Determination method and system of prediction speed model, storage medium and equipment

Abstract

The embodiment of the invention provides a method and system for determining a prediction speed model, a storage medium and equipment, and relates to the field of oil exploration seismic data processing, and the method comprises the steps: obtaining a plurality of speed model labels through controlling a speed longitudinal control line to slide up and down on a VSP zero offset speed model; performing ray tracing forward modeling on the speed model label to obtain corresponding first arrival travel time data; marking the respective corresponding first arrival travel time data through each speed model label, and constructing a training data set; training an initial neural network model through the training data set to obtain a target neural network model which is qualified in training; and inputting the actual first arrival travel time data into the target neural network model for speed model prediction, and obtaining a predicted speed model corresponding to the actual first arrival travel time data. According to the method, the prediction precision of the target neural network model is improved.

Description

Technical Field

[0001] This application relates to the field of seismic data processing in petroleum exploration, and more specifically, to a method, system, storage medium, and device for determining a prediction velocity model. Background Technology

[0002] Velocity model establishment is a crucial step in seismic exploration. The accuracy of the velocity model directly impacts the quality of subsequent migration imaging and the precision of structural interpretation. Common methods for velocity model establishment include migration velocity analysis, tomography, and full-waveform inversion. However, these methods often require extensive human experience, and inversion-based velocity model establishment methods are typically highly dependent on the initial model. If the initial model deviates significantly from the true solution, it may get trapped in local optima, failing to guarantee global convergence and ultimately resulting in a less accurate velocity model. Summary of the Invention

[0003] This application aims to provide a method, system, storage medium, and device for determining a predictive velocity model, with the goal of improving the accuracy of the target neural network model and the velocity model.

[0004] The first aspect of this application provides a method for determining a predictive velocity model, the method comprising: Multiple velocity model labels are obtained by controlling the vertical speed control line to slide up and down on the VSP zero-bias velocity model. Ray tracing forward modeling was performed on the velocity model labels to obtain the corresponding first arrival travel time data; The initial arrival time data for each velocity model is labeled to construct a training dataset; The initial neural network model is trained using the training dataset to obtain a qualified target neural network model, wherein the activation function of the last layer in the initial neural network model is the Tanh activation function. The actual initial arrival time data is input into the target neural network model to predict the speed model, thereby obtaining a predicted speed model corresponding to the actual initial arrival time data.

[0005] Optionally, the initial arrival time data for each velocity model is labeled to construct a training dataset, including: Normalize the velocity model labels to obtain the corresponding first velocity model labels; The first arrival time data for each of the first velocity models are labeled to construct a training dataset.

[0006] Optionally, training the initial neural network model using the training dataset to obtain a qualified target neural network model includes: The initial arrival time data from the training dataset is input into the initial neural network model to predict the velocity model and obtain the corresponding first predicted velocity model. The loss function algorithm is used to calculate the loss value for the first predicted velocity model and the velocity model label of the corresponding first arrival travel time data. Plot the corresponding loss function curves based on multiple loss values; When the trend of the loss function curve tends to be stable, the training is deemed successful, and the corresponding target neural network model is obtained.

[0007] Optionally, the method further includes: Ray tracing forward modeling was performed on the predicted velocity model to obtain the corresponding predicted first arrival travel time data; The prediction accuracy of the target neural network model is determined based on the difference between the predicted initial arrival time data and the actual initial arrival time data.

[0008] Optionally, the prediction accuracy of the target neural network model is determined based on the difference between the predicted initial arrival time data and the actual initial arrival time data, including: The difference between the predicted initial arrival time data and the actual initial arrival time data is compared with a preset time range to obtain the comparison result; Based on the comparison results, the prediction accuracy of the target neural network model is determined.

[0009] Optionally, based on the comparison results, the prediction performance of the target neural network model is determined, including: If the comparison result shows that the difference data is within a preset time range, it is determined that the prediction accuracy of the target neural network model meets the standard. If the comparison results show that there are differences in the difference data that are outside the preset time range, it is determined that the prediction accuracy of the target neural network model has not met the standard.

[0010] Optionally, determining the initial neural network model includes: A batch normalization layer is set after the convolutional layer and the transposed convolutional layer of the neural network model to obtain the first neural network model; The activation functions of the convolutional and transposed convolutional layers of the first neural network model are determined to be LeakyReLU functions, and the activation function of the last layer is determined to be Tanh activation function, thus obtaining the initial neural network model.

[0011] A second aspect of this application provides a system for determining a predictive velocity model, the system comprising: The velocity model label determination module is used to obtain multiple velocity model labels by controlling the vertical speed control line to slide up and down on the VSP zero-bias velocity model. The first arrival time data determination module is used to perform ray tracing forward modeling on the velocity model labels to obtain the corresponding first arrival time data; The training dataset determination module is used to label the first arrival time data corresponding to each velocity model label and construct the training dataset. The target neural network model determination module is used to train the initial neural network model using the training dataset to obtain a qualified target neural network model, wherein the activation function of the last layer in the initial neural network model is the Tanh activation function; The predicted speed model determination module is used to input the actual initial arrival time data into the target neural network model to predict the speed model and obtain the predicted speed model corresponding to the actual initial arrival time data.

[0012] A third aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for displaying the state of a device as described in any of the first aspects.

[0013] A fourth aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the device state display method as described in any of the first aspects.

[0014] Beneficial effects: This application provides a method for determining a predicted velocity model. The method includes: controlling a longitudinal velocity control line to slide up and down on a VSP zero-bias velocity model to obtain multiple velocity model labels; performing ray tracing forward modeling on the velocity model labels to obtain corresponding first-arrival travel times (FATs); labeling the FATs corresponding to each velocity model label to construct a training dataset; training an initial neural network model using the training dataset to obtain a qualified target neural network model, wherein the activation function of the last layer in the initial neural network model is a Tanh activation function; and inputting actual FATs into the target neural network model to perform velocity model prediction to obtain a predicted velocity model corresponding to the actual FATs.

[0015] Multiple velocities are obtained from the VSP zero-biased velocity model, and these velocities are used to construct velocity model labels. Ray tracing forward modeling is then performed on these velocity model labels to obtain the first-arrival travel times (FATs) corresponding to these labels. A training dataset is constructed using the velocity model labels and FATs, and this training dataset is input into an initial neural network model for training, enabling the initial neural network model to accurately identify the velocity models in the training dataset. Then, the actual FATs are picked up and input into the trained target neural network model for velocity model prediction, ultimately obtaining the predicted velocity model. This method effectively establishes velocity models, and the FAT curves obtained from the forward modeling of the predicted velocity model fit well with the actual FAT curves. Attached Figure Description

[0016] 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.

[0017] Figure 1 This is a flowchart of a method for determining a predicted velocity model according to an embodiment of this application; Figure 2 This is a schematic diagram of a predicted velocity model provided in an embodiment of this application; Figure 3 This is a schematic diagram of a loss function curve provided in an embodiment of this application; Figure 4 This is a comparison diagram of a predicted first arrival time curve and an actual first arrival time curve provided in an embodiment of this application; Figure 5 This is a three-dimensional display diagram of initial travel data provided in one embodiment of this application; Figure 6 This is a schematic diagram of the structure of an initial neural network model provided in one embodiment of this application; Figure 7 This is a schematic diagram of a system for determining a predicted velocity model according to an embodiment of this application. Detailed Implementation

[0018] 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.

[0019] This application provides a flowchart of a method for determining a predicted velocity model, as shown in the embodiments below. Figure 1 As shown. Specifically, this application provides a method for determining a predictive velocity model, the method comprising: S11: By controlling the longitudinal speed control line to slide up and down on the VSP zero-bias speed model, multiple speed model labels can be obtained.

[0020] In this embodiment, zero-bias VSP technology refers to seismic observations performed when the seismic source and detector are located in the same well and at extremely close distances (theoretically zero offset). In this observation method, seismic waves propagate almost vertically downwards and are reflected back, thus minimizing complex path effects and multiple wave interference during propagation. Consequently, VSP data obtained through zero-bias VSP technology has a higher signal-to-noise ratio and higher resolution compared to ordinary seismic data, and it can get closer to the target layer. VSP data includes multiple VSP zero-bias velocities, which are used to construct a VSP zero-bias velocity model.

[0021] This embodiment also includes a longitudinal velocity control line extending from the surface to the subsurface. Specifically, to simulate the undulations of underground structures, the longitudinal velocity control line can be slid up and down on a two-dimensional VSP zero-bias velocity model to reflect the true morphology of the underground strata. During the sliding process, the velocity values ​​at each point on the control line remain constant, but their positions change on the two-dimensional VSP zero-bias velocity model, thereby simulating structural features such as strata tilt, folds, and faults. Each slide yields a corresponding two-dimensional velocity model, which serves as the label for the training dataset used to train the neural network model in this embodiment. In this application, the velocity models used as labels for the training dataset of the neural network model are defined as velocity model labels. That is, by controlling the longitudinal velocity control line to slide up and down multiple times on the VSP zero-bias velocity model, multiple velocity model labels can be obtained.

[0022] S12: Perform ray tracing forward modeling on the velocity model labels to obtain the corresponding first arrival travel time data.

[0023] In this embodiment, ray tracing forward modeling is a method used in seismic exploration to simulate the propagation path of seismic waves in the subsurface medium. Specifically, in seismic exploration, seismic waves are excited at the surface, and then the time, amplitude, and phase of these seismic waves after reflection or refraction at different subsurface layers are recorded. First arrival travel time (FAT) data refers to the time record of the first arrival of a seismic wave from its source to the receiving point. By simulating the actual propagation path of seismic waves in the subsurface medium through ray tracing forward modeling and calculating the shortest path from the source to the receiving point, the arrival time of the seismic wave at the receiving point can be calculated. Therefore, by performing ray tracing forward modeling on each velocity point in the velocity model label, the FAT corresponding to each velocity point can be obtained. Multiple FAT sets constitute the FAT data. Correspondingly, each velocity model label has its own corresponding FAT data.

[0024] S13: Label the first arrival time data of each velocity model with its corresponding labels to construct a training dataset.

[0025] S14: Train the initial neural network model using the training dataset to obtain a qualified target neural network model, wherein the activation function of the last layer in the initial neural network model is the Tanh activation function.

[0026] Specifically, after matching and labeling the corresponding initial arrival time data using velocity model labels, a corresponding training dataset is obtained. This training dataset is then input into an initial neural network model for training, enabling the initial neural network model to accurately identify features in the training dataset; in this embodiment, this means accurately identifying and extracting velocity models from the training dataset. The successfully trained initial neural network model becomes the target neural network model. Simultaneously, because the values ​​of each velocity in the velocity model labels are relatively large, the velocity model labels are normalized. To ensure that the initial neural network model can better extract the normalized velocity model labels, the activation function of the last layer in the initial neural network model is determined to be the Tanh activation function.

[0027] S15: Input the actual initial arrival time data into the target neural network model to predict the speed model and obtain the predicted speed model corresponding to the actual initial arrival time data.

[0028] Specifically, after obtaining a target neural network model capable of accurately identifying the velocity model, actual first-arrival travel time (FAT) data obtained during seismic exploration is collected and input into the target neural network model for velocity model prediction, thereby obtaining an accurate predicted velocity model corresponding to the actual FAT data. In this embodiment, a schematic diagram of the predicted velocity model is also provided, as shown below. Figure 2 As shown, by inputting the actual initial travel time data into the target neural network model, a predictive velocity model as shown in the figure can be predicted. The predictive velocity model obtained by the method of this application can be directly used for migration imaging, providing reliable velocity information for migration imaging.

[0029] In conjunction with the above embodiments, in one implementation, the present invention also provides a method for determining a predicted velocity model. In this method for determining the predicted velocity model, step S13 includes steps S21 to S23: S21: Normalize the velocity model labels to obtain the corresponding first velocity model labels.

[0030] S22: Label the first arrival travel time data corresponding to each first velocity model label to construct a training dataset.

[0031] Specifically, during model training, especially when using optimization algorithms such as gradient descent, if the numerical ranges of velocities in the velocity model labels vary significantly, it can lead to inconsistent step sizes in gradient updates across different directions, thus affecting the convergence speed. Normalizing the velocity model labels can transform all velocity values ​​to a similar scale, thereby accelerating the convergence process. Therefore, in this embodiment, since the velocities in the velocity model labels represent the propagation speed of seismic waves underground, and the numerical differences in velocity are significant, normalization is performed on the velocity model labels to stabilize and accelerate the convergence process of the neural network model. After normalization, the velocity data is transformed to a uniform scale, such as [-1, 1], and a first velocity model label is obtained. The first velocity model label is then used to label the initial travel time data to obtain a training dataset. This training dataset is then used as input data into the initial neural network model. The Tanh activation function in this initial neural network model can map the input data to the range [-1, 1], ensuring the scale consistency of the input data within the neural network.

[0032] In conjunction with the above embodiments, in one implementation, the present invention also provides a method for determining a predicted velocity model. In this method for determining the predicted velocity model, step S14 includes steps S31 to S34: S31: Input the initial arrival travel time data from the training dataset into the initial neural network model to predict the velocity model and obtain the corresponding first predicted velocity model.

[0033] S32: Calculate the corresponding loss value by using a loss function algorithm on the first predicted velocity model and the velocity model label of the corresponding first arrival travel time data.

[0034] S33: Plot the corresponding loss function curves based on multiple loss values.

[0035] S34: When the trend of the loss function curve tends to be stable, the training is deemed qualified and the corresponding target neural network model is obtained.

[0036] Specifically, to determine whether the initial neural network model has been trained successfully, this embodiment inputs the initial arrival time data from the training data into the initial neural network model for velocity model prediction, and the result is determined as the first predicted velocity model. Then, the first predicted velocity model and the corresponding velocity model labels of the initial arrival time data are substituted into the loss function algorithm to obtain the corresponding loss value. In this embodiment, the loss value is an indicator that measures the difference or error between the model's prediction result and the actual result; a smaller loss value usually indicates that the model's prediction result is closer to the actual situation. Specifically, the loss function algorithm is as follows:

[0037] in, This refers to the loss value. This refers to the velocity model label. This represents the first predicted velocity model. Multiple first predicted velocity models can be obtained by using multiple initial arrival times from the training dataset. When the difference between the first predicted velocity model identified by the initial neural network model and its corresponding velocity model label is small (i.e., the difference between the predicted result and the actual result is small), the initial neural network model is considered to have passed training. Specifically, this is demonstrated by plotting multiple loss values ​​obtained from multiple first predicted velocity models and their corresponding velocity model labels onto a loss function curve, with the trend of the loss function gradually stabilizing. For example... Figure 3 As shown in the diagram, this embodiment also provides a schematic diagram of the loss function curve. With the training of the initial neural network model using a large amount of initial arrival time data in the training dataset, the loss function curve gradually stabilizes, and the target neural network model becomes qualified, ultimately resulting in a successfully trained target neural network model.

[0038] In conjunction with the above embodiments, in one implementation, the present invention further provides a method for determining a predicted velocity model. In this method for determining the predicted velocity model, the method further includes steps S41 to S42: S41: Perform ray tracing forward modeling on the predicted velocity model to obtain the corresponding predicted first arrival travel time data.

[0039] S42: Determine the prediction accuracy of the target neural network model based on the difference between the predicted initial arrival time data and the actual initial arrival time data.

[0040] Specifically, the actual arrival time data is input into the target neural network model for velocity model prediction, resulting in a predicted velocity model corresponding to the actual arrival time data. To verify the prediction accuracy of the target neural network model, this embodiment also performs ray tracing forward modeling on the predicted velocity model to obtain the predicted arrival time data corresponding to the predicted velocity model. Then, the difference between the predicted arrival time data and the actual arrival time data is calculated to obtain the difference between the two. The magnitude of this difference determines the degree of difference between the predicted and actual arrival time data. The smaller the difference, the more similar the predicted and actual arrival time data are, and the better the prediction accuracy of the target neural network model.

[0041] In conjunction with the above embodiments, in one implementation, the present invention also provides a method for determining a predicted velocity model. In this method for determining the predicted velocity model, step S42 includes steps S51 to S52: S51: Compare the difference between the predicted initial arrival time data and the actual initial arrival time data with a preset time range to obtain the comparison result.

[0042] S52: Based on the comparison results, determine the prediction accuracy of the target neural network model.

[0043] Specifically, this embodiment also sets a preset time range, which is used to determine whether the difference between the predicted initial arrival time data and the actual initial arrival time data is large or small. That is, the difference between the predicted initial arrival time data and the actual initial arrival time data obtained in the above steps is compared with the preset time range, and the corresponding comparison result is obtained. The prediction accuracy of the target neural network model can be determined by the comparison result.

[0044] In conjunction with the above embodiments, in one implementation, the present invention also provides a method for determining a predicted velocity model. In this method for determining the predicted velocity model, step S52 includes steps S61 to S62: S61: If the comparison result shows that the difference data is within a preset time range, determine that the prediction accuracy of the target neural network model meets the standard.

[0045] S62: If the comparison result shows that there are differences in the difference data that are outside the preset time range, it is determined that the prediction accuracy of the target neural network model has not met the standard.

[0046] Specifically, if the comparison result shows that the difference in the data is within a preset time range, it indicates that the difference between the predicted arrival time data and the actual arrival time data is small, meaning that the prediction accuracy of the corresponding target neural network model is high and meets the standard for prediction accuracy of the target neural network model. If the comparison result shows that some of the differences in the data are outside the preset time range, it indicates that the difference between the predicted arrival time data and the actual arrival time data is large, meaning that the prediction accuracy of the corresponding target neural network model is low and does not meet the standard for prediction accuracy of the target neural network model. In this embodiment, a comparison chart of the predicted arrival time curve and the actual arrival time curve is also provided, such as... Figure 4 As shown. The predicted first-arrival time data is plotted as a predicted first-arrival time curve, as shown below. Figure 4 The dashed line in the figure shows the actual first-arrival travel time; the actual first-arrival travel time data is plotted as an actual first-arrival travel time curve, as shown in the figure. Figure 4 The solid lines in the diagram are shown. Furthermore, this embodiment also provides a three-dimensional display of initial travel time data, as shown in... Figure 5 As shown, Figure 5 The first 3D plot in the image is a 3D plot of the predicted first-arrival travel data. Figure 5 The second 3D display is a 3D display of the actual initial arrival travel time data. Figure 5 The third 3D display plot is a 3D display plot of the difference between the predicted first arrival time data and the actual first arrival time data.

[0047] In conjunction with the above embodiments, in one implementation, the present invention also provides a method for determining a predicted velocity model. In this method for determining the predicted velocity model, the determination of the initial neural network model includes steps S71 to S72: S71: Set a batch normalization layer after the convolutional layer and transposed convolutional layer of the neural network model to obtain the first neural network model; S72: Determine that the activation functions of the convolutional layers and transposed convolutional layers of the first neural network model are LeakyReLU functions, and determine that the last layer is Tanh activation function, to obtain the initial neural network model.

[0048] This embodiment also provides a schematic diagram of the structure of an initial neural network model, such as... Figure 6As shown. In this embodiment, a batch normalization layer is set after the convolutional layer and the transposed convolutional layer of the neural network model. The purpose of the batch normalization layer is to normalize the mean and variance of each mini-batch of data, making the input distribution of the network layers in the neural network model more stable, thereby reducing the internal covariate shift problem. The neural network model with the batch normalization layer is the first neural network model. In the first neural network model, the activation function of the convolutional layer and the transposed convolutional layer is set to the LeakyReLU function. However, the last layer in the neural network model is set to the Tanh activation function to obtain the initial neural network model. The purpose of setting the last layer to the Tanh activation function is that the Tanh activation function can map the input data to the range of [-1, 1]. This limitation of the output range helps to control the output amplitude of the neural network model, so that the output value is within the interpretable range.

[0049] Based on the same inventive concept, one embodiment of this application provides a system for determining a predictive velocity model, such as... Figure 7 As shown in the diagram, this embodiment also provides a schematic diagram of a system for determining a predicted velocity model, the system comprising: The velocity model label determination module 701 is used to obtain multiple velocity model labels by controlling the longitudinal velocity control line to slide up and down on the VSP zero-bias velocity model. The first arrival time data determination module 702 is used to perform ray tracing forward modeling on the velocity model labels to obtain the corresponding first arrival time data. The training dataset determination module 703 is used to label the first arrival time data corresponding to each velocity model label to construct the training dataset; The target neural network model determination module 704 is used to train the initial neural network model using the training dataset to obtain a qualified target neural network model, wherein the activation function of the last layer in the initial neural network model is the Tanh activation function; The predicted speed model determination module 705 is used to input the actual initial arrival time data into the target neural network model to predict the speed model and obtain the predicted speed model corresponding to the actual initial arrival time data.

[0050] Based on the same inventive concept, one embodiment of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any step in the method for determining a predictive speed model.

[0051] Based on the same inventive concept, one embodiment of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any step in the method for determining a predictive speed model.

[0052] In the method for determining a predicted speed model provided in this application embodiment, this application uses speed model labels to create a training dataset and trains the training dataset through an optimized neural network model. The target neural network model obtained after successful training can accurately identify the predicted speed model corresponding to the actual arrival time data. This method can effectively establish a predicted speed model, and the predicted arrival time data obtained by forward modeling the predicted speed model has little difference from the actual arrival time data.

[0053] 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.

[0054] Those skilled in the art will understand that embodiments of the present invention can provide methods, apparatus, electronic devices, storage media, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[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] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0057] 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 limitation, 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 that element.

[0058] The above provides a detailed description of the method, system, storage medium, and device for determining a predictive speed model 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 method of determining a predictive velocity model, characterized by, The method includes: Multiple velocity model labels are obtained by controlling the vertical speed control line to slide up and down on the VSP zero-bias velocity model. Ray tracing forward modeling was performed on the velocity model labels to obtain the corresponding first arrival travel time data; The initial arrival time data for each velocity model is labeled to construct a training dataset; The initial neural network model is trained using the training dataset to obtain a qualified target neural network model, wherein the activation function of the last layer in the initial neural network model is the Tanh activation function. The actual initial arrival time data is input into the target neural network model to predict the speed model, thereby obtaining a predicted speed model corresponding to the actual initial arrival time data.

2. The method of determining a predictive velocity model according to claim 1, wherein, The initial arrival time data for each velocity model is labeled to construct a training dataset, including: Normalize the velocity model labels to obtain the corresponding first velocity model labels; The first arrival time data for each of the first velocity models are labeled to construct a training dataset.

3. The method for determining the predicted velocity model according to claim 1, characterized in that, The initial neural network model is trained using the training dataset to obtain a qualified target neural network model, including: The initial arrival time data from the training dataset is input into the initial neural network model to predict the velocity model and obtain the corresponding first predicted velocity model. The loss function algorithm is used to calculate the loss value for the first predicted velocity model and the velocity model label of the corresponding first arrival travel time data. Plot the corresponding loss function curves based on multiple loss values; When the trend of the loss function curve tends to be stable, the training is deemed successful, and the corresponding target neural network model is obtained.

4. The method for determining the predicted velocity model according to claim 1, characterized in that, The method further includes: Ray tracing forward modeling was performed on the predicted velocity model to obtain the corresponding predicted first arrival travel time data; The prediction accuracy of the target neural network model is determined based on the difference between the predicted initial arrival time data and the actual initial arrival time data.

5. The method for determining the predicted velocity model according to claim 4, characterized in that, The prediction accuracy of the target neural network model is determined based on the difference between the predicted initial arrival time data and the actual initial arrival time data, including: The difference between the predicted initial arrival time data and the actual initial arrival time data is compared with a preset time range to obtain the comparison result; Based on the comparison results, the prediction accuracy of the target neural network model is determined.

6. The method for determining the predicted velocity model according to claim 5, characterized in that, Based on the comparison results, the prediction performance of the target neural network model is determined, including: If the comparison result shows that the difference data is within a preset time range, it is determined that the prediction accuracy of the target neural network model meets the standard. If the comparison results show that there are differences in the difference data that are outside the preset time range, it is determined that the prediction accuracy of the target neural network model has not met the standard.

7. The method for determining the predicted velocity model according to claim 1, characterized in that, The determination of the initial neural network model includes: A batch normalization layer is set after the convolutional layer and the transposed convolutional layer of the neural network model to obtain the first neural network model; The activation functions of the convolutional and transposed convolutional layers of the first neural network model are determined to be LeakyReLU functions, and the activation function of the last layer is determined to be Tanh activation function, thus obtaining the initial neural network model.

8. A system for determining a predictive velocity model, characterized in that, The system includes: The velocity model label determination module is used to obtain multiple velocity model labels by controlling the vertical speed control line to slide up and down on the VSP zero-bias velocity model. The first arrival time data determination module is used to perform ray tracing forward modeling on the velocity model labels to obtain the corresponding first arrival time data; The training dataset determination module is used to label the first arrival time data corresponding to each velocity model label and construct the training dataset. The target neural network model determination module is used to train the initial neural network model using the training dataset to obtain a qualified target neural network model, wherein the activation function of the last layer in the initial neural network model is the Tanh activation function; The predicted speed model determination module is used to input the actual initial arrival time data into the target neural network model to predict the speed model and obtain the predicted speed model corresponding to the actual initial arrival time data.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps in the method for determining the predicted velocity model as described in any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for determining the predicted velocity model as described in any one of claims 1 to 7.

Images