A three-dimensional seismic phase intelligent identification method and system based on V-Net

Abstract

The present invention belongs to but is not limited to the technical field of intelligent identification of seismic phases, and in particular, relates to a three-dimensional seismic phase intelligent identification method and system based on V‑Net, including: S1: acquisition of three-dimensional seismic data and corresponding seismic phase interpretation results; S2: establishment of learning samples for three-dimensional seismic phase identification; S3: construction of a seismic phase intelligent identification model based on V‑Net; S4: three-dimensional seismic phase identification of unknown data. The present invention regards seismic phase identification as a three-dimensional image segmentation problem, and at the same time adds residuals to the modeling, optimizes the rationality of the model's assumption space, and can effectively improve the accuracy of intelligent identification of seismic phases. The present invention adds a residual layer on the basis of three-dimensional U‑Net modeling, optimizes the structure of the intelligent model, helps to ensure the rationality of the model's assumption space, and thus improves the overall feature learning ability of the network.

Description

Technical Field

[0001] The present invention belongs to but is not limited to the technical field of intelligent identification of seismic phases, and in particular relates to a three-dimensional seismic phase intelligent identification method and system based on V-Net. Background Art

[0002] Seismic phases are the reflection of sedimentary phases in seismic data. Analysis and identification of seismic phases can provide an effective basis for the exploration and development of underground resources, especially oil and gas resources. In recent years, oil and gas artificial intelligence research has developed rapidly. Domestic and foreign scholars have proposed a variety of methods for intelligent identification of seismic phases, among which convolutional neural networks have shown excellent performance. The subsequent fully convolutional neural network replaces the fully connected layer in the traditional convolutional neural network with a convolutional layer, retains the spatial information of the original input, and can even achieve pixel-level seismic phase classification. However, most of the existing fully convolutional neural networks treat seismic phase identification as a two-dimensional image segmentation problem, ignoring the continuity of seismic phases in three-dimensional space. This leads to the shortcomings of existing seismic phase intelligent identification methods such as low prediction accuracy and unclear seismic phase boundaries.

[0003] Although Chinese patent CN113805235A and Chinese patent CN115128671A respectively disclose a three-dimensional seismic phase identification method based on deep learning technology, the former's model structure contains a fully connected layer and cannot achieve pixel-level seismic phase prediction, while the latter adopts the traditional U-Net structure.

[0004] In view of the above analysis, the technical problems that urgently need to be solved in the existing technology are: the existing intelligent identification methods of seismic phases have shortcomings such as low prediction accuracy and unclear seismic phase boundaries. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a three-dimensional seismic phase intelligent identification method and system based on V-Net, which adds a residual layer on the basis of three-dimensional U-Net modeling, optimizes the structure of the intelligent model, helps to ensure the rationality of the model's assumption space, and thus improves the overall feature learning ability of the network. Treating seismic phase identification as a three-dimensional image segmentation problem and adding residuals to the modeling can effectively solve the problems of inaccurate boundary identification and insufficient noise resistance of existing seismic phase intelligent identification methods, and can provide more accurate guidance for the subsequent description of reservoir heterogeneity and reservoir space.

[0006] The present invention is implemented as follows: a three-dimensional seismic phase intelligent identification method based on V-Net, comprising:

[0007] S1: Acquisition of 3D seismic data and corresponding seismic phase interpretation results.

[0008] S2: Establishment of learning samples for 3D seismic phase identification.

[0009] S3: Construction of seismic phase intelligent identification model based on V-Net.

[0010] S4: 3D seismic phase identification of unknown data.

[0011] Furthermore, the acquisition of the three-dimensional seismic data and the corresponding seismic phase interpretation results in S1 specifically includes the following steps:

[0012] S11, collecting three-dimensional post-stack seismic data covering the reservoir to be predicted, and performing denoising processing on the data to improve the signal-to-noise ratio of the seismic data.

[0013] S12, normalize the denoised seismic data to [-1, 1].

[0014] S13, performing seismic phase interpretation on the seismic data to obtain corresponding seismic phase labels.

[0015] Furthermore, the establishment of learning samples for 3D seismic phase identification in S2 specifically includes the following steps:

[0016] S21, determine the size (L, W, H) of the three-dimensional interception window and the step length STRIDE of the window movement.

[0017] S22, overlap the left front vertex of the three-dimensional interception window with the left front vertex of the three-dimensional seismic data body, then slide the interception window from left to right and from top to bottom, and save the seismic data in the interception window during each sliding process as a sample until the entire seismic data is covered.

[0018] S23, repeating the process similar to S21-S22 for the three-dimensional seismic phase label data to obtain a seismic phase label data block corresponding to the seismic data sample.

[0019] S24, performing one-hot encoding on the seismic phase label data block obtained in S23 according to the number of seismic phases.

[0020] S25, matching the seismic data sample set and the corresponding seismic phase label data set to form a learning sample for V-Net modeling.

[0021] Furthermore, the construction of the V-Net-based seismic phase intelligent identification model in S3 specifically includes the following steps:

[0022] S31, input three-dimensional seismic data blocks of size (L, W, H) into an input module and two compression modules in batches, where the input module consists of a three-dimensional convolution with a size of (k1, k1, k1), a step size of 1, and a channel number of c, and a three-dimensional convolution with a size of (k2, k2, k2) and a step size of 2; the two compression modules consist of two three-dimensional convolutions with a size of (k1, k1, k1) and a step size of 1, and one three-dimensional convolution with a size of (k2, k2, k2) and a step size of 2, and the number of channels of each three-dimensional convolution is doubled module by module.

[0023] S32, inputting the output of the last compression module into two decompression modules and an output module in sequence, wherein the decompression module is composed of a three-dimensional transposed convolution and two to three three-dimensional convolutions of size (k1, k1, k1); the output module structure is similar to the decompression module structure, except that its convolution kernel size is (1, 1, 1) and the activation function is softmax.

[0024] It should be noted that the three-dimensional convolution with a stride of 2 is used to downsample its input, and the three-dimensional transposed convolution is used to upsample the input; except for the output module, the activation function after other convolutions is PreLU.

[0025] S33, in the downsampling and upsampling layers of each module, residual operations are performed on the features before and after sampling for input to the lower layer; and residual operations are also performed on the head end of each decompression module and the tail end of each compression module.

[0026] Furthermore, the 3D seismic phase identification of unknown data in S4 specifically includes the following steps:

[0027] S41, obtain three-dimensional post-stack seismic data unknown to V-Net.

[0028] S42, performing denoising and normalization processing on the seismic data as described in S11-S12.

[0029] S43, determining a suitable STRIDE (which may be different from that in S21), and performing three-dimensional interception on the seismic data according to the interception window size determined in S21, to obtain a three-dimensional test data set T.

[0030] S44, input T into the trained V-Net in batches to obtain prediction results.

[0031] S45, splicing the prediction result blocks obtained in S44 to obtain the overall prediction result.

[0032] Another object of the present invention is to provide a V-Net-based three-dimensional seismic phase intelligent identification system for realizing the V-Net-based three-dimensional seismic phase intelligent identification method, comprising:

[0033] Seismic data acquisition module: acquisition of three-dimensional seismic data and corresponding seismic phase interpretation results.

[0034] Learning sample establishment module: establishment of learning samples for three-dimensional seismic phase identification.

[0035] Intelligent identification model building module: Building a seismic phase intelligent identification model based on V-Net.

[0036] 3D seismic phase identification module: used for 3D seismic phase identification of unknown data.

[0037] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the V-Net-based three-dimensional seismic phase intelligent identification method.

[0038] Another object of the present invention is to provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the V-Net-based three-dimensional seismic phase intelligent identification method.

[0039] Another object of the present invention is to provide an information data processing terminal, which includes the three-dimensional seismic phase intelligent identification system based on V-Net.

[0040] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0041] The present invention regards seismic phase identification as a three-dimensional image segmentation problem, and simultaneously adds residuals into the modeling, thereby optimizing the rationality of the model spanning hypothesis space and effectively improving the accuracy of intelligent seismic phase identification.

[0042] The V-Net-based 3D seismic phase intelligent identification method of the present invention solves several key problems in seismic phase identification in the prior art through innovative model structure and sample processing flow, and achieves significant technical progress, as follows:

[0043] 1) Improve the accuracy of 3D seismic phase identification.

[0044] Traditional seismic phase identification methods mostly use 2D convolution or shallow neural networks, which are difficult to fully capture the spatial characteristics of three-dimensional seismic data, resulting in low recognition accuracy. The present invention adopts a V-Net network structure to extract the spatial characteristics of seismic phases through multi-layer three-dimensional convolution operations, ensuring that the model accurately recognizes three-dimensional seismic data. Compared with models such as 3D U-Net, the V-Net structure design of the present invention has a higher performance in the complexity of three-dimensional seismic phase distribution, achieving a higher accuracy of seismic phase prediction.

[0045] 2) Improve the utilization efficiency of seismic data.

[0046] The present invention processes seismic data in blocks by sliding a three-dimensional interception window, which not only effectively improves data utilization, but also captures more spatial information in the data, allowing the model to use more valid samples for training. By inputting data blocks in batches, it is ensured that the model can efficiently process large-scale three-dimensional seismic data. This data interception and block strategy improves the adaptability of the model in large-scale seismic data and meets the needs of efficient exploration.

[0047] 3) Enhance the generalization ability and stability of the model.

[0048] The present invention adopts a variety of optimization techniques such as residual connection, batch regularization and PReLU activation function, which significantly improves the generalization ability and stability of the V-Net model, reduces the risk of overfitting, and enables the model to maintain high recognition accuracy under different geological conditions. These improvements solve the problem of poor adaptability of traditional models to geological environments, making the application effect of the present invention more stable in various geological environments and able to meet the needs of seismic phase identification under complex geological conditions.

[0049] 4) Simplify the seismic phase interpretation process and improve exploration efficiency.

[0050] The present invention incorporates the intelligent identification function of seismic phases into the V-Net model, which can greatly simplify the seismic phase interpretation process, making the processing and interpretation of seismic data more efficient and automated. Compared with the traditional process that relies on manual interpretation by experts, the method of the present invention not only improves work efficiency, but also reduces the errors caused by human factors. The intelligent identification method can be applied to fields such as oil and gas exploration and mineral resource development, which helps to accelerate the process of reservoir analysis and resource assessment, and has broad industrial application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flow chart of a three-dimensional seismic phase intelligent identification method based on V-Net provided in an embodiment of the present invention.

[0052] Figure 23 is a schematic diagram of modeling data provided by an embodiment of the present invention; wherein part (a) is three-dimensional seismic data, and part (b) is seismic phase data corresponding to part (a).

[0053] Figure 3 is a schematic diagram of a learning sample provided by an embodiment of the present invention; wherein parts (a), (b), (c) and (d) are three-dimensional seismic data blocks with a size of 128×128×64, and parts (e), (f), (g) and (h) are corresponding three-dimensional seismic phase blocks.

[0054] Figure 4 1 is a schematic diagram of the structure of a V-Net established according to an embodiment of the present invention; wherein part (a), part (b), part (c) and part (d) are respectively an input module, an output module, a compression module and a decompression module in the network model.

[0055] Figure 5 Schematic diagram of learning loss curve provided by an embodiment of the present invention; (a) part of 3D U-Net and (b) part of V-Net.

[0056] Figure 6 The figure shows a comparison of seismic phase identification results of 3D U-Net and V-Net in one embodiment of the present invention; wherein part (a) is 3D U-Net and part (b) is V-Net.

[0057] Figure 7 It is a structural diagram of a three-dimensional seismic phase intelligent identification system based on V-Net provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0059] like Figure 1 As shown, the present invention proposes a three-dimensional seismic phase intelligent identification method based on V-Net, and the method comprises the following steps:

[0060] S1: Acquisition of 3D seismic data and corresponding seismic phase interpretation results.

[0061] S2: Establishment of learning samples for 3D seismic phase identification.

[0062] S3: Construction of seismic phase intelligent identification model based on V-Net.

[0063] S4: 3D seismic phase identification of unknown data.

[0064] In a specific implementation, the acquisition of three-dimensional seismic data and corresponding seismic phase interpretation results specifically includes the following steps:

[0065] S11, collecting three-dimensional post-stack seismic data covering the reservoir to be predicted, and performing denoising processing on the data to improve the signal-to-noise ratio of the seismic data.

[0066] S12, normalize the denoised seismic data to [-1, 1].

[0067] S13, performing seismic phase interpretation on the seismic data to obtain corresponding seismic phase labels.

[0068] like Figure 2 Shown are three-dimensional seismic data and corresponding seismic phases used in one embodiment of the present invention.

[0069] In a specific implementation, the establishment of learning samples for three-dimensional seismic phase identification specifically includes the following steps:

[0070] S21, determining the size of the three-dimensional interception window and the step length of the window movement to be (128, 128, 64) and (60, 60, 30) respectively.

[0071] S22, overlap the left front vertex of the three-dimensional interception window with the left front vertex of the three-dimensional seismic data body, then slide the interception window from left to right and from top to bottom, and save the seismic data in the interception window during each sliding process as a sample until the entire seismic data is covered.

[0072] S23, repeating the process similar to S21-S22 for the three-dimensional seismic phase label data to obtain a seismic phase label data block corresponding to the seismic data sample.

[0073] S24, performing one-hot encoding on the seismic phase label data block obtained in S23 according to the number of seismic phases.

[0074] Since the seismic phase data in this embodiment is 6, the one-hot encoding of the seismic phase data block encoded as 1 is [1, 0, 0, 0, 0, 0], and other encodings are similar.

[0075] S25, matching the seismic data sample set and the corresponding seismic phase label data set to form a learning sample for V-Net modeling.

[0076] like Figure 3 Shown are four learning samples for V-Net training obtained according to the above steps in one embodiment of the present invention.

[0077] In the specific implementation, the construction of the V-Net-based seismic phase intelligent identification model specifically includes the following steps:

[0078] S31, input three-dimensional seismic data blocks of size (128, 128, 64) into an input module and two compression modules in batches, where the input module consists of a three-dimensional convolution with a size of (3, 3, 3), a step size of 1, and 16 channels, and a three-dimensional convolution with a size of (2, 2, 2) and a step size of 2; the remaining two compression modules consist of two three-dimensional convolutions with a size of (3, 3, 3) and a step size of 1, and one three-dimensional convolution with a size of (2, 2, 2) and a step size of 2, and the number of channels of each three-dimensional convolution is doubled module by module.

[0079] S32, inputting the output of the last compression module into two decompression modules and an output module in sequence, wherein the decompression module is composed of a three-dimensional transposed convolution and two to three three-dimensional convolutions of size (3, 3, 3); the output module structure is similar to the decompression module structure, except that its convolution kernel size is (1, 1, 1) and the activation function is softmax.

[0080] It should be noted that the three-dimensional convolution with a stride of 2 is used to downsample its input, and the three-dimensional transposed convolution is used to upsample the input; except for the output module, the activation function after other convolutions is PreLU.

[0081] S33, in the downsampling and upsampling layers of each module, residual operations are performed on the features before and after sampling for input to the lower layer; and residual operations are also performed on the head end of each decompression module and the tail end of each compression module.

[0082] In order to prevent the model from overfitting, a batch normalization layer is added after each 3D convolutional layer in this embodiment.

[0083] like Figure 4 The figure shows a V-Net established in one embodiment of the present invention.

[0084] In a specific implementation, the 3D seismic phase identification of unknown data specifically includes the following steps:

[0085] S41, obtain three-dimensional post-stack seismic data unknown to V-Net.

[0086] S42, performing denoising and normalization processing on the seismic data as described in S11-S12.

[0087] S43, determining STRIDE to be (120, 120, 60), and performing three-dimensional interception on the seismic data according to the interception window size determined in S21, to obtain a three-dimensional test data set T.

[0088] S44, input T into the trained V-Net in batches to obtain prediction results.

[0089] S45, splicing the prediction result blocks obtained in S44 to obtain the overall prediction result.

[0090] like Figure 5 3D U-Net in one embodiment of the present invention Figure 5 Part (a) and V-Net Figure 5 Part (b) is a diagram of the loss curve during the learning process. During the training process, 1 / 3 of the data is set as verification data, and the performance of the model on the verification data is monitored. When the loss value of the model on the verification data does not change after three iterations, the model training is stopped and the model is saved.

[0091] like Figure 6 FIG. 4 is a comparison of seismic phase identification results of 3D U-Net and V-Net in one embodiment of the present invention, wherein Figure 6 Part (a) is 3D U-Net. Figure 6 Part (b) is V-Net. The comparison results show that the proposed method has better three-dimensional seismic phase recognition ability than 3d U-Net.

[0092] The V-Net-based 3D seismic phase intelligent identification method proposed in this invention realizes high-precision identification of seismic phases through multi-step preprocessing, sample construction, model training and prediction. The specific working principle is as follows:

[0093] 1) Acquisition of 3D seismic data and seismic phase labels.

[0094] First, 3D post-stack seismic data is collected from the reservoir area to be identified, and the data is denoised to improve the signal-to-noise ratio and ensure the quality of the seismic data. Next, the denoised data is normalized to the interval [-1, 1] to ensure the consistency of the model input. Then, geological experts interpret the data for seismic phases, obtain seismic phase labels, and use them for subsequent model training. The acquisition of this data and labels provides the basis for intelligent identification of seismic phases.

[0095] 2) Construction of 3D seismic phase learning samples.

[0096] In 3D seismic phase identification, the 3D seismic data volume is gradually cut into multiple sample data blocks by specifying the interception window (size is 128×128×64) and the moving step size (60×60×30) to ensure that the entire seismic data area is covered. For each intercepted sample data block, the corresponding seismic phase label data block is extracted and one-hot encoded (for example, the seismic phase is encoded as [1, 0, 0, 0, 0, 0] to represent one of the 6 types of seismic phases). After pairing these sample data blocks and their labels, they are constructed into the training data set of the model to ensure that the model can learn based on abundant samples.

[0097] 3) Construction of intelligent recognition model based on V-Net.

[0098] The constructed samples are input into the V-Net model. The V-Net structure includes input module, compression module, decompression module and output module. The input module receives the three-dimensional data block, performs convolution and downsampling, and gradually compresses the data to a lower dimension. In the decompression module, the data is gradually restored through three-dimensional transposed convolution and upsampling operations, and the output module uses the softmax activation function to generate the seismic phase prediction probability distribution of each data block. The residual connection in the V-Net structure enables the feature information in the compression and decompression processes to be correlated with each other, enhancing the generalization ability of the model.

[0099] 4) Multi-layer convolution and feature extraction.

[0100] A batch normalization layer is added after each convolutional layer of the V-Net model, and the PReLU activation function is used to reduce the possibility of overfitting and improve the stability of the model. The downsampling layer is used to extract deep features, enabling the model to capture higher-level seismic phase features in space, while the decompression module restores the features to the original data scale to ensure that the output results can retain sufficient spatial details. Each convolution operation further expands the model's feature representation capabilities, thereby enhancing the accuracy of seismic phase identification.

[0101] 5) Model training and verification.

[0102] During the training process, 1 / 3 of the training data is used as validation data to monitor the model performance. When the loss value of the model on the validation data set does not change after three iterations, the training is stopped and the optimal model parameters are saved. The loss curve is recorded during the training process to visually observe the convergence of the model. Compared with the traditional 3D U-Net, the V-Net structure of the present invention is more suitable for the complex feature recognition of three-dimensional seismic phases and has higher stability and accuracy.

[0103] 6) Seismic phase identification of unknown seismic data.

[0104] In the recognition stage, the unknown 3D seismic data is first preprocessed (denoising and normalization). The seismic data is sliced ​​using the same window and step size as in training to form a 3D test data set. The test data is input into the trained V-Net model in batches to obtain local seismic phase recognition results. Finally, these local prediction results are spliced ​​and restored to the overall 3D seismic phase recognition map, and high-precision 3D seismic phase distribution results are output to provide support for reservoir analysis and geological exploration.

[0105] Through the above steps, the present invention achieves high-precision identification of three-dimensional seismic phases while maintaining computational efficiency. Verification results show that compared with 3D U-Net, the method of the present invention performs better in identifying complex seismic phase structures.

[0106] like Figure 7 As shown in the figure, the V-Net-based 3D seismic phase intelligent identification system includes:

[0107] Seismic data acquisition module: acquisition of three-dimensional seismic data and corresponding seismic phase interpretation results.

[0108] Learning sample establishment module: establishment of learning samples for three-dimensional seismic phase identification.

[0109] Intelligent identification model building module: Building a seismic phase intelligent identification model based on V-Net.

[0110] 3D seismic phase identification module: used for 3D seismic phase identification of unknown data.

[0111] An application embodiment of the present invention provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of a three-dimensional seismic phase intelligent identification method based on V-Net.

[0112] An application embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of a three-dimensional seismic phase intelligent identification method based on V-Net.

[0113] An application embodiment of the present invention provides an information data processing terminal, which includes a three-dimensional seismic phase intelligent identification system based on V-Net.

[0114] The seismic data acquisition module is responsible for collecting 3D seismic data and corresponding seismic phase interpretation results. These data are usually obtained through seismic exploration equipment or seismic simulation software. The module preprocesses the raw seismic data, including denoising, data format conversion and coordinate alignment, to ensure the integrity and consistency of the data. The interpretation results are annotated by expert geologists based on experience or existing analysis tools and used as label data in the training phase to provide guidance for the model to learn seismic phase characteristics.

[0115] The learning sample establishment module pairs seismic data with the corresponding seismic phase interpretation results to generate high-quality training samples. The module first divides the three-dimensional seismic data into multiple small-volume data blocks (Patch) to ensure that the model can handle local geological features. Then, for each data block, its label is mapped to the seismic phase category to form an input data and labeled sample pair. In order to improve the generalization ability of the model, the module also performs data enhancement operations on the samples, such as random rotation, flipping, and noise addition.

[0116] The intelligent identification model building module builds an intelligent seismic phase identification model based on the V-Net architecture. V-Net is a three-dimensional convolutional neural network that is particularly suitable for three-dimensional data segmentation tasks. The module first initializes the network weights and defines the network structure, including the encoder and decoder. The encoder extracts the high-dimensional features of the seismic data through multiple layers of three-dimensional convolution, and the decoder gradually restores the high-dimensional features to the input data size and generates seismic phase prediction results. After the model is built, the module uses learning samples to train the network and optimize the network parameters to minimize the error between the prediction results and the true labels.

[0117] The 3D seismic phase recognition module is responsible for seismic phase classification of unknown data. The module inputs new 3D seismic data into the trained V-Net model, which predicts the seismic phase category of each data point based on the spatial characteristics of the data. The decoder outputs a 3D seismic phase classification result of the same size as the input data. The module also performs post-processing on the recognition results, such as smoothing and regional consistency checks, to ensure the coherence and accuracy of the output seismic phase distribution.

[0118] The computer device of the present invention realizes the three-dimensional seismic phase intelligent identification process based on V-Net by executing the program stored in the memory. The processor is responsible for calling the seismic data acquisition module, the learning sample establishment module, the intelligent identification model building module and the three-dimensional seismic phase identification module to complete the reading of seismic data, the training sample generation, the model training and the automatic identification of unknown seismic data. During the execution process, the processor adjusts the training parameters according to the dynamic tuning results (such as the learning rate and the convergence of the loss function) to ensure the continuous optimization of the model performance.

[0119] The information data processing terminal combines hardware and software resources to provide full-process automation support from data acquisition to seismic phase prediction. The terminal runs a three-dimensional seismic phase intelligent identification system based on V-Net, loads three-dimensional seismic data through a network interface or local storage, and completes data processing, feature extraction, model training, and result output. Users can view the three-dimensional visualization results of seismic phase distribution in real time through the interactive interface, and analyze the seismic phase distribution characteristics based on the evaluation report generated by the terminal, thereby assisting geological exploration decisions.

[0120] The system of the present invention utilizes the three-dimensional convolution characteristics of the V-Net network architecture to achieve high-precision automatic segmentation of three-dimensional seismic data. The modules work together to make the intelligent identification of seismic phases more efficient and accurate through data acquisition, sample generation, model training and prediction, significantly improving the work efficiency and decision-making quality of geological exploration.

[0121] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0122] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A three-dimensional seismic phase intelligent identification method based on V-Net, characterized in that: include: S1: Acquisition of 3D seismic data and corresponding seismic phase interpretation results; S2: Establishment of learning samples for 3D seismic phase identification; S3: Construction of seismic phase intelligent identification model based on V-Net; S4: 3D seismic phase identification of unknown data; The establishment of learning samples for 3D seismic phase identification in S2 specifically includes the following steps: S21, determining the size of the three-dimensional interception window and the step length STRIDE of the window movement; S22, overlapping the left front vertex of the three-dimensional interception window with the left front vertex of the three-dimensional seismic data volume, then sliding the interception window from left to right and from top to bottom, and saving the seismic data in the interception window during each sliding process as a sample until the entire three-dimensional seismic data volume is covered; S23, repeating the process of S21-S22 for the three-dimensional seismic phase label data to obtain a seismic phase label data block corresponding to the seismic data sample; S24, performing one-hot encoding on the seismic phase label data block obtained in S23 according to the number of seismic phases; S25, matching the seismic data sample set and the corresponding seismic phase label data set to form a learning sample for V-Net modeling.

2. The V-Net-based 3D seismic phase intelligent identification method according to claim 1, characterized in that: The acquisition of 3D seismic data and corresponding seismic phase interpretation results in S1 specifically includes the following steps: S11, collecting three-dimensional post-stack seismic data covering the reservoir to be predicted, and performing denoising on the data to improve the signal-to-noise ratio of the seismic data; S12, normalize the denoised seismic data to [-1, 1]; S13, performing seismic phase interpretation on the normalized seismic data to obtain corresponding seismic phase labels.

3. The V-Net-based 3D seismic phase intelligent identification method according to claim 1, characterized in that: The construction of the V-Net-based seismic phase intelligent identification model in S3 specifically includes the following steps: S31, inputting a three-dimensional seismic data volume of size (L, W, H) into an input module and two compression modules in batches, wherein the input module consists of a three-dimensional convolution with a size of (k1, k1, k1), a step size of 1, and a channel number of c, and a three-dimensional convolution with a size of (k2, k2, k2) and a step size of 2; the two compression modules each consist of two three-dimensional convolutions with a size of (k1, k1, k1) and a step size of 1, and one three-dimensional convolution with a size of (k2, k2, k2) and a step size of 2, and the number of channels of each three-dimensional convolution is doubled module by module; S32, inputting the output of the last compression module into two decompression modules and an output module in sequence, wherein the decompression module is composed of a three-dimensional transposed convolution and two to three three-dimensional convolutions of size (k1, k1, k1); the output module structure is similar to the decompression module structure, except that its convolution kernel size is (1, 1, 1) and the activation function is softmax; The 3D convolution with a stride of 2 is used to downsample its input, and the 3D transposed convolution is used to upsample the input. Except for the output module, the activation function after other convolutions is PReLU. S33, in the downsampling and upsampling layers of each module, residual operations are performed on the features before and after sampling for input to the lower layer; and residual operations are also performed on the head end of each decompression module and the tail end of each compression module.

4. The V-Net-based 3D seismic phase intelligent identification method according to claim 2, characterized in that: The 3D seismic phase identification of unknown data in S4 specifically includes the following steps: S41, obtaining unknown three-dimensional post-stack seismic data of V-Net; S42, performing denoising and normalization processing on the three-dimensional post-stack seismic data as described in S11-S12; S43, determining a suitable STRIDE, and performing three-dimensional interception on the normalized seismic data according to the interception window size determined in S21, to obtain a three-dimensional test data set T; S44, input T into the trained V-Net in batches to obtain prediction results; S45, splicing the prediction result blocks obtained in S44 to obtain the overall prediction result.

5. A V-Net-based 3D seismic phase intelligent identification system for implementing the V-Net-based 3D seismic phase intelligent identification method according to any one of claims 1 to 4, characterized in that: include: Seismic data acquisition module: acquisition of three-dimensional seismic data and corresponding seismic phase interpretation results; Learning sample establishment module: establishment of learning samples for 3D seismic phase identification; Intelligent identification model building module: Building of seismic phase intelligent identification model based on V-Net; 3D seismic phase identification module: used for 3D seismic phase identification of unknown data.

6. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the V-Net-based three-dimensional seismic phase intelligent identification method as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the V-Net-based three-dimensional seismic phase intelligent identification method as described in any one of claims 1 to 4.

8. An information data processing terminal, comprising the V-Net-based three-dimensional seismic phase intelligent identification system as claimed in claim 5.

Images