Surface wave apparent velocity analysis method, device and equipment based on deep learning and storage medium

Abstract

The invention discloses a surface wave apparent velocity analysis method and device based on deep learning, equipment and a storage medium, and the method comprises the steps: extracting part of seismic data as a data set, and taking the seismic data which is not extracted as to-be-processed data; analyzing the data set to obtain a corresponding apparent velocity label; the data set and the corresponding apparent velocity label are converted to a t-v domain through RT transformation; the converted data set is preprocessed; according to a preprocessing result, extracting a part of the data set as a training set, and taking the remaining data set as a verification set; constructing a deep neural network, and training a U-net model; converting the to-be-processed data to a t-v domain through RT conversion; using the U-net model to predict the data to be processed; a prediction result is converted to a t-x domain through inverse RT transformation; converting the t-x domain prediction result into apparent velocity information; and performing surface wave suppression on the surface wave apparent velocity result. The method is well suitable for surface wave analysis work of seismic data of different terrains, the manual workload is reduced, and the analysis efficiency and the analysis result accuracy are improved.

Description

Technical Field

[0001] This application relates to the field of seismic data processing technology in oil and gas exploration, and more specifically, to a surface wave apparent velocity analysis method, apparatus, device, and storage medium based on deep learning. Background Technology

[0002] During seismic data acquisition, a lot of noise interference is encountered. Based on the propagation mechanism of noise, it can be roughly divided into two categories: coherent noise and random noise. Coherent noise is characterized by interference with certain regularity in the spatial and temporal directions. Surface waves are a typical type of coherent noise. To effectively and accurately remove this type of noise, it is an important prerequisite to accurately describe the noise characteristics.

[0003] Currently, commonly used surface wave suppression methods both domestically and internationally primarily utilize the differences between surface waves and effective waves in terms of apparent velocity, frequency, and propagation time, employing techniques such as filtering and various transformations for suppression. These methods require the use of parameters such as the maximum apparent velocity and frequency of the surface wave in practical applications. The accuracy of these parameters plays a decisive role in the noise reduction effect. Therefore, accurately obtaining these characteristics of surface waves becomes a crucial step in accurately achieving noise suppression. Among these noise characteristics, apparent velocity varies significantly depending on factors such as underground conditions and propagation patterns, exhibiting substantial differences at different locations within the same work area.

[0004] Currently, in traditional processing, the maximum apparent velocity of surface waves is usually determined manually by analyzing several representative data points. Typically, only three to four different sets of parameters are used to process a single 3D seismic data area, or even just one apparent velocity parameter is used for all data. This method struggles to account for all data, and when data changes drastically, it affects noise suppression and may even damage the valid signal. However, requiring processors to perform detailed analysis of data from each shot is too time-consuming and labor-intensive. To achieve accurate surface wave apparent velocity analysis down to the individual shot level using reasonable time and labor costs, we introduced artificial intelligence methods, implementing a surface wave apparent velocity analysis method based on deep neural networks. This aims to provide more accurate surface wave apparent velocity information to traditional denoising algorithms, thereby improving data processing efficiency and accuracy. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a deep learning-based surface wave apparent velocity analysis method, apparatus, device, and storage medium, aiming to provide more accurate surface wave apparent velocity information for traditional denoising algorithms, thereby improving data processing efficiency and accuracy.

[0006] The first aspect of this invention provides a surface wave apparent velocity analysis method based on deep learning, comprising: A portion of the earthquake data is extracted to form a dataset, while the remaining earthquake data is data to be processed. Analyze the dataset to obtain the corresponding visual speed labels; Transform the dataset and corresponding visual velocity labels into the TV domain using RT transform; The transformed dataset is preprocessed; Based on the preprocessing results, a portion of the dataset is extracted as the training set, and the remainder is used as the validation set. Build a deep neural network and train the U-net model; The data to be processed is transformed into the TV domain using RT transformation; The trained U-net model is used to make predictions on the data to be processed. The prediction results are transformed to the tx domain using an inverse RT transform. The tx domain prediction results are converted into apparent velocity information to obtain the surface wave apparent velocity results; The surface wave apparent velocity results are then subjected to surface wave suppression.

[0007] In one optional implementation, before extracting seismic data, static correction processing is performed on the seismic data. Based on the statically corrected seismic data, extraction is performed using three keywords: shot number, receiver line number, and trace number, to obtain several datasets.

[0008] In one optional implementation, the method further includes: extracting additional location information from the data in the dataset, the additional location information including: shot point coordinates, receiver point coordinates, elevation, and projected shot-receiver distance information; Building a deep neural network and training the U-net model includes: The training set and additional location information are input into a deep neural network to train the U-net model.

[0009] In one optional implementation, the projection gun ranging information is classified into two parts, left and right, based on whether it is positive or negative. The projection gun ranging with negative values ​​is denoted as T1, T2, T3, ... Tn, and the projection gun ranging with positive values ​​is denoted as T`1, T`2, T`3, ... T`n. The dataset was analyzed to obtain the corresponding visual velocity labels, including: The extracted dataset is analyzed using interactive software. The analysis determines the data in the dataset by shot number and receiver line number. For the positive and negative parts of the projected shot-receiver distance, the envelope of the maximum apparent velocity of the surface wave is picked out. The envelope is the apparent velocity label of this data.

[0010] In one alternative implementation, the preprocessing involves performing gain and filtering operations on the transformed dataset.

[0011] In one optional implementation, the loss function in the trained U-net model is a binary cross-entropy loss function:

[0012] Where loss is the loss function. Let (i, j) be the view velocity label. The training result is the output of the U-net model to be trained at point (i, j).

[0013] In one alternative implementation, before transforming the prediction result to the tx domain through the inverse RT transform, the method further includes: smoothing the prediction result to remove outliers.

[0014] A second aspect of this application provides a surface wave apparent velocity analysis device based on deep learning, the device comprising: Data acquisition module: used to extract a portion of the earthquake data. The extracted earthquake data is the dataset, and the unextracted earthquake data is the data to be processed. Data analysis module: used to analyze the dataset and obtain the corresponding visual speed label; Data Transformation Module: Used for mutual conversion between the TX domain and the TV domain, including: converting the dataset and corresponding visual velocity labels to the TV domain through RT transformation; converting the data to be processed to the TV domain through RT transformation; and converting the prediction results to the TX domain through inverse RT transformation. Data processing module: used to perform gain and filtering operations on the transformed dataset; smooth the prediction results and remove outliers; Data training module: used to train the U-net model, and then use the trained U-net model to make predictions on the data to be processed; Surface wave suppression module: Used to suppress the surface wave apparent velocity results.

[0015] A third aspect of the present invention provides an electronic device, characterized in that it includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a deep learning-based surface wave apparent velocity analysis method.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program, when executed by a processor, implements a surface wave apparent velocity analysis method based on deep learning.

[0017] In this embodiment, the surface wave apparent velocity analysis method based on a deep neural network specifically utilizes a limited number of analyzed surface wave apparent velocity results as training labels to construct an efficient deep neural network model. This deep neural network model is trained to identify and learn key features of surface wave apparent velocities, thereby enabling automated prediction of massive amounts of unlabeled seismic data. This process improves processing efficiency and ensures that prediction results are accurately mapped to each individual gather, providing unprecedentedly detailed apparent velocity information for fields such as geological exploration and oil and gas exploration.

[0018] The significant advantage of this method lies in its ability to ensure accurate analysis and prediction regardless of the type of dataset, the complex and varied subsurface structure, or the seismic records acquired under different conditions. The surface wave apparent velocity analysis method in this embodiment not only simplifies the traditional analysis process but also greatly improves analysis accuracy and efficiency, providing solid data support for subsequent exploration decisions and resource assessments. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a surface wave apparent velocity analysis method based on deep learning proposed in an embodiment of this application; Figure 2 This is a flowchart of a deep neural network training process proposed in an embodiment of this application; Figure 3 This is a diagram showing the RT transformation result proposed in an embodiment of this application; Figure 4 This is a flowchart of a deep neural network prediction process proposed in an embodiment of this application; Figure 5 This is a diagram showing the prediction results of a deep neural network proposed in an embodiment of this application; Figure 6 This is a schematic diagram of an electronic device according to this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] Please refer to Figure 1 , Figure 1 This is a flowchart of a surface wave apparent velocity analysis method based on deep learning proposed in Embodiment 1 of this application. Figure 1 As shown, the deep learning-based surface wave apparent velocity analysis method includes: A portion of the earthquake data is extracted to form a dataset, while the remaining earthquake data is data to be processed. In this embodiment, considering factors such as data scale, computing resources, and expected analysis accuracy, the extracted seismic data volume is approximately 5% of the total data volume. When surface conditions are more complex, the characteristics of seismic data may be more diverse, increasing the difficulty of data analysis and processing. It may be necessary to increase the amount of data extracted, with the remaining data remaining to be processed.

[0023] Analyze the dataset to obtain the corresponding visual speed labels; In this embodiment, the extracted dataset was manually analyzed using interactive software to obtain the corresponding visual speed labels.

[0024] Transform the dataset and corresponding visual velocity labels into the TV domain using RT transform; In this embodiment, the data is converted to the TV domain using RT transformation, such as... Figure 3 The two figures on the left depict the continuous changes in the apparent velocity distribution at different time points; the envelope label data of the surface wave with the highest apparent velocity is read and transformed to the TV domain using the RT transform, as shown below. Figure 3 The two figures on the right are shown in the middle. The Radon Transform (RT) is used to transform the original dataset and its corresponding visual velocity labels from the traditional tx domain (spatiotemporal domain) to the more intuitive tv domain (time-visual velocity domain).

[0025] The transformed dataset is preprocessed; Based on the preprocessing results, a portion of the dataset is extracted as the training set, and the remainder is used as the validation set. In this embodiment, in order to effectively utilize the preprocessed dataset for model training and evaluation, a strategy of dividing the dataset into training and validation sets is adopted. Based on random sampling, 80% of the data is selected from the entire preprocessed dataset and used as the training set for the model training process. In order to evaluate the model's generalization ability and performance, the remaining 20% ​​of the data is reserved as the validation set.

[0026] Build a deep neural network and train the U-net model; In this embodiment, the dataset and its corresponding apparent velocity labels are randomly divided into three parts in a ratio of 8:1:1. The first part (80%) is explicitly designated as the training set, and its core function is to provide sufficient sample data to drive the training process of the U-net model. The model can gradually learn and identify key features in the data, thereby establishing a complex mapping relationship between the input data (TV domain seismic data) and the output target (TV domain apparent velocity labels). The second part (10%) is the remaining data extracted as the validation set. Some of this data does not directly participate in the model training, but it will be used to test the model's performance at the end of each training cycle so as to adjust and optimize the model parameters in a timely manner and ensure that the model can also show good predictive ability on unseen data. The third part (10%) is the remaining data extracted as the test set, which is used after the model training is completed. The performance on the test set can provide a more accurate understanding of the model's final performance. Throughout the process, the TV domain label data and the corresponding TV domain seismic data are used for neural network training.

[0027] The data to be processed is transformed into the TV domain using RT transformation; In this embodiment, during seismic exploration, the Relative-Range Transform (RT Transform) can convert wavefield information in seismic records from a time-offset coordinate system to a time-apparent velocity coordinate system. This conversion helps to separate and identify different components in seismic waves, such as reflected waves, refracted waves, and multiples, thereby improving the accuracy of seismic data interpretation. The RT Transform can also more effectively suppress coherent noise, improving the signal-to-noise ratio of seismic data. For example, the principle of the RT Transform is used in the Radial Trace Transform to separate coherent noise from the effective signal.

[0028] The trained U-net model is used to make predictions on the data to be processed. The prediction results are transformed to the tx domain using an inverse RT transform. In this embodiment, during seismic exploration, seismic data (such as time-velocity domain data) processed by RT transformation needs to undergo inverse RT transformation to restore the original time-space domain data, which facilitates geological interpretation and structural analysis.

[0029] The tx domain prediction results are converted into apparent velocity information to obtain the surface wave apparent velocity results; The surface wave apparent velocity results are then subjected to surface wave suppression.

[0030] In this embodiment, surface wave suppression technology is used to reduce or eliminate surface wave interference in seismic data. The apparent velocity results are passed to the surface wave suppression module, which can be any surface wave suppression module or algorithm that requires the maximum apparent velocity as an input parameter. This method uses the high-precision surface wave prediction and elimination (SurfWvRemv) module in GeoEast as a verification object, effectively predicting and suppressing surface wave components in seismic data based on the input apparent velocity information. This module uses this apparent velocity data as a reference to analyze and identify surface wave characteristics in the seismic record, and then takes appropriate algorithmic measures to suppress or eliminate these surface wave interferences.

[0031] Furthermore, before extracting seismic data, static correction processing is performed on the seismic data. Based on the seismic data after static correction processing, several datasets are obtained by extracting data according to three keywords: shot number, receiver line number, and trace number.

[0032] In this embodiment, static correction is used to process seismic data. This aims to eliminate or mitigate seismic wave propagation time errors caused by geological factors such as surface undulations and changes in low-velocity zones. This allows for the interpretation and analysis of seismic data to be performed on a unified reference surface, typically a horizontal plane or a specific reference plane. Static correction eliminates the influence of surface undulations and changes in subsurface geological conditions on seismic data, improving the signal-to-noise ratio and resolution, thereby more accurately reflecting subsurface geological structures and lithological variations. This makes the subsurface geological structure clearer, ensuring the accuracy and reliability of the seismic data. Trace numbers, used in conjunction with parameters such as receiver line numbers and shot numbers, allow for precise location of each data trace within the seismic data volume. Trace numbers are crucial for data sorting, organization, and analysis, helping researchers quickly locate and access specific data segments within complex seismic datasets. Specifically, shot numbers uniquely identify each firing (shot) during the seismic data acquisition process; receiver line numbers identify the location of the seismic detector (the receiving unit of a seismograph, also called a detector or seismic trace) on the ground. Trace numbers uniquely identify each record in the seismic data.

[0033] Furthermore, it also includes: extracting additional location information from the data in the dataset, the additional location information including: shot point coordinates, receiver point coordinates, elevation, and projected shot-receiver distance information; Building a deep neural network and training the U-net model includes: The training set and additional location information are input into a deep neural network to train the U-net model.

[0034] In this embodiment, the shot point coordinates, receiver coordinates, elevation, and projected shot-receiver distance information of each trace in the dataset are extracted as additional location information and paired one-to-one with the seismic trace data to construct a deep neural network. The U-net model is selected as the core architecture. The input data consists of the processed dataset after RT transformation and the additional location information, and the labels are the apparent velocity labels after RT transformation. The U-net model needs to learn how to combine the waveform of the RT-transformed seismic data with specific spatial location information to accurately predict the corresponding apparent velocity labels.

[0035] Furthermore, the projection gun ranging information is classified into two parts based on whether it is positive or negative: negative projection gun ranging is denoted as T1, T2, T3, ... Tn, and positive projection gun ranging is denoted as T`1, T`2, T`3, ... T`n. The dataset was analyzed to obtain the corresponding visual velocity labels, including: The extracted dataset is analyzed using interactive software. The analysis determines the data in the dataset by shot number and receiver line number. For the positive and negative parts of the projected shot-receiver distance, the envelope of the maximum apparent velocity of the surface wave is picked out. The envelope is the apparent velocity label of this data.

[0036] In this embodiment, to process the dataset more meticulously and extract the apparent velocity labels more accurately, the projected shot-receiver distance information was classified. A dataset, defined by shot number and receiver line number, typically consists of two parts: the left part has negative projected shot-receiver distances (denoted as T1, T2, T3, ..., Tn), and the right part has positive projected shot-receiver distances (denoted as T`1, T`2, T`3, ..., T`n). During manual analysis, the left and right parts of the data are processed separately. Because the propagation characteristics of surface waves can be affected by various factors such as underground medium inhomogeneity and terrain undulations, the data on the left and right sides may exhibit different apparent velocity characteristics. Therefore, by separately picking the maximum apparent velocity envelopes of the surface waves on the left and right sides, these subtle differences can be captured more accurately, providing more refined and accurate label information for subsequent deep neural network training.

[0037] Furthermore, the preprocessing involves performing gain and filtering operations on the transformed dataset.

[0038] In this embodiment, gain operation is a dynamic balancing process of signal strength. By intelligently analyzing the amplitude distribution in the data, the signal strength is enhanced or weakened in a timely manner to ensure that the entire dataset is uniformly displayed within a suitable dynamic range. This process not only helps to highlight weak underground reflection signals but also effectively avoids distortion caused by signal saturation, providing a clearer and more accurate image basis for subsequent seismic interpretation work.

[0039] Filtering operations involve processing data in the frequency or time domain to remove unnecessary noise and interference. Based on the characteristics of seismic data, various filters, such as bandpass filters and notch filters, have been designed to specifically suppress high-frequency noise, low-frequency background noise, and interference waves of specific frequencies. Through filtering, the signal-to-noise ratio of seismic data is improved, making the information about subsurface geological structures more prominent and providing more reliable data support for geological interpretation and reservoir prediction.

[0040] Furthermore, the loss function used in training the U-net model is the binary cross-entropy loss function:

[0041] Where loss is the loss function. Let (i, j) be the view velocity label. The training result is the output of the U-net model to be trained at point (i, j).

[0042] In this embodiment, the binary cross-entropy loss function is designed to evaluate the difference between the U-net model's predicted visual velocity and the actual visual velocity label at each pixel (i, j). For each such pixel, the loss function calculates the difference in probability distribution between the predicted and actual values, representing the difference between the probability that the U-net model considers the pixel to belong to a certain velocity category and the true probability. In this way, the loss function can capture the model's predictive performance across the entire image or data volume and guide the model to continuously optimize its parameters during training to reduce this difference.

[0043] Furthermore, before transforming the prediction results to the tx domain through the inverse RT transform, the process also includes: smoothing the prediction results and removing outliers.

[0044] In this embodiment, the prediction results are smoothed to eliminate potential outliers or noise points. The smoothing process uses mathematical algorithms or filters to fine-tune the prediction results, reducing or eliminating sharp fluctuations and unreasonable jumps in the data, making the apparent velocity curve smoother and more continuous. Then, an inverse RT transform is used to convert the data back to the tx domain, remapping the seismic data to a time and space coordinate system to recover the true propagation path and velocity structure of seismic waves underground.

[0045] In summary, for reference Figure 2 , Figure 2This is a flowchart of a deep neural network training process proposed in one embodiment of this application. The specific steps of the neural network training method are as follows: First, prepare the dataset and construct the corresponding observation system to ensure that the data can be arranged in an orderly manner according to the three core keywords: shot number, receiver line number, and track number. Then, focus on processing single-shot data, specifically, accurately extract the dataset of a certain common receiver line and perform a series of preprocessing operations such as gain adjustment and filtering optimization to improve data quality. Next, use an interactive software platform to manually analyze the preprocessed dataset. During this process, focus on identifying and picking the surface wave envelopes that maximize apparent velocity; these envelopes serve as key labels and are crucial for subsequent training. To maintain dimensionality consistency with the seismic data, RT transform technology is used to convert the picked apparent velocity envelope labels and the original seismic data to the TV domain (time-velocity domain), referencing... Figure 3 , Figure 3 This is a diagram showing the RT transformation result proposed in an embodiment of this application, as follows: Figure 3 The left and right illustrations show the data formats before and after the transformation. After the transformation, the data reading and label picking steps were repeated until all labeled data and their corresponding labels were properly processed. Subsequently, a scientific partitioning strategy was adopted, randomly dividing the data into training, validation, and test sets in a ratio of 8:1:1 to ensure the independence and effectiveness of the model training, validation, and testing processes. In the deep neural network training phase, labeled data from the TV domain and seismic data were used to train the model through a deep neural network. During this process, the binary cross-entropy loss function was selected as the optimization objective. This function can effectively evaluate the difference between the model's predicted values ​​and the true labels, and drive the model parameters to adjust towards the optimal direction. Through continuous iterative training, it is expected that the model can learn the complex laws of surface wave apparent velocity changes in seismic data, thereby achieving accurate prediction and suppression of surface wave interference.

[0046] refer to Figure 4 , Figure 4This is a flowchart of a deep neural network prediction process proposed in an embodiment of this application. The specific steps of the neural network prediction and application method are as follows: First, the prediction process is started, and a pre-constructed and fully trained neural network model is loaded. This model is trained using a large amount of labeled data through a deep learning algorithm and has the ability to predict the apparent velocity of surface waves in seismic data. Next, the seismic data to be processed is selected, and a data segment of a specific common receiver line is extracted from a single shot record. To improve the accuracy of the prediction, a series of preprocessing operations are performed on these data, including gain adjustment to enhance signal visibility, filtering to remove unnecessary noise interference, and transforming these data from the original tx domain (time-space domain) to the tv domain (time-velocity domain) through RT transformation to better adapt to the input requirements of the neural network model. Subsequently, the preprocessed and domain-transformed data is input into the loaded neural network model, which then calculates and analyzes the data and outputs the predicted apparent velocity results. These prediction results initially reflect the changing trend of the apparent velocity of surface waves, but may still contain some outliers or noise points, affecting the accuracy of the results. Therefore, after obtaining the prediction results, refer to Figure 5 , Figure 5 This is a deep neural network prediction result image proposed in one embodiment of this application. The image is smoothed by using mathematical methods or filter techniques to eliminate or reduce the influence of outliers, making the prediction results smoother and more continuous. Then, using inverse RT transform technology, the processed prediction results are converted from the TV domain back to the original TX domain, so as to intuitively display the changes in apparent velocity in both time and space dimensions. Finally, based on the prediction results converted back to the TX domain, specific apparent velocity values ​​are further calculated, providing strong support for seismic data processing, stratigraphic interpretation, and structural analysis.

[0047] This disclosure provides a surface wave apparent velocity analysis device based on deep learning, the device comprising: Data acquisition module: used to extract a portion of the earthquake data. The extracted earthquake data is the dataset, and the unextracted earthquake data is the data to be processed. The data acquisition module is also used for: Before extracting seismic data, static correction processing is performed on the seismic data. Based on the seismic data after static correction processing, several datasets are extracted according to three keywords: shot number, receiver line number, and trace number. Extract additional location information from the dataset, including: shot point coordinates, receiver point coordinates, elevation, and projected shot-receiver distance information; The projection gun ranging information is classified into two parts based on whether it is positive or negative: negative projection gun ranging is denoted as T1, T2, T3, ... Tn, and positive projection gun ranging is denoted as T`1, T`2, T`3, ... T`n. Data analysis module: used to analyze the dataset and obtain the corresponding visual speed label; The data analysis module is also used for: The extracted dataset is analyzed using interactive software. The analysis determines the data in the dataset by shot number and receiver line number. For the positive and negative parts of the projected shot-receiver distance, the envelope of the maximum apparent velocity of the surface wave is picked out. The envelope is the apparent velocity label of this data. Data Transformation Module: Used for mutual conversion between the TX domain and the TV domain, including: converting the dataset and corresponding visual velocity labels to the TV domain through RT transformation; converting the data to be processed to the TV domain through RT transformation; and converting the prediction results to the TX domain through inverse RT transformation. Data processing module: used to perform gain and filtering operations on the transformed dataset; smooth the prediction results and remove outliers; Data training module: used to train the U-net model, and then use the trained U-net model to make predictions on the data to be processed; The data training module is also used for: The training set and additional location information are input into a deep neural network to train the U-net model; The loss function used in training the U-net model is the binary cross-entropy loss function; Surface wave suppression module: Used to suppress the surface wave apparent velocity results.

[0048] This disclosure also provides an electronic device, please refer to... Figure 6 , Figure 6 This is a schematic diagram of an electronic device illustrated in an embodiment of this disclosure. For example... Figure 6 As shown, the electronic device 100 includes a memory 110 and a processor 120. The memory 110 and the processor 120 are connected via a bus for communication. The memory 110 stores a computer program that can run on the processor 120 to implement the steps in the deep learning-based surface wave apparent velocity analysis method disclosed in this embodiment.

[0049] The disclosed embodiments also provide a computer-readable storage medium that, when the instructions in the computer-readable storage medium are executed by a processor of a computer device, enables the computer device to perform the steps in the deep learning-based surface wave apparent velocity analysis method of the present disclosure embodiments.

[0050] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices, and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0051] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0052] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0053] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0054] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0055] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0056] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0057] The above provides a detailed description of a deep learning-based surface wave apparent velocity analysis method, apparatus, device, and storage medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A surface wave apparent velocity analysis method based on deep learning, characterized in that, include: A portion of the earthquake data is extracted to form a dataset, while the remaining earthquake data is data to be processed. Analyze the dataset to obtain the corresponding visual speed labels; Transform the dataset and corresponding visual velocity labels into the TV domain using RT transform; The transformed dataset is preprocessed; Based on the preprocessing results, a portion of the dataset is extracted as the training set, and the remainder is used as the validation set. Build a deep neural network and train the U-net model; The data to be processed is transformed into the TV domain using RT transformation; The trained U-net model is used to make predictions on the data to be processed. The prediction results are transformed to the tx domain using an inverse RT transform. The tx domain prediction results are converted into apparent velocity information to obtain the surface wave apparent velocity results; The surface wave apparent velocity results are then subjected to surface wave suppression.

2. The surface wave apparent velocity analysis method based on deep learning according to claim 1, characterized in that, Before extracting seismic data, static correction processing is performed on the seismic data. Based on the seismic data after static correction processing, several datasets are obtained by extracting data according to three keywords: shot number, receiver line number, and trace number.

3. The surface wave apparent velocity analysis method based on deep learning according to claim 2, characterized in that, Also includes: Extract additional location information from the dataset, including: shot point coordinates, receiver point coordinates, elevation, and projected shot-receiver distance information; Building a deep neural network and training the U-net model includes: The training set and additional location information are input into a deep neural network to train the U-net model.

4. The surface wave apparent velocity analysis method based on deep learning according to claim 3, characterized in that, The projection gun ranging information is classified into two parts based on whether it is positive or negative: negative projection gun ranging is denoted as T1, T2, T3, ... Tn, and positive projection gun ranging is denoted as T`1, T`2, T`3, ... T`n. The dataset was analyzed to obtain the corresponding visual velocity labels, including: The extracted dataset is analyzed using interactive software. The analysis determines the data in the dataset by shot number and receiver line number. For the positive and negative parts of the projected shot-receiver distance, the envelope of the maximum apparent velocity of the surface wave is picked out. The envelope is the apparent velocity label of this data.

5. The surface wave apparent velocity analysis method based on deep learning according to claim 1, characterized in that, The preprocessing involves performing gain and filtering operations on the transformed dataset.

6. The surface wave apparent velocity analysis method based on deep learning according to claim 1, characterized in that, The loss function used in the training U-net model is the binary cross-entropy loss function: Where loss is the loss function. Let (i, j) be the view velocity label. The training result is the output of the U-net model to be trained at point (i, j).

7. The surface wave apparent velocity analysis method based on deep learning according to claim 1, characterized in that, Before transforming the prediction results to the tx domain using the inverse RT transform, the process also includes smoothing the prediction results and removing outliers.

8. A surface wave apparent velocity analysis device based on deep learning, characterized in that, The device includes: Data acquisition module: used to extract a portion of the earthquake data. The extracted earthquake data is the dataset, and the unextracted earthquake data is the data to be processed. Data analysis module: used to analyze the dataset and obtain the corresponding visual speed label; Data Transformation Module: Used for mutual conversion between the TX domain and the TV domain, including: converting the dataset and corresponding visual velocity labels to the TV domain through RT transformation; converting the data to be processed to the TV domain through RT transformation; and converting the prediction results to the TX domain through inverse RT transformation. Data processing module: used to perform gain and filtering operations on the transformed dataset; smooth the prediction results and remove outliers; Data training module: used to train the U-net model, and then use the trained U-net model to make predictions on the data to be processed; Surface wave suppression module: Used to suppress the surface wave apparent velocity results.

9. An electronic device, characterized in that, include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the deep learning-based surface wave apparent velocity analysis method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the deep learning-based surface wave apparent velocity analysis method as described in any one of claims 1 to 7.

Images