An intelligent fault identification method and system based on an improved U-Net network

Abstract

This invention relates to an intelligent fault identification method and system based on an improved U-Net network. The method includes the following steps: acquiring raw seismic profile data and processing it to obtain fault data; establishing an improved residual module and a channel attention mechanism module based on the fault data; constructing a U-Net model based on the improved residual module and the channel attention mechanism module; optimizing the U-Net model using a weighted cross-entropy function, and training the optimized U-Net model based on the fault data to obtain a final model; inputting the raw seismic profile data into the final model to obtain the fault identification result. This invention improves the discriminative power of features at each stage, increases the network depth, enhances the consistency of fault prediction, and improves the accuracy of fault identification; and uses a weighted loss function to alleviate the problem of extreme imbalance between fault data and seismic data, thereby improving the accuracy of network fault identification.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas exploration technology, and particularly relates to an intelligent fault identification method and system based on an improved U-Net network. Background Technology

[0002] Faults play a crucial role in hydrocarbon accumulation and development. As a vital component of seismic data processing and interpretation, rapid and effective fault identification, based on seismic data and existing geological information, has been a key research focus both domestically and internationally. Early seismic data interpretation involved professionals manually interpreting faults by establishing fault identification markers based on reflection axis characteristics. Later, seismic data was combined with well logging data for joint well-seismic identification, and seismic attributes such as coherence volume and curvature were used to identify faults. Currently, researchers employ machine learning algorithms such as ant colony decomposition, extreme gradient descent, edge detection, and support vector machines to characterize faults, continuously improving the accuracy and efficiency of fault identification.

[0003] While automatic fault identification methods have achieved some success, many problems remain. First, automatic fault identification methods based on seismic attributes often weaken the lateral resolution of seismic profiles to emphasize fault characteristics, resulting in low accuracy. Second, these methods require extensive computation and emphasize local features, making them unsuitable for constructing faults holistically. Furthermore, these methods are highly sensitive to noise, easily leading to false fault identifications that require significant post-construction corrections by professionals. Finally, various improvements to automatic fault identification methods also face new challenges; for example, incorporating dip estimation into variance attributes still fails to eliminate the influence of stratigraphic layers.

[0004] Deep learning, as a key core technology in machine learning, is widely used in natural language processing, computer vision, and other interdisciplinary fields. When deep learning methods are used for fault identification, they continuously optimize the network's parameters and weights to extract hidden low-dimensional and high-dimensional features of the fault, including features such as low continuity and high discontinuity of reflection phase axes, thereby optimizing the network and ultimately achieving accurate fault identification. Furthermore, deep learning itself requires a large amount of data to train the model and can process large amounts of data quickly and effectively. Seismic exploration precisely meets the prerequisites for deep learning, and the exploration and development of oil and gas fields urgently require the powerful feature extraction capabilities of deep learning. For these reasons, many researchers have introduced various theories and methods of deep learning into fault identification.

[0005] In existing technologies, the accuracy of manual fault picking or seismic attribute detection in complex seismic profiles is low. U-Net has achieved excellent fault identification results. It uses an encoder-decoder structure combined with a convolutional neural network to predict each pixel in the seismic profile image, determining whether each pixel represents a fault. This approach yields good identification results on real data. U-Net can not only be used directly for automatic fault identification but also as a high-performance network framework. The specific approach involves designing corresponding extraction modules based on actual fault characteristics and building a model using U-Net as the framework to improve fault identification accuracy. A multi-scale attention convolutional neural network is introduced into the autoencoder structure of U-Net to merge and refine different spatial features of the fault. The network was tested with synthetic and field data, and the results show that it outperforms conventional convolutional neural network methods in identifying faults in complex seismic profiles.

[0006] Although the application of U-Net network to tomography identification has achieved some results, some problems still exist, such as gradient explosion and vanishing during gradient backpropagation, which leads to a decrease in network performance and poor accuracy. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent tomography identification method and system based on an improved U-Net network in order to solve the above-mentioned problems.

[0008] The present invention achieves the above objectives through the following technical solutions:

[0009] A smart tomography identification method based on an improved U-Net network includes the following steps:

[0010] Obtain raw seismic profile data and process it to obtain fault data;

[0011] An improved residual module and channel attention mechanism module are established based on the aforementioned fault data;

[0012] A U-Net model is constructed based on the improved residual module and the channel attention mechanism module;

[0013] The U-Net model is optimized using a weighted cross-entropy function, and the optimized U-Net model is trained based on the tomographic data to obtain the final model.

[0014] The original seismic profile data is input into the final model to obtain the fault identification results.

[0015] As a further optimization of the present invention, the fault data includes seismic grayscale images and mask label data. The specific process of obtaining the original seismic profile data and processing it to obtain the fault data is as follows:

[0016] Seismic grayscale maps are obtained by converting raw seismic profile data containing faults into grayscale maps.

[0017] Based on the aforementioned seismic grayscale map, faults are marked to generate mask label data;

[0018] The original seismic profiles and their corresponding masked label data are combined to form training and testing sets.

[0019] As a further optimization of the present invention, the improved residual module consists of a dual-channel structure comprising one 1×1 convolution, two 3×3 convolutions, batch normalization, and an activation layer; wherein the padding of the convolution is 1 and the stride is 1; firstly, the seismic grayscale image is divided into two channels after being processed by the 1×1 convolution; the first channel is sequentially processed by the 3×3 convolution, batch normalization, ELU activation function, and another 3×3 convolution, and then added to the result of the 1×1 convolution; finally, the feature tensor is output using the ELU activation function.

[0020] The formula for the ELU activation function is:

[0021]

[0022] In the formula, α is the adjustment coefficient; y' is the characteristic output; The input is the feature.

[0023] As a further optimization of the present invention, the channel attention mechanism module consists of two 3×3 convolutions, batch normalization, ELU activation function, and sigmoid activation function; wherein the padding of the convolution is 1 and the stride is 1; firstly, the shallow features output by the improved residual module and the deep features output by the deconvolution module are fused; the fused features are then activated by the ELU activation function after a 3×3 convolution; the activated result is then activated by another 3×3 convolution and the sigmoid activation function; the activated result is multiplied by the shallow features; finally, the result of the multiplication is added to the deep features and output.

[0024] The formula for the sigmoid activation function is:

[0025]

[0026] In the formula, x is the input of the feature; σ(x) is the output of the feature.

[0027] As a further optimization of the present invention, the specific process of constructing the U-Net model based on the improved residual module and the channel attention mechanism module is as follows:

[0028] The system constructs an input layer, a hidden layer, and an output layer. The input layer is the input of the seismic grayscale image; the output layer is the predicted probability distribution of the fault; the hidden layer is divided into two parts: upsampling and downsampling.

[0029] The downsampling includes four of the improved residual modules and four max-pooling layers;

[0030] Upsampling consists of four deconvolutional layers and four residual convolutional modules;

[0031] Upsampling and downsampling are connected by four channel attention mechanism modules.

[0032] As a further optimization of the present invention, the U-Net model is optimized using a weighted cross-entropy function, and the optimized U-Net model is trained based on the tomographic data to obtain the final model. The specific process is as follows:

[0033] The training set is input into the U-Net model, gradient backpropagation is used, and the weight cross-entropy is used as the loss function to measure the distance between the true value and the predicted value.

[0034] The formula for the weighted cross-entropy is:

[0035]

[0036] In the formula, α is the proportion of tomographic data in the entire dataset; γ is the attention parameter; The modulation coefficient;

[0037] In each iteration of the dataset, an adaptive algorithm is used to adjust the weights between neurons to reduce the loss function, thereby optimizing the network. The performance of the U-Net model is qualitatively judged by accuracy, recall, and F1 score. The parameters of the network model are saved initially. The iteration is repeated until the requirements for tomography are met and the final model is saved.

[0038] The formulas for the accuracy (Prec), recall (Rec), and F1 score are as follows:

[0039]

[0040]

[0041]

[0042] In the formula, TP is the number of pixels that are correctly detected as faults; FP is the number of pixels that are not faults but are predicted as faults; and FN is the number of pixels that are not actually faults but are not detected as faults.

[0043] As a further optimization of the present invention, the fault identification result is a probability distribution map of the fault.

[0044] A smart tomography system based on an improved U-Net network includes:

[0045] The data acquisition module is used to acquire raw seismic profile data and process it to obtain fault data;

[0046] An improved module construction module is used to build an improved residual module and a channel attention mechanism module based on the tomographic data;

[0047] A model building module is used to build a U-Net model based on the improved residual module and the channel attention mechanism module;

[0048] The model optimization module is used to optimize the U-Net model using the weighted cross-entropy function, and to train the optimized U-Net model based on the tomographic data to obtain the final model.

[0049] The fault identification module is used to input the original seismic profile data into the final model to obtain the fault identification results.

[0050] An electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0051] Memory, used to store computer programs;

[0052] The processor, when executing programs stored in memory, implements an intelligent tomography identification method based on an improved U-Net network.

[0053] A computer-readable storage medium storing a computer program that, when executed by a processor, implements an intelligent tomography identification method based on an improved U-Net network.

[0054] The beneficial effects of this invention are as follows:

[0055] This invention replaces the convolutional layers of the U-Net network with an improved residual module, solving the problem of gradient explosion and vanishing during gradient backpropagation that leads to network performance degradation. It also refines features and integrates channel information, improving the discriminative power of features at each stage and increasing the network depth. In upsampling, a channel attention mechanism is added to enhance the consistency of fault prediction and improve the accuracy of fault identification. Finally, a weighted loss function is used to alleviate the extreme imbalance between fault data and seismic data, further improving the accuracy of network fault identification. Attached Figure Description

[0056] Figure 1 This is a flowchart of the method of the present invention;

[0057] Figure 2 This is a schematic diagram of the seismic profile cut in an embodiment of the present invention;

[0058] Figure 3 This is a schematic diagram of mask label generation in an embodiment of the present invention;

[0059] Figure 4 This is a schematic diagram of the improved residual module in an embodiment of the present invention;

[0060] Figure 5 This is a schematic diagram of the channel attention mechanism in an embodiment of the present invention;

[0061] Figure 6 This is a schematic diagram of the improved U-Net structure in an embodiment of the present invention;

[0062] Figure 7 This is a flowchart of an embodiment of the present invention;

[0063] Figure 8 This is a diagram illustrating the improved U-Net tomography identification effect in an embodiment of the present invention;

[0064] Figure 9 This is a system structure block diagram in an embodiment of the present invention;

[0065] Figure 10 This is a block diagram of the device structure in an embodiment of the present invention. Detailed Implementation

[0066] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content.

[0067] like Figure 1 As shown, an intelligent tomography identification method based on an improved U-Net network includes the following steps:

[0068] Obtain raw seismic profile data and process it to obtain fault data;

[0069] An improved residual module and channel attention mechanism module are established based on the aforementioned fault data;

[0070] A U-Net model is constructed based on the improved residual module and the channel attention mechanism module;

[0071] The U-Net model is optimized using a weighted cross-entropy function, and the optimized U-Net model is trained based on the tomographic data to obtain the final model.

[0072] The original seismic profile data is input into the final model to obtain the fault identification results.

[0073] In this embodiment, to efficiently identify faults in seismic profiles, the present invention employs a combination of an improved residual module and a channel attention mechanism, such as... Figure 4 and Figure 5 As shown, an improved U-Net tomography identification network model was constructed. (Refer to...) Figure 7 The flowchart of the present invention is shown, which includes the following steps:

[0074] Step 1, Preparation of fault input data:

[0075] (1) Conversion and cropping of seismic grayscale profiles, such as Figure 2 As shown; the fault identification accuracy is 16×16, with a total of 500 profiles. Each interpretation profile contains a different number of seismic traces, and the number of sampling points per trace also varies, with a sampling interval of 2ms. The amplitude data of the seismic profiles were truncated using the largest rectangle. To further reduce computational load, the extracted seismic data was normalized and a single-channel grayscale image was generated, as shown below. Figure 5 ;

[0076] (2) Generation of mask data, such as Figure 3 As shown; after cropping, each seismic profile was divided into several sub-profiles, expanding the entire dataset from 500 to 6000 images. The input to the deep learning model became single-channel grayscale values. The labelme annotation tool in the Python compiler was used for fault annotation, generating layer mask labels. The size of the fault mask label data matrix is ​​the same as the size of the seismic data, where faults are marked with 1 and non-faults with 0, as shown. Figure 7 ;

[0077] (3) The original seismic sub-sections and corresponding fault mask data constitute the dataset of the network model.

[0078] Step 2: Construct an improved U-Net tomography network.

[0079] (1) Construct an improved residual module.

[0080] This module consists of a dual-channel architecture comprising one 1×1 convolution, two 3×3 convolutions, batch normalization, and an activation layer. The convolutions have a padding of 1 and a stride of 1. First, the seismic grayscale image is divided into two channels after a 1×1 convolution. The first channel then undergoes a 3×3 convolution, batch normalization, ELU activation, and another 3×3 convolution, before being added to the result of the 1×1 convolution. Finally, the ELU activation function is used to output the feature tensor.

[0081] The formula for the ELU activation function is:

[0082]

[0083] In the formula, α is the adjustment coefficient; y' is the characteristic output; The input is the feature.

[0084] (2) Construct a channel attention mechanism.

[0085] This module consists of two 3×3 convolutions, batch normalization, ELU activation, and sigmoid activation. The convolution padding is 1, and the stride is 1. First, the shallow features output from the improved residual module are fused with the deep features output from the deconvolution module. The fused features are then activated using the ELU activation function after a 3×3 convolution. The activated result is then further activated using a 3×3 convolution and the sigmoid activation function. This activated result is multiplied by the shallow features. Finally, the multiplied result is added to the deep features and output.

[0086] The formula for the sigmoid activation function is:

[0087]

[0088] In the formula, x is the input of the feature; σ(x) is the output of the feature.

[0089] Step 3: Build the U-Net model using the designed residual module and channel attention mechanism module.

[0090] A three-layer network is constructed: an input layer, hidden layers, and an output layer. The input layer is the seismic grayscale image; the output layer is the predicted probability distribution of faults. The hidden layer consists of two parts: upsampling and downsampling. Downsampling includes four improved residual modules and four max-pooling layers; upsampling consists of four deconvolutional layers and four residual convolutional modules. Upsampling and downsampling are connected by four channel attention modules, such as... Figure 6 .

[0091] Step 4: Input the preprocessed training set from Step 1 into the network model built in Step 3 by queuing, use gradient backpropagation, and use weight cross-entropy as the loss function to measure the distance between the true value and the predicted value.

[0092] The formula for weighted cross-entropy is:

[0093]

[0094] In the formula, α is the proportion of tomographic data in the entire dataset, which can be set through label data statistics; γ is the attention parameter; This is the modulation coefficient, which reduces the weight of the background, thus allowing the model to focus more on the difficult-to-classify faults.

[0095] Step 5: Setting the initial parameters of the model.

[0096] After the dataset and network model were built, the model parameters were initialized. Specifically, they were set as follows:

[0097]

[0098] Step 6: Optimize the network model and save the model parameters.

[0099] In each iteration of the dataset, an adaptive algorithm is used to adjust the weights between neurons to reduce the loss function and optimize the network. The accuracy, recall, and F1 score are used to qualitatively judge the performance of the network model. The parameters of the network model are saved initially. The model iteration is repeated. If it does not meet the requirements of tomography recognition, training continues or the network is trained again after adjusting the parameters in step 5. If it meets the requirements, the iteration is stopped and the final network model is saved.

[0100] The formulas for precision, recall, and F1 score are:

[0101]

[0102]

[0103]

[0104] In the formula, TP is the number of pixels that are correctly detected as faults; FP is the number of pixels that are not faults but are predicted as faults; and FN is the number of pixels that are not actually faults but are not detected as faults.

[0105] Step 7: Use the optimized network to identify faults in the original seismic profile and output the probability distribution map of the faults.

[0106] Figure 8 Examples of tomography identification according to the present invention are shown below, where (a) is the result of manual interpretation; (b) is the tomography identification effect of FCN; (c) is the tomography identification effect of U-Net; (d) is the tomography identification effect of U-Net++; (e) is the tomography identification effect of Deeplap V3; and (f) is the tomography identification effect of the method of the present invention.

[0107] like Figure 9 As shown, embodiments of this disclosure provide an intelligent tomography system based on an improved U-Net network, comprising:

[0108] The data acquisition module is used to acquire raw seismic profile data and process it to obtain fault data;

[0109] An improved module construction module is used to build an improved residual module and a channel attention mechanism module based on the tomographic data;

[0110] A model building module is used to build a U-Net model based on the improved residual module and the channel attention mechanism module;

[0111] The model optimization module is used to optimize the U-Net model using the weighted cross-entropy function, and to train the optimized U-Net model based on the tomographic data to obtain the final model.

[0112] The fault identification module is used to input the original seismic profile data into the final model to obtain the fault identification results.

[0113] The implementation process of the functions and roles of each module in the above system is detailed in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0114] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0115] In the above embodiments, any number of modules can be combined into one module, or any one module can be split into multiple modules. Alternatively, at least some functionality of one or more modules can be combined with at least some functionality of other modules and implemented in one module. At least one of the modules can be at least partially implemented as hardware circuitry, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a System-on-Chip, a System-on-Substrate, a System-on-Package, an Application-Specific Integrated Circuit (ASIC), or any other reasonable method of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three methods. Alternatively, at least one of the modules can be at least partially implemented as a computer program module that, when run, performs a corresponding function.

[0116] See Figure 10The electronic device provided in the embodiments of this disclosure includes a processor 1110, a communication interface 1120, a memory 1130 and a communication bus 1140, wherein the processor 1110, the communication interface 1120 and the memory 1130 communicate with each other through the communication bus 1140.

[0117] Memory 1130 is used to store computer programs;

[0118] When the processor 1110 executes the program stored in the memory 1130, it implements the intelligent tomography identification method based on the improved U-Net network as shown below.

[0119] The aforementioned communication bus 1140 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 1140 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus.

[0120] The communication interface 1120 is used for communication between the above-mentioned electronic device and other devices.

[0121] The memory 1130 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory 1130 may also be at least one storage device located remotely from the aforementioned processor 1110.

[0122] The processor 1110 mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0123] Embodiments of this disclosure also provide a computer-readable storage medium. The computer-readable storage medium stores a computer program that, when executed by a processor, implements the intelligent tomography identification method based on the improved U-Net network as described above.

[0124] The computer-readable storage medium may be included in the device / apparatus described in the above embodiments; or it may exist independently and not assembled into the device / apparatus. The computer-readable storage medium carries one or more programs that, when executed, implement the intelligent tomography method based on the improved U-Net network according to embodiments of this disclosure.

[0125] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0126] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. An intelligent fault identification method based on an improved U-Net network, characterized in that, Includes the following steps: The process involves acquiring and processing raw seismic profile data to obtain fault data, which includes seismic grayscale images and mask label data. Specifically, the process includes: converting the raw seismic profile data containing faults into grayscale images; labeling the faults based on the seismic grayscale images to generate mask label data; and combining the raw seismic profile data with the corresponding mask label data to form a training set and a test set. An improved residual module and channel attention mechanism module are established based on the fault data. The improved residual module consists of a dual-channel architecture comprising one 1×1 convolution, two 3×3 convolutions, batch normalization, and an activation layer. The convolution padding is 1, and the stride is 1. First, the seismic grayscale image is divided into two channels after a 1×1 convolution. The first channel is then sequentially processed by a 3×3 convolution, batch normalization, ELU activation function, and another 3×3 convolution, before being added to the result of the 1×1 convolution. Finally, the ELU activation function is used to output the feature tensor. The channel attention mechanism module... It consists of two 3×3 convolutions, batch normalization, ELU activation function, and sigmoid activation function; the padding of the convolution is 1, and the stride is 1. First, the shallow features output by the improved residual module are fused with the deep features output by the deconvolution module. The fused features are then activated by the ELU activation function after a 3×3 convolution. The activated result is then activated by the sigmoid activation function again. The activated result is multiplied with the shallow features. Finally, the result of the multiplication is added to the deep features and output. The U-Net model is constructed based on the improved residual module and the channel attention mechanism module, including: constructing an input layer, a hidden layer, and an output layer; the hidden layer is divided into upsampling and downsampling parts; the downsampling part includes 4 of the improved residual modules and 4 max pooling layers; the upsampling part includes 4 deconvolution layers and 4 residual convolution modules; the upsampling and downsampling parts are connected by 4 of the channel attention mechanism modules. The U-Net model is optimized using a weighted cross-entropy function, and the optimized U-Net model is trained based on the tomographic data to obtain the final model. The process includes: inputting the training set into the U-Net model, using gradient backpropagation, and using weighted cross-entropy as the loss function to measure the distance between the true value and the predicted value. The formula for the weighted cross-entropy is: ； wherein is the ratio of the fault data over the entire data set; is the attention parameter; is the modulation coefficient; In each iteration of the dataset, an adaptive algorithm is used to adjust the weights between neurons to reduce the loss function, thereby optimizing the network. The performance of the U-Net model is qualitatively judged by accuracy, recall, and F1 score. The parameters of the network model are saved initially. The iteration is repeated until the requirements for tomography are met and the final model is saved. The original seismic profile data is input into the final model to obtain the fault identification results.

2. The intelligent fault identification method based on the improved U-Net network according to claim 1, characterized in that, The formula for the ELU activation function of the improved residual module is: ； wherein is a tuning coefficient; is a characteristic output; is an input to the characteristic. 3.The intelligent fault identification method based on improved U-Net network according to claim 1, characterized in that, The formula for the sigmoid activation function of the channel attention mechanism module is: ； wherein input characteristic; output characteristic.

4. The intelligent fault identification method based on the improved U-Net network according to claim 1, characterized in that, The input layer is the input of the seismic grayscale image; the output layer is the predicted probability distribution of the fault.

5. The intelligent fault identification method based on improved U-Net network according to claim 1, characterized in that, The accuracy Prec The recall rate Rec The formula for F1 value is: ； ； ； In the formula, TP is the number of pixels in which the fault is correctly detected. FP the number of pixels that are not faults that are predicted to be faults; FN the number of pixels that are actual faults that are not detected as faults. 6.The intelligent fault identification method based on improved U-Net network according to claim 1, characterized in that, The fault identification result is a probability distribution map of the fault.

7. An intelligent fault identification system based on improved U-Net network, characterized in that, include: The data acquisition module is used to acquire raw seismic profile data and process it to obtain fault data. The fault data includes seismic grayscale images and mask label data. The specific process includes: converting the raw seismic profile data containing faults into grayscale images to obtain seismic grayscale images; labeling the faults based on the seismic grayscale images to generate mask label data; and combining the raw seismic profiles and the corresponding mask label data to form a training set and a test set. An improved module construction module is used to establish an improved residual module and a channel attention mechanism module based on the fault data. The improved residual module consists of a dual-channel structure composed of one 1×1 convolution, two 3×3 convolutions, batch normalization, and an activation layer. The convolution padding is 1, and the stride is 1. First, the seismic grayscale image is divided into two channels after a 1×1 convolution. The first channel is then sequentially processed by a 3×3 convolution, batch normalization, ELU activation function, and another 3×3 convolution, and then added to the result of the 1×1 convolution. Finally, the ELU activation function is used to output the feature tensor. The channel attention mechanism... The force mechanism module consists of two 3×3 convolutions, batch normalization, ELU activation function, and sigmoid activation function; the padding of the convolution is 1, and the stride is 1. First, the shallow features output by the improved residual module are fused with the deep features output by the deconvolution module. The fused features are then activated by the ELU activation function after a 3×3 convolution. The activated result is then activated by the sigmoid activation function again. The activated result is multiplied with the shallow features. Finally, the result of the multiplication is added to the deep features and output. The model building module is used to build a U-Net model based on the improved residual module and the channel attention mechanism module, including: building an input layer, a hidden layer, and an output layer; the hidden layer is divided into upsampling and downsampling parts; the downsampling part contains 4 of the improved residual modules and 4 max pooling layers; the upsampling part contains 4 deconvolution layers and 4 residual convolution modules; the upsampling and downsampling parts are connected by 4 of the channel attention mechanism modules. The model optimization module is used to optimize the U-Net model using the weighted cross-entropy function and train the optimized U-Net model based on the tomographic data to obtain the final model. The module includes: inputting the training set into the U-Net model, using gradient backpropagation, and using the weighted cross-entropy as the loss function to measure the distance between the true value and the predicted value. The formula for the weighted cross-entropy is: ； wherein is the ratio of fault data over the entire data set; is the attention parameter; is the modulation coefficient; In each iteration of the dataset, an adaptive algorithm is used to adjust the weights between neurons to reduce the loss function, thereby optimizing the network. The performance of the U-Net model is qualitatively judged by accuracy, recall, and F1 score. The parameters of the network model are saved initially. The iteration is repeated until the requirements for tomography are met and the final model is saved. The fault identification module is used to input the original seismic profile data into the final model to obtain the fault identification results.

8. An electronic device, comprising: It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the intelligent tomography method based on the improved U-Net network as described in any one of claims 1-6.

9. A computer readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent tomography method based on the improved U-Net network as described in any one of claims 1-6.

Images