Seismic attribute fusion method and device

Through weakening methods and neural network technology, the seismic attribute degradation and fusion problem is solved, and the problem of poor seismic attribute fusion analysis efficiency and feature extraction effect is achieved, and more accurate reservoir prediction is achieved.

CN114624767BActive Publication Date: 2024-11-01CHINA NAT PETROLEUM CORP +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011424695.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-08
Publication Date
2024-11-01
Estimated Expiration
2040-12-08

AI Technical Summary

Technical Problem

The efficiency of existing seismic attribute fusion analysis and feature extraction effects still need to be improved, making it difficult to solve multi-solution problems when predicting reservoirs.

Method used

The weakening method is used to degrade the target seismic attributes, and the single-attribute neural network is used to extract attribute features, and the multi-attribute fusion network is used to constrain the loss function of unsupervised similarity, and the fusion attribute features are output, combining single-attribute and multi-attribute neural networks. Perform feature extraction and attribute fusion.

Benefits of technology

The efficiency and feature extraction effect of seismic attribute fusion analysis are improved, multi-solvency is reduced, and target geological characteristics are highlighted.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114624767B_ABST
    Figure CN114624767B_ABST
Patent Text Reader

Abstract

The present invention discloses a seismic attribute fusion method and device, which comprises: using a weakening method to perform a degradation operation on each clear seismic attribute, determining a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute, using each clear seismic attribute and its corresponding plurality of fuzzy degraded attributes as inputs of a trained single-attribute neural network, and extracting attribute features of each clear seismic attribute; using the attribute features of each clear seismic attribute as inputs of a trained multi-attribute fusion network, and outputting fused attribute features corresponding to a plurality of clear seismic attributes with a loss function based on unsupervised similarity as a constraint; the present invention uses a single-attribute neural network combined with a multi-attribute neural network to perform feature extraction and attribute fusion on a seismic attribute set, and outputs fused attribute features, which can improve the efficiency of seismic attribute fusion analysis and the effect of seismic attribute feature extraction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of seismic exploration technology, and in particular to a seismic attribute fusion method and device. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention recited in the claims. No admission is made that the description herein is prior art by inclusion in this section.

[0003] Seismic attributes have evolved from the earlier amplitude attributes to the dozens of commonly used seismic attributes today. These seismic attributes reflect the underground geological conditions, but each seismic attribute is only sensitive to certain geological features. For example, reflection intensity reflects the thickness of sandstone, and relative wave impedance reflects the gas content of the reservoir. Therefore, using a single seismic attribute to predict reservoirs will result in multiple solutions.

[0004] The joint analysis method of seismic attributes is proposed to solve the problem of multiple solutions for predicting reservoirs with a single attribute. This method combines multiple attributes on the basis of certain mathematical operations, while considering the factors affecting the reservoir of each attribute, and finally obtains the best result by combining these factors.

[0005] The main purpose of seismic multi-attribute joint analysis is to eliminate multiple solutions and highlight the target geological characteristics. However, multi-attribute fusion cannot be performed blindly, and it is necessary to select from the attribute set. Attribute optimization requires that the geological information contained in the selected seismic attributes is maximized and can establish a correlation with the geological target. There are many studies on seismic attribute fusion. Commonly used attribute fusion methods include proportional fusion technology based on mathematical methods, cluster analysis technology, and multivariate linear regression fusion technology; there is also seismic attribute fusion based on the multi-attribute RGB method of color space. These methods have their own advantages and disadvantages and their scope of application.

[0006] However, the analysis efficiency of existing seismic attribute fusion and the feature extraction effect of seismic attributes still need to be further improved. Summary of the invention

[0007] The embodiment of the present invention provides a seismic attribute fusion method for improving the efficiency of seismic attribute fusion analysis and improving the effect of seismic attribute feature extraction. The method includes:

[0008] A degradation operation is performed on each clear seismic attribute in the target seismic attribute set by using a weakening method to determine a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set;

[0009] Each clear seismic attribute in the target seismic attribute set and its corresponding multiple fuzzy degraded attributes are used as inputs of the trained single-attribute neural network to extract the attribute features of each clear seismic attribute in the target seismic attribute set; the single-attribute neural network reflects the mapping relationship between the attribute features of each clear seismic attribute and the corresponding multiple fuzzy degraded attributes;

[0010] The attribute features of each clear seismic attribute in the target seismic attribute set are taken as the input of the trained multi-attribute fusion network, and the loss function based on unsupervised similarity is used as a constraint to output the fused attribute features corresponding to multiple clear seismic attributes in the target seismic attribute set; the multi-attribute fusion network reflects the mapping relationship between the fused attribute features and the attribute features of each clear seismic attribute.

[0011] The embodiment of the present invention further provides a seismic attribute fusion device, which is used to improve the efficiency of seismic attribute fusion analysis and improve the effect of seismic attribute feature extraction. The device includes:

[0012] A target attribute weakening module is used to perform a degradation operation on each clear seismic attribute in the target seismic attribute set by using a weakening method, and determine a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set;

[0013] The target single attribute extraction module is used to take each clear seismic attribute in the target seismic attribute set and its corresponding multiple fuzzy degraded attributes as inputs of the trained single attribute neural network to extract the attribute features of each clear seismic attribute in the target seismic attribute set; the single attribute neural network reflects the mapping relationship between the attribute features of each clear seismic attribute and the corresponding multiple fuzzy degraded attributes;

[0014] The target attribute fusion module is used to take the attribute features of each clear seismic attribute in the target seismic attribute set as the input of the trained multi-attribute fusion network, and output the fused attribute features corresponding to multiple clear seismic attributes in the target seismic attribute set under the constraint of the loss function based on unsupervised similarity; the multi-attribute fusion network reflects the mapping relationship between the fused attribute features and the attribute features of each clear seismic attribute.

[0015] An embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned seismic attribute fusion method when executing the computer program.

[0016] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for executing the above-mentioned seismic attribute fusion method.

[0017] In the embodiment of the present invention, a weakening method is used to perform a degradation operation on each clear seismic attribute in the target seismic attribute set to determine multiple fuzzy degraded attributes corresponding to each clear seismic attribute, and then a trained single-attribute neural network is used to extract the attribute features of each clear seismic attribute in the target seismic attribute set. Finally, the attribute features of each clear seismic attribute in the target seismic attribute set are fused using a trained multi-attribute fusion network with an unsupervised similarity loss function as a constraint to output a fused attribute feature. The embodiment of the present invention is based on an unsupervised neural network, and uses a single-attribute neural network combined with a multi-attribute neural network to perform feature extraction and attribute fusion on a seismic attribute set to output a fused attribute feature, which can improve the efficiency of seismic attribute fusion analysis and improve the effect of seismic attribute feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0019] Figure 1 A flowchart of the implementation of the seismic attribute fusion method provided in the embodiment of the present invention;

[0020] Figure 1-1 A schematic diagram of a specific structure of a multi-attribute fusion network provided by an embodiment of the present invention;

[0021] Figure 1-2 (a) and Figure 1-2 (b) is a schematic diagram showing two clear seismic attributes provided by an embodiment of the present invention;

[0022] Figure 1-2 (c) is a schematic diagram showing fusion attribute features corresponding to two clear seismic attributes provided by an embodiment of the present invention;

[0023] Figure 2 Another implementation flow chart of the seismic attribute fusion method provided in an embodiment of the present invention;

[0024] Figure 3 A flowchart for implementing the training of a single attribute neural network in the seismic attribute fusion method provided in an embodiment of the present invention;

[0025] Figure 4 A flowchart for implementing training of a multi-attribute fusion network in the seismic attribute fusion method provided in an embodiment of the present invention;

[0026] Figure 5A functional module diagram of a seismic attribute fusion device provided by an embodiment of the present invention;

[0027] Figure 6 Another functional module diagram of the earthquake attribute fusion device provided by an embodiment of the present invention;

[0028] Figure 7 A structural block diagram of the training of a single attribute neural network in the seismic attribute fusion method provided in an embodiment of the present invention;

[0029] Figure 8 A structural block diagram of the training of a multi-attribute fusion network in the seismic attribute fusion method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0030] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0031] Figure 1 The implementation process of the seismic attribute fusion method provided by the embodiment of the present invention is shown. For the convenience of description, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0032] like Figure 1 As shown, the seismic attribute fusion method includes:

[0033] Step 101, performing a degradation operation on each clear seismic attribute in the target seismic attribute set by using a weakening method, and determining a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set;

[0034] Step 102, taking each clear seismic attribute in the target seismic attribute set and its corresponding multiple fuzzy degraded attributes as inputs of the trained single attribute neural network, extracting the attribute features of each clear seismic attribute in the target seismic attribute set; the single attribute neural network reflects the mapping relationship between the attribute features of each clear seismic attribute and the corresponding multiple fuzzy degraded attributes;

[0035] Step 103, taking the attribute features of each clear seismic attribute in the target seismic attribute set as the input of the trained multi-attribute fusion network, and outputting the fused attribute features corresponding to multiple clear seismic attributes in the target seismic attribute set with the loss function based on unsupervised similarity as a constraint; the multi-attribute fusion network reflects the mapping relationship between the fused attribute features and the attribute features of each clear seismic attribute.

[0036] The target seismic attribute set is a seismic attribute set to be processed, including multiple clear seismic attributes. Among them, the type of each clear seismic attribute in the target seismic attribute set is consistent. When performing seismic attribute fusion (i.e., fusing each clear seismic attribute in the target seismic attribute set to obtain a fused seismic attribute), firstly, a weakening method (also called a degradation method) is used to perform a degradation operation on each clear seismic attribute in the target seismic attribute set to obtain multiple fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set.

[0037] The weakening method includes at least a mean weakening method and a maximum and minimum weakening method. Those skilled in the art can understand that the weakening method may also include other weakening methods besides the mean weakening method and the maximum and minimum weakening method, and the embodiment of the present invention does not impose any particular limitation on this.

[0038] The mean weakening formula can be expressed as:

[0039] m c = 0.2, 0.4, 0.5;

[0040]

[0041] Among them, I t represents a clear seismic attribute in the target seismic attribute set, Indicates clear earthquake attribute I t The corresponding fuzzy degenerate attribute obtained by the mean weakening method, m c is a constant.

[0042] The maximum and minimum weakening formula can be expressed as:

[0043] λ f = three values ​​of 0.3, 0.4, and 0.5;

[0044]

[0045] Among them, I t represents a clear seismic attribute in the target seismic attribute set, represents the corresponding fuzzy degenerate attribute obtained by maximum and minimum weakening, λ f is a constant.

[0046] Through the above degradation operation, for each clear earthquake attribute I c , we can get the corresponding 6 fuzzy degenerate attributes.

[0047] Accordingly, after obtaining multiple fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set, each clear seismic attribute in the target seismic attribute set and the multiple fuzzy degraded attributes corresponding to each clear seismic attribute are used as inputs of the trained single attribute neural network, so that the trained single attribute neural network can extract the attribute features of each clear seismic attribute in the target seismic attribute set. Among them, the trained single attribute neural network reflects the mapping relationship between the attribute features of the clear seismic attribute and the corresponding multiple fuzzy degraded attributes, so that the attribute features of each clear seismic attribute in the target seismic attribute set can be extracted.

[0048] Among them, each clear attribute I c Single attribute neural network net Ic The model output error can be expressed as follows:

[0049]

[0050] in, represents the single attribute neural network output error, f(*) represents the neural network model between each clear seismic attribute and a single corresponding fuzzy degraded attribute, W c Represents the link weight vector and bias vector of a single-attribute neural network. Represents the clear seismic attribute I in the target seismic attribute set c The corresponding fuzzy degenerate attribute, I c Represents a clear seismic attribute in the target seismic attribute set.

[0051] U-net autoencoder type network for each clear earthquake attribute The model mainly consists of convolutional layers, pooling layers, upsampling layers and skip layers, where the skip layer can merge multi-scale features. After a series of operations, the single attribute neural network F tThe output score map is half the size of the source image (input), and then, we will get a focus map of the same size as the original image through bilinear interpolation. The specific structure is: conv1 (including three convolution kernels, using 3*3 convolution kernel, same scale convolution), conv2 (including three convolution kernels, using 3*3 convolution kernel, same scale convolution), maxpool (using maximum pooling), conv3 (including three convolution kernels, using 3*3 convolution kernel, same scale convolution), uppool1 (using bilinear interpolation upsampling), fuse1 (concat operation of uppool upsampling result with conv2 result), conv4 (4 convolution kernels, using 3*3 convolution kernel, same scale convolution), fuse2 (concat operation of uppool2 upsampling result with conv1 result), conv5 (5 convolution kernels, using 3*3 convolution kernel, same scale convolution), using 1*1*c (c is the number of channels) full convolution to the final fusion result (attribute features of each clear seismic attribute) The Network It mainly consists of two parts, namely conv1+conv2+maxpool+conv3 downsampling network And uppool1+fuse1+conv4+fuse2+conv5 upsampling network composition.

[0052] Above The network structure is similar to an end-to-end self-organizing type network structure. Through network learning, a mapping relationship between fuzzy degraded attributes and clear seismic attributes (high-precision attributes) can be established. This method is used to establish an extraction process from unclear features to clear features in an unsupervised manner, that is, to establish an extraction process from fuzzy degraded attributes to clear seismic attributes. Each clear seismic attribute I c There is a network Ic Corresponding to it.

[0053] After extracting the attribute features of each clear seismic attribute in the target seismic attribute set using the trained single-attribute neural network, the attribute features of each clear seismic attribute in the target seismic attribute set are then fused using the trained multi-attribute fusion network with the loss function based on unsupervised similarity as a constraint, so as to output the fused attribute features corresponding to multiple clear seismic attributes in the target seismic attribute set. The trained multi-attribute fusion network reflects the mapping relationship between the fused attribute features and the attribute features of each clear seismic attribute.

[0054] After extracting the attribute features of each clear seismic attribute in the target seismic attribute set, the attribute features of each clear seismic attribute in the target seismic attribute set are used as the input of the trained multi-attribute fusion network, and the attribute features of each clear seismic attribute in the target seismic attribute set are fused with the loss function based on unsupervised similarity as a constraint to obtain and output the fused attribute features corresponding to multiple clear seismic attributes in the target seismic attribute set. Among them, the trained multi-attribute fusion network reflects the mapping relationship between the fused attribute features and the attribute features of each clear seismic attribute.

[0055] The function of the multi-attribute fusion network is to fuse the attribute features of each clear seismic attribute extracted above, and establish the attribute features of multiple clear seismic attributes in the target seismic attribute set. and fusion attribute features The unsupervised nonlinear mapping relationship between θ is the network parameter of the multi-attribute fusion network. Therefore, traditional deep learning methods cannot be directly used for feature extraction, and new objective functions are needed to constrain and optimize the algorithm.

[0056] Specifically, the loss function of unsupervised similarity is determined by the following formula:

[0057]

[0058]

[0059]

[0060]

[0061] in, and represents any two clear seismic attributes in the target seismic attribute set, w represents a 9×9 window, and δ represents a constant 1×10 -8 , Represents clear earthquake attributes The average value within the window w, Represents clear earthquake attributes The average value in window w; Represents clear earthquake attributes The variance within window w, Represents clear earthquake attributes The variance within window w, Represents clear earthquake attributes With clear earthquake attributes The covariance in window w, W represents the set of time windows w obtained by scanning from left to right and from top to bottom to each point, and the function ||W|| represents the number of all windows w in W, S i,j (W) indicates clear earthquake attributes With clear earthquake attributes Unsupervised similarity between

[0062] Represents the clear seismic attributes in the target seismic attribute set The variance in window w, represents the maximum variance of all clear seismic attributes in the target seismic attribute set in window w, represents the target earthquake attribute, Represents the earthquake attribute c and the earthquake attribute similarity between them; local_loss(w) represents the local loss function within window w, and loss represents the loss function based on unsupervised similarity.

[0063] Figure 1-1 A specific structural diagram of a multi-attribute fusion network provided by an embodiment of the present invention is shown. Figure 1-1 It can be seen that the example is taken when the target seismic attribute set includes two clear seismic attributes. When the target seismic attribute set includes multiple clear seismic attributes, the number of branch networks in the front of the multi-attribute fusion network can be increased, and the network in the second half remains unchanged. Figure 1-2 (a) and Figure 1-2 (b) shows a schematic diagram of two clear seismic attributes provided by an embodiment of the present invention, Figure 1-2 (c) shows a schematic diagram of fused attribute features corresponding to two clear seismic attributes provided by an embodiment of the present invention.

[0064] Specifically, the multi-attribute fusion network includes:

[0065] Input path p1 on the left: the first conv from top to bottom (including 64 convolution kernels, using 3*3 convolution kernels, and convolution of the same scale), the second conv from top to bottom (including 128 convolution kernels, using 3*3 convolution kernels, and convolution of the same scale), maxpool (using 2*2 maximum pooling), and the third conv from top to bottom (including 256 convolution kernels, using 3*3 convolution kernels, and convolution of the same scale).

[0066] Input path p2 on the right: the first conv from top to bottom (including 64 convolution kernels, using 3*3 convolution kernels, and convolution of the same scale), the second conv from top to bottom (including 128 convolution kernels, using 3*3 convolution kernels, and convolution of the same scale), maxpool (using 2*2 maximum pooling), and the third conv from top to bottom (including 256 convolution kernels, using 3*3 convolution kernels, and convolution of the same scale).

[0067] Full_1 (including 256 fully linked nodes) performs feature merging.

[0068] Based on this, using the above loss function as a constraint, the multi-attribute fusion network is used to fuse the attribute features of each clear seismic attribute in the target seismic attribute set, and finally the fused attribute features are output.

[0069] In an embodiment of the present invention, a weakening method is used to perform a degradation operation on each clear seismic attribute in the target seismic attribute set, and multiple fuzzy degraded attributes corresponding to each clear seismic attribute are determined, and then a trained single-attribute neural network is used to extract the attribute features of each clear seismic attribute in the target seismic attribute set, and finally, the attribute features of each clear seismic attribute in the target seismic attribute set are fused using a trained multi-attribute fusion network with an unsupervised similarity loss function as a constraint to output a fused attribute feature. The embodiment of the present invention is based on an unsupervised neural network, and uses a single-attribute neural network combined with a multi-attribute neural network to perform feature extraction and attribute fusion on a seismic attribute set to output a fused attribute feature, which can improve the efficiency of seismic attribute fusion analysis and improve the effect of seismic attribute feature extraction.

[0070] Figure 2 Another implementation process of the seismic attribute fusion method provided by an embodiment of the present invention is shown. For ease of description, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0071] In one embodiment of the present invention, in order to further improve the effect of seismic attribute fusion, as Figure 2 As shown, based on the method steps shown in FIG. 1 above, the seismic attribute fusion method further includes:

[0072] Step 201, normalizing each clear seismic attribute in the target seismic attribute set to obtain each clear seismic attribute in the normalized target seismic attribute set;

[0073] Step 101, using a weakening method to perform a degradation operation on each clear seismic attribute in the target seismic attribute set, and determining a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set, including:

[0074] Step 202: Degrade each clear seismic attribute in the normalized target seismic attribute set using a weakening method to determine a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set.

[0075] Among them, the normalization processing at least includes one or more of minimum-maximum normalization, zero-mean normalization and decimal scaling normalization. Those skilled in the art can understand that other normalization methods besides the above-mentioned minimum-maximum normalization, zero-mean normalization and decimal scaling normalization can also be used. The embodiment of the present invention does not impose any special limitation on this. The normalization processing can further improve the effect of seismic attribute fusion.

[0076] In the embodiment of the present invention, each clear seismic attribute in the target seismic attribute set is normalized to obtain each clear seismic attribute in the target seismic attribute set after normalization, which can further improve the effect of seismic attribute fusion.

[0077] Figure 3 The implementation process of training a single attribute neural network in the seismic attribute fusion method provided in the embodiment of the present invention is shown. For ease of description, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0078] In one embodiment of the present invention, in order to improve the feature extraction effect of a single attribute neural network, as Figure 3 As shown in Figure 1, the process of training a single attribute neural network in the seismic attribute fusion method includes:

[0079] Step 301, normalizing each clear seismic attribute in the sample seismic attribute set to obtain each clear seismic attribute in the normalized sample seismic attribute set;

[0080] Step 302, performing a degradation operation on each clear seismic attribute in the sample seismic attribute set by using a weakening method, and determining a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the sample seismic attribute set;

[0081] Step 303, taking each clear seismic attribute in the sample seismic attribute set and its corresponding multiple fuzzy degraded attributes as training samples, and iteratively training a single attribute neural network;

[0082] Step 304, when the conditions for iteration termination are met, a trained single-attribute neural network is obtained; the trained single-attribute neural network reflects the mapping relationship between the attribute characteristics of each clear seismic attribute in the sample seismic attribute set and its corresponding multiple fuzzy degraded attributes.

[0083] Among them, before using the single-attribute neural network to extract the attribute characteristics of each clear seismic attribute in the target seismic attribute set, the single-attribute neural network needs to be trained.

[0084] Specifically, a sample seismic attribute set consisting of a large number of clear seismic attributes is first collected, and then each clear seismic attribute in the sample seismic attribute set is normalized to obtain each clear seismic attribute in the normalized sample seismic attribute set to improve the effect of seismic attribute fusion.

[0085] Then, a degradation operation is performed on each clear seismic attribute in the sample seismic attribute set using a weakening method to determine a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the sample seismic attribute set. This part can refer to the above-mentioned embodiment of performing a degradation operation on each clear seismic attribute in the target seismic attribute set, and will not be described in detail.

[0086] Then, each clear seismic attribute in the sample seismic attribute set and the multiple fuzzy degraded attributes corresponding to each clear seismic attribute obtained by weakening are used as the input of the single attribute neural network, and the single attribute neural network is iteratively trained. When the conditions for iterative termination are met, the trained single attribute neural network can be obtained. The trained single attribute neural network can reflect the mapping relationship between each clear seismic attribute and the corresponding multiple fuzzy degraded attributes.

[0087] Among them, the condition for terminating the iteration may include that the number of iterations reaches a predetermined number of iterations, such as 100,000 times or 80,000 times, etc.; in addition, the condition for terminating the iteration may also include that the accuracy of the trained single-attribute neural network reaches a predetermined accuracy, such as reaching 98% or reaching 95%, etc.

[0088] In an embodiment of the present invention, each clear seismic attribute in the sample seismic attribute set is normalized to improve the effect of attribute fusion, and then a weakening method is used to degrade each clear seismic attribute in the sample seismic attribute set, and then each clear seismic attribute and its corresponding multiple fuzzy degraded attributes are used to iteratively train a single-attribute neural network. When the conditions for iterative termination are met, the trained single-attribute neural network is obtained, which can improve the feature extraction effect of the single-attribute neural network.

[0089] Figure 4 The implementation process of training a multi-attribute fusion network in the seismic attribute fusion method provided in an embodiment of the present invention is shown. For ease of description, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0090] In one embodiment of the present invention, in order to improve the attribute fusion effect of the multi-attribute fusion network, as Figure 4 As shown in Figure 2, the process of training a multi-attribute fusion network includes:

[0091] Step 401, taking the attribute features of each clear earthquake attribute in the sample earthquake attribute set as input, iteratively training a multi-attribute fusion network;

[0092] Step 402, in the process of iteratively training the multi-attribute fusion network, using the loss function based on unsupervised similarity as a constraint;

[0093] Step 403, when the conditions for iteration termination are met, a trained multi-attribute fusion network is obtained, and fusion attribute features corresponding to multiple clear seismic attributes in the sample seismic attribute set are output; the trained multi-attribute fusion network reflects the mapping relationship between the fusion attribute features and the attribute features of each clear seismic attribute in the sample seismic attribute set.

[0094] Among them, before using the multi-attribute fusion network to fuse the attribute characteristics of each clear seismic attribute in the target seismic attribute set, the multi-attribute fusion network needs to be trained.

[0095] First, the attribute features of each clear earthquake attribute in the sample earthquake attribute set obtained above are used as input to iteratively train the multi-attribute fusion network. In the process of iteratively training the multi-attribute fusion network, the loss function based on unsupervised similarity (see above) is used. Figure 1 The trained multi-attribute fusion network is used as a constraint until the condition for iterative termination is met, and the fusion attribute features corresponding to multiple clear seismic attributes in the sample seismic attribute set are output. The trained multi-attribute fusion network reflects the mapping relationship between the fusion attribute features and the attribute features of each clear seismic attribute.

[0096] Among them, the condition for terminating the iteration may include that the number of iterations reaches a predetermined number of iterations, such as 100,000 times or 80,000 times, etc.; in addition, the condition for terminating the iteration may also include that the accuracy of the trained multi-attribute fusion network reaches a predetermined accuracy, such as reaching 98% or reaching 95%, etc.

[0097] In an embodiment of the present invention, the attribute features of each clear seismic attribute in the sample seismic attribute set are taken as input, and a multi-attribute fusion network is iteratively trained. During the iterative training of the multi-attribute fusion network, a loss function based on unsupervised similarity is used as a constraint. When the conditions for iterative termination are met, the trained multi-attribute fusion network is obtained, and fusion attribute features corresponding to multiple clear seismic attributes in the sample seismic attribute set are output. The multi-attribute fusion network is iteratively trained, which can improve the attribute fusion effect of the multi-attribute fusion network.

[0098] The present invention also provides a seismic attribute fusion device, as described in the following embodiments. Since the principles of solving problems by these devices are similar to those of the seismic attribute fusion method, the implementation of these devices can refer to the implementation of the method, and the repeated parts will not be repeated.

[0099] Figure 5 The functional modules of the seismic attribute fusion device provided by the embodiment of the present invention are shown. For the convenience of description, only the parts related to the embodiment of the present invention are shown, which are described in detail as follows:

[0100] refer to Figure 5 The modules included in the seismic attribute fusion device are used to execute Figure 1 For details of each step in the corresponding embodiment, please refer to Figure 1 as well as Figure 1 In the embodiment of the present invention, the seismic attribute fusion device includes a target attribute weakening module 501 , a target single attribute extraction module 502 and a target attribute fusion module 503 .

[0101] The target attribute weakening module 501 is used to perform a degradation operation on each clear seismic attribute in the target seismic attribute set by using a weakening method, and determine a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set.

[0102] The target single attribute extraction module 502 is used to use each clear seismic attribute in the target seismic attribute set and its corresponding multiple fuzzy degraded attributes as inputs of the trained single attribute neural network to extract the attribute features of each clear seismic attribute in the target seismic attribute set; the single attribute neural network reflects the mapping relationship between the attribute features of each clear seismic attribute and the corresponding multiple fuzzy degraded attributes.

[0103] The target attribute fusion module 503 is used to take the attribute features of each clear seismic attribute in the target seismic attribute set as the input of the trained multi-attribute fusion network, and output the fused attribute features corresponding to multiple clear seismic attributes in the target seismic attribute set with the loss function based on unsupervised similarity as a constraint; the multi-attribute fusion network reflects the mapping relationship between the fused attribute features and the attribute features of each clear seismic attribute.

[0104] Among them, the weakening method includes at least a mean weakening method and a maximum and minimum weakening method.

[0105] The loss function of unsupervised similarity is determined by the following formula:

[0106]

[0107]

[0108]

[0109]

[0110] in, and represents any two clear seismic attributes in the target seismic attribute set, w represents a 9×9 window, and δ represents a constant 1×10 -8 , Represents clear earthquake attributes The average value within the window w, Represents clear earthquake attributes The average value within the window w; Represents clear earthquake attributes The variance within window w, Represents clear earthquake attributes The variance within window w, Represents clear earthquake attributes With clear earthquake attributes The covariance in window w, W represents the set of time windows w obtained by scanning from left to right and from top to bottom to each point, and the function ||W|| represents the number of all windows w in W, S i,j (W) indicates clear earthquake attributes With clear earthquake attributes Unsupervised similarity between

[0111] Represents the clear seismic attributes in the target seismic attribute set The variance in window w, represents the maximum variance of all clear seismic attributes in the target seismic attribute set in window w, represents the target earthquake attribute, Represents the earthquake attribute c and the earthquake attribute similarity between them; local_loss(w) represents the local loss function within window w, and loss represents the loss function based on unsupervised similarity.

[0112] In an embodiment of the present invention, the target attribute weakening module 501 uses a weakening method to perform a degradation operation on each clear seismic attribute in the target seismic attribute set, and determines multiple fuzzy degraded attributes corresponding to each clear seismic attribute. Then, the target single attribute extraction module 502 uses a trained single attribute neural network to extract the attribute features of each clear seismic attribute in the target seismic attribute set. Finally, the target attribute fusion module 503 uses a trained multi-attribute fusion network to fuse the attribute features of each clear seismic attribute in the target seismic attribute set with an unsupervised similarity loss function as a constraint to output a fused attribute feature. The embodiment of the present invention is based on an unsupervised neural network, and uses a single attribute neural network combined with a multi-attribute neural network to perform feature extraction and attribute fusion on a seismic attribute set to output a fused attribute feature, which can improve the efficiency of seismic attribute fusion analysis and improve the effect of seismic attribute feature extraction.

[0113] Figure 6 Another functional module of the seismic attribute fusion device provided by an embodiment of the present invention is shown. For ease of description, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0114] In one embodiment of the present invention, in order to further improve the effect of seismic attribute fusion, reference Figure 6 The various units included in the seismic attribute fusion device are used to perform Figure 2 For details of each step in the corresponding embodiment, please refer to Figure 2 as well as Figure 2 The relevant descriptions in the corresponding embodiments are not repeated here. Figure 1 Based on the module structure shown, the seismic attribute fusion device further includes a target attribute normalization module 601 .

[0115] The target attribute normalization module 601 is used to perform normalization processing on each clear seismic attribute in the target seismic attribute set to obtain each clear seismic attribute in the target seismic attribute set after normalization.

[0116] The target attribute weakening module 501 is further used to perform a degradation operation on each clear seismic attribute in the normalized target seismic attribute set using a weakening method, and determine a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set.

[0117] The normalization process includes at least one or more of minimum-maximum normalization, zero-mean normalization and fractional calibration normalization.

[0118] In the embodiment of the present invention, the target attribute normalization module 601 normalizes each clear seismic attribute in the target seismic attribute set to obtain each clear seismic attribute in the normalized target seismic attribute set, which can further improve the effect of seismic attribute fusion.

[0119] Figure 7 The structure of the training single attribute neural network provided by the embodiment of the present invention is shown. For the convenience of explanation, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0120] In one embodiment of the present invention, in order to improve the feature extraction effect of the single attribute neural network, reference Figure 7 The modules included in the training of the single attribute neural network are used to perform Figure 3 For details of each step in the corresponding embodiment, please refer to Figure 3 as well as Figure 3 In the embodiment of the present invention, the training single attribute neural network includes a sample attribute normalization module 701, a sample attribute weakening module 702, a single attribute network training module 703 and a single attribute network training termination module 704.

[0121] The sample attribute normalization module 701 is used to perform normalization processing on each clear seismic attribute in the sample seismic attribute set to obtain each clear seismic attribute in the normalized sample seismic attribute set.

[0122] The sample attribute weakening module 702 is used to perform a degradation operation on each clear seismic attribute in the sample seismic attribute set by using a weakening method, and determine a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the sample seismic attribute set.

[0123] The single attribute network training module 703 is used to iteratively train the single attribute neural network by taking each clear seismic attribute and its corresponding multiple fuzzy degraded attributes in the sample seismic attribute set as training samples.

[0124] The single attribute network training termination module 704 is used to obtain a trained single attribute neural network when the iteration termination condition is met; the trained single attribute neural network reflects the mapping relationship between the attribute characteristics of each clear seismic attribute in the sample seismic attribute set and its corresponding multiple fuzzy degraded attributes.

[0125] In an embodiment of the present invention, the sample attribute normalization module 701 improves the effect of attribute fusion by normalizing each clear seismic attribute in the sample seismic attribute set, and then the sample attribute weakening module 702 uses a weakening method to degrade each clear seismic attribute in the sample seismic attribute set, and then the single attribute network training module 703 uses each clear seismic attribute and its corresponding multiple fuzzy degraded attributes to iteratively train the single attribute neural network, and the single attribute network training termination module 704 obtains the trained single attribute neural network when the conditions for iterative termination are met, which can improve the feature extraction effect of the single attribute neural network.

[0126] Figure 8 The structure diagram of training a multi-attribute fusion network in the seismic attribute fusion method provided in the embodiment of the present invention is shown. For ease of explanation, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0127] In one embodiment of the present invention, in order to improve the attribute fusion effect of the multi-attribute fusion network, reference Figure 8 The various units included in the training multi-attribute fusion network are used to perform Figure 4 For details of each step in the corresponding embodiment, please refer to Figure 4 as well as Figure 4 In the embodiment of the present invention, the training of the multi-attribute fusion network includes a multi-attribute fusion network training module 801 , a multi-attribute fusion constraint module 802 and a multi-attribute network training termination module 803 .

[0128] The multi-attribute fusion network training module 801 is used to iteratively train the multi-attribute fusion network by taking the attribute features of each clear seismic attribute in the sample seismic attribute set as input.

[0129] The multi-attribute fusion constraint module 802 is used to use the loss function based on unsupervised similarity as a constraint during the iterative training of the multi-attribute fusion network.

[0130] The multi-attribute network training termination module 803 is used to obtain the trained multi-attribute fusion network when the conditions for iteration termination are met, and output the fusion attribute features corresponding to multiple clear seismic attributes in the sample seismic attribute set; the trained multi-attribute fusion network reflects the mapping relationship between the fusion attribute features and the attribute features of each clear seismic attribute in the sample seismic attribute set.

[0131] In an embodiment of the present invention, the multi-attribute fusion network training module 801 takes the attribute features of each clear seismic attribute in the sample seismic attribute set as input, and iteratively trains the multi-attribute fusion network. The multi-attribute fusion constraint module 802 uses the loss function based on unsupervised similarity as a constraint during the iterative training of the multi-attribute fusion network. The multi-attribute network training termination module 803 obtains the trained multi-attribute fusion network when the conditions for iterative termination are met, and outputs the fused attribute features corresponding to multiple clear seismic attributes in the sample seismic attribute set. Iterative training of the multi-attribute fusion network can improve the attribute fusion effect of the multi-attribute fusion network.

[0132] An embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned seismic attribute fusion method when executing the computer program.

[0133] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for executing the above-mentioned seismic attribute fusion method.

[0134] In summary, in the embodiment of the present invention, a weakening method is used to perform a degradation operation on each clear seismic attribute in the target seismic attribute set, and multiple fuzzy degraded attributes corresponding to each clear seismic attribute are determined, and then the attribute features of each clear seismic attribute in the target seismic attribute set are extracted using a trained single-attribute neural network, and finally, the attribute features of each clear seismic attribute in the target seismic attribute set are fused using a trained multi-attribute fusion network with an unsupervised similarity loss function as a constraint to output a fused attribute feature. The embodiment of the present invention is based on an unsupervised neural network, and uses a single-attribute neural network combined with a multi-attribute neural network to perform feature extraction and attribute fusion on a seismic attribute set to output a fused attribute feature, which can improve the efficiency of seismic attribute fusion analysis and improve the effect of seismic attribute feature extraction.

[0135] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0139] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A seismic attribute fusion method, characterized in that: include: A degradation operation is performed on each clear seismic attribute in the target seismic attribute set by using a weakening method to determine a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set; Each clear seismic attribute in the target seismic attribute set and its corresponding multiple fuzzy degraded attributes are used as inputs of the trained single-attribute neural network to extract the attribute features of each clear seismic attribute in the target seismic attribute set; The single-attribute neural network reflects the mapping relationship between the attribute characteristics of each clear seismic attribute and the corresponding multiple fuzzy degraded attributes; The attribute features of each clear seismic attribute in the target seismic attribute set are used as the input of the trained multi-attribute fusion network, and the fusion attribute features corresponding to multiple clear seismic attributes in the target seismic attribute set are outputted with the loss function based on unsupervised similarity as the constraint; The multi-attribute fusion network reflects the mapping relationship between the fused attribute features and the attribute features of each clear seismic attribute.

2. The earthquake attribute fusion method according to claim 1, characterized in that: The weakening method includes at least a mean weakening method and a maximum and minimum weakening method.

3. The earthquake attribute fusion method according to claim 1, characterized in that: Also includes: Normalizing each clear seismic attribute in the target seismic attribute set to obtain each clear seismic attribute in the normalized target seismic attribute set; The weakening method is used to perform degradation operation on each clear seismic attribute in the target seismic attribute set, and multiple fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set are determined, including: A weakening method is used to perform a degradation operation on each clear seismic attribute in the normalized target seismic attribute set to determine multiple fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set.

4. The earthquake attribute fusion method according to claim 3, characterized in that: The normalization process includes at least one or more of minimum-maximum normalization, zero-mean normalization and fractional calibration normalization.

5. The earthquake attribute fusion method according to claim 1, characterized in that: The loss function of unsupervised similarity is determined by the following formula: in, and represents any two clear seismic attributes in the target seismic attribute set, w represents a 9×9 window, and δ represents a constant 1×10 -8 , Represents clear earthquake attributes The average value within the window w, Represents clear earthquake attributes The average value within the window w; Represents clear earthquake attributes The variance within window w, Represents clear earthquake attributes The variance within window w, Represents clear earthquake attributes With clear earthquake attributes The covariance in window w, W represents the set of time windows w obtained by scanning from left to right and from top to bottom to each point, and the function ||W|| represents the number of all windows w in W, S i,j (W) indicates clear earthquake attributes With clear earthquake attributes Unsupervised similarity between Represents the clear seismic attributes in the target seismic attribute set The variance in window w, represents the maximum variance of all clear seismic attributes in the target seismic attribute set in window w, represents the target earthquake attribute, Represents the earthquake attribute c and the earthquake attribute similarity between them; local_loss(w) represents the local loss function within window w, and loss represents the loss function based on unsupervised similarity.

6. The earthquake attribute fusion method according to claim 1, characterized in that: The process of training a single-attribute neural network includes: Normalizing each clear seismic attribute in the sample seismic attribute set to obtain each clear seismic attribute in the normalized sample seismic attribute set; A degradation operation is performed on each clear seismic attribute in the sample seismic attribute set by using a weakening method to determine a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the sample seismic attribute set; Each clear seismic attribute and its corresponding multiple fuzzy degraded attributes in the sample seismic attribute set are used as training samples to iteratively train the single attribute neural network; When the conditions for iteration termination are met, the trained single-attribute neural network is obtained.

7. The earthquake attribute fusion method according to claim 6, characterized in that: The process of training a multi-attribute fusion network includes: The attribute features of each clear earthquake attribute in the sample earthquake attribute set are taken as input to iteratively train the multi-attribute fusion network; In the process of iteratively training the multi-attribute fusion network, the loss function based on unsupervised similarity is used as a constraint; When the conditions for iterative termination are met, the trained multi-attribute fusion network is obtained, and the fusion attribute features corresponding to multiple clear seismic attributes in the sample seismic attribute set are output.

8. A seismic attribute fusion device, characterized in that: include: A target attribute weakening module is used to perform a degradation operation on each clear seismic attribute in the target seismic attribute set by using a weakening method, and determine a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set; A target single attribute extraction module is used to use each clear seismic attribute and its corresponding multiple fuzzy degraded attributes in the target seismic attribute set as inputs of the trained single attribute neural network to extract attribute features of each clear seismic attribute in the target seismic attribute set; The single-attribute neural network reflects the mapping relationship between the attribute characteristics of each clear seismic attribute and the corresponding multiple fuzzy degraded attributes; The target attribute fusion module is used to take the attribute features of each clear seismic attribute in the target seismic attribute set as the input of the trained multi-attribute fusion network, and output the fusion attribute features corresponding to multiple clear seismic attributes in the target seismic attribute set based on the loss function of unsupervised similarity as the constraint; The multi-attribute fusion network reflects the mapping relationship between the fused attribute features and the attribute features of each clear seismic attribute.

9. The earthquake attribute fusion device according to claim 8, characterized in that: The weakening method includes at least a mean weakening method and a maximum and minimum weakening method.

10. The earthquake attribute fusion device according to claim 8, characterized in that: Also includes: A target attribute normalization module is used to normalize each clear seismic attribute in the target seismic attribute set to obtain each clear seismic attribute in the normalized target seismic attribute set; The target attribute weakening module is also used to use a weakening method to perform a degradation operation on each clear seismic attribute in the normalized target seismic attribute set to determine multiple fuzzy degraded attributes corresponding to each clear seismic attribute in the target seismic attribute set.

11. The earthquake attribute fusion device according to claim 10, characterized in that: The normalization process includes at least one or more of minimum-maximum normalization, zero-mean normalization and fractional calibration normalization.

12. The earthquake attribute fusion device according to claim 8, characterized in that: The loss function of unsupervised similarity is determined by the following formula: in, and represents any two clear seismic attributes in the target seismic attribute set, w represents a 9×9 window, and δ represents a constant 1×10 -8 , Represents clear earthquake attributes The average value within the window w, Represents clear earthquake attributes The average value within the window w; Represents clear earthquake attributes The variance within window w, Represents clear earthquake attributes The variance within window w, Represents clear earthquake attributes With clear earthquake attributes The covariance in window w, W represents the set of time windows w obtained by scanning from left to right and from top to bottom to each point, and the function ||W|| represents the number of all windows w in W, S i,j (W) indicates clear earthquake attributes With clear earthquake attributes Unsupervised similarity between Represents the clear seismic attributes in the target seismic attribute set The variance in window w, represents the maximum variance of all clear seismic attributes in the target seismic attribute set in window w, represents the target earthquake attribute, Represents the earthquake attribute c and the earthquake attribute similarity between them; local_loss(w) represents the local loss function within window w, and loss represents the loss function based on unsupervised similarity.

13. The earthquake attribute fusion device according to claim 8, characterized in that: Also includes: A sample attribute normalization module is used to normalize each clear seismic attribute in the sample seismic attribute set to obtain each clear seismic attribute in the normalized sample seismic attribute set; A sample attribute weakening module is used to perform a degradation operation on each clear seismic attribute in the sample seismic attribute set by using a weakening method, and determine a plurality of fuzzy degraded attributes corresponding to each clear seismic attribute in the sample seismic attribute set; A single attribute network training module is used to iteratively train a single attribute neural network by taking each clear seismic attribute and its corresponding multiple fuzzy degraded attributes in the sample seismic attribute set as training samples; A single-attribute network training termination module is used to obtain a trained single-attribute neural network when the conditions for iterative termination are met; The trained single-attribute neural network reflects the mapping relationship between the attribute characteristics of each clear seismic attribute in the sample seismic attribute set and its corresponding multiple fuzzy degraded attributes.

14. The earthquake attribute fusion device according to claim 13, characterized in that: Also includes: A multi-attribute fusion network training module is used to take the attribute features of each clear seismic attribute in the sample seismic attribute set as input and iteratively train the multi-attribute fusion network; The multi-attribute fusion constraint module is used to use the unsupervised similarity-based loss function as a constraint in the iterative training of the multi-attribute fusion network; The multi-attribute network training termination module is used to obtain the trained multi-attribute fusion network when the conditions for iterative termination are met, and output the fusion attribute features corresponding to multiple clear seismic attributes in the sample seismic attribute set; The trained multi-attribute fusion network reflects the mapping relationship between the fused attribute features and the attribute features of each clear earthquake attribute in the sample earthquake attribute set.

15. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the seismic attribute fusion method according to any one of claims 1 to 7 is implemented.

16. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program for executing the seismic attribute fusion method according to any one of claims 1 to 7.