Machine learning based seismic event identification method, apparatus, device and medium

Abstract

The present application relates to the field of seismic exploration, and more particularly to a method and device for identifying seismic events based on machine learning, equipment and medium, which comprises: obtaining seismic signals, using a thin interlayer velocity model to perform forward numerical simulation on the seismic signals to obtain training data to be processed; performing single-channel segmentation and data reorganization on the training data to be processed to obtain training data; performing feature processing on the training data to obtain a feature vector; performing feature screening on the feature vector to obtain a screened feature vector; using gradient boosting decision tree learning to learn the screened feature vector to obtain a seismic event identification model; obtaining test data, and using the seismic event identification model to identify the test data to obtain seismic event data. The present application can accurately and quickly extract seismic event data from a large number of seismic signals, improving the processing efficiency and accuracy of the seismic event identification model.

Description

Technical Field

[0001] This invention relates to the field of seismic exploration, and in particular to a method, apparatus, equipment, and medium for identifying seismic phase axes based on machine learning. Background Technology

[0002] Seismic exploration is an important technique in Earth science, typically used to study the structure and properties of the Earth's interior. By analyzing the propagation and reflection characteristics of seismic waves within the Earth's crust, scientists can reveal crucial information such as Earth's structure, rock layer distribution, and crustal movements.

[0003] In the field of seismic exploration, automatic extraction of seismic phase axes is a crucial step in data processing. However, in this step, because seismic phase axis signals are usually weak and easily masked by background noise, it is difficult to accurately detect the first arrival of the signal, and this also poses a challenge to determining the precise location of the signal. Therefore, how to quickly and accurately identify seismic phase axis data in large amounts of seismic data has become an urgent problem to be solved. Summary of the Invention

[0004] To address the aforementioned problems, embodiments of the present invention provide a method, apparatus, device, and medium for identifying seismic phase axes based on machine learning.

[0005] In a first aspect, embodiments of the present invention provide a machine learning-based method for identifying seismic phase axes, comprising:

[0006] Seismic signals are acquired, and forward numerical simulations are performed on the seismic signals using a thin interlayer velocity model to obtain training data to be processed.

[0007] The training data to be processed is segmented into single channels and reassembled to obtain training data.

[0008] The training data is subjected to feature processing to obtain a feature vector;

[0009] The feature vector is subjected to feature filtering to obtain the filtered feature vector;

[0010] The selected feature vectors are learned using the gradient boosting decision tree learning method to obtain the seismic phase axis identification model;

[0011] The test data is acquired, and the earthquake phase axis identification model is used to identify the test data to obtain earthquake phase axis data.

[0012] According to an embodiment of the present invention, the seismic signal is subjected to forward numerical simulation using a thin interlayer velocity model to obtain training data to be processed, including:

[0013] The wavefield is obtained by calculating the seismic signal using the finite difference method and the thin interlayer velocity model.

[0014] Position information is extracted from the wave field to obtain forward-modeled synthetic shot record data;

[0015] The forward modeling composite gun recording data is divided into preset values ​​to obtain training data.

[0016] According to an embodiment of the present invention, the step of performing single-channel segmentation and data recombination on the training data to be processed to obtain training data includes:

[0017] The training data is segmented to obtain segmented data;

[0018] The segmented data is divided equally to obtain reconstructed data;

[0019] Noise is added to the reconstructed data to obtain training data.

[0020] According to an embodiment of the present invention, the step of performing feature processing on the training data to obtain a feature vector includes:

[0021] Single-channel feature extraction is performed on the training data to obtain single-number feature data;

[0022] Gray-level co-occurrence matrix features are extracted from the training data to obtain gray-level co-occurrence matrix feature data;

[0023] Inter-track feature extraction is performed on the training data to obtain inter-track feature data;

[0024] Feature vectors are constructed from the gray-level co-occurrence matrix feature data, the single-channel feature extraction, and the inter-channel feature data to obtain feature data.

[0025] The feature data is standardized to obtain a feature vector.

[0026] According to an embodiment of the present invention, the step of performing feature filtering on the feature vector to obtain a filtered feature vector includes:

[0027] The eigenvectors were subjected to univariate feature analysis using analysis of variance to obtain univariate feature vectors.

[0028] The univariate feature vectors are selected using the external model of the support vector machine to obtain the filtered feature vectors.

[0029] According to an embodiment of the present invention, the step of performing univariate feature analysis on the feature vector using analysis of variance to obtain a univariate feature vector includes:

[0030] Calculate the sum of squared deviations between the eigenvectors to obtain the vector deviation;

[0031] The training bias is obtained by calculating the sum of squared deviations of the feature vectors.

[0032] Based on the training bias and the vector bias, the test statistic of each feature vector is calculated.

[0033] The test statistics are sorted in descending order, and the features corresponding to the test statistics that meet the preset conditions are selected as univariate feature vectors.

[0034] According to an embodiment of the present invention, the step of using a support vector machine external model to perform feature selection on the univariate feature vector to obtain a filtered feature vector includes:

[0035] The importance score of each univariate feature is calculated using a pre-defined support vector external model.

[0036] Univariate features are filtered out based on importance scores to obtain the selected feature vector.

[0037] Secondly, embodiments of the present invention provide a seismic phase axis identification device based on machine learning, characterized in that it includes:

[0038] The data acquisition module is used to acquire seismic signals and perform forward numerical simulation on the seismic signals using a thin interlayer velocity model to obtain training data to be processed.

[0039] The data processing module is used to perform single-channel segmentation and data recombination on the training data to be processed to obtain training data;

[0040] The feature processing module is used to perform feature processing on the training data to obtain feature vectors;

[0041] The feature filtering module is used to filter the feature vector to obtain the filtered feature vector;

[0042] The model training module is used to learn the selected feature vectors using the gradient boosting decision tree learning method to obtain the earthquake phase axis identification model.

[0043] The model application module is used to acquire test data and use the seismic phase axis identification model to identify the test data to obtain seismic phase axis data.

[0044] Thirdly, embodiments of the present invention provide an electronic device, which includes:

[0045] processor;

[0046] Memory used to store the processor's executable instructions;

[0047] The processor is configured to execute the instructions to implement the steps of a machine learning-based method for identifying seismic phase axes as described in the first aspect above.

[0048] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a machine learning-based method for identifying seismic phase axes as described in the first aspect above.

[0049] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial effects:

[0050] This invention, through data segmentation of training data to obtain segmented data, equal distribution of the segmented data to obtain reconstructed data, and noise addition to the reconstructed data to obtain training data, can reduce the dimensionality of the obtained training data and improve the recognition efficiency of the seismic phase axis identification model. By acquiring a decision tree model, initializing the decision tree model to obtain a strong learner, calculating the negative gradient based on the strong learner and selected feature vectors, training a weak learner based on the negative gradient, and training and optimizing the strong learner based on the weak learner, strong learner, and selected feature vectors, the residual can be gradually reduced to decrease the model's variance, thereby reducing the risk of overfitting. This allows the obtained seismic phase axis identification model to quickly process low signal-to-noise ratio data or seismic signals with phase axes obscured by background noise, improving the processing efficiency and accuracy of the seismic phase axis identification model. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 The flowchart of the seismic phase axis identification method based on machine learning according to Embodiment 1 of the present invention is shown.

[0053] Figure 2 The thin interlayer velocity model used for forward modeling in Embodiment 1 of the present invention is shown;

[0054] Figure 3 This shows a composite gun recording data graph obtained through forward modeling in Embodiment 1 of the present invention;

[0055] Figure 4The image shows the training data obtained by performing single-channel segmentation and data recombination according to Embodiment 1 of the present invention.

[0056] Figure 5 This shows a univariate feature analysis diagram obtained using analysis of variance in Embodiment 1 of the present invention;

[0057] Figure 6 This shows a schematic diagram of the external model of the support vector machine according to Embodiment 1 of the present invention;

[0058] Figure 7 The diagram shows test data used for testing and optimizing the model according to Embodiment 1 of the present invention.

[0059] Figure 8 This shows a graph of seismic phase axis data obtained by verifying the test data of Embodiment 1 of the present invention through the seismic phase axis identification model;

[0060] Figure 9 This shows a comparison diagram of single-shot records from Embodiment 1 of the present invention;

[0061] Figure 10 This shows a functional block diagram of the seismic phase axis identification device based on machine learning according to Embodiment 3 of the present invention;

[0062] Figure 11 The diagram shows the composition of an electronic device that implements the machine learning-based seismic phase axis identification method according to Embodiment 5 of the present invention. Detailed Implementation

[0063] The present disclosure will be further described below with reference to the embodiments shown in the accompanying drawings.

[0064] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0065] This invention proposes a machine learning-based method for identifying seismic phase axes. Based on forward modeling theory and combined with feature selection methods, a seismic phase axis identification model is built to identify seismic phase axis data. Compared with traditional methods, deep learning-based seismic facies analysis technology is more efficient and reduces human subjectivity, showing great potential and application prospects in the field of seismic exploration.

[0066] Example 1

[0067] like Figure 1 As shown, this invention proposes a machine learning-based method for identifying seismic phase axes, comprising the following steps:

[0068] S1. Acquire seismic signals and perform forward numerical simulation on the seismic signals using a thin interlayer velocity model to obtain training data to be processed.

[0069] In this embodiment of the invention, the seismic signal is subjected to forward numerical simulation using a thin interlayer velocity model to generate training data to be processed, which can be used to train monitoring and prediction models. This enriches the training data dataset and makes the obtained training data more reliable.

[0070] In one optional embodiment of the present invention, forward numerical simulation can be performed using the finite difference method. When using the finite difference method, its basic idea is relatively simple and easy to implement. The mesh size and time step can be flexibly adjusted to balance numerical stability and simulation accuracy, thus resulting in higher accuracy of the training data to be processed. In another optional embodiment of the present invention, forward numerical simulation can also be performed using the finite element method. When using the finite element method, its matrix representation allows for a very simple description of the problem, making the solution method faster and thus improving the efficiency of obtaining the training data to be processed.

[0071] In this embodiment of the invention, the seismic signal is subjected to forward numerical simulation using a thin interlayer velocity model to obtain training data to be processed, including:

[0072] The wavefield is obtained by calculating the seismic signal using the finite difference method and the thin interlayer velocity model.

[0073] Position information is extracted from the wave field to obtain forward-modeled synthetic shot record data;

[0074] The forward modeling composite gun recording data is divided into preset values ​​to obtain training data.

[0075] See Figure 2 As shown, this is the thin interbedded velocity model used for forward modeling. The thin interbedded layer refers to the repeated occurrence of two types of rock layers. The thin interbedded velocity model is a discrete grid structure that decomposes the underground rock layers into a series of thin layers and assigns a velocity value to each layer. The middle 10 layers are low- to high-velocity thin interbedded bodies.

[0076] See Figure 3 As shown, this is a composite gun recording data graph obtained through forward modeling. The composite gun recording data graph can provide more richness and diversity to the training data, so as to cover more different situations and thus improve the robustness of the trained model.

[0077] In this embodiment of the invention, performing forward numerical simulation on the designed seismic signal can obtain more accurate data. By dividing the data into signal and noise data using preset values, the noise data in the training set can be excluded, thereby reducing the model's learning and adaptation to noise, simplifying the model training process, and improving training efficiency.

[0078] S2. Perform single-channel segmentation and data recombination on the training data to be processed to obtain training data.

[0079] In this embodiment of the invention, by performing single-channel segmentation and data recombination on the training data to be processed, the dimensionality of the obtained training data can be reduced, thereby improving the efficiency of subsequent feature processing.

[0080] In this embodiment of the invention, the step of performing single-channel segmentation and data recombination on the training data to be processed to obtain training data includes:

[0081] The training data is segmented to obtain segmented data;

[0082] The segmented data is divided equally to obtain reconstructed data;

[0083] Noise is added to the reconstructed data to obtain training data.

[0084] See Figure 4 The image shown is a training data graph used in this invention. By segmenting, reorganizing, and adding random noise to the training data to be processed, data augmentation can be performed on the training data to be processed, preventing the risk of data overfitting.

[0085] In this embodiment of the invention, a threshold-based segmentation method can be used for data segmentation. When using a threshold-based segmentation method for segmentation, the accuracy of the segmented data is higher because the threshold-based segmentation method can segment the training data based on specific features.

[0086] In this embodiment of the invention, by performing single-channel segmentation and data recombination on the training data to be processed, the dimensionality and complexity of the data can be reduced, the computational efficiency can be improved, and the risk of overfitting can be reduced.

[0087] S3. Perform feature processing on the training data to obtain feature vectors.

[0088] In this embodiment of the invention, by performing feature processing on the data, the distribution of the data can be adjusted and key information of the data can be captured, thereby ensuring that subsequent model training and analysis can be carried out effectively.

[0089] In this embodiment of the invention, the min-max normalization method can be used for feature processing. When the min-max normalization method is used for feature processing, since the min-max normalization method maps the training data to the range between 0 and 1, the method is simple and therefore the resulting feature vector is more efficient.

[0090] In this embodiment of the invention, the step of performing feature processing on the training data to obtain a feature vector includes:

[0091] Single-channel feature extraction is performed on the training data to obtain single-number feature data;

[0092] Gray-level co-occurrence matrix features are extracted from the training data to obtain gray-level co-occurrence matrix feature data;

[0093] Inter-track feature extraction is performed on the training data to obtain inter-track feature data;

[0094] Feature vectors are constructed from the gray-level co-occurrence matrix feature data, the single-channel feature extraction, and the inter-channel feature data to obtain feature data.

[0095] The feature data is standardized to obtain a feature vector.

[0096] In this embodiment of the invention, the feature data can be standardized using the following formula:

[0097]

[0098] Where x represents the feature vector, x' represents the feature vector, min(x) represents the minimum value in the feature data, and max(x) represents the maximum value in the feature data.

[0099] In this embodiment of the invention, referring to Appendix Table 1, the preset feature table includes 191 features, including 63 single-channel features such as mean, energy, skewness and kurtosis, and 128 inter-channel features, gray-level co-occurrence matrix contrast, correlation, energy and homogeneity.

[0100] Appendix 1

[0101]

[0102]

[0103] In this embodiment of the invention, by constructing new features, the nonlinear relationship in the original training data can be transformed into a linear relationship, thereby enabling the subsequent model to better fit the data. Standardizing the feature data can ensure that the data is within a uniform distribution range, thereby reducing the training error of the subsequent model.

[0104] S4. Perform feature filtering on the feature vector to obtain the filtered feature vector.

[0105] In this embodiment of the invention, feature filtering can be performed on the feature vector to extract features, reduce the dimension of the feature space, remove irrelevant or redundant features, improve data utilization efficiency and the efficiency of subsequent model training.

[0106] In this embodiment of the invention, the step of performing feature filtering on the feature vector to obtain a filtered feature vector includes:

[0107] The eigenvectors were subjected to univariate feature analysis using analysis of variance to obtain univariate feature vectors.

[0108] The univariate feature vectors are selected using the external model of the support vector machine to obtain the filtered feature vectors.

[0109] In this embodiment of the invention, the external support vector machine model is a supervised learning algorithm for classification, including a mathematical expression of the hyperplane and a constraint mathematical expression. In this invention, it is used to perform a regression task on the univariate feature vectors to obtain the contribution value of each univariate feature, and to filter based on the contribution value to obtain filtered feature vectors. In this invention, the feature vectors are assigned corresponding weights by the recursive external support vector machine model, and features are selected from the recursively reduced feature set to obtain filtered feature vectors.

[0110] In this embodiment of the invention, the step of performing univariate feature analysis on the feature vector using analysis of variance to obtain a univariate feature vector includes:

[0111] Calculate the sum of squared deviations between the eigenvectors to obtain the vector deviation;

[0112] The training bias is obtained by calculating the sum of squared deviations of the feature vectors.

[0113] Based on the training bias and the vector bias, the test statistic of each feature vector is calculated.

[0114] The test statistics are sorted in descending order, and the features corresponding to the test statistics that meet the preset conditions are selected as univariate feature vectors.

[0115] In this embodiment of the invention, the test statistic can be calculated using the following formula:

[0116]

[0117] Where F represents the test characteristic statistic, S A S represents the deviation of the first A vectors.E The first E training biases are represented by r, the total number of feature vectors, and n is a preset fixed constant.

[0118] See appendix Figure 5 The diagram shown illustrates univariate feature analysis, illustrating the scores of univariate feature selection. Analysis of variance (ANOVA) is used to calculate the F-statistic, enabling univariate feature selection. The resulting features are then dimensionality-reduced to decrease the probability of overfitting. By calculating the F-values ​​among the features, the feature vectors with the highest F-values ​​are selected as the univariate feature vectors.

[0119] In this embodiment of the invention, analysis of variance (ANOVA) can be used for univariate feature analysis. Since ANOVA is based on whether different levels of feature vectors have different effects on classification, ANOVA can accurately determine whether each feature vector has a significant impact on the classification accuracy of the training data, thereby achieving the purpose of analyzing univariate feature vectors. Therefore, the univariate feature vectors obtained by ANOVA are more accurate.

[0120] In this embodiment of the invention, the step of using a support vector machine external model to perform feature selection on the univariate feature vector to obtain a filtered feature vector includes:

[0121] The importance score of each univariate feature is calculated using a pre-defined support vector external model.

[0122] Univariate features are filtered out based on importance scores to obtain the selected feature vector.

[0123] In this embodiment of the invention, the basic formula of the support vector machine is:

[0124]

[0125] y i (w T x i +b)≥1-ξ i

[0126] ξ i ≥0, i=1,2,…,l

[0127] Where w is the importance score of the univariate feature, T is the transpose operation, C is the first preset constant, and ξ i Let y be the slack variable of the i-th univariate feature. i x is the first preset label for the i-th univariate feature. i Let b be the i-th univariate feature, and b be the preset error amount. The decision condition for the support vector machine is, when... When the time is reached, optimization stops, where l is the number of the univariate features.

[0128] See appendix Figure 6 The diagram illustrates the external model of a Support Vector Machine (SVM). The external SVM calculates the contribution weight of each univariate feature vector and simulates a classification hyperplane based on the univariate features (i.e., γ = 0). Based on the classification hyperplane and the weight of each contribution value, the univariate features furthest from the classification hyperplane are filtered out, thus achieving feature selection. In the diagram, x1 and x2 are the univariate features, forming a two-dimensional plane. Each univariate feature can be considered a point in space. In the diagram, + represents the positive class, and - represents the negative class. The positive and negative classes are the preset labels. The solid line in the diagram represents the classification hyperplane. γ represents the margin, and the dashed line represents the margin boundary. Based on the margin boundary, the hyperplane can classify all selected feature vectors. In this embodiment of the invention, by using analysis of variance to obtain univariate features and using the external SVM model to perform feature selection on the univariate feature vectors, irrelevant or redundant features are removed, improving the accuracy of the selected feature vectors and the data utilization rate.

[0129] S5. The selected feature vectors are learned using the gradient boosting decision tree learning method to obtain the seismic phase axis identification model.

[0130] In this embodiment of the invention, the gradient boosting decision tree learning method is used to learn the training data, which enables the obtained seismic phase axis identification model to adaptively handle the problem of seismic data being masked due to weak signals, thereby improving the identification accuracy of the seismic phase axis identification model.

[0131] In this embodiment of the invention, the step of learning the training data using the gradient boosting decision tree learning method to obtain the seismic phase axis identification model includes:

[0132] Obtain the decision tree model, initialize the parameters based on the decision tree model, and obtain a strong learner;

[0133] A function is constructed on the strong learner and the training data to obtain the strong loss function of the strong learner;

[0134] The negative gradient of the training data is calculated based on the strong loss function and the strong prediction value.

[0135] The negative gradient is used as the target label, and a weak learner is trained based on the target label.

[0136] Calculate the weak prediction value of the weak learner, and add the weighted weak prediction value to the prediction value of the strong learner to obtain a new strong learner;

[0137] Construct the loss function for the new strong learner;

[0138] If the loss function of the new strong learner is greater than a preset threshold, then return to the step of constructing a function for the strong learner and the training data to obtain the strong loss function of the strong learner;

[0139] If the loss function of the new strong learner is less than the threshold, then the recognition model is obtained;

[0140] The identification model is optimized using preset verification data to obtain an earthquake phase axis identification model.

[0141] In this embodiment of the invention, the decision tree model is a tree structure and serves as the basic model in the gradient boosting decision tree learning method. This model iterates and corrects each error to finally obtain the seismic phase axis identification model.

[0142] In this embodiment of the invention, the weak learner and the strong learner can be weighted and accumulated using the following formula:

[0143] F m (x)=F m-1 (x)+β m h(x;a m )

[0144] Among them, F m β is the strong prediction value of the strong learner. m As a preset learning constant, h(x; a) m ) represents the weak prediction value of the weak trainer, a m is the parameter of the weak learner in the m-th round, x is the selected feature vector, and m is the preset number of iterations.

[0145] In this embodiment of the invention, the parameters of the weak learner can be determined using the following formula:

[0146]

[0147] Among them, a m These are the parameters of the weak learner in the m-th round, where α and β are the second and third preset constants, respectively, and g m (x i ) represents the i-th selection feature vector x i The prediction result on the m-th weak learner, x i For the i-th filtering feature vector, h(x; a) m) represents the weak prediction value of the weak trainer, m is the preset number of iterations, and i is the selected feature vector.

[0148] In this embodiment of the invention, the negative gradient value of the new strong learner can be obtained using the following formula:

[0149]

[0150] Where, k m (x) represents the negative gradient value obtained in the m-th iteration, x is the selected feature vector, F(x) is the prediction function of the accumulated strong learner, L(y, F(x)) is the loss function, z is the second preset label, and E is the expected value corresponding to the selected feature vector.

[0151] In this embodiment of the invention, the gradient boosting decision tree learning method is used to learn the training data. Since each step in the gradient boosting decision tree learning method is to perform error training on previous errors, thereby reducing errors, the obtained seismic phase axis identification model has high accuracy and stability when identifying seismic phase axes. It can effectively capture the characteristics of seismic phase axes, quickly and accurately discover seismic phase axis data, and improve the identification efficiency and accuracy of seismic phase axis data.

[0152] S6. Obtain test data, and use the earthquake phase axis identification model to identify the test data to obtain earthquake phase axis data.

[0153] In this embodiment of the invention, the test data is identified using the seismic phase axis identification model. The identified phase axis data can be quickly identified from a large amount of data, thereby improving the identification efficiency of seismic phase axis data.

[0154] In this embodiment of the invention, enhanced data is obtained by segmenting and augmenting the test data. The enhanced data is then further used to identify the test data using the seismic phase axis identification model to obtain the seismic phase axis data.

[0155] See appendix Figure 7 As shown, Figure 7 (a) and Figure 7 (b) The test data used in this invention, where the horizontal axis represents distance and the vertical axis represents velocity. The test data is input into the seismic phase axis identification model for verification, and the test result diagram is obtained. See Appendix. Figure 8 As shown, where, Figure 8 (a) The precision rate for noise detection is 96% and the recall rate is 99%, while the precision rate for in-phase shaft detection is 92% and the recall rate is 70%. Figure 8(b) The precision of noise detection is 96% and the recall is 99%, while the precision of detection for in-phase axes is 91% and the recall is 69%. Therefore, from... Figure 8 The test results show that the effective signal of the in-phase axis is well detected, whether it is a direct wave or a reflected wave.

[0156] In a practical application scenario of this invention, to compare the detection performance of the seismic phase axis identification model, the following datasets were extracted: single-shot records at a shot-receiver distance of 560m, 1760m, and 2860m, ​​respectively. (See reference...) Figure 9 The figure shows a comparison of single-shot records, where the vertical axis represents amplitude and the horizontal axis represents time. As can be seen from the figure, using the seismic phase axis identification model to identify the test data can more quickly identify seismic phase axis data.

[0157] Example 2

[0158] To better understand the present invention, a second embodiment is provided below to further explain the situation of performing feature filtering on the feature vector to obtain the filtered feature vector.

[0159] In this embodiment of the invention, the preset L1 regularization model is a linear regression model. According to the preset L1 regularization model, the weights of unimportant features can be gradually reduced to zero, thereby achieving feature selection.

[0160] In this embodiment of the invention, the step of performing feature filtering on the feature vector to obtain a filtered feature vector includes:

[0161] The feature vector is weighted using a pre-defined L1 regularization model to obtain weight data.

[0162] Extract the feature vectors corresponding to the non-zero weights in the weighted data to obtain the regularized feature vectors;

[0163] Calculate the covariance matrix of the regular eigenvectors;

[0164] Based on the eigenvalues ​​of the covariance matrix and the corresponding matrix vector;

[0165] The feature values ​​are sorted in descending order to obtain a sorting result. Based on the sorting result, a preset number of matrix vectors corresponding to the feature values ​​are selected as principal components.

[0166] The regularized feature vector is projected onto the principal component to obtain the filtered feature vector.

[0167] In this invention, by using preset L1 regularization to perform feature filtering to obtain regular feature vectors, and then calculating the principal components of the regular feature vectors, the accuracy of the selected features can be greatly improved, and the effectiveness of the data can be enhanced.

[0168] Example 3

[0169] like Figure 10 As shown in the diagram, this embodiment also provides a functional block diagram of a machine learning-based seismic phase axis identification device. The machine learning-based seismic phase axis identification device 100 described in this embodiment can be installed in an electronic device. Depending on the functions implemented, the machine learning-based seismic phase axis identification device 100 may include a data acquisition module 101, a data processing module 102, a feature processing module 103, a feature filtering module 104, a model training module 105, and a model application module 106. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and which are stored in the memory of the electronic device.

[0170] In this embodiment, the functions of each module / unit are as follows:

[0171] Data acquisition module 101 is used to acquire seismic signals and perform forward numerical simulation on the seismic signals using a thin interlayer velocity model to obtain training data to be processed.

[0172] Data processing module 102 is used to perform single-channel segmentation and data recombination on the training data to be processed to obtain training data;

[0173] Feature processing module 103 is used to perform feature processing on the training data to obtain feature vectors;

[0174] Feature filtering module 104 is used to filter the feature vector to obtain a filtered feature vector;

[0175] The model training module 105 is used to learn the selected feature vectors using the gradient boosting decision tree learning method to obtain the earthquake phase axis identification model.

[0176] The model application module 106 is used to acquire test data and use the seismic phase axis identification model to identify the test data to obtain seismic phase axis data.

[0177] In detail, each module in the machine learning-based seismic phase axis identification device 100 described in this embodiment of the invention uses the same technical means as the machine learning-based seismic phase axis identification method described in Embodiments 1 and 2, and can produce the same technical effect, which will not be repeated here.

[0178] Example 4

[0179] like Figure 11 As shown, this embodiment also provides a computer electronic device, which may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and capable of running on the processor 10, such as a machine learning-based seismic phase axis identification program.

[0180] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., executing a machine learning-based seismic phase axis identification program) and calls data stored in the memory 11 to perform various functions of the electronic device and process data.

[0181] The memory 11 includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device, such as a plug-in portable hard drive, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. Furthermore, the memory 11 can include both internal and external storage units of the electronic device. The memory 11 can be used not only to store application software and various types of data installed on the electronic device, such as the code of a machine learning-based seismic phase axis identification program, but also to temporarily store data that has been output or will be output.

[0182] The communication bus 12 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable communication between the memory 11 and at least one processor 10, etc.

[0183] The communication interface 13 is used for communication between the aforementioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, Bluetooth interface, etc.), typically used to establish communication connections between the electronic device and other electronic devices. The user interface may be a display, an input unit (such as a keyboard), or, optionally, a standard wired or wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device and to display a visual user interface.

[0184] The figure only shows an electronic device with components. Those skilled in the art will understand that the structure shown in the figure does not constitute a limitation on the electronic device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0185] For example, although not shown, the electronic device may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0186] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0187] The machine learning-based seismic phase axis identification program stored in the memory 11 of the electronic device is a combination of multiple instructions. When run in the processor 10, it can achieve the following:

[0188] Seismic signals are acquired, and forward numerical simulations are performed on the seismic signals using a thin interlayer velocity model to obtain training data to be processed.

[0189] The training data to be processed is segmented into single channels and reassembled to obtain training data.

[0190] The training data is subjected to feature processing to obtain a feature vector;

[0191] The feature vector is subjected to feature filtering to obtain the filtered feature vector;

[0192] The selected feature vectors are learned using the gradient boosting decision tree learning method to obtain the seismic phase axis identification model;

[0193] The test data is acquired, and the earthquake phase axis identification model is used to identify the test data to obtain earthquake phase axis data.

[0194] Specifically, the specific implementation method of the processor 10 for the above instructions can be referred to the description of the relevant steps in the corresponding embodiment of the accompanying drawings, and will not be repeated here.

[0195] Furthermore, if the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0196] Example 5

[0197] This embodiment provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the machine learning-based seismic phase axis identification method described above.

[0198] This program code can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable device for implementing the process. Figure 1 Steps of a specified function in one or more processes.

[0199] Storage media include permanent and non-permanent, removable and non-removable media, and can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by computing devices.

[0200] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application. When the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0201] It should be understood that the terms used in this way can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in a sequence other than those illustrated or described herein.

[0202] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0203] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0204] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0205] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0206] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0207] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0208] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0209] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A machine learning-based method for identifying seismic phase axes, characterized in that, The method includes: Seismic signals are acquired, and forward numerical simulations are performed on the seismic signals using a thin interlayer velocity model to obtain training data to be processed. The training data to be processed is segmented into single channels and reassembled to obtain training data. The training data is subjected to feature processing to obtain a feature vector; The feature vector is subjected to feature filtering to obtain the filtered feature vector; The selected feature vectors are learned using the gradient boosting decision tree learning method to obtain the seismic phase axis identification model; Acquire test data, and use the earthquake phase axis identification model to identify the test data to obtain earthquake phase axis data; The step of filtering the feature vector to obtain the filtered feature vector includes: performing univariate feature analysis on the feature vector using analysis of variance to obtain univariate feature vectors; and performing feature selection on the univariate feature vectors using an external support vector machine model to obtain the filtered feature vectors. The step of using a support vector machine external model to select features from the univariate feature vectors to obtain a filtered feature vector includes: calculating the importance score of each univariate feature using a preset support vector external model; and filtering out univariate features based on the importance score to obtain a filtered feature vector.

2. The seismic phase axis identification method based on machine learning as described in claim 1, characterized in that, The forward numerical simulation of the seismic signal using a thin interlayer velocity model yields training data to be processed, including: The wavefield is obtained by calculating the seismic signal using the finite difference method and the thin interlayer velocity model. Position information is extracted from the wave field to obtain forward-modeled synthetic shot record data; The forward modeling composite gun recording data is divided into preset values ​​to obtain training data.

3. The seismic phase axis identification method based on machine learning as described in claim 1, characterized in that, The step of performing single-channel segmentation and data recombination on the training data to be processed to obtain training data includes: The training data is segmented to obtain segmented data; The segmented data is divided equally to obtain reconstructed data; Noise is added to the reconstructed data to obtain training data.

4. The seismic phase axis identification method based on machine learning as described in claim 1, characterized in that, The step of performing feature processing on the training data to obtain a feature vector includes: Single-channel feature extraction is performed on the training data to obtain single-number feature data; Gray-level co-occurrence matrix features are extracted from the training data to obtain gray-level co-occurrence matrix feature data; Inter-track feature extraction is performed on the training data to obtain inter-track feature data; Feature vectors are constructed from the gray-level co-occurrence matrix feature data, the single-channel feature extraction, and the inter-channel feature data to obtain feature data. The feature data is standardized to obtain a feature vector.

5. The seismic phase axis identification method based on machine learning as described in claim 1, characterized in that, The step of performing univariate feature analysis on the feature vector using analysis of variance to obtain univariate feature vectors includes: Calculate the sum of squared deviations between the eigenvectors to obtain the vector deviation; The training bias is obtained by calculating the sum of squared deviations of the feature vectors. Based on the training bias and the vector bias, the test statistic of each feature vector is calculated. The test statistics are sorted in descending order, and the features corresponding to the test statistics that meet the preset conditions are selected as univariate feature vectors.

6. A seismic phase axis identification device based on machine learning, characterized in that, The device includes: The data acquisition module is used to acquire seismic signals and perform forward numerical simulation on the seismic signals using a thin interlayer velocity model to obtain training data to be processed. The data processing module is used to perform single-channel segmentation and data recombination on the training data to be processed to obtain training data; The feature processing module is used to perform feature processing on the training data to obtain feature vectors; The feature filtering module is used to filter the feature vector to obtain the filtered feature vector; The model training module is used to learn the selected feature vectors using the gradient boosting decision tree learning method to obtain the earthquake phase axis identification model. The model application module is used to acquire test data and use the seismic phase axis identification model to identify the test data to obtain seismic phase axis data. The feature selection module is used to perform univariate feature analysis on the feature vector using analysis of variance to obtain a univariate feature vector; and to select features from the univariate feature vector using an external support vector machine model to obtain a selected feature vector. The feature filtering module is used to calculate the importance score of each univariate feature using a preset support vector external model; and to filter out univariate features based on the importance score to obtain the filtered feature vector.

7. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the steps of the machine learning-based seismic phase axis identification method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the machine learning-based seismic phase axis identification method as described in any one of claims 1 to 5.

Images