Active source seismic exploration background noise suppression method and device based on deep learning

Abstract

The invention provides an active source seismic exploration background noise suppression method and device based on deep learning, and is applied to the technical field of geoscience exploration seismic data processing. The method comprises the following steps: acquiring background noise data and noisy seismic data of a target exploration area; performing fusion processing on the background noise data of the target exploration area and the seismic forward modeling data to obtain a training sample set; training a neural network model by adopting the training sample set to obtain a background noise suppression model; and inputting the noisy seismic data into the background noise suppression model to obtain a noise suppression result of the noisy seismic data, thereby improving the processing efficiency and reducing errors.

Description

Technical Field

[0001] The present application belongs to the technical field of seismic data processing for geological exploration, and in particular, relates to a method and device for suppressing background noise in active source seismic exploration based on deep learning. Background Art

[0002] Seismic data can reveal underground rock structure and fluid distribution, predict earthquake risks, provide a basis for energy exploration, and promote the development of the energy industry. Seismic data acquisition is inevitably interfered by background noise, which comes from the natural environment and human activities. The noise energy is strong and masks the characteristics of seismic signals. Therefore, effectively suppressing background noise is crucial to improving the quality of seismic exploration data.

[0003] In the prior art, the main methods for suppressing background noise in seismic exploration are superposition method, filter method, sparse transformation method, and matrix rank reduction method. The superposition method suppresses random noise by summing the event axis signal, but it is only suitable for random noise suppression. The filter method uses different types of filters to remove noise in specific frequency bands, but has limited effect in the face of strong aliasing noise. The sparse transformation method uses the difference in signal-to-noise characteristics to separate noise, and it is difficult to suppress coherent noise similar to the signal. The matrix rank reduction method is based on the difference in signal and noise ranks, and denoises by reducing the data rank number, but is limited by the event axis curvature and rank number setting.

[0004] However, the existing seismic exploration background noise suppression methods mostly use artificially simulated noise to produce neural network training data sets. However, the complex and changeable background noise is difficult to be effectively simulated, which makes it difficult to obtain high-precision and high-fidelity seismic exploration background noise denoising results. Summary of the invention

[0005] The present application provides a method, device, and equipment for suppressing background noise in active source seismic exploration based on deep learning, so as to solve the defects of low efficiency and large error of the background noise suppression method in seismic exploration in the prior art.

[0006] In a first aspect, the present application provides a method for suppressing background noise in active source seismic exploration based on deep learning, the method comprising:

[0007] Obtain background noise data and noisy seismic data in the target exploration area;

[0008] The background noise data of the target exploration area is fused with the seismic forward modeling data to obtain a training sample set, wherein the seismic forward modeling data is computer-generated data of seismic waves propagating in underground media;

[0009] Using the training sample set to train a neural network model to obtain a background noise suppression model, wherein the neural network model is used to determine the correlation between the background noise data and the seismic forward simulation data;

[0010] The noisy seismic data is input into the background noise suppression model to obtain the noise suppression result of the noisy seismic data.

[0011] Optionally, the step of obtaining background noise data and noisy seismic data in the target exploration area includes:

[0012] Acquire first sensor data of a target exploration area within a first time period before source excitation, and determine the background noise data based on the first sensor data;

[0013] Second sensor data of the target exploration area within a first time period after the earthquake source is excited is acquired, and the noisy seismic data is determined based on the second sensor data.

[0014] Optionally, the fusing of the background noise data of the target exploration area with the seismic forward modeling data to obtain a training sample set includes:

[0015] According to the geographical structure and physical characteristics of the target exploration area, the propagation process of seismic waves in the underground medium is simulated to obtain a plurality of seismic forward modeling data, wherein the seismic forward modeling data is seismic data that does not contain background noise;

[0016] The background noise data and the seismic forward modeling data are segmented by using overlapping sliding windows to obtain a plurality of target noise blocks corresponding to the background noise data and a plurality of target signal blocks corresponding to the seismic forward modeling data, wherein the lengths of the noise blocks and the signal blocks are consistent;

[0017] The multiple noise blocks and the multiple signal blocks are randomly combined to obtain a training sample set.

[0018] Optionally, the randomly combining the multiple noise blocks and the multiple signal blocks to obtain a training sample set includes:

[0019] Randomly pairing any one of the multiple noise blocks with any one of the multiple signal blocks;

[0020] Superimposing the paired noise blocks onto the corresponding signal blocks to obtain noisy signal blocks;

[0021] A signal block corresponding to each of the noisy signal blocks is used as a label of the noisy signal block to obtain a training sample set.

[0022] Optionally, the training of the neural network model to obtain a background noise suppression model includes:

[0023] Determining a loss function of the neural network model, wherein the loss function indicates a difference between output data predicted by the neural network model and actual data;

[0024] When it is determined that the loss function converges, the neural network model is controlled to stop iterative processing, and it is determined that the background noise suppression model training is completed.

[0025] Optionally, the step of inputting the noisy seismic data into the background noise suppression model to obtain a noise suppression result of the noisy seismic data includes:

[0026] Inputting the noisy seismic data into the background noise suppression model;

[0027] The background noise suppression model determines the seismic signal data corresponding to the background noise data based on the learned association relationship between the background noise data and the seismic forward simulation data;

[0028] Based on the seismic signal data, the background noise data is extracted and processed to obtain a noise suppression result of the noisy seismic data, and the noise suppression result is used to indicate a target seismic signal contained in the noisy seismic data.

[0029] In a second aspect, the present application provides a device for suppressing background noise in active source seismic exploration based on deep learning, the device comprising:

[0030] An acquisition module is used to acquire background noise data and noisy seismic data in the target exploration area;

[0031] A processing module, used for fusing the background noise data of the target exploration area with the seismic forward modeling data to obtain a training sample set, wherein the seismic forward modeling data is data generated by computer simulation of seismic waves propagating in underground media;

[0032] The processing module is further used to train a neural network model using the training sample set to obtain a background noise suppression model, wherein the neural network model is used to determine the correlation between the background noise data and the seismic forward simulation data;

[0033] The processing module is further used to input the noisy seismic data into the background noise suppression model to obtain a noise suppression result of the noisy seismic data.

[0034] Optionally, the active source seismic exploration background noise suppression device based on deep learning further includes: a determination module;

[0035] The acquisition module is further used to acquire first sensor data of the target exploration area within a first time period before the source is excited;

[0036] The determination module is used to determine the background noise data according to the first sensor data;

[0037] The acquisition module is further used to acquire second sensor data of the target exploration area within a first time period after the earthquake source is excited;

[0038] The determination module is further used to determine the noisy seismic data based on the second sensor data.

[0039] Optionally, the processing module is further used to simulate the propagation process of seismic waves in the underground medium according to the geographical structure and physical characteristics of the target exploration area to obtain a plurality of seismic forward simulation data, wherein the seismic forward simulation data is seismic data that does not contain background noise;

[0040] The processing module is further used to segment the background noise data and the seismic forward modeling data using overlapping sliding windows to obtain a plurality of target noise blocks corresponding to the background noise data and a plurality of target signal blocks corresponding to the seismic forward modeling data, wherein the lengths of the noise blocks and the signal blocks are consistent;

[0041] The processing module is further used to randomly combine the multiple noise blocks and the multiple signal blocks to obtain a training sample set.

[0042] Optionally, the processing module is further used to randomly pair any one of the multiple noise blocks with any one of the multiple signal blocks;

[0043] The processing module is further used to superimpose the paired noise blocks onto the corresponding signal blocks to obtain noisy signal blocks;

[0044] The processing module is further configured to use a signal block corresponding to each of the noisy signal blocks as a label of the noisy signal block to obtain a training sample set.

[0045] Optionally, the determination module is further used to determine a loss function of the neural network model, wherein the loss function indicates a difference between output data predicted by the neural network model and actual data;

[0046] The determination module is also used to control the neural network model to stop iterative processing when it is determined that the loss function has converged, and to determine that the background noise suppression model training is completed.

[0047] Optionally, the processing module is further used to input the noisy seismic data into the background noise suppression model;

[0048] The determination module is further used for the background noise suppression model to determine the seismic signal data corresponding to the background noise data based on the learned association relationship between the background noise data and the seismic forward simulation data;

[0049] The processing module is also used to extract and process the background noise data based on the seismic signal data to obtain a noise suppression result of the noisy seismic data, and the noise suppression result is used to indicate a target seismic signal contained in the noisy seismic data.

[0050] In a third aspect, the present application provides an active source seismic exploration background noise suppression device based on deep learning, comprising:

[0051] Memory;

[0052] processor;

[0053] Wherein, the memory stores computer-executable instructions;

[0054] The processor executes the computer-executable instructions stored in the memory to implement the active source seismic exploration background noise suppression method based on deep learning as described in the above-mentioned first aspect and various possible implementation methods of the first aspect.

[0055] In a fourth aspect, the present application provides a computer storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the deep learning-based active source seismic exploration background noise suppression method as described in the first aspect and various possible implementation methods of the first aspect.

[0056] The present application provides a method and device for suppressing background noise in active source seismic exploration based on deep learning. The method obtains background noise data and noisy seismic data in the target exploration area; fuses the background noise data in the target exploration area with the seismic forward simulation data to obtain a training sample set; uses the training sample set to train a neural network model to obtain a background noise suppression model; inputs the noisy seismic data into the background noise suppression model to obtain the noise suppression result of the noisy seismic data, thereby improving processing efficiency and reducing errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0058] Figure 1 Schematic diagram of the process of background noise suppression method for active source seismic exploration based on deep learning provided in this application Figure 1 ;

[0059] Figure 2Schematic diagram of the process of background noise suppression method for active source seismic exploration based on deep learning provided in this application Figure 2 ;

[0060] Figure 3 A schematic diagram of the structure of the active source seismic exploration background noise suppression device based on deep learning provided in this application;

[0061] Figure 4 A schematic diagram of the structure of the active source seismic exploration background noise suppression device based on deep learning provided in this application.

[0062] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of this application clearer, the technical solutions in this application will be clearly and completely described below in conjunction with the drawings in this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0064] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein.

[0065] In the embodiments of the present application, the words "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0066] Seismic data is the key to gaining insight into underground rock structures and fluids. They can warn of potential earthquake dangers, assist in the exploration of energy resources, and promote progress in the energy field. However, in the process of seismic data acquisition, it is inevitable to encounter interference from background noise. This noise comes from changes in the natural environment and human activities, and its powerful energy often obscures the key features of seismic signals. Therefore, effectively suppressing background noise is extremely important for improving the quality of seismic exploration data.

[0067] In the prior art, a variety of technologies are used in the field of seismic data processing to reduce noise interference and improve data quality. The main methods for suppressing background noise in seismic exploration are superposition method, filter method, sparse transformation method, and matrix rank reduction method. The superposition method suppresses noise by summing the event axis, which is suitable for suppressing random noise, but can only obtain post-stack data. The filter method uses bandpass filters, low-pass filters, and high-pass filters to remove noise in a specific frequency band, but the suppression effect on strong aliasing noise is limited. The sparse transformation method separates the signal and noise according to the difference in the characteristics of the signal and noise in a specific domain, but it is difficult to effectively suppress coherent noise that is very similar to the signal. The matrix rank reduction method is based on the linear event axis assumption and the signal and noise rank difference assumption. It suppresses noise by reducing the rank of noisy data, but its denoising effect is limited by the event axis curvature and rank setting.

[0068] However, the existing seismic exploration background noise suppression methods usually rely on artificially generated noise data to train neural networks. However, the background noise in the real world is extremely complex and varied, which makes it difficult for the simulated noise to accurately reflect the real situation. Therefore, the existing technology has limitations in achieving high-precision and high-fidelity denoising effects, and it is difficult to achieve ideal seismic exploration data quality.

[0069] In response to the above problems, the present application proposes a background noise suppression method for active source seismic exploration based on deep learning. The method obtains background noise data and noisy seismic data in the target exploration area; fuses the background noise data of the target exploration area with the seismic forward simulation data to obtain a training sample set; uses the training sample set to train a neural network model to obtain a background noise suppression model; inputs the noisy seismic data into the background noise suppression model to obtain the noise suppression result of the noisy seismic data, thereby improving processing efficiency and reducing errors.

[0070] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0071] Figure 1 A process diagram of a method for suppressing background noise in active source seismic exploration based on deep learning provided in an embodiment of the application Figure 1 .like Figure 1 As shown, the active source seismic exploration background noise suppression method based on deep learning provided in this embodiment includes:

[0072] S101. Obtain background noise data and noisy seismic data in a target exploration area.

[0073] Among them, the background noise suppression effect of seismic exploration based on deep learning depends largely on the distribution similarity between training data and actual data. Therefore, in order to enable the neural network model to learn the real background noise distribution more effectively, for active source seismic exploration, the background noise data and noisy seismic data of the target exploration area are obtained, and the real background noise data and noisy seismic data collected separately at the seismic exploration construction site are directly used to create the training data set.

[0074] One possible implementation method is to conduct on-site surveys in the seismic exploration area, understand the topography and possible noise, and deploy noise data acquisition equipment to ensure that the equipment distribution can cover the target exploration area. Before the source is excited, the noise data acquisition equipment is started to record the background noise data for a certain period of time; after the source is excited, the noisy seismic data is also recorded for a certain period of time, where the noisy seismic data includes the signal generated by the propagation of seismic waves through the underground medium and the accompanying background noise data.

[0075] Before the source is excited, recording the background noise data for a certain period of time helps to capture the noise generated by the natural environment and human activities, and provide basic data for subsequent noise suppression; after the source is excited, recording the noisy seismic data for a certain period of time ensures that the real-time seismic wave data is collected. Because the noisy seismic data contains the actual seismic signal and background noise, this enables the model obtained by subsequent training to play a better role in practical applications.

[0076] S102: fusing the background noise data of the target exploration area with the seismic forward modeling data to obtain a training sample set.

[0077] Among them, seismic forward simulation software or algorithms are used to generate a series of simulated data of seismic waves propagating in the underground medium of the target exploration area based on the geological structure parameters of the target exploration area, such as velocity model, density and other parameters. The background noise data of the target exploration area and the seismic forward simulation data are superimposed in time series, and data preprocessing is performed, such as standardization and normalization, to ensure the compatibility of the two data types. The background noise data of the target exploration area and the seismic forward simulation data are fused. The fused data are closer to the actual seismic observation data, which helps to train a model that is closer to the actual situation. Appropriate fragments are extracted from the fused data set as training samples, and each sample is labeled to obtain a training sample set.

[0078] The background noise of the target exploration area and the seismic forward simulation data are integrated to make the training sample set closer to the actual seismic exploration situation. An accurate training sample set helps to train a higher-precision noise suppression model, thereby improving the generalization ability of the model.

[0079] S103: Use the training sample set to train the neural network model to obtain a background noise suppression model.

[0080] Among them, select a neural network architecture suitable for processing time series data, such as recurrent neural networks, long short-term memory networks, or gated recurrent units. Design the model structure, including input layer, hidden layer, output layer, etc., and determine the connection method between each layer. Define the loss function, such as mean square error, cross entropy, etc., to measure the difference between the model prediction results and the actual results. Select an optimizer to update the model parameters. The time series model can effectively capture the correlation between background noise data and seismic forward simulation data.

[0081] The training set is used to train the neural network model, and the weights and biases are continuously adjusted to minimize the loss function. During the training process, the validation set is used to monitor the performance of the neural network model to avoid overfitting. According to the results of the validation set, the hyperparameters of the neural network model are adjusted, such as learning rate, batch size, etc. The changes in the loss function during the training process are monitored. When the loss function no longer decreases significantly, the training of the neural network is stopped to obtain the background noise suppression model.

[0082] Neural networks can efficiently learn key features in input data, improve training efficiency, and obtain background noise suppression models through training, which can improve the clarity of seismic signals and thus improve the accuracy of seismic background noise suppression results.

[0083] S104, inputting the noisy seismic data into a background noise suppression model to obtain a noise suppression result of the noisy seismic data.

[0084] Among them, the preprocessed noisy seismic data is input into the trained background noise suppression model. Based on the correlation between the background noise data and the seismic forward simulation data learned during training, the model automatically identifies the background noise data in the input noisy seismic data and extracts the seismic signal that filters out the background noise data, thereby obtaining the noise suppression result of the noisy seismic data.

[0085] The actual noisy seismic data is input into the background noise suppression model to obtain seismic wave data with background noise removed, which improves the clarity and accuracy of seismic signals and provides stronger support for seismic exploration and monitoring.

[0086] This embodiment proposes a background noise suppression method for active source seismic exploration based on deep learning. The method obtains background noise data and noisy seismic data in the target exploration area; fuses the background noise data in the target exploration area with the seismic forward simulation data to obtain a training sample set; uses the training sample set to train a neural network model to obtain a background noise suppression model; inputs the noisy seismic data into the background noise suppression model to obtain a noise suppression result for the noisy seismic data. Compared with the traditional method of using simulated noise data, this method can better reflect the actual situation and improve the suppression efficiency and the overall quality of seismic data.

[0087] Figure 2 This is a schematic diagram of a process of suppressing background noise in active source seismic exploration based on deep learning provided in an embodiment of the present application. Figure 2 This embodiment is in Figure 1 Based on the embodiment, the background noise suppression method of active source seismic exploration based on deep learning is described in detail. Figure 2 As shown, the active source seismic exploration background noise suppression method based on deep learning provided in this embodiment includes:

[0088] S201, acquiring first sensor data of a target exploration area within a first time period before source excitation, and determining background noise data based on the first sensor data.

[0089] Among them, the geographical location and exploration area of ​​the target exploration area are determined, the seismic sensor array is deployed to ensure that the sensors can cover the entire exploration area, and the time point of source excitation is determined to facilitate the collection of background noise data before excitation.

[0090] In one possible implementation, within a first time period before the source is excited, for example, from the second to the second, the first sensor data of the target exploration area is recorded, and the data recorded during this period does not contain any artificially excited seismic signals, but only natural background noise. The collected first sensor data is preprocessed, including but not limited to data cleaning, removal of outliers, standardization, etc., and the background noise data is determined based on the preprocessed first sensor data.

[0091] The first sensor data of the target exploration area within the first time period before the source is excited is obtained, and the background noise data is determined based on these data. The background noise data can not only truly reflect the natural noise environment of the target exploration area, but also provide high-quality training samples for subsequent noise suppression work, thereby improving the clarity of seismic signals and the accuracy of seismic exploration.

[0092] S202: Acquire second sensor data of a target exploration area within a first time period after source excitation, and determine noisy seismic data based on the second sensor data.

[0093] Among them, the geographical location and exploration area of ​​the target exploration area are determined, the seismic sensor array is deployed to ensure that the sensors can cover the entire exploration area, and the time point of source excitation is determined to facilitate the collection of background noise data before excitation.

[0094] In one possible implementation, within a first time period after the source is excited, for example, from the second to the second, the second sensor data of the target exploration area is recorded. The data recorded during this period includes seismic signals and background noise. The collected second sensor data is preprocessed, including but not limited to data cleaning, outlier removal, standardization, etc., and noisy seismic data is determined based on the preprocessed second sensor data.

[0095] The second sensor data of the target exploration area within the first time period after the source is excited is obtained, and the noisy seismic data is determined based on these data. The noisy seismic data can not only truly reflect the mixing of seismic signals and background noise in the target exploration area, but also provide high-quality training samples for subsequent noise suppression work, thereby improving the clarity of seismic signals and the accuracy of seismic exploration.

[0096] S203. According to the geographical structure and physical characteristics of the target exploration area, the propagation process of seismic waves in the underground medium is simulated to obtain a plurality of seismic forward modeling data.

[0097] Among them, according to the geographical structure and physical characteristics of the target exploration area, seismic forward simulation software is used to simulate the propagation process of seismic waves in the underground medium according to the geological model and source settings. Simulation parameters such as time step, spatial grid size, etc. are set, and multiple simulations are performed by changing the parameters to obtain diversified seismic forward simulation data. Seismic forward simulation data is clean seismic data without background noise.

[0098] By simulating the propagation process of seismic waves in the underground medium of the target exploration area, multiple seismic forward simulation data without background noise are obtained. The simulated data not only helps to build a high-quality training sample set, but also improves the training effect of the subsequent noise suppression model, thereby improving the clarity of seismic signals and the accuracy of seismic exploration.

[0099] S204, using overlapping sliding windows to segment the background noise data and the seismic forward simulation data to obtain a plurality of target noise blocks corresponding to the background noise data and a plurality of target signal blocks corresponding to the seismic forward simulation data.

[0100] Among them, determine the length of the sliding window and set the sliding step size, which is usually smaller than the window length to ensure data overlap. By setting a smaller sliding step size, overlap can be generated between adjacent data blocks, which helps to capture the continuity features in the data. Data overlap can improve data utilization and ensure that the model can learn more information from the data.

[0101] Apply a sliding window to the background noise data to split it into multiple noise blocks of the same length. Apply the same sliding window to the seismic forward simulation data to split it into multiple signal blocks of the same length. Ensuring that the lengths of the noise blocks and signal blocks are consistent helps the model learn and predict.

[0102] The background noise data and seismic forward simulation data are segmented using overlapping sliding windows to obtain multiple target noise blocks corresponding to the background noise data and multiple target signal blocks corresponding to the seismic forward simulation data. These noise blocks and signal blocks not only help to build a high-quality training sample set, but also improve the training effect of the subsequent noise suppression model, thereby improving the clarity of the seismic signal and the accuracy of seismic exploration.

[0103] S205: Randomly pair any one of the multiple noise blocks with any one of the multiple signal blocks.

[0104] Among them, a noise block is randomly selected from multiple noise blocks, and a signal block is randomly selected from multiple signal blocks, and the selected noise block is paired with the signal block. Random pairing can increase the diversity of training samples and help the model learn different types of noise and signal combinations. Repeat the above random pairing process until all noise blocks and signal blocks have been paired at least once. By repeating the pairing, more training samples can be generated, which helps to improve the generalization ability of the model.

[0105] S206: Superimpose the paired noise blocks onto the corresponding signal blocks to obtain noisy signal blocks.

[0106] S207: Use the signal block corresponding to each noisy signal block as a label of the noisy signal block to obtain a training sample set.

[0107] The paired noise blocks are superimposed on the signal blocks to generate noisy signal blocks. The superposition may be, for example, simple addition or mixing according to a certain ratio or weight. By superimposing the noise blocks on the signal blocks, the situation in which the signals are interfered by noise in the actual environment can be simulated. For each noisy signal block, its corresponding original signal block, i.e., the signal block without noise, is used as the label of the noisy signal block, and all such noisy signal blocks and their corresponding labels form a training sample set.

[0108] A training sample set for training a machine learning model is constructed by using a series of noisy signal blocks and their corresponding clean signal blocks as labels. This method helps to improve the robustness and generalization ability of the model, and can better handle signal processing tasks in various noisy environments in practical applications.

[0109] S208: Use the training sample set to train a neural network model and determine a loss function of the neural network model.

[0110] The neural network model architecture is defined, including the input layer, hidden layer, output layer, etc., and the model hyperparameters are set, such as learning rate, optimizer type, etc. A suitable loss function is selected to measure the difference between the model's predicted output data and the actual data. Common loss functions include mean square error, root mean square error, cross entropy loss, etc. Use the noisy signal blocks in the training set as input and the corresponding clean signal blocks as labels to start training the neural network model. Calculate the loss value of each batch, and use the optimization algorithm to adjust the model parameters to minimize the loss.

[0111] One possible implementation method is that in each training cycle, a certain number of samples of seismic data containing background noise are first input into a neural network model with randomly initialized network weight coefficients and bias matrix to obtain predicted seismic data; the distribution consistency loss between the predicted seismic data and the clean seismic data is calculated; the gradient is automatically obtained based on the distribution consistency loss, and the back-propagation algorithm is used to update the deep neural network weight coefficients and bias matrix until the training loss curve converges.

[0112] Specifically, the noisy seismic data is input into the deep neural network after parameter update to obtain the predicted seismic data, and the distribution consistency loss between it and the clean seismic data is calculated; then, through multiple network verification processes, the verification loss curve is obtained, and compared with the training loss curve to effectively evaluate whether the deep neural network is overfitting, properly fitting or underfitting, so as to facilitate the artificial adjustment of deep learning hyperparameters and optimize the network parameter update process.

[0113] S209: When it is determined that the loss function converges, the neural network model is controlled to stop iterative processing, and the training of the background noise suppression model is determined to be completed.

[0114] Among them, after each training iteration cycle, the loss function value of the model on the training set and the validation set is recorded. For example, a threshold can be set or the change trend can be used to determine whether the loss function is stable. Monitoring the loss function helps to understand the progress of model training. When the loss no longer decreases significantly, it indicates that the model may have reached the optimal state. The neural network model is controlled to stop iterative processing and determine that the background noise suppression model training is completed.

[0115] S210, inputting the noisy seismic data into a background noise suppression model.

[0116] S211. The background noise suppression model determines the seismic signal data corresponding to the background noise data based on the learned correlation between the background noise data and the seismic forward simulation data.

[0117] Among them, the pre-processed noisy seismic data is input into the trained background noise suppression model, and the model can automatically process the input data without manual intervention, simplifying the processing flow. After the data is input, the model can start processing immediately. The model determines the seismic signal data corresponding to the background noise data based on the learned correlation between the background noise data and the seismic forward simulation data.

[0118] S212. Extract and process the background noise data based on the seismic signal data to obtain a noise suppression result of the noisy seismic data.

[0119] Wherein, the background noise data is extracted and processed based on the seismic signal data. Specifically, the background noise is separated from the background noise data, and the noise suppression result is used to indicate the target seismic signal contained in the noisy seismic data.

[0120] The target seismic signal is extracted from the noisy seismic data to obtain the result of noise suppression. This process not only improves the signal clarity and signal-to-noise ratio, but also helps to improve the accuracy of seismic data analysis, thus providing more powerful support for seismic exploration and monitoring.

[0121] This embodiment proposes a background noise suppression method for active source seismic exploration based on deep learning. The method obtains background noise and seismic wave data before and after source excitation, simulates the propagation process of seismic waves in underground media, and obtains clean seismic data without background noise; divides the data and randomly pairs them to form a training sample set, and uses the sample set to train a neural network model to obtain a background noise suppression model; applies the background noise suppression model to actual seismic wave data to extract clear seismic signals, reduces manual participation, improves suppression efficiency, improves the quality of model training data and denoising accuracy in practical applications, and thereby improves the quality of seismic data.

[0122] Figure 3 A schematic diagram of the structure of a background noise suppression device for active source seismic exploration based on deep learning provided in this application, such as Figure 3 As shown, the active source seismic exploration background noise suppression device 300 based on deep learning provided in this embodiment includes:

[0123] An acquisition module 301 is used to acquire background noise data and noisy seismic data in a target exploration area;

[0124] A processing module 302 is used to fuse the background noise data of the target exploration area with the seismic forward modeling data to obtain a training sample set, wherein the seismic forward modeling data is data generated by computer simulation of seismic waves propagating in underground media;

[0125] The processing module 302 is further used to train a neural network model using the training sample set to obtain a background noise suppression model, wherein the neural network model is used to determine the correlation between the background noise data and the seismic forward modeling data;

[0126] The processing module 302 is further used to input the noisy seismic data into the background noise suppression model to obtain a noise suppression result of the noisy seismic data.

[0127] Optionally, the active source seismic exploration background noise suppression device based on deep learning further includes: a determination module 303;

[0128] The acquisition module 301 is further used to acquire first sensor data of the target exploration area within a first time period before the source is excited;

[0129] The determination module 303 is used to determine the background noise data according to the first sensor data;

[0130] The acquisition module 301 is also used to acquire second sensor data of the target exploration area within a first time period after the source is excited;

[0131] The determination module 303 is further configured to determine the noisy seismic data according to the second sensor data.

[0132] Optionally, the processing module 302 is further used to simulate the propagation process of seismic waves in the underground medium according to the geographical structure and physical characteristics of the target exploration area to obtain a plurality of seismic forward modeling data, wherein the seismic forward modeling data is seismic data that does not contain background noise;

[0133] The processing module 302 is further used to segment the background noise data and the seismic forward modeling data using overlapping sliding windows to obtain a plurality of target noise blocks corresponding to the background noise data and a plurality of target signal blocks corresponding to the seismic forward modeling data, wherein the lengths of the noise blocks and the signal blocks are consistent;

[0134] The processing module 302 is further configured to perform random combination processing on the multiple noise blocks and the multiple signal blocks to obtain a training sample set.

[0135] Optionally, the processing module 302 is further configured to perform random pairing processing on any one of the multiple noise blocks and any one of the multiple signal blocks;

[0136] The processing module 302 is further used to superimpose the paired noise blocks onto the corresponding signal blocks to obtain noisy signal blocks;

[0137] The processing module 302 is further configured to use a signal block corresponding to each of the noisy signal blocks as a label of the noisy signal block to obtain a training sample set.

[0138] Optionally, the determination module 303 is further used to determine a loss function of the neural network model, where the loss function indicates the difference between the output data predicted by the neural network model and the actual data;

[0139] The determination module 303 is further used to control the neural network model to stop iterative processing when it is determined that the loss function converges, and to determine that the background noise suppression model training is completed.

[0140] Optionally, the processing module 302 is further used to input the noisy seismic data into the background noise suppression model;

[0141] The determination module 303 is further used for the background noise suppression model to determine the seismic signal data corresponding to the background noise data based on the learned association relationship between the background noise data and the seismic forward simulation data;

[0142] The processing module 302 is further used to extract and process the background noise data based on the seismic signal data to obtain a noise suppression result of the noisy seismic data, and the noise suppression result is used to indicate a target seismic signal contained in the noisy seismic data.

[0143] Figure 4 This is a schematic diagram of the structure of a deep learning-based active source seismic exploration background noise suppression device provided in this application. Figure 4 As shown, the present application provides an active source seismic exploration background noise suppression device based on deep learning. The active source seismic exploration background noise suppression device based on deep learning 400 includes: a receiver 401, a transmitter 402, a processor 403 and a memory 404.

[0144] Receiver 401, for receiving instructions and data;

[0145] A transmitter 402, used for sending instructions and data;

[0146] Memory 404, for storing computer-executable instructions;

[0147] The processor 403 is used to execute the computer-executable instructions stored in the memory 404 to implement the various steps performed by the method for suppressing background noise in active source seismic exploration based on deep learning in the above-mentioned embodiment. For details, please refer to the relevant description in the embodiment of the method for suppressing background noise in active source seismic exploration based on deep learning.

[0148] Optionally, the memory 404 may be independent or integrated with the processor 403 .

[0149] When the memory 404 is independently provided, the electronic device further includes a bus for connecting the memory 404 and the processor 403 .

[0150] The present application also provides a computer storage medium, in which computer execution instructions are stored. When a processor executes the computer execution instructions, an active source seismic exploration background noise suppression method based on deep learning as performed by the above-mentioned active source seismic exploration background noise suppression device based on deep learning is implemented.

[0151] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0152] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0153] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for suppressing background noise in active source seismic exploration based on deep learning, characterized in that: The method comprises: Obtain background noise data and noisy seismic data in the target exploration area; The background noise data of the target exploration area is fused with the seismic forward modeling data to obtain a training sample set, wherein the seismic forward modeling data is computer-generated data of seismic waves propagating in underground media; Using the training sample set to train a neural network model to obtain a background noise suppression model, wherein the neural network model is used to determine the correlation between the background noise data and the seismic forward simulation data; The noisy seismic data is input into the background noise suppression model to obtain the noise suppression result of the noisy seismic data.

2. The method according to claim 1, characterized in that The step of obtaining background noise data and noisy seismic data in the target exploration area includes: Acquire first sensor data of a target exploration area within a first time period before source excitation, and determine the background noise data based on the first sensor data; Second sensor data of the target exploration area within a first time period after the earthquake source is excited is acquired, and the noisy seismic data is determined based on the second sensor data.

3. The method according to claim 1, characterized in that The fusing of the background noise data of the target exploration area with the seismic forward modeling data to obtain a training sample set includes: According to the geographical structure and physical characteristics of the target exploration area, the propagation process of seismic waves in the underground medium is simulated to obtain a plurality of seismic forward modeling data, wherein the seismic forward modeling data is seismic data that does not contain background noise; The background noise data and the seismic forward modeling data are segmented by using overlapping sliding windows to obtain a plurality of target noise blocks corresponding to the background noise data and a plurality of target signal blocks corresponding to the seismic forward modeling data, wherein the lengths of the noise blocks and the signal blocks are consistent; The multiple noise blocks and the multiple signal blocks are randomly combined to obtain a training sample set.

4. The method according to claim 3, characterized in that The step of randomly combining the multiple noise blocks and the multiple signal blocks to obtain a training sample set includes: Randomly pairing any one of the multiple noise blocks with any one of the multiple signal blocks; Superimposing the paired noise blocks onto the corresponding signal blocks to obtain noisy signal blocks; A signal block corresponding to each of the noisy signal blocks is used as a label of the noisy signal block to obtain a training sample set.

5. The method according to claim 1, characterized in that The training of the neural network model to obtain a background noise suppression model includes: Determining a loss function of the neural network model, wherein the loss function indicates a difference between output data predicted by the neural network model and actual data; When it is determined that the loss function converges, the neural network model is controlled to stop iterative processing, and it is determined that the background noise suppression model training is completed.

6. The method according to claim 1, characterized in that The step of inputting the noisy seismic data into the background noise suppression model to obtain a noise suppression result of the noisy seismic data includes: Inputting the noisy seismic data into the background noise suppression model; The background noise suppression model determines the seismic signal data corresponding to the background noise data based on the learned association relationship between the background noise data and the seismic forward simulation data; Based on the seismic signal data, the background noise data is extracted and processed to obtain a noise suppression result of the noisy seismic data, and the noise suppression result is used to indicate a target seismic signal contained in the noisy seismic data.

7. A background noise suppression device for active source seismic exploration based on deep learning, characterized in that: The device comprises: An acquisition module is used to acquire background noise data and noisy seismic data in the target exploration area; A processing module, used for fusing the background noise data of the target exploration area with the seismic forward modeling data to obtain a training sample set, wherein the seismic forward modeling data is data generated by computer simulation of seismic waves propagating in underground media; The processing module is further used to train a neural network model using the training sample set to obtain a background noise suppression model, wherein the neural network model is used to determine the correlation between the background noise data and the seismic forward simulation data; The processing module is further used to input the noisy seismic data into the background noise suppression model to obtain a noise suppression result of the noisy seismic data.

8. The device according to claim 7, characterized in that The processing module is further used to simulate the propagation process of seismic waves in the underground medium according to the geographical structure and physical characteristics of the target exploration area to obtain a plurality of seismic forward simulation data, wherein the seismic forward simulation data is seismic data that does not contain background noise; The processing module is further used to segment the background noise data and the seismic forward modeling data using overlapping sliding windows to obtain a plurality of target noise blocks corresponding to the background noise data and a plurality of target signal blocks corresponding to the seismic forward modeling data, wherein the lengths of the noise blocks and the signal blocks are consistent; The processing module is further used to randomly combine the multiple noise blocks and the multiple signal blocks to obtain a training sample set.

9. A deep learning-based active source seismic exploration background noise suppression device, characterized in that: include: Memory; processor; Wherein, the memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the active source seismic exploration background noise suppression method based on deep learning as described in any one of claims 1-6.

10. A computer storage medium, characterized in that: The computer storage medium stores computer execution instructions, which, when executed by a processor, are used to implement the active source seismic exploration background noise suppression method based on deep learning as described in any one of claims 1 to 6.

Images