Seismic attribute fusion method and device based on deep learning, and program product

Abstract

The invention discloses a seismic attribute fusion method and device based on deep learning and a program product. The method comprises the following steps: firstly, extracting various first seismic attribute data related to a geological target from a seismic stack section; normalizing the first seismic attribute data, and performing correlation analysis processing on the normalized data to obtain second seismic attribute data; performing data splitting and data enhancement processing on the second seismic attribute data by using a preset sliding time window, and constructing a third seismic attribute data set; then, based on a weighted feature fusion auto-encoder network, convolutional features and variational features of the third seismic attribute data set are extracted respectively, weighted feature fusion is performed by using the convolutional features and the variational features, and weighted fusion features are obtained; according to the seismic attribute fusion method and device based on deep learning and the program product disclosed by the invention, efficient and accurate seismic attribute extraction and fusion are realized, and the interpretable precision and fusion quality of seismic attribute fusion are effectively improved.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and seismic exploration, and in particular to a seismic attribute fusion method, device and program product based on deep learning. Background Art

[0002] Seismic attributes are a kind of characteristics that describe and quantify seismic data. They are calculated from seismic data through mathematical transformation techniques. They can highlight the characteristics of seismic data and reflect the knowledge and experience accumulated by humans in the field of seismic exploration. At present, hundreds of seismic attributes have been extracted, each of which reflects the underground geological structure from different angles. However, a single seismic attribute often contains characteristic information of multiple geological targets and is a comprehensive reflection of the geological structure. In addition, due to the inevitable multi-solution of exploration technology, a single seismic attribute has multi-solutions and limitations in seismic interpretation, making it difficult to fully characterize the underground geological structure. To solve this problem and make full use of different seismic attributes, researchers have proposed a seismic attribute fusion method. Under the guidance of geological laws, attributes that are sensitive to geological targets are selected, and multiple factors are combined through mathematical operations to fully explore the useful information contained in different seismic attribute data to obtain the best results. Seismic attribute fusion aims to integrate the physical or statistical significance of different seismic attributes so that the fusion results can more comprehensively and reliably reflect the characteristics of the geological body.

[0003] The existing commonly used seismic attribute fusion methods include: superposition method, which is simple to operate but has low accuracy and may interfere with each other; RGB fusion method, which is based on color theory but uses a limited number of attributes; data fusion methods such as multiple linear regression, principal component analysis, and isometric mapping. Although these existing methods are more reliable than single seismic attributes, they still have limitations. For example, multiple linear regression requires the prediction of attribute correlation; linear methods such as principal component analysis have difficulty in processing high-dimensional complex nonlinear data; nonlinear methods such as isometric mapping have high computational requirements and may not converge or be sensitive to noise.

[0004] Nowadays, deep learning and artificial intelligence are becoming the major trends of geophysical technology innovation, providing new strategies for complex nonlinear data fusion. Autoencoders can effectively perform low-dimensional representation, which is better than the dimensionality reduction fusion effect of principal component analysis. Its variants, such as sparse autoencoders, obtain more scale features through sparse restrictions, and shared deep autoencoders use shared layers to extract features. As a deep neural network, autoencoders can obtain deeper intrinsic features and have been widely used in various fields. Therefore, how to apply deep learning and autoencoder technology to the field of seismic exploration is a key research direction.

[0005] Therefore, in order to solve the problems of poor fusion effect and low efficiency of existing seismic attribute fusion methods, it is urgent to invent a seismic attribute fusion method based on deep learning, so as to fully obtain the effective information of seismic data and improve the interpretable accuracy and fusion quality of seismic attribute fusion. Summary of the invention

[0006] In view of this, embodiments of the present invention provide a seismic attribute fusion method, device and program product based on deep learning, which at least partially solve the problems existing in the prior art.

[0007] Other features and advantages of the present invention will become apparent from the following detailed description, or may be learned in part by practice of the present invention.

[0008] In order to achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0009] According to a first aspect of an embodiment of the present invention, a seismic attribute fusion method based on deep learning is provided, wherein the method comprises:

[0010] Extracting various first seismic attribute data related to geological targets from seismic stack sections;

[0011] Normalizing the first seismic attribute data, and performing correlation analysis on the normalized data to obtain second seismic attribute data;

[0012] Using a preset sliding time window to perform data splitting processing on the second seismic attribute data, and performing data enhancement on the split data to construct a third seismic attribute data set;

[0013] Based on the weighted feature fusion autoencoder network, the convolutional features and the variational features of the third seismic attribute data set are extracted respectively, and the weighted feature fusion is performed using the convolutional features and the variational features to obtain the weighted fusion features.

[0014] Further, the first seismic attribute data is normalized, the normalized data is subjected to correlation analysis to obtain second seismic attribute data, the second seismic attribute data is subjected to data splitting processing using a preset sliding time window, and the split data is subjected to data enhancement to construct a third seismic attribute data set, including:

[0015] Normalizing the first seismic attribute data to obtain normalized seismic attribute data;

[0016] Performing correlation coefficient calculation and principal component analysis optimization processing on the normalized seismic attribute data, and selecting second seismic attribute data with a higher correlation with the geological target from the normalized seismic attribute data;

[0017] The second seismic attribute data is subjected to data splitting processing using a preset sliding time window, and data enhancement is performed on the split data to construct a third seismic attribute data set, wherein the data enhancement processing includes random rotation and flipping.

[0018] Further, the weighted feature fusion autoencoder network includes an encoder;

[0019] The encoder includes a shared layer, a convolutional coding branch layer, a variational coding branch layer and a feature fusion layer;

[0020] The shared layer includes a first convolutional layer, a first batch normalization layer, and a first ReLU activation function layer;

[0021] The convolutional coding branch layer includes a second convolutional layer, a second batch normalization layer, a second ReLU activation function layer, a first flattening layer and a first fully connected layer;

[0022] The variational coding branch layer includes a second flattening layer, a second fully connected layer and a variational layer;

[0023] The feature fusion layer includes a weighted feature fusion layer, a third fully connected layer, a third batch normalization layer and a third ReLU activation function layer.

[0024] Furthermore, the weighted feature fusion autoencoder network also includes a decoder;

[0025] The decoder includes a transposed convolution layer, a fourth batch normalization layer, and a fourth ReLU activation function layer.

[0026] Further, respectively extracting the convolutional features and the variational features of the third seismic attribute data set, and performing weighted feature fusion using the convolutional features and the variational features to obtain weighted fusion features, including:

[0027] Based on the encoder, using the shared layer, a preliminary feature extraction process is performed on the third seismic attribute data set to obtain preliminary attribute features;

[0028] Using the convolutional coding branch layer and the variational coding branch layer to perform convolutional feature extraction processing and variational feature extraction processing on the attribute preliminary features, respectively, to obtain attribute convolutional features and attribute variational features;

[0029] According to the feature fusion layer, weighted feature fusion processing is performed using the attribute convolution feature and the attribute variational feature to obtain a weighted fusion feature.

[0030] Furthermore, the training process of the weighted feature fusion autoencoder network includes:

[0031] extracting first earthquake attribute training data from the earthquake stack section;

[0032] Normalizing the first seismic attribute training data, and performing correlation analysis on the normalized data to obtain second seismic attribute training data;

[0033] Using a preset sliding time window to perform data splitting processing on the second seismic attribute data, and performing data enhancement processing on the split data to construct a third seismic attribute training data set;

[0034] Using the encoder to extract convolutional features and variational features of the third seismic attribute training data set respectively, and using the convolutional features and variational features to perform weighted feature fusion to obtain training weighted fusion features;

[0035] The decoder is used to decode and reconstruct the training weighted fusion features, and the weighted feature fusion autoencoder network is iteratively optimized through a loss function according to the decoding results to obtain a trained weighted feature fusion autoencoder network.

[0036] Furthermore, the loss function of the convolutional coding branch layer is a mean square error loss function, and the loss function of the variational coding branch layer is a KL divergence loss function;

[0037] Based on Elastic Net regularization, the loss function of the convolutional coding branch layer and the loss function of the variational coding branch layer are weighted to obtain the comprehensive loss function of the weighted feature fusion autoencoder network. The formula of the comprehensive loss function is as follows:

[0038]

[0039] in, is the original data, is the reconstructed data, n is the amount of training set data, and represent the probability distribution functions of the original and reconstructed data respectively, and are the weights of the convolutional coding branch layer and the variational coding branch layer, respectively. and Used to control the strength of regularization, is the weight parameter of the network.

[0040] Furthermore, the training process of the weighted feature fusion autoencoder network also includes:

[0041] According to the hierarchical adaptive transfer learning strategy, the seismic attribute data of different regions are used to perform transfer learning on the weighted feature fusion autoencoder network, and a weighted feature fusion autoencoder network suitable for seismic attribute fusion in different regions is obtained.

[0042] According to a second aspect of an embodiment of the present invention, there is provided a seismic attribute fusion device based on deep learning, the device comprising: a processor and a memory;

[0043] The memory is used to store one or more program instructions;

[0044] The processor is used to run one or more program instructions to execute the steps of a seismic attribute fusion method based on deep learning as described in any one of the above items.

[0045] According to a third aspect of an embodiment of the present invention, a computer program product is provided, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, and when the program instructions are executed by a computer, the computer implements the steps of a deep learning-based seismic attribute fusion method as described in any one of the above items.

[0046] The deep learning-based seismic attribute fusion method, device and program product provided in the embodiments of the present invention extract the convolution and variational features of seismic attributes through a constructed weighted feature fusion autoencoder network to obtain fused seismic attributes, and through hierarchical adaptation-based transfer learning, the weighted feature fusion autoencoder network can quickly adapt to seismic data under different geological conditions, thereby realizing efficient and accurate seismic attribute extraction and seismic attribute fusion in actual seismic exploration. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0048] Figure 1 A schematic diagram of a flow chart of a seismic attribute fusion method based on deep learning provided in an embodiment of the present invention;

[0049] Figure 2 A schematic block diagram of the principle of a seismic attribute fusion method based on deep learning provided in an embodiment of the present invention;

[0050] Figure 3 A schematic diagram of the effect of the third seismic attribute data provided by an embodiment of the present invention;

[0051] Figure 4 A schematic diagram of the structure of a weighted feature fusion autoencoder network provided in an embodiment of the present invention;

[0052] Figure 5 A schematic diagram of the effect of weighted fusion features provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0054] In addition, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or elements is not necessarily limited to those steps or elements explicitly listed, but may include other steps or elements not explicitly listed or inherent to such process, method, product, or apparatus.

[0055] Figure 1 The figure shows a flow chart of a seismic attribute fusion method based on deep learning according to an embodiment of the present invention.

[0056] like Figure 1 As shown, the deep learning-based seismic attribute fusion method according to an embodiment of the present invention may include step S100, step S200, step S300 and step S400.

[0057] In step S100, a plurality of first seismic attribute data related to the geological target are extracted from the seismic stack section.

[0058] Specifically, the above steps include:

[0059] First, determine the geological target of interest and extract various first seismic attribute data of the geological target. Among them, seismic attributes are seismic signals that can highlight seismic data and are calculated from seismic data through mathematical transformation technology based on seismic exploration theory. They are the carrier of knowledge and experience accumulated in the field of seismic exploration.

[0060] Next, in step S200, the first seismic attribute data is normalized, and correlation analysis is performed on the normalized data to obtain second seismic attribute data.

[0061] Specifically, the above steps include:

[0062] First, the first seismic attribute data is normalized to obtain normalized seismic attribute data. The normalization process is used to convert the value of the seismic data into a value between 0 and 1, so that the seismic attributes can be compared and analyzed on the same scale.

[0063] Then, the correlation coefficient and principal component analysis are calculated, and the first seismic attribute data after normalization are subjected to sensitive attribute optimization processing, so as to select the sensitive seismic attribute data with higher correlation with the geological target, that is, the second seismic attribute data;

[0064] In step S300, the second seismic attribute data is split using a preset sliding time window, and the split data is enhanced to construct a third seismic attribute data set.

[0065] Specifically, the above steps include:

[0066] The second seismic attribute data is split using a preset sliding time window. To ensure effective data processing, the data edge is first filled with zeros. Then, at each point center of the second seismic attribute data, a small data of size b*b is intercepted using a step size of b / 2 to obtain the split third seismic attribute data.

[0067] The split third seismic attribute data are subjected to data enhancement processing, including random rotation and flipping, to construct a third seismic attribute dataset. Figure 3 The figure shows a schematic diagram of the effect of the third seismic attribute data.

[0068] The embodiment of the present invention captures local features of data through data splitting processing, and at the same time maintains the continuity of adjacent regions through overlap (step length is b / 2), and the uniform data block size is conducive to subsequent batch processing and network model training. The embodiment of the present invention also uses data enhancement technology to increase data diversity.

[0069] In step S400, based on the weighted feature fusion autoencoder network, convolutional features and variational features of the third seismic attribute data set are extracted respectively, and weighted feature fusion is performed using the convolutional features and variational features to obtain weighted fusion features.

[0070] Specifically, the above steps include:

[0071] Figure 4 The figure shows a schematic diagram of the structure of the weighted feature fusion autoencoder network.

[0072] The weighted feature fusion autoencoder network includes an encoder and a decoder.

[0073] The encoder includes a shared layer, a convolutional coding branch layer, a variational coding branch layer and a feature fusion layer.

[0074] Among them, the shared layer includes the first convolution layer, the first batch normalization layer and the first ReLU activation function layer; the convolutional coding branch layer includes the second convolution layer, the second batch normalization layer, the second ReLU activation function layer, the first flattening layer and the first fully connected layer, wherein the above-mentioned flattening layer is used to convert multidimensional data into one dimension to prepare for the subsequent fully connected layer; the fully connected layer is used to perform comprehensive processing on all inputs.

[0075] The variational coding branch layer includes a second flattening layer, a second fully connected layer, and a variational layer. The variational coding branch layer encodes the mean and logarithmic variance and uses a reparameterization technique to generate a latent vector.

[0076] The feature fusion layer includes a weighted feature fusion layer, a third fully connected layer, a third batch normalization layer, and a third ReLU activation function layer.

[0077] The decoder includes a transposed convolution layer, a fourth batch normalization layer, and a fourth ReLU activation function layer, wherein the transposed convolution layer is used to upsample and reconstruct the feature map, increase the spatial dimension of the feature map, and reduce the number of channels.

[0078] The training process of the weighted feature fusion autoencoder network includes:

[0079] Extract the first seismic attribute training data from the seismic stacking profile; normalize the first seismic attribute training data, perform correlation analysis on the normalized data, and obtain the second seismic attribute training data; use a preset sliding time window to perform data splitting on the second seismic attribute data, and perform data enhancement on the split data to construct a third seismic attribute training data set; use an encoder to extract the convolutional features and variational features of the third seismic attribute training data set respectively, use the convolutional features and variational features to perform weighted feature fusion to obtain training weighted fusion features; use a decoder to decode and reconstruct the training weighted fusion features, and iteratively optimize the weighted feature fusion autoencoder network according to the decoding results through a loss function to obtain a trained weighted feature fusion autoencoder network.

[0080] Among them, the loss function of the convolutional coding branch layer in the weighted feature fusion autoencoder network is the mean squared error (MSE) loss function, and the loss function of the variational coding branch layer is the KL divergence loss function calculated by the mean and variance.

[0081] Based on Elastic Net regularization, the loss function of the convolutional coding branch layer and the loss function of the variational coding branch layer are weighted to form a comprehensive loss function of the weighted feature fusion autoencoder network. The formula of the comprehensive loss function is as follows:

[0082]

[0083] in, is the original data, is the reconstructed data, n is the amount of training set data, and represent the probability distribution functions of the original and reconstructed data respectively, and are the weights of the convolutional coding branch layer and the variational coding branch layer, respectively. and Used to control the strength of regularization, is the weight parameter of the network.

[0084] The embodiment of the present invention utilizes the loss functions of the convolutional coding branch and the variational coding branch to form a comprehensive loss function of the overall network. At the same time, Elastic Net regularization is adopted to effectively prevent overfitting and improve the generalization ability of the model, especially when processing high-dimensional seismic attribute data, it can produce sparse solutions and small and dispersed weights.

[0085] Each attribute (such as amplitude, frequency, phase, etc.) of the third seismic attribute data set is input into the trained weighted feature fusion autoencoder network.

[0086] Based on the weighted feature fusion of the encoder in the autoencoder network, the shared layer is used to perform preliminary feature extraction on the third seismic attribute dataset to obtain preliminary attribute features.

[0087] The above steps specifically include: extracting convolution features by the first convolution layer to capture spatial structure and local features; the first batch normalization layer is applied after each first convolution layer to reduce internal covariate shift, which helps to speed up the training process and improve model stability; the first ReLU activation function layer is used to introduce nonlinearity, allowing the network to learn more complex patterns, enhance the network's expressiveness, and alleviate the gradient vanishing problem.

[0088] The convolutional coding branch layer and the variational coding branch layer are used to perform convolutional feature extraction and variational feature extraction on the preliminary attribute features, respectively, to obtain attribute convolutional features and attribute variational features, respectively.

[0089] The convolutional coding branch layer is used to capture the local and global characteristics of seismic attributes; the variational coding branch layer is used to generate a potential representation of probabilistic seismic attributes, capturing the statistical characteristics and uncertainty of the data.

[0090] The new features (attribute convolution and variational features) extracted from the seismic attribute profile in the embodiment of the present invention can be regarded as new seismic attributes, thereby achieving the purpose of extracting new seismic attributes.

[0091] Then use the feature fusion layer to weightedly merge the attribute convolutional features and the attribute variational features to obtain weighted fusion features, and use the third connection layer to further process the fused features, and use the third batch normalization layer to standardize the data again.

[0092] Figure 5 The figure shows a schematic diagram of the effect of earthquake fusion features after feature weighted fusion.

[0093] The embodiment of the present invention combines the advantages of convolution and variational features to provide a more comprehensive and robust representation of seismic attributes. The fused seismic attributes can better reflect complex geological structures. At the same time, the fused features contain seismic information of different scales, which helps to analyze large-scale structures and small-scale details at the same time, can capture complex nonlinear relationships between attributes, and can discover hidden patterns that are difficult to identify with traditional methods.

[0094] The embodiment of the present invention uses a convolutional autoencoder (CAE) to extract convolutional features to capture spatial structures and local features; and uses a variational autoencoder (VAE) to extract variational features to learn the probability distribution of data, which helps to generate new samples and handle uncertainty. The features extracted by CAE and VAE are then weighted and fused to form a richer data feature representation, achieving the purpose of seismic attribute fusion. The decoder of the reconstructed data ensures that the features learned by the encoder can effectively represent the original data.

[0095] In addition, the training process of the weighted feature fusion autoencoder network also includes:

[0096] When using the weighted feature fusion autoencoder network to fuse seismic attributes in different regions, it is necessary to enhance the universality of the weighted feature fusion autoencoder network.

[0097] According to the hierarchical adaptive transfer learning strategy, the seismic attribute data of different regions are used to perform transfer learning on the weighted feature fusion autoencoder network, and a weighted feature fusion autoencoder network suitable for seismic attribute fusion in different regions is obtained.

[0098] Specifically, for the autoencoder hierarchical structure of the weighted feature fusion autoencoder network, a hierarchical adaptive transfer (HAT) learning strategy is proposed to optimize the network performance by gradually increasing the trainable part of the network model.

[0099] The specific steps of the hierarchical adaptation-based transfer learning strategy HAT include:

[0100] Feature extraction, using the pre-trained encoder part as a fixed feature extractor, adding multiple new task-specific layers on the encoder output, and only training the newly added layers.

[0101] Fine-tune the top layers, unfreeze the top encoding layers in the encoder layer by layer, and fine-tune these layers and the newly added layers using a smaller learning rate.

[0102] Unfreeze layer by layer, gradually unfreeze more encoding layers, and perform short-term fine-tuning after each unfreeze.

[0103] Fine-tune the entire model, unfreeze all layers, and fine-tune the entire model using a very small learning rate.

[0104] Finally, the adjusted weighted feature fusion autoencoder network is obtained, which can be applied to the fusion of seismic attributes in different regions.

[0105] In order to increase the universality of the weighted feature fusion autoencoder network, the embodiment of the present invention adds a hierarchical adaptive transfer learning module after the trained network model. Through hierarchical adaptive transfer learning, the trained model parameters are migrated to the new model, and the shared knowledge between related data or tasks is utilized to accelerate the training speed and improve the model generalization ability based on the iteratively trained network.

[0106] Through the hierarchical adaptation-based transfer learning strategy proposed in the embodiment of the present invention, the network model can gradually adapt to find a better parameter configuration. The slow adaptation process helps to retain the original useful features, and the hierarchical progressive method helps the model learn more robust feature representations, and the strategy of each stage can be adjusted according to the specific tasks and data characteristics.

[0107] The hierarchical adaptation-based transfer learning strategy provided by the embodiment of the present invention provides a powerful and flexible framework for the transfer learning of autoencoders, and opens up a new way to improve the universality of seismic data analysis models. By systematically unfreezing layers, fine-tuning learning rates, and gradually increasing model complexity, efficient migration of pre-trained autoencoder knowledge to new seismic data tasks is achieved. Its core advantage is not only reflected in the significant improvement of model performance, but more importantly, it enhances the adaptability and generalization ability of the model in a diverse seismic environment. Through this method, not only can the adaptability problem of traditional models in the face of new data be solved, but also a more intelligent and flexible new method is proposed for seismic data analysis. With continuous practice and optimization, the hierarchical adaptation-based transfer learning strategy will become a key tool to improve the universality of seismic data analysis models.

[0108] In addition, an embodiment of the present invention also provides a seismic attribute fusion device based on deep learning, which includes: a processor and a memory; the memory is used to store one or more program instructions; the processor is used to run one or more program instructions to execute the steps of a seismic attribute fusion method based on deep learning as described above.

[0109] In addition, an embodiment of the present invention further provides a computer program product, which includes computer program instructions, and when the computer program instructions are executed by a processor, the steps of the seismic attribute fusion method based on deep learning as described above are implemented.

[0110] The seismic attribute fusion method, device and program product provided in the embodiments of the present invention extract the convolutional and variational features of seismic attributes through a weighted feature fusion autoencoder network, fuse the features of the two at the feature fusion layer, and enable the weighted feature fusion autoencoder network to quickly adapt to seismic data under different geological conditions through hierarchical adaptation-based transfer learning, thereby realizing efficient and accurate seismic attribute extraction and seismic attribute fusion in actual seismic exploration.

[0111] In the embodiment of the present invention, the processor may be an integrated circuit chip having the ability to process signals. The processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, or discrete hardware components. The methods, steps, and logic block diagrams disclosed in the embodiments of the present invention may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present invention may be directly embodied as being executed by a hardware decoding processor, or may be executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The processor reads the information in the storage medium and completes the steps of the above method in combination with its hardware. The storage medium may be a memory, for example, a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct RAM bus random access memory (DRRAM).The storage medium described in the embodiment of the present invention is intended to include but is not limited to these and any other suitable types of memory. Those skilled in the art should be aware that in one or more of the above examples, the functions described in the present invention can be implemented by a combination of hardware and software. When the software is applied, the corresponding function can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein the communication medium includes any medium that is convenient for transmitting a computer program from one place to another. The storage medium can be any available medium that a general or special-purpose computer can access. Although the present invention has been described in detail above with general descriptions and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to it on the basis of the present invention. Therefore, these modifications or improvements made on the basis of not departing from the spirit of the present invention all belong to the scope of protection claimed in the present invention.

[0112] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Those skilled in the art may make some simple modifications, equivalent changes or modifications using the technical contents disclosed above, which all fall within the protection scope of the present invention.

Claims

1. A seismic attribute fusion method based on deep learning, characterized in that: The method comprises: Extracting various first seismic attribute data related to geological targets from seismic stack sections; Normalizing the first seismic attribute data, and performing correlation analysis on the normalized data to obtain second seismic attribute data; Using a preset sliding time window to perform data splitting processing on the second seismic attribute data, and performing data enhancement on the split data to construct a third seismic attribute data set; Based on the weighted feature fusion autoencoder network, the convolutional features and the variational features of the third seismic attribute data set are extracted respectively, and the weighted feature fusion is performed using the convolutional features and the variational features to obtain the weighted fusion features.

2. The seismic attribute fusion method based on deep learning according to claim 1, characterized in that: Normalizing the first seismic attribute data, performing correlation analysis on the normalized data to obtain second seismic attribute data, performing data splitting processing on the second seismic attribute data using a preset sliding time window, and performing data enhancement on the split data to construct a third seismic attribute data set, including: Normalizing the first seismic attribute data to obtain normalized seismic attribute data; Performing correlation coefficient calculation and principal component analysis optimization processing on the normalized seismic attribute data, and selecting second seismic attribute data with a higher correlation with the geological target from the normalized seismic attribute data; The second seismic attribute data is subjected to data splitting processing using a preset sliding time window, and data enhancement is performed on the split data to construct a third seismic attribute data set, wherein the data enhancement processing includes random rotation and flipping.

3. The seismic attribute fusion method based on deep learning according to claim 1, characterized in that: The weighted feature fusion autoencoder network includes an encoder; The encoder includes a shared layer, a convolutional coding branch layer, a variational coding branch layer and a feature fusion layer; The shared layer includes a first convolutional layer, a first batch normalization layer, and a first ReLU activation function layer; The convolutional coding branch layer includes a second convolutional layer, a second batch normalization layer, a second ReLU activation function layer, a first flattening layer and a first fully connected layer; The variational coding branch layer includes a second flattening layer, a second fully connected layer and a variational layer; The feature fusion layer includes a weighted feature fusion layer, a third fully connected layer, a third batch normalization layer and a third ReLU activation function layer.

4. The seismic attribute fusion method based on deep learning according to claim 3, characterized in that: The weighted feature fusion autoencoder network also includes a decoder; The decoder includes a transposed convolution layer, a fourth batch normalization layer, and a fourth ReLU activation function layer.

5. The seismic attribute fusion method based on deep learning according to claim 3, characterized in that: Extracting convolution features and variational features of the third seismic attribute data set respectively, and performing weighted feature fusion using the convolutional features and the variational features to obtain weighted fusion features, including: Based on the encoder, using the shared layer, a preliminary feature extraction process is performed on the third seismic attribute data set to obtain preliminary attribute features; Using the convolutional coding branch layer and the variational coding branch layer to perform convolutional feature extraction processing and variational feature extraction processing on the attribute preliminary features, respectively, to obtain attribute convolutional features and attribute variational features; According to the feature fusion layer, weighted feature fusion processing is performed using the attribute convolution feature and the attribute variational feature to obtain a weighted fusion feature.

6. The seismic attribute fusion method based on deep learning according to claim 4, characterized in that: The training process of the weighted feature fusion autoencoder network includes: extracting first earthquake attribute training data from the earthquake stack section; Normalizing the first seismic attribute training data, and performing correlation analysis on the normalized data to obtain second seismic attribute training data; Using a preset sliding time window to perform data splitting processing on the second seismic attribute data, and performing data enhancement processing on the split data to construct a third seismic attribute training data set; Using the encoder to extract convolutional features and variational features of the third seismic attribute training data set respectively, and using the convolutional features and variational features to perform weighted feature fusion to obtain training weighted fusion features; The decoder is used to decode and reconstruct the training weighted fusion features, and the weighted feature fusion autoencoder network is iteratively optimized through a loss function according to the decoding results to obtain a trained weighted feature fusion autoencoder network.

7. The seismic attribute fusion method based on deep learning according to claim 6, characterized in that: The loss function of the convolutional coding branch layer is a mean square error loss function, and the loss function of the variational coding branch layer is a KL divergence loss function; Based on Elastic Net regularization, the loss function of the convolutional coding branch layer and the loss function of the variational coding branch layer are weighted to obtain the comprehensive loss function of the weighted feature fusion autoencoder network. The formula of the comprehensive loss function is as follows: in, is the original data, is the reconstructed data, n is the amount of training set data, and represent the probability distribution functions of the original and reconstructed data respectively, and are the weights of the convolutional coding branch layer and the variational coding branch layer, respectively. and Used to control the strength of regularization, is the weight parameter of the network.

8. The seismic attribute fusion method based on deep learning according to claim 6, characterized in that: The training process of the weighted feature fusion autoencoder network also includes: According to the hierarchical adaptive transfer learning strategy, the seismic attribute data of different regions are used to perform transfer learning on the weighted feature fusion autoencoder network, and a weighted feature fusion autoencoder network suitable for seismic attribute fusion in different regions is obtained.

9. A seismic attribute fusion device based on deep learning, characterized in that: The device comprises: a processor and a memory; The memory is used to store one or more program instructions; The processor is used to run one or more program instructions to execute the steps of a seismic attribute fusion method based on deep learning as described in any one of claims 1 to 8.

10. A computer program product, characterized in that The computer program product comprises computer program instructions, which, when executed by a processor, implement the steps of a seismic attribute fusion method based on deep learning as described in any one of claims 1 to 8.

Images