.
[0109] Seismic horizon identification for the transition horizon identification network is achieved by acquiring known seismic profiles, manually interpreted seismic horizons obtained after annotating the known seismic profiles, and at least four auxiliary seismic horizons associated with the manually interpreted seismic horizons.
[0110] (3) Substitute the known seismic profile into the transition horizon identification network to obtain the predicted seismic horizon and at least four auxiliary seismic horizons related to the predicted seismic horizon.
[0111] By obtaining the predicted seismic horizon and at least four auxiliary seismic horizons associated with the predicted seismic horizon, the target loss value can be calculated.
[0112] (4) The target loss value is calculated based on the predicted seismic horizon and at least four auxiliary seismic horizons related to the predicted seismic horizon, as well as the artificially interpreted seismic horizon and at least four auxiliary seismic horizons related to the artificially interpreted seismic horizon obtained after annotating the known seismic profile.
[0113] The training performance of the transition layer recognition network can be judged by using the target loss value.
[0114] Based on multiple seismic horizons trained from the initial data, the loss value is determined according to the following formula:
[0115]
[0116] Among them, L loss The loss value is mask. q (*) represents the position index function, which indexes the position along the x-axis of the survey line. i The yth online i The location of each layer on the track, i is the subscript index of the manually interpreted seismic layer sample, q is the subscript index of the manually interpreted seismic layer and the auxiliary seismic layer, N is the maximum value of i, K is the maximum value of q, l(f known ,f q (x i The expression ,o)) represents the loss function of the transition layer identification network, f known This represents the artificially interpreted seismic horizon in a known seismic profile, f q (x i ,o) represents the predicted seismic horizon and at least four auxiliary seismic horizons associated with the predicted seismic horizon, where o represents the parameters of the loss function. This represents the constraint matrix of the loss function.
[0117] It should be noted that when there is a multi-target sample set When including the original artificially interpreted seismic horizon and at least four original auxiliary seismic horizons, the maximum value of K is 5.
[0118] (5) In response to the target loss value being less than or equal to the threshold, the transition layer identification network is used as the multi-target layer identification network Net. L .
[0119] The loss value can be used to judge the training effect and control the training process. When the loss value is less than or equal to the threshold, the transition layer recognition network is used as the multi-target layer recognition network Net. L When the loss value exceeds the threshold, the transition layer recognition network needs to be trained again until the loss value is less than or equal to the threshold. The transition layer recognition network with a loss value less than or equal to the threshold is then used as the multi-target layer recognition network Net. L .
[0120] Step 103: Substitute the seismic profile to be identified into the multi-target horizon identification network Net. L In the process, transitional seismic horizons and at least four auxiliary seismic horizons associated with the transitional seismic horizons were obtained.
[0121] By substituting the seismic profile to be identified into the multi-target horizon identification network Net L By performing identification, the transitional seismic horizon corresponding to the seismic profile to be identified and at least four auxiliary seismic horizons associated with the transitional seismic horizon can be obtained.
[0122] Step 104: Perform fusion processing on the transitional seismic horizon and at least four auxiliary seismic horizons related to the transitional seismic horizon to determine the target seismic horizon.
[0123] The target seismic horizon is determined through fusion processing.
[0124] See Figure 4 This step may include the following sub-steps:
[0125] Step 1041: Obtain the adaptive adjustment operator α.
[0126] Obtain the adaptive adjustment operator α to prepare for subsequent calculation of the target seismic horizon.
[0127] Step 1042: Based on the adaptive adjustment operator α, the set of horizons along the survey line direction, and the set of horizons perpendicular to the survey line direction, the target seismic horizon is calculated.
[0128] In some embodiments, the set of horizons along the survey line includes the target seismic horizons along the survey line. The set of horizons perpendicular to the survey line includes the target seismic horizons perpendicular to the survey line direction.
[0129] By analyzing the target seismic horizon along the survey line and the target seismic horizon perpendicular to the survey line direction The target seismic horizon can be obtained by performing calculations.
[0130] In some embodiments, the target seismic horizon along the survey line is determined based on the adaptive adjustment operator α. and the target seismic horizon perpendicular to the survey line direction Perform a weighted summation to obtain the target seismic horizon. Among them, the target seismic horizon The following formula is used for calculation:
[0131]
[0132] In some embodiments, after obtaining the target seismic horizon After that, the target seismic horizon can be determined. Fault correction processing is performed to obtain continuous seismic horizons after fault correction.
[0133] Due to the target seismic horizon There may be faults in the middle, so it is necessary to determine the target seismic horizon. Fault correction is performed to ensure that the obtained seismic horizons are continuous.
[0134] This step can be implemented as follows: Obtain the seismic horizon correction network; substitute the known seismic profiles and the seismic horizons located on the known seismic profiles into the seismic horizon correction network for training, to obtain the trained continuous seismic horizon network; and then, for the target seismic horizon... The data is input into a continuous seismic horizon model and corrected to obtain the continuous seismic horizon after fault correction.
[0135] Figure 5 This is a rendering of the seismic horizon identified using the seismic horizon identification method provided in this application. See also... Figure 5 This allows us to see the location of seismic layers in the seismic profile, which facilitates seismic interpretation.
[0136] This application provides a method for identifying seismic horizons by substituting the seismic profile to be identified into a multi-target horizon identification network Net. LSeismic horizon identification is performed to obtain transitional seismic horizons and at least four auxiliary seismic horizons associated with them. These are then fused to obtain the target seismic horizon. The multi-target horizon identification sample set in this method includes manually interpreted seismic horizons obtained by annotating sample seismic profiles and at least four auxiliary seismic horizons associated with these manually interpreted seismic horizons. This indicates that the training sample is sufficiently rich, enabling the multi-target horizon identification network Net trained based on this sample to achieve the desired results. L It has high recognition accuracy and is suitable for the detection and identification stage of complex seismic horizons in the process of interpreting 3D seismic data. It can help seismic interpreters to quickly, conveniently and accurately identify complex seismic horizons, thereby helping them to make correct judgments on seismic data.
[0137] Figure 6 A schematic diagram of a seismic horizon identification device provided in an embodiment of this application. See also... Figure 6 The device includes:
[0138] Module 601 is used to acquire seismic data volumes. and multi-target layer identification sample set Among them, earthquake data volume Used to generate sample seismic profiles W r Multi-target hierarchical identification sample set This includes manually interpreted seismic horizons obtained after annotating the sample seismic profile W. and artificial interpretation of seismic horizon At least four relevant auxiliary seismic horizons;
[0139] Training module 602 is used to train samples based on seismic profiles W. r and multi-target layer identification sample set Train a convolutional neural network to obtain a multi-target hierarchical recognition network Net. L Among them, the sample seismic profile W r and multi-target layer identification sample set These correspond to the input and output of a convolutional neural network;
[0140] The identification module 603 is used to input the seismic profile to be identified into the multi-target horizon identification network Net. L In the process, transitional seismic horizons and at least four auxiliary seismic horizons associated with the transitional seismic horizons were obtained;
[0141] Fusion module 604 is used to fuse transitional seismic horizons and at least four auxiliary seismic horizons associated with the transitional seismic horizons to determine the target seismic horizon.
[0142] In some embodiments, the training module 602 includes:
[0143] A network building unit is used to build a convolutional neural network, wherein the convolutional neural network includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a branch network, a seventh convolutional layer, and an eighth convolutional layer connected in series, wherein the branch network includes at least five sub-networks connected in parallel, and each sub-network includes a first upsampling layer, a fourth convolutional layer, a second upsampling layer, a fifth convolutional layer, a third upsampling layer, and a sixth convolutional layer connected in series;
[0144] Network training unit, used to train sample seismic profiles W r and multi-target layer identification sample set The input is substituted into a convolutional neural network for training, resulting in the multi-object hierarchical recognition network Net. L .
[0145] In some embodiments, the network training unit includes:
[0146] Training sub-units are used to train sample seismic profiles W. r and multi-target layer identification sample set Substitute the data into a convolutional neural network for training to obtain a transition layer recognition network;
[0147] The first acquisition sub-unit is used to acquire known seismic profiles, artificially interpreted seismic horizons obtained after annotating the known seismic profiles, and at least four auxiliary seismic horizons related to the artificially interpreted seismic horizons.
[0148] The prediction sub-unit is used to substitute known seismic profiles into the transition horizon identification network to obtain the predicted seismic horizon and at least four auxiliary seismic horizons related to the predicted seismic horizon.
[0149] The loss value calculation subunit is used to calculate the target loss value based on the predicted seismic horizon and at least four auxiliary seismic horizons related to the predicted seismic horizon, as well as the artificially interpreted seismic horizon obtained after annotating the known seismic profile and at least four auxiliary seismic horizons related to the artificially interpreted seismic horizon.
[0150] The network determines the sub-unit, which is used to identify the transition layer network as a multi-target layer identification network Net in response to the target loss value being less than or equal to a threshold. L .
[0151] In some embodiments, the fusion module 604 includes:
[0152] Obtain the operator unit, which is used to obtain the adaptive adjustment operator α;
[0153] The horizon calculation unit is used to calculate the target seismic horizon based on the adaptive adjustment operator α, the horizon set along the survey line direction, and the horizon set perpendicular to the survey line direction.
[0154] In some embodiments, the set of horizons along the survey line includes the target seismic horizons along the survey line. The set of horizons perpendicular to the survey line includes the target seismic horizons perpendicular to the survey line direction. The stratification calculation unit includes:
[0155] The horizon calculation sub-unit is used to calculate the target seismic horizon along the survey line direction based on the adaptive adjustment operator α. and the target seismic horizon perpendicular to the survey line direction Perform a weighted summation to obtain the target seismic horizon. Among them, the target seismic horizon The following formula is used for calculation:
[0156]
[0157] In some embodiments, the fusion module 604 is followed by:
[0158] The correction processing unit is used for the target seismic horizon. Fault correction processing is performed to obtain continuous seismic horizons after fault correction.
[0159] In some embodiments, the acquisition module 601 includes:
[0160] The acquisition unit is used to acquire the seismic data volume D and the original manually interpreted seismic horizons. The original manually interpreted seismic horizons are the labeled horizons obtained after the seismic profiles generated from the seismic data volume D are annotated.
[0161] The interpolation processing unit is used to interpolate the original manually interpreted seismic horizons to obtain at least four original auxiliary seismic horizons, and to use the original manually interpreted seismic horizons and at least four original auxiliary seismic horizons as a multi-objective sample set.
[0162] The augmentation processing unit is used to augment the seismic data volume D to obtain the augmented seismic data volume.
[0163] Augmentation processing unit for processing multi-target sample sets and the expanded seismic data volume Augmentation processing is performed to obtain the seismic data volume. and multi-target layer identification sample set
[0164] In some embodiments, the interpolation processing unit includes:
[0165] The second acquisition subunit is used to acquire the information at the center point p. i =(x i ,y i ,t i Centered on t i Time along survey line x i Direction and perpendicular to the survey line y i Earthquake traces in the direction The maximum and minimum values of , where i is the subscript index of the original manually interpreted seismic horizon, and the value of i is . This indicates the number of center points;
[0166] Obtain sequence sub-units for use based on the seismic traces Find the maxima and minima to obtain the sequence of local maxima points. and local minimum point sequence Among them, u j This means that at t j Time along survey line x i Direction and perpendicular to the survey line y i Earthquake traces in the direction The maximum point of v k This means that at t k Time along survey line x i Direction and perpendicular to the survey line y i Earthquake traces in the direction The minimum point, The number of maxima. Let j be the number of local minima, and j take the value of 1. The value of k is
[0167] The extreme point sub-units are obtained and used for the local maximum point sequence. and local minimum point sequence After sorting, the maximum point u is obtained. j and the minimum point v k ;
[0168] Auxiliary seismic horizon sub-units are used to locate the maximum point u. j and the minimum point v k The corresponding center point p i The four sets and These are the four original auxiliary seismic horizons.
[0169] This application provides a seismic horizon identification device that inputs the seismic profile to be identified into a multi-target horizon identification network (Net). L Seismic horizon identification is performed to obtain transitional seismic horizons and at least four auxiliary seismic horizons associated with them. These are then fused to obtain the target seismic horizon. The multi-target horizon identification sample set in this device includes manually interpreted seismic horizons obtained by annotating sample seismic profiles and at least four auxiliary seismic horizons associated with these manually interpreted seismic horizons. This indicates that the training samples are sufficiently rich, enabling the multi-target horizon identification network Net trained based on these samples to achieve the desired results. L It has high identification accuracy, enabling accurate identification of seismic horizons and improving the accuracy of seismic interpretation.
[0170] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory including program code that can be executed by a processor in an electronic device to complete the seismic horizon identification method in the above embodiments. For example, the non-volatile computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.
[0171] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.
[0172] In this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The term "multiple" refers to two or more unless otherwise expressly defined.
[0173] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only.
[0174] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.
Claims
1. A method for identifying seismic horizons, characterized in that, The method includes: Acquiring earthquake data and multi-target layer identification sample set The earthquake data body Used to generate sample seismic profiles The multi-target hierarchical identification sample set Including the sample seismic profile Manually interpreted seismic horizons obtained after annotation processing and the artificially interpreted seismic horizon At least four relevant auxiliary seismic horizons; Based on the sample seismic profile and multi-target layer identification sample set Train a convolutional neural network to obtain a multi-target hierarchical recognition network. The sample seismic profile and multi-target layer identification sample set These correspond to the input and output of the convolutional neural network; The seismic profile to be identified is substituted into the multi-target horizon identification network. In the process, a transitional seismic horizon and at least four auxiliary seismic horizons associated with the transitional seismic horizon are obtained; The transitional seismic horizon and at least four auxiliary seismic horizons associated with it are fused to determine the target seismic horizon. ; Among them, the acquisition of earthquake data body D and multi-target layer identification sample set include: Acquiring earthquake data D and the original manually interpreted seismic horizon, wherein the original manually interpreted seismic horizon is derived from the seismic data volume D The labeled horizons are obtained after the generated seismic profile has been annotated. Interpolation processing is performed on the original manually interpreted seismic horizons to obtain at least four original auxiliary seismic horizons. The original manually interpreted seismic horizons and the at least four original auxiliary seismic horizons are then used as a multi-objective sample set. ; For the earthquake data volume D The data volume is augmented to obtain the augmented seismic data. ; For the multi-target sample set and the augmented seismic data volume Augmentation processing is performed to obtain the seismic data volume. and the multi-target layer identification sample set ; The original manually interpreted seismic horizon includes multiple center points. The interpolation process performed on the original manually interpreted seismic horizons to obtain at least four original auxiliary seismic horizons includes: Obtain the center point Centered on, in Along the survey line at all times Direction and perpendicular to the survey line Earthquake traces in the direction The maximum and minimum values of , where i is the subscript index of the original manually interpreted seismic horizon, and the value of i is 1, 2, ... , This indicates the number of center points; Based on the earthquake trace Find the maxima and minima to obtain the sequence of local maxima points. and local minimum point sequence ,in, It means that in Along the survey line at all times Direction and perpendicular to the survey line Earthquake traces in the direction The maximum point, It means that in Along the survey line at all times Direction and perpendicular to the survey line Earthquake traces in the direction The minimum point, The number of maxima. Let j be the number of local minima, and j take values of 1, 2, ... The value of k is 1, 2, ... ; For the local maximum point sequence and local minimum point sequence The maximum points are obtained by sorting the data. and minimum point ; The maximum point and the minimum point Corresponding center point The four sets , , and These are the four original auxiliary seismic horizons.
2. The method for identifying seismic horizons according to claim 1, characterized in that, The seismic profile based on the sample and multi-target layer identification sample set Train a convolutional neural network to obtain a multi-target hierarchical recognition network. include: Convolutional neural networks are constructed, wherein the convolutional neural network includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a branch network, a seventh convolutional layer, and an eighth convolutional layer connected in series, wherein the branch network includes at least five sub-networks connected in parallel, and each sub-network includes a first upsampling layer, a fourth convolutional layer, a second upsampling layer, a fifth convolutional layer, a third upsampling layer, and a sixth convolutional layer connected in series; The sample seismic profile and multi-target layer identification sample set The input is substituted into the convolutional neural network for training to obtain the multi-target hierarchical recognition network. .
3. The method for identifying seismic horizons according to claim 2, characterized in that, The sample seismic profile and multi-target layer identification sample set The input is substituted into the convolutional neural network for training to obtain the multi-target hierarchical recognition network. include: The sample seismic profile and multi-target layer identification sample set Substitute the data into the convolutional neural network for training to obtain the transition layer recognition network; Obtain known seismic profiles and artificially interpreted seismic horizons obtained after annotating the known seismic profiles, and at least four auxiliary seismic horizons related to the artificially interpreted seismic horizons; Substituting the known seismic profile into the transition horizon identification network yields the predicted seismic horizon and at least four auxiliary seismic horizons associated with the predicted seismic horizon. The target loss value is calculated based on the predicted seismic horizon and at least four auxiliary seismic horizons associated with the predicted seismic horizon, as well as the artificially interpreted seismic horizon and at least four auxiliary seismic horizons associated with the artificially interpreted seismic horizon obtained after annotating the known seismic profile. In response to the target loss value being less than or equal to a threshold, the transition layer identification network is used as the multi-target layer identification network. .
4. The method for identifying seismic horizons according to claim 1, characterized in that, The transitional seismic horizon and at least four auxiliary seismic horizons associated with it include a set of horizons along the survey line and a set of horizons perpendicular to the survey line. The process of fusing the transitional seismic horizon and the at least four auxiliary seismic horizons associated with it to determine the target seismic horizon is described. include: Obtain the adaptive adjustment operator ; Based on the adaptive adjustment operator The target seismic horizon is calculated from the set of horizons along the survey line and the set of horizons perpendicular to the survey line. .
5. The method for identifying seismic horizons according to claim 4, characterized in that, The set of horizons along the survey line includes the target seismic horizons along the survey line. The set of horizons perpendicular to the survey line direction includes the target seismic horizons perpendicular to the survey line direction. The adaptive adjustment operator The target seismic horizon is calculated from the set of horizons along the survey line and the set of horizons perpendicular to the survey line. include: Based on the adaptive adjustment operator For the target seismic horizon along the survey line direction and the target seismic horizon perpendicular to the survey line direction The target seismic horizon is obtained by performing a weighted summation. The target seismic horizon The following formula is used for calculation: 。 6. The method for identifying seismic horizons according to claim 1, characterized in that, The transitional seismic horizon and at least four auxiliary seismic horizons associated with it are fused to determine the target seismic horizon. Subsequently, the method further includes: For the target seismic horizon Fault correction processing is performed to obtain continuous seismic horizons after fault correction. .
7. A seismic horizon identification device, characterized in that, The device includes: The acquisition module is used to acquire seismic data volumes. and multi-target layer identification sample set The earthquake data body Used to generate sample seismic profiles The multi-target hierarchical identification sample set Including sample seismic profiles Manually interpreted seismic horizons obtained after annotation processing and the artificially interpreted seismic horizon At least four relevant auxiliary seismic horizons; Training module, used to train based on the sample seismic profile and multi-target layer identification sample set Train a convolutional neural network to obtain a multi-target hierarchical recognition network. The sample seismic profile and multi-target layer identification sample set These correspond to the input and output of the convolutional neural network; The identification module is used to input the seismic profile to be identified into the multi-target horizon identification network. In the process, a transitional seismic horizon and at least four auxiliary seismic horizons associated with the transitional seismic horizon are obtained; The fusion module is used to fuse the transitional seismic horizon and at least four auxiliary seismic horizons related to the transitional seismic horizon to determine the target seismic horizon. ; Specifically, the acquisition module is used to acquire seismic data volumes. D and the original manually interpreted seismic horizon, wherein the original manually interpreted seismic horizon is derived from the seismic data volume D The labeled horizons are obtained after the generated seismic profile has been annotated. Interpolation processing is performed on the original manually interpreted seismic horizons to obtain at least four original auxiliary seismic horizons. The original manually interpreted seismic horizons and the at least four original auxiliary seismic horizons are then used as a multi-objective sample set. ; For the earthquake data volume D The data volume is augmented to obtain the augmented seismic data. ; For the multi-target sample set and the augmented seismic data volume Augmentation processing is performed to obtain the seismic data volume. and the multi-target layer identification sample set ; The original manually interpreted seismic horizon includes multiple center points. The interpolation process performed on the original manually interpreted seismic horizons to obtain at least four original auxiliary seismic horizons includes: Obtain the center point Centered on, in Along the survey line at all times Direction and perpendicular to the survey line Earthquake traces in the direction The maximum and minimum values of , where i is the subscript index of the original manually interpreted seismic horizon, and the value of i is 1, 2, ... , This indicates the number of center points; Based on the earthquake trace Find the maxima and minima to obtain the sequence of local maxima points. and local minimum point sequence ,in, It means that in Along the survey line at all times Direction and perpendicular to the survey line Earthquake traces in the direction The maximum point, It means that in Along the survey line at all times Direction and perpendicular to the survey line Earthquake traces in the direction The minimum point, The number of maxima. Let j be the number of local minima, and j take values of 1, 2, ... The value of k is 1, 2, ... ; For the local maximum point sequence and local minimum point sequence The maximum points are obtained by sorting the data. and minimum point ; The maximum point and the minimum point Corresponding center point The four sets , , and These are the four original auxiliary seismic horizons.
8. A non-volatile computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform the seismic horizon identification method as described in any one of claims 1 to 6.