Automatic velocity spectrum picking method based on mean shift clustering analysis

Abstract

The application provides a speed spectrum automatic picking method based on MeanShift clustering analysis, comprising the following steps: step 1, pre-processing artificial picking speed control points; step 2, training an initial constraint model of a work area speed; step 3, cleaning a speed spectrum according to the speed constraint model; step 4, picking a time-speed pair based on the MeanShift method; step 5, updating and optimizing the work area speed constraint model according to the speed picking result in step 4; step 6, repeating steps 3 to 5 until an iteration condition is reached; and step 7, outputting a speed spectrum automatic picking result. The speed spectrum automatic picking method based on MeanShift clustering analysis can effectively reduce the labor and time cost consumed in manual picking, and further improve the precision of speed automatic picking by introducing a three-dimensional speed field nonlinear multivariate regression model based on DNN as a constraint model for pre-processing the speed spectrum.

Description

Technical Field

[0001] This invention relates to the field of oilfield development technology, and in particular to an automatic velocity spectrum picking method based on MeanShift clustering analysis. Background Technology

[0002] Velocity is a crucial parameter in seismic wave propagation and a vital attribute for seismic exploration. To facilitate its application in various processing stages, exploration geophysicists have introduced different types of velocity concepts, such as layer velocity, apparent velocity, average velocity, root-mean-square velocity, instantaneous velocity, phase velocity, group velocity, and migration velocity. Velocity picking has always been a critical issue in processing, and velocity analysis is fundamental to conventional seismic data processing. Velocity analysis is multi-layered, including near-surface velocity analysis and refraction interface velocity analysis. The velocity spectrum is a key manifestation of velocity analysis, and picking stacking velocities is a crucial aspect. The accuracy of picked velocities directly impacts the effectiveness of dynamic correction, stacking, and even seismic data migration imaging, leading to inaccurate interpretation of seismic data. As the geological bodies encountered in current seismic exploration become increasingly complex, the requirements for imaging accuracy also increase, necessitating careful attention to velocity analysis. In the early stages of seismic data processing, suitable velocity models are generally unavailable, requiring conventional velocity analysis to obtain the necessary velocities.

[0003] Picking up effective velocity spectra is a crucial task in velocity analysis. Currently, velocity analysis methods based on common center point (CMP) gathers are the most widely used in practical processing. Spectrum-based velocity analysis is an important means of determining stacking velocities in seismic data processing and a fundamental tool for velocity analysis. This method picks up stacking velocities based on the strength of energy clusters in the velocity spectrum. Picking is usually done manually. Manual picking is relatively flexible because the human eye has a strong ability to identify velocity spectra with low focusing precision. However, manual picking also has certain shortcomings, mainly that it relies on the experience of the processor, which is currently the most labor-intensive step in conventional seismic data processing, resulting in low efficiency, high cost, and low accuracy. In this context, automated velocity picking methods have been continuously developed. The application of automated technology can not only solve the loss of effective qualitative information caused by partial picking but also improve work efficiency and save costs. With the deepening of seismic exploration and the development of computer science, research on automated picking has made significant progress. Several new velocity extraction methods have been proposed: intelligent velocity picking using iterative principles and layer tracking, automatic layer velocity picking using the Monte Carlo method, automatic velocity analysis of non-hyperbolic structures, automatic picking using the Viterbi algorithm, and so on. The implementation of these methods not only proves the possibility of intelligently picking and stacking velocities, but also improves work efficiency and reduces work costs, thus possessing certain economic value and practical significance.

[0004] Machine learning is a branch of computer science with wide applications across various fields. Deep learning, a subfield of machine learning, is one of the hottest areas in data science today, with many successful applications in robotics, image recognition, and artificial intelligence (AI) (Caté et al., 2017). In the oil and gas industry, machine learning and neural networks have also been developing for some time and can be viewed as a set of data analysis methods including classification, clustering, and regression (Hall, 2016). These methods can be used to discover features and trends in seismic data, supplemented by the automatic acquisition of velocity spectra.

[0005] Traditional velocity picking is primarily achieved manually, with accuracy heavily reliant on the experience of the processing personnel. However, with the increasing density and quantification of seismic acquisition data, the inefficiencies, high costs, and inconsistent accuracy of manual picking have become increasingly apparent. Existing automated velocity picking technologies, such as nonlinear optimization automatic picking and Monte Carlo automatic layer velocity picking, have achieved some success in acquiring velocities from relatively simple geological bodies. However, they still face the challenge of obtaining accurate solutions for velocity analysis of complex geological bodies.

[0006] Chinese patent application CN201910840514.0 discloses an automatic velocity spectrum picking method based on a convolutional neural network. The method describes the following implementation process: First, a velocity picking curve and a horizontal gather are randomly generated. Then, a reaction correction method is applied to generate a simulated gather, and a Radon transform is performed on this gather to generate a velocity spectrum. Repeating this process can generate any number of training labels. The invention designs a neural network with four convolutional layers and two fully connected layers. Additionally, maxpooling and dropout layers are added to reduce computation and prevent overfitting to some extent. Finally, the randomly generated labels are used to train the neural network, which is then used to identify velocity spectra not involved in the training. The advantages of this invention are: it can effectively pick up the positions of velocity energy clusters and generate corresponding velocity curves with high accuracy compared to theoretical curves; it also avoids the manual energy cluster picking process, saving manpower and mitigating errors introduced by human factors. The training and application objects of this patented technology are both numerical simulation seismic data, and the noise added is also simulated Gaussian random noise. It uses a trained convolutional neural network to directly predict and pick up velocity energy clusters.

[0007] Chinese patent application CN202011209040.9 discloses a method and apparatus for automatically picking seismic velocity spectra based on unsupervised learning. The method includes: acquiring a seismic stacking velocity spectrum; performing cropping processing on the acquired seismic stacking velocity spectrum according to the maximum and minimum reference stacking velocities of a preset velocity scanning corridor; normalizing and threshold filtering the cropped seismic stacking velocity spectrum to obtain a scattered seismic stacking velocity spectrum; and automatically picking the stacking velocity from the scattered seismic stacking velocity spectrum by sampling at each time point along the time direction, performing the following steps: constructing two overlapping sub-temporal windows within the main spatiotemporal window at each time sampling point; clustering the data within the two sub-temporal windows using a weighted K-means algorithm; and determining the cluster center of the main spatiotemporal window based on the cluster centers of the two sub-temporal windows within the main spatiotemporal window. This invention can adaptively determine the energy cluster centers of the velocity spectrum.

[0008] Chinese patent application CN202110284488.5 discloses a method for picking up accurate stacked velocity spectra, comprising the following steps: S1, determining the relevant stacked channel layer data, filtering the gather data according to the designed frequency division parameters, and extracting the channel set data volume at the velocity analysis point; S2, introducing layer data into different channel set data volumes at the velocity analysis point to obtain layer data on each channel set at the relevant velocity analysis point; S3, using the layer data and initial velocity spectrum on each channel set at the relevant velocity analysis point as constraints, iteratively calculating the relevant velocity spectrum on each channel set data volume at each velocity analysis point to obtain the stacked velocity spectrum at each velocity analysis point. This invention solves the problem of low accuracy in picking up velocity spectra in the prior art.

[0009] The existing technologies described above are significantly different from the present invention and have failed to solve the technical problem we want to address. Therefore, we have invented a new method for automatic velocity spectrum picking based on MeanShift clustering analysis. Summary of the Invention

[0010] The purpose of this invention is to provide an automatic velocity spectrum acquisition method based on MeanShift clustering analysis, addressing the shortcomings of existing velocity acquisition methods in terms of accuracy and applicability in meeting actual production needs.

[0011] The objective of this invention can be achieved through the following technical measures: an automatic velocity spectrum picking method based on MeanShift clustering analysis, which includes:

[0012] Step 1: Preprocess the manual pickup speed control points;

[0013] Step 2: Train the initial constraint model for the speed of the work area;

[0014] Step 3: Clean the velocity spectrum according to the velocity constraint model;

[0015] Step 4: Pick time-velocity pairs based on the MeanShift method;

[0016] Step 5: Update and optimize the work area speed constraint model based on the speed picking results in Step 4.

[0017] Step 6: Repeat steps 3 to 5 until the iteration condition is met;

[0018] Step 7: Output the automatic velocity spectrum picking results.

[0019] The objective of this invention can also be achieved through the following technical measures:

[0020] In step 1, manually picked velocity spectrum control point data are input and normalized to obtain preprocessed velocity spectrum control point data.

[0021] In step 1, the seismic data is processed to obtain the corresponding velocity spectrum. A certain number of manually picked control point time-velocity pairs are selected to form a feature vector. The feature vector uses geodetic coordinates X, Y, and two-way travel time T as input training data, and velocity V as a label. The vector is preprocessed to standardize it.

[0022] In step 2, the initial deep neural network is trained based on the preprocessed velocity spectrum control point data to obtain the trained initial constraint model for the work area velocity.

[0023] In step 2, a deep neural network is established, and the network is trained using the feature vectors from step 1. After training, the multivariate nonlinear regression of the velocity at each common depth point (CDP) is obtained. That is, the geodetic coordinates X, Y and the round-trip time T are input into the trained work area velocity constraint model to obtain the velocity V at that spatial location and predict the initial velocity field.

[0024] In step 3, the velocity values ​​in the work area velocity constraint model are used as the reference values, and the upper and lower limit ranges of the values ​​are set as conditions to clean the velocity spectrum of the work area. Data outside the range is set to zero, and data within the range is kept at their original values.

[0025] In step 4, the cleaned velocity spectrum is subjected to MeanShift clustering analysis, and the center point of the cluster is used as the value of the picked time-velocity pair.

[0026] In step 4, the MeanShift vector is the sum and average of the offset vectors of all sample points in the specified space relative to the specified point. This means that the MeanShift vector is the sum of the offset vectors of all sample points x in the specified space. i The MeanShift vector M at point x, where i = 1, ..., n. h (x) can be defined as:

[0027]

[0028] Where h is the radius of the specified space, w(x) i ≥0 is an assignment to the sampling point x i The weight of is inversely proportional to the offset from point x, and G is a unit uniform kernel function, which can be specifically expressed as:

[0029]

[0030] Since a non-zero probability density gradient points in the direction of maximum increase in probability density, on average, sample points in a given space tend to fall more along the direction of the probability density gradient; therefore, the corresponding MeanShift vector should point in the direction of the probability density gradient.

[0031] Let m be the first term on the right side of expression 1. h (x), that is

[0032]

[0033] Given an initial point x, a kernel function G, and a tolerance error ε, the MeanShift algorithm iteratively executes the following three steps until the termination condition is met:

[0034] a) Calculate m h (x);

[0035] b) Put m h (x) is assigned to x;

[0036] c) If ||m h If (x)-x||<ε, the loop ends; otherwise, continue executing a).

[0037] Because m h (x)=x+M h (x), so the point moves continuously along the gradient direction of the probability density. The step size is not only related to the magnitude of the gradient, but also to the probability density of the point. Where the probability density is high, the MeanShift algorithm calculates that the step size should be smaller, and vice versa. Finally, the MeanShift algorithm will converge to the peak near the point.

[0038] In step 5, the input parameters are geodetic coordinates and two-way travel time, and the output parameter is speed. Based on the speed automatically picked up in step 4, a deep neural network is used to train and optimize the work area speed constraint model.

[0039] In step 6, steps 3 to 5 are repeated, and the work area velocity prediction model is iterated based on the new velocity spectrum picking results. The stopping condition for the iteration is that the root mean square error of both the velocity spectrum picking results before and after the iteration and the work area velocity constraint model is less than the set threshold.

[0040] In step 7, after exiting the iteration loop, the final velocity spectrum auto-picking result is output.

[0041] This invention presents an automatic velocity spectrum picking method based on MeanShift clustering analysis. Both the training and application objects are real, acquired seismic data. A deep neural network (DNN) is used to predict the three-dimensional velocity field and constrain and clean the velocity spectrum accordingly. Then, the MeanShift clustering method is used to achieve high-precision automatic picking of time-velocity pairs on the cleaned velocity spectrum. DNN is a supervised learning algorithm that requires training data to learn the relationships between observations (or features). The MeanShift algorithm, on the other hand, is an unsupervised learning algorithm, a vector quantization method, and a widely used clustering analysis method in data mining. Therefore, this invention employs a novel automatic velocity picking method based on supervised and unsupervised learning. The three-dimensional velocity field predicted by the deep neural network constrains the velocity spectrum, and the time-velocity pairs on the velocity spectrum are automatically picked based on clustering analysis. This constraint-then-picking approach effectively improves the accuracy of automatic picking when applied to actual seismic data. The advanced nature of deep learning algorithms can further ensure the robustness and practicality of the method of this invention. At the same time, this constraint-then-pickup mode can effectively accelerate the convergence of the picking process, reduce the error rate, and provide effective technical support and data guarantee for subsequent imaging and other processing processes.

[0042] Compared with existing speed acquisition technologies, this invention has three main advantages:

[0043] (1) A new method for automatic speed picking based on machine learning algorithm is provided, which can effectively reduce the labor and time costs of manual picking;

[0044] (2) The clustering analysis algorithm based on machine learning is suitable for the analysis and application of big data, which meets the current trend and demand for increasingly high-density earthquake data and is easy to promote and apply.

[0045] (3) Introducing a deep neural network to predict the three-dimensional velocity field and using it to clean and optimize the velocity spectrum can effectively accelerate the convergence process of the MeanShift algorithm and further improve the accuracy of automatic velocity picking. Attached Figure Description

[0046] Figure 1 This is a flowchart of a specific embodiment of the automatic velocity spectrum picking method based on MeanShift clustering analysis of the present invention;

[0047] Figure 2 This is a time-velocity overlay display diagram of manually picked control points in one embodiment of the present invention;

[0048] Figure 3 This is a schematic diagram of a deep neural network architecture in one embodiment of the present invention;

[0049] Figure 4 This is a slice of the predicted initial velocity field in one embodiment of the present invention;

[0050] Figure 5 This is a comparison of velocity spectra before and after cleaning in one embodiment of the present invention;

[0051] Figure 6 This is the MeanShift auto-picking result of the velocity spectrum after cleaning in one embodiment of the present invention;

[0052] Figure 7 This is a diagram illustrating the motion correction effect based on MeanShift automatic picking in one embodiment of the present invention. Detailed Implementation

[0053] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0054] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, and / or combinations thereof.

[0055] The automatic velocity spectrum picking method based on MeanShift clustering analysis of the present invention mainly includes the following steps:

[0056] (1) Preprocessing manual pickup speed control points;

[0057] (2) Training the initial speed constraint model for the work area;

[0058] (3) Clean the velocity spectrum according to the velocity constraint model;

[0059] (4) Automatically pick time-velocity pairs based on the MeanShift method;

[0060] (5) Update and optimize the work area speed constraint model based on the speed picking results in (4);

[0061] (6) Repeat steps (3)-(5) until the iteration condition is met;

[0062] (7) Output the automatic velocity spectrum picking results.

[0063] This invention provides a novel method for automatic velocity picking based on machine learning algorithms. It enables automatic velocity spectrum picking based on MeanShift clustering analysis, effectively reducing the manual labor and time costs associated with manual picking. Furthermore, it introduces a three-dimensional nonlinear multiple regression model of velocity field based on DNN as a constraint model for preprocessing the velocity spectrum, overcoming the difficulties in selecting cluster centers during MeanShift clustering and further improving the accuracy of automatic velocity picking.

[0064] The following are several specific embodiments of the application of the present invention.

[0065] Example 1

[0066] In a specific embodiment of the present invention, the automatic velocity spectrum picking method based on MeanShift clustering analysis of the present invention specifically includes the following steps:

[0067] (1) Input manually picked velocity spectrum control point data and normalize it to obtain preprocessed velocity spectrum control point data;

[0068] (2) Based on the preprocessed velocity spectrum control point data, the initial deep neural network is trained to obtain the trained initial constraint model of the work area velocity.

[0069] (3) Using the velocity values ​​in the work area velocity constraint model as the benchmark values, and setting the upper and lower limits of the numerical range as conditions, the velocity spectrum of the work area is cleaned. Data outside the range is set to zero, and data within the range is kept to its original value.

[0070] (4) Perform mean-shift clustering analysis on the cleaned velocity spectrum, and use the center point of the cluster as the value of the time-velocity pair picked;

[0071] (5) The input parameters are geodetic coordinates and two-way travel time, and the output parameter is speed. Based on the speed automatically picked up in step (4), a deep neural network is used to train and optimize the work area speed constraint model.

[0072] (6) Repeat steps (3)-(5) and iterate the work area speed prediction model based on the new velocity spectrum picking results. The stopping condition for the iteration is that the root mean square error of the velocity spectrum picking results before and after the iteration and the work area speed constraint model are both less than the set threshold.

[0073] (7) After exiting the iteration loop, output the final velocity spectrum automatically picked result.

[0074] Example 2

[0075] In a specific embodiment 2 of the present invention, the automatic velocity spectrum picking method based on MeanShift clustering analysis specifically includes the following steps:

[0076] (1) Perform necessary processing on the seismic data to obtain the corresponding velocity spectrum, and select a certain number of manually picked control points with time-velocity pairs (e.g., ...). Figure 2 As shown), the feature vector is formed by using the geodetic coordinates X, Y, and round-trip travel time T as input training data, and the speed V as the label. The vector is preprocessed to standardize it.

[0077] (2) Establish a deep neural network. The network architecture diagram is shown below. Figure 3 As shown, the feature vectors from step (1) are used to train the network. After training, the multivariate nonlinear regression of the velocity at each CDP point is obtained. That is, the geodetic coordinates X, Y and the two-way travel time T are input into the trained work area velocity constraint model to obtain the velocity V at that spatial location. The initial velocity field is predicted as follows. Figure 4 As shown;

[0078] (3) The predicted output value of the work area velocity constraint model is used as the initial constraint value of the work area velocity. Using this initial constraint value as the benchmark value, and setting a range not exceeding ±15% of the upper and lower limits of the benchmark value as a condition, the velocity spectrum of the work area is cleaned. Data outside the range is set to zero, while data within the range is kept unchanged. A comparison of the velocity spectra before and after cleaning is shown below. Figure 5 As shown.

[0079] (4) Perform mean-shift clustering analysis on the velocity spectrum obtained in step (3). The coordinates of the cluster center point correspond to the automatically picked time-velocity pairs. The automatically picked results after cleaning are as follows: Figure 6 As shown; the MeanShift vector is the sum and average of the offset vectors of all sample points in the specified space relative to the specified point, and is the result of considering n sample points x in the specified space. i The MeanShift vector M at point x, where i = 1, ..., n. h (x) can be defined as:

[0080]

[0081] Where h is the radius of the specified space, w(x) i ≥0 is an assignment to the sampling point x i The weight of is inversely proportional to the offset from point x, and G is a unit uniform kernel function, which can be specifically expressed as:

[0082]

[0083] Since a non-zero probability density gradient points in the direction of maximum increase in probability density, on average, sample points in a given space tend to fall more along the direction of the probability density gradient. Therefore, the corresponding MeanShift vector should point in the direction of the probability density gradient.

[0084] Let m be the first term on the right side of expression 1. h (x), that is

[0085]

[0086] Given an initial point x, a kernel function G, and a tolerance error ε, the MeanShift algorithm iteratively executes the following three steps until the termination condition is met:

[0087] d) Calculate m h (x);

[0088] e) put m h (x) is assigned to x;

[0089] f) If ||m h (x)-x||<ε, the loop ends; otherwise, continue executing a).

[0090] Because m h (x)=x+M h (x), therefore, the point continuously moves along the gradient direction of the probability density. The step size depends not only on the magnitude of the gradient but also on the probability density at that point. Where the probability density is high, the MeanShift algorithm calculates a smaller step size, and vice versa. Ultimately, the MeanShift algorithm converges to the peak near that point.

[0091] (5) Using geodetic coordinates and two-way travel time as inputs and speed as output, a deep neural network is used to train the optimized and updated work area speed constraint model based on the speed automatically picked up in step (4).

[0092] (6) Repeat steps (3)-(5) and iterate the work area speed prediction model based on the new velocity spectrum picking results. The stopping condition for the iteration is that the root mean square error of the velocity spectrum picking results and the work area speed constraint model before and after the iteration is less than the set threshold.

[0093] (7) Output the final velocity spectrum auto-picking result, and the dynamic correction effect is as follows: Figure 7 As shown.

[0094] Example 3:

[0095] In a specific embodiment 3 of the present invention, the automatic velocity spectrum picking method based on MeanShift clustering analysis of the present invention specifically includes the following steps:

[0096] (1) Input manually picked velocity spectrum control point data and normalize it to obtain preprocessed velocity spectrum control point data;

[0097] (2) Based on the preprocessed velocity spectrum control point data, the initial deep neural network is trained to obtain the trained initial constraint model of the work area velocity.

[0098] (3) Using the velocity values ​​in the work area velocity constraint model as the benchmark values, and setting the upper and lower limits of the numerical range as conditions, the velocity spectrum of the work area is cleaned. Data outside the range is set to zero, and data within the range is kept to its original value.

[0099] (4) Perform mean-shift clustering analysis on the cleaned velocity spectrum, and use the center point of the cluster as the value of the time-velocity pair picked;

[0100] (5) The input parameters are geodetic coordinates and two-way travel time, and the output parameter is speed. Based on the speed automatically picked in step (4), a deep neural network is used to train and optimize the speed constraint model of the work area; the output speed spectrum is automatically picked.

[0101] This invention introduces machine learning methods to establish an interdisciplinary solution suitable for automatic velocity spectrum picking, reducing manual costs and improving picking efficiency. It fully combines supervised and unsupervised learning methods, realizing velocity spectrum cleaning constraints under supervised learning and, on this basis, realizing automatic velocity spectrum picking under unsupervised learning, thus improving the overall robustness and accuracy of the method.

[0102] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0103] Except for the technical features described in the specification, all other technologies are known to those skilled in the art.

Claims

1. An automatic velocity spectrum picking method based on MeanShift clustering analysis, characterized in that, This automatic velocity spectrum picking method based on MeanShift clustering analysis includes: Step 1: Preprocess the manually picked speed control point data; Step 2: Train the initial constraint model for the speed of the work area; Step 3: Clean the velocity spectrum according to the velocity constraint model; Step 4: Pick time-velocity pairs based on the MeanShift method; Step 5: Update and optimize the work area speed constraint model based on the speed picking results in Step 4. Step 6: Repeat steps 3 to 5 until the iteration stopping condition is met; Step 7: Output the automatic velocity spectrum picking results.

2. The automatic velocity spectrum picking method based on MeanShift clustering analysis according to claim 1, characterized in that, In step 1, the manually picked speed control point data is normalized to obtain preprocessed speed control point data.

3. The automatic velocity spectrum picking method based on MeanShift clustering analysis according to claim 2, characterized in that, Step 1 also includes: processing the seismic data to obtain the corresponding velocity spectrum, selecting a certain number of manually picked control points with time-velocity pairs to form a feature vector, the feature vector containing the geodetic coordinates X, Y and two-way travel time T of the time-velocity pair as input training data, using velocity V as the label, and standardizing the feature vector.

4. The automatic velocity spectrum picking method based on MeanShift clustering analysis according to claim 3, characterized in that, In step 2, a deep neural network is used to train the initial constraint model of the work area speed using the preprocessed speed control point data as training data.

5. The automatic velocity spectrum picking method based on MeanShift clustering analysis according to claim 4, characterized in that, The network is trained using the aforementioned feature vectors to achieve multivariate nonlinear regression of the velocity at each common depth point (CDP) and generate an initial velocity field.

6. The automatic velocity spectrum picking method based on MeanShift clustering analysis according to claim 1, characterized in that, In step 3, the velocity values ​​of the work area velocity constraint model are used as the benchmark, and the upper and lower limit ranges of the values ​​are set. The velocity spectrum data outside the range is set to zero, while the velocity spectrum data within the range is retained.

7. The automatic velocity spectrum picking method based on MeanShift clustering analysis according to claim 1, characterized in that, In step 4, MeanShift clustering analysis is performed on the cleaned velocity spectrum, and the center point of the cluster is used as the picked time-velocity pair.

8. The automatic velocity spectrum picking method based on MeanShift clustering analysis according to claim 1, characterized in that, The execution process of the MeanShift method includes: (Equation 1) Where h is the radius of the specified space, It is an assignment to the sampling point The weight of is inversely proportional to the offset from point x, and G is a unit uniform kernel function, which can be specifically expressed as: (Equation 2) The corresponding MeanShift vector should point in the direction of the probability density gradient; The first term on the right side of expression 1 is ,Right now (Equation 3) Given an initial point x, a kernel function G, and a tolerance error The MeanShift algorithm iteratively executes the following three steps until the termination condition is met: a) Calculation ; b) Put Assigned ; c) If If the loop terminates, then execute a); otherwise, continue executing a). because Therefore, the point moves continuously along the gradient direction of the probability density. The step size is not only related to the magnitude of the gradient, but also to the probability density of the point. Where the probability density is high, the MeanShift algorithm calculates that the step size should be smaller, and vice versa. Finally, the MeanShift algorithm will converge to the peak near the point.

9. The automatic velocity spectrum picking method based on MeanShift clustering analysis according to claim 1, characterized in that, In step 5, the geodetic coordinates and round-trip travel time are used as input parameters, and the speed is used as the output parameter. The speed data automatically picked up in step 4 is used as training data, and the work area speed constraint model is updated and optimized through deep neural network training.

10. The automatic velocity spectrum picking method based on MeanShift clustering analysis according to claim 1, characterized in that, In step 6, the iteration stopping condition is that the root mean square error of the velocity spectrum picking results before and after the iteration and the velocity constraint model of the work area are both less than a set threshold.

11. The automatic velocity spectrum picking method based on MeanShift clustering analysis according to claim 1, characterized in that, In step 7, after exiting the iteration loop, the final velocity spectrum auto-picking result is output.

Images