Seismic wave continuity noise automatic identification method and system

Abstract

This invention discloses an automatic method and system for identifying continuity noise in seismic waves, comprising: acquiring seismic wave data to be identified; calculating the characteristic values ​​of the seismic wave data at each time point according to a constructed seismic wave continuity noise index function, forming a feature set; inputting the feature set into a trained continuity noise identification model, and outputting the prediction results at each time point on the seismic wave; weighting and summing the prediction results and characteristic values ​​at each time point on the seismic wave to obtain the noise value at each time point; calculating the judgment value at each time point, and identifying time points on the seismic wave with noise values ​​greater than the judgment value as noise points. This invention achieves automatic identification of continuity noise in seismic wave signals by constructing a seismic wave continuity noise identification index function and a hybrid autoencoder-attention model.

Description

Technical Field

[0001] This invention relates to the application of deep learning algorithms in the field of geophysical observation signal processing, and particularly to the automatic identification of seismic wave signal continuity noise by constructing a custom seismic wave continuity noise identification index and a hybrid autoencoder and attention network model. Specifically, it is a method and system for automatic identification of seismic wave continuity noise. Background Technology

[0002] Operational Earthquake Forecasting (OEF) is a crucial tool supporting post-earthquake emergency response and scientific investigation. Among these methods, time-evolution tracking analysis based on the "contagious aftershock sequence" model has become the mainstream approach for predicting short-term aftershock probabilities. This method predicts the probability of aftershocks of different magnitudes occurring within the next day by real-time fitting of the attenuation parameters (p-value) and sub-event triggering capability (α-value) of the aftershock sequence. The prediction results can be quantified and evaluated using the Brier score. However, the accuracy of this technology heavily relies on the reliability of the earthquake catalog, especially in the early post-earthquake stages. Monitoring equipment is susceptible to interference from factors such as unstable power supply, digital acquisition system malfunctions, and environmental vibrations, resulting in a large amount of continuous noise mixed in the signal, directly affecting the scientific determination of the minimum integrity magnitude. Therefore, when analyzing short-term aftershock signals, noise caused by the above factors should be eliminated as much as possible to restore the undisturbed seismic wave signal to the greatest extent possible.

[0003] Most existing studies employ mean-shift clustering for noise identification, defaulting to treating data points far from the cluster center as noise. However, in real-world scenarios, continuous noise from seismic waves is often superimposed on valid seismic signals. This continuity leads to two types of misclassifications: noise points are misidentified as normal signals, and valid seismic signals are misidentified as noise. This makes continuous noise identification reliant on prior knowledge and ineffective, failing to meet the adaptive and accurate requirements of post-earthquake emergency response and post-earthquake analysis. Summary of the Invention

[0004] To address the challenge of automatic identification of continuity noise in seismic waves, this invention provides a method for automatic identification of continuity noise in seismic waves. This method involves constructing a seismic wave continuity noise identification index function and a hybrid autoencoder-attention model to automatically identify continuity noise in seismic wave signals.

[0005] According to one aspect of the present invention, an automatic method for identifying seismic wave continuity noise is provided, comprising: Acquire seismic wave data to be identified; Based on the constructed seismic wave continuity noise index function, the characteristic values ​​of the seismic wave data to be identified at each time point are calculated to form a feature set; The feature set is input into the trained continuous noise recognition model, and the prediction results at each time point on the seismic wave are output. The predicted results and characteristic values ​​at each time point on the seismic wave are weighted and summed point by point to obtain the noise value at each time point; Calculate the judgment value at each time point, and identify the time points where the noise value on the seismic wave is greater than the judgment value as noise points.

[0006] As a further technical solution, the training of the continuous noise recognition model includes: Construct a training sample feature set; A hybrid autoencoder-attention model is constructed. Temporal features of the feature set are extracted by sliding convolutional layers. The temporal features are input into the reconstruction branch and the classification branch in parallel. The reconstruction branch reconstructs the signal through a deconvolutional layer. The classification branch generates a dynamic weight matrix by introducing an attention module. Weights are set for the reconstruction branch and the classification branch respectively, and the dynamic weight matrix is ​​calculated. The features are recalibrated by matrix multiplication using the dynamic weight matrix. The noise prediction result is output after passing through a fully connected layer. Construct a bi-branch dynamic loss function, iterate until the loss function converges, and output the trained continuous noise recognition model.

[0007] As a further technical solution, the training of the continuous noise recognition model also includes: For the reconstruction branch, calculate the mean square error of the reconstructed signal and the original signal multiplied by the adaptive factor. Losses from reconstruction , This represents the difference between the reconstructed signal and the original signal value; For the classification branch, the classification loss is obtained by calculating the binary cross-entropy between the noisy prediction results and the true results. ; Reconstruction losses and classification loss The two-branch dynamic loss function is obtained by weighting them separately. ; The iteration process brings the loss function to its minimum value, and then outputs the prediction results for each point on the seismic wave at this point.

[0008] As a further technical solution, the constructed seismic wave continuity noise index function is: , in, Let be the eigenvalue at time t. For logarithmically corrected difference operators, For local information entropy, For local standard deviation, This is the adaptive amplification factor.

[0009] As a further technical solution, the method also includes: Using the mean, standard deviation, maximum value, minimum value, and gradient mean as new scale features, the new scale features are calculated within a time window, and the new scale features and feature values ​​at the same moment are combined to form a comprehensive feature set.

[0010] As a further technical solution, the determination value at each time point is calculated, including: Based on the model prediction results Calculate the adaptive decision value for each point on the seismic wave. .

[0011] According to one aspect of the present invention, an automatic seismic wave continuity noise identification system is provided, comprising: The first main module is used to acquire the seismic wave data to be identified; The second main module is used to calculate the characteristic values ​​of the seismic wave data to be identified at each time point based on the constructed seismic wave continuity noise index function, and form a feature set. The third main module is used to input the feature set into the trained continuous noise recognition model and output the prediction results at each time point on the seismic wave. The fourth main module is used to weight and sum the prediction results and characteristic values ​​at each time point on the seismic wave to obtain the noise value at each time point; The fifth main module is used to calculate the judgment value at each time point and identify the time points where the noise value on the seismic wave is greater than the judgment value as noise points.

[0012] As a further technical solution, the third main module also includes: The first submodule is used to construct the feature set of training samples; The second submodule is used to construct a hybrid autoencoder-attention model. It extracts temporal features of the feature set by sliding convolutional layers. The temporal features are input into the reconstruction branch and the classification branch in parallel. The reconstruction branch reconstructs the signal through a deconvolutional layer. The classification branch generates a dynamic weight matrix by introducing an attention module. Weights are set for the reconstruction branch and the classification branch respectively, and the dynamic weight matrix is ​​calculated. The features are recalibrated by matrix multiplication using the dynamic weight matrix. The noise prediction result is output after passing through a fully connected layer. The third submodule is used to construct a two-branch dynamic loss function, iterate until the loss function converges, and output the trained continuous noise recognition model.

[0013] According to one aspect of the present invention, an automatic seismic wave continuity noise identification device is provided, comprising a memory and a processor, wherein the memory stores program instructions that are executed by the processor, and the processor invokes the program instructions to execute the automatic seismic wave continuity noise identification method.

[0014] According to one aspect of the present invention, a non-transitory computer-readable storage medium is provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute the described automatic seismic wave continuity noise identification method.

[0015] This invention provides a method for identifying continuity noise in seismic waves. Specifically, it constructs a feature set by defining a novel noise identification index function, and then builds a hybrid autoencoder-attention model based on this feature set to automatically identify continuity noise in seismic waves. Compared with existing technologies (such as mean-shift clustering methods), the advantages of this invention are: (1) It can automatically identify continuous noise without relying on prior knowledge; (2) It greatly avoids false detection of effective seismic signals and missed detection of continuous noise, and has high accuracy in identifying continuous noise. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. 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 illustrating an automatic identification method for seismic wave continuity noise provided in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the recognition results based on the index constructed in this invention and the hybrid autoencoder-attention model, provided as an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram of standard noise labeling based on prior knowledge, provided for an embodiment of the present invention.

[0020] Figure 4 This diagram illustrates the missed detection points and false detection points based on the index constructed in this invention and the hybrid autoencoder-attention model, as provided in an embodiment of the invention.

[0021] Figure 5 This diagram illustrates the missed detection points and false detection points of the conventional mean-shift clustering method provided in this embodiment of the invention. Detailed Implementation

[0022] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0024] In view of the shortcomings of existing methods for identifying continuous noise in seismic waves, namely, two types of misjudgments in the automatic identification of continuous noise in seismic waves: (1) noise points are misidentified as normal signals and (2) valid seismic signals are misidentified as noise, this invention proposes the following solutions.

[0025] The purpose of this invention is to provide a novel method for identifying continuous noise in seismic waves. By defining a new index function to construct a feature set, the efficiency of continuous noise identification in seismic waves is improved. The seismic wave noise identification method described in this invention can more accurately identify continuous noise, with the ultimate goal of achieving automatic identification of continuous noise and providing a complete and noise-free seismic record for subsequent seismic wave analysis.

[0026] like Figure 1As shown in the figure, this embodiment of the invention provides an automatic method for identifying seismic wave continuity noise. First, the seismic wave data to be identified is acquired. Then, based on the constructed seismic wave continuity noise index function, the characteristic values ​​of the seismic wave data to be identified at each time point are calculated to form a feature set. Subsequently, the feature set is input into a trained continuity noise identification model, and the prediction results at each time point on the seismic wave are output. Next, the prediction results and characteristic values ​​at each time point on the seismic wave are weighted and summed point by point to obtain the noise value at each time point. Finally, the judgment value at each time point is calculated, and the time points on the seismic wave with noise values ​​greater than the judgment value are identified as noise points.

[0027] The specific steps of the method described in this embodiment of the invention are as follows: (1) Seismic wave continuity noise is characterized by small amplitude, strong randomness, and susceptibility to environmental interference. Moreover, it is similar to the effective seismic signal in terms of time scale, which makes it easy for the two to overlap and difficult to distinguish effectively. In order to solve the problem of seismic wave continuity noise identification, this invention needs to construct a new seismic wave continuity noise diagnostic index system, use the index values ​​as feature set to train the continuity noise identification model, and complete the automatic identification of continuity noise by adaptively adjusting the threshold.

[0028] First, a new seismic wave continuity noise identification index function (Equation (1)) is proposed, the derivation of which depends on the intermediate parameters defined by Equations (2)(3)(4)(5). and The function has the following form:

[0029] Equation (1) depends on the joint calculation of equations (2)(3)(4)(5): In equation (2), For the logarithmically corrected difference operator, considering the small amplitude of continuous noise, logarithmic enhancement is used to extract low-to-medium amplitude noise. The gradient of the signal value at time t is obtained by using a sliding window to get the time window. Internal signal value The calculated value, which is the difference between the maximum and minimum values, is used to measure the continuous change within the dynamic range of the signal value. This calculated value is then compared with the signal value at that moment. relative to the signal value at the previous moment The mutations are added together and the result is normalized (coefficient 1 / 3). In equation (3), Local information entropy is used to distinguish random noise from normal signals, taking into account the strong randomness of continuous noise. The absolute value of the signal at time k within the time window [tw,t] is used, and weights are applied based on the absolute value of the signal. ; In equation (4), To address the local standard deviation and the susceptibility of continuous noise to environmental interference, the robustness of the indicator is enhanced (the indicator remains unaffected when environmental factors interfere with the noise amplitude or frequency). For time windows Signal value at time k The calculated mean; In equation (5), As an adaptive amplification factor, considering the tendency for continuous noise and effective seismic signals to alias, a hyperbolic tangent function is used to amplify noise characteristics during low-amplitude periods and suppress noise misjudgments during high-amplitude periods. Gradient within the time window [tw,t] mean For time windows Inner gradient The standard deviation, expressed by the hyperbolic tangent function, is... The degree of deviation from the mean is mapped to the [0,1] interval to achieve adaptive amplification (when the calculated value within tanh is less than or equal to 1, af approaches 0, suppressing background fluctuations, i.e. suppressing false detections of effective seismic signals; conversely, it approaches 1, enabling the identification of noise enhancement).

[0030] (2) The index function constructed in step (1) Perform joint calculation: The input data is the seismic wave data to be identified, the time window w is set (e.g., 100), and then the equation (1) is used to calculate the time at time t. The calculated values ​​are then used to form a feature set in the form of column vectors, providing input features for the subsequent hybrid autoencoder and attention model.

[0031] In addition to feature values, this invention can also add other feature sets. Taking the addition of five scale features—mean, standard deviation, maximum value, minimum value, and gradient mean—as an example, the method is to add them within a time window. The above features are calculated internally, and these five scale features at the same instant are compared with... Value combinations form a comprehensive feature set. This addresses the issue of missing scale features or... Features with a value of 0 are removed to ensure that all feature sets are valid.

[0032] (3) Use the feature set obtained in step (2) as input features to construct a hybrid autoencoder-attention model. First, use 1D-CNN layers (32 / 64 channels, The convolutional kernel slides to extract temporal features from the feature set. These temporal features are high-dimensional abstract features, which are input into the reconstruction branch and the classification branch in parallel. In the reconstruction branch, the signal features are reconstructed through a deconvolutional layer (64→32 channels, 2x upsampling) to obtain... In the classification branch, an attention module for seismic waves is inserted into the CNN feature layer (i.e., after extracting temporal features). A dynamic weight matrix A is generated through two fully connected layers, with the formula as follows:

[0033] in For the output of the CNN feature layer, These correspond to the weights of the reconstruction and classification branches, respectively, and can be adjusted as needed. The calculated dynamic weight matrix A is used to recalibrate the features through matrix multiplication, focusing the model's attention on these features. In the high-value region, the noise prediction result is output after passing through the fully connected layer. This refers to the probability value that the model predicts the sample as noise.

[0034] Further calculate the bi-branch dynamic loss function The reconstruction branch and the classification branch are balanced. For the reconstruction branch, the mean square error of the reconstructed signal and the original signal are calculated and multiplied by an adaptive factor. Losses from reconstruction , This represents the difference between the reconstructed signal and the original signal value; for the classification branch, the classification loss is obtained by calculating the binary cross-entropy between the predicted and true results. , and The two-branch dynamic loss function is obtained by weighting them separately. (Weights and above) (Consistent). Iteration brings the loss function to its minimum, and the predicted results for each point on the seismic wave at this point are output. (The probability that a seismic wave is classified as a noise point at each time point).

[0035] (4) Prediction results of each point on the seismic wave obtained in step (3) and the points on the seismic waves obtained in step (2) The values ​​are weighted and summed (e.g., 0.4, 0.6) to obtain the calculated value, which is then used as the noise value V at each point on the seismic wave. Simultaneously, the model prediction results are used again... Calculate the adaptive decision value for each point on the seismic wave. The purpose of this judgment value is to evaluate the noise points predicted by the model. Lowering the threshold avoids missed detections, while raising the threshold for normal points avoids false detections. Finally, points with noise values ​​V on the seismic wave greater than the adaptive judgment value r are identified as noise points, and those with V > r are not noise points (i.e., if V > r, the point is identified as noise, otherwise it is not noise).

[0036] from Figure 2 and Figure 3 As can be seen from this, the recognition result proposed in this invention does not rely on prior knowledge. Figure 2The recognition results after basic and superimposed prior knowledge ( Figure 3 Maintaining consistency, it can accurately display the spatial distribution of noise points without relying on prior knowledge. From Figure 4 and Figure 5 As can be seen from the above, the method proposed in this invention compares the missed detection points and false detection points with the standard noise point annotation. Figure 4 The number of missed detections and false detections is significantly less than that of conventional mean-shift clustering methods. Figure 5 Furthermore, the accuracy of the proposed method and the conventional mean-shift clustering method are 97.28% and 47.16%, respectively. The proposed method improves the accuracy of seismic wave continuity noise identification by 50.12%, enabling more accurate automatic identification of seismic wave continuity noise.

[0037] As one implementation method, this embodiment of the invention uses a raw seismic wave data containing continuous noise with a total duration of 86,400 s as an example to illustrate the specific implementation method of the invention.

[0038] In step one, using the constructed seismic wave continuity noise identification index function, the original seismic wave data to be identified is input. It is appropriate to set a time window that is two orders of magnitude smaller than the total duration of the seismic wave data (100 s in this example). The feature values ​​at each time t are calculated, such as [0.15376, 0.81390, 0.72745, ...]. This set of feature values ​​is stored as a column vector in an array of length 86400 as a feature set.

[0039] In step two, the feature set from step one is used as input features, and these features are input in parallel to the reconstruction and classification branches of the hybrid autoencoder-attention model. Specifically, in the reconstruction branch, the signal features are reconstructed through a deconvolutional layer (64→32 channels, 2x upsampling) to obtain... For example, [0.39998, 0.43295, 0.43326, ...]. In the classification branch, using the calculation formula of matrix A, the reconstruction branch and the classification branch are weighted by 0.7 and 0.3 respectively. The calculated matrix A is then multiplied with the feature set to reconstruct the features. The model prediction result is then output through a fully connected layer. This step calculates the loss functions for the reconstruction branch and the classification branch separately. In the reconstruction branch, the reconstructed features are calculated. The loss function is obtained by multiplying the mean squared error of the original feature by an adaptive factor; in the classification branch, the model prediction result is calculated. The loss function is obtained by calculating the binary cross-entropy between the actual result and the calculated result. Weights of 0.7 and 0.3 are applied separately and then summed to obtain the final loss function. This loss function is iterated until it converges to its minimum value of 0.124, yielding the final model prediction result. ([0.15670, 0.53985, 0.63189, ...]).

[0040] In step three, the prediction results obtained in step two are... The Feature values ​​obtained in step one are weighted by 0.4 and 0.6 respectively at each time step and then summed to obtain the noise values ​​V at each point on the seismic wave, such as [0.15495, 0.70428, 0.68922,...]. The decision value r at each point on the seismic wave is calculated using the adaptive decision value formula, such as [0.50784, 0.52699, 0.53159]. Points on the seismic wave with V values ​​greater than r values ​​are determined as noise points, such as 0.70428>0.52699, and conversely, points with V values ​​less than r values ​​are not noise points, such as 0.15495<0.50784.

[0041] The implementation of the various embodiments of the present invention is based on programmed processing by a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide an automatic seismic wave continuity noise identification system, which is used to execute an automatic seismic wave continuity noise identification method from the above method embodiments.

[0042] The system includes: a first main module for acquiring seismic wave data to be identified; a second main module for calculating the characteristic values ​​of the seismic wave data to be identified at each time point based on a constructed seismic wave continuity noise index function, forming a feature set; a third main module for inputting the feature set into a trained continuity noise identification model and outputting the prediction results at each time point on the seismic wave; a fourth main module for weighted summation of the prediction results and characteristic values ​​at each time point on the seismic wave to obtain the noise value at each time point; and a fifth main module for calculating the judgment value at each time point and identifying time points on the seismic wave with noise values ​​greater than the judgment value as noise points.

[0043] This invention provides an automatic seismic wave continuity noise identification system, addressing the challenge of automatic seismic wave continuity noise identification. It employs several modules described above and automatically identifies seismic wave signal continuity noise by constructing a seismic wave continuity noise identification index function and a hybrid autoencoder-attention model.

[0044] It should be noted that the system embodiments provided by this invention, in addition to implementing the methods in the above method embodiments, are also used to implement the methods in other method embodiments provided by this invention. The difference lies only in setting corresponding functional modules, and their principles are basically the same as those of the above system embodiments provided by this invention. As long as those skilled in the art, based on the above system embodiments and referring to the specific technical solutions in other method embodiments, obtain corresponding technical means and technical solutions composed of these technical means by combining technical features, and improve the modules in the above system embodiments while ensuring the practicality of the technical solutions, they can obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments. For example: Based on the above system embodiments, as a preferred embodiment, the automatic seismic wave continuity noise identification system provided in this embodiment of the invention further includes, in its third main module: The first submodule is used to construct the feature set of training samples; The second submodule is used to construct a hybrid autoencoder-attention model. It extracts temporal features of the feature set by sliding convolutional layers. The temporal features are input into the reconstruction branch and the classification branch in parallel. The reconstruction branch reconstructs the signal through a deconvolutional layer. The classification branch generates a dynamic weight matrix by introducing an attention module. Weights are set for the reconstruction branch and the classification branch respectively, and the dynamic weight matrix is ​​calculated. The features are recalibrated by matrix multiplication using the dynamic weight matrix. The noise prediction result is output after passing through a fully connected layer. The third submodule is used to construct a two-branch dynamic loss function, iterate until the loss function converges, and output the trained continuous noise recognition model.

[0045] Based on the same inventive concept as the foregoing embodiments, this embodiment of the invention also provides an automatic seismic wave continuity noise identification device, including a memory and a processor. The memory stores program instructions that are executed by the processor, and the processor calls the program instructions to execute the automatic seismic wave continuity noise identification method.

[0046] In embodiments of the present invention, the memory can be non-volatile memory, such as a hard disk drive (HDD) or a solid-state drive (SSD), or it can be volatile memory, such as random-access memory (RAM). Memory is any other medium capable of carrying or storing desired program code having an instruction or data structure form and accessible by a computer, but is not limited thereto. The memory in embodiments of the present invention can also be a circuit or any other device capable of implementing a storage function for storing program instructions and / or data.

[0047] In this embodiment of the invention, the processor may be a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in this embodiment of the invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in this embodiment of the invention can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0048] Based on the same inventive concept as the foregoing embodiments, this embodiment of the invention also provides a non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute the automatic seismic wave continuity noise identification method as follows: Acquire seismic wave data to be identified; Based on the constructed seismic wave continuity noise index function, the characteristic values ​​of the seismic wave data to be identified at each time point are calculated to form a feature set; The feature set is input into the trained continuous noise recognition model, and the prediction results at each time point on the seismic wave are output. The predicted results and characteristic values ​​at each time point on the seismic wave are weighted and summed point by point to obtain the noise value at each time point; Calculate the judgment value at each time point, and identify the time points where the noise value on the seismic wave is greater than the judgment value as noise points.

[0049] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

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

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

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

[0053] In summary, the present invention proposes an innovative solution to the problem of automatic identification of seismic wave continuity noise. By constructing a custom seismic wave continuity noise identification index and a hybrid autoencoder and attention network model, the present invention achieves good results in automatic identification of seismic wave signal continuity noise and makes up for the shortcomings of existing methods.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatic identification of seismic wave continuity noise, characterized in that, include: Acquire seismic wave data to be identified; Based on the constructed seismic wave continuity noise index function, the characteristic values ​​of the seismic wave data to be identified at each time point are calculated to form a feature set; the constructed seismic wave continuity noise index function is: ,in, Let be the eigenvalue at time t. For logarithmically corrected difference operators, For local information entropy, For local standard deviation, This is an adaptive amplification factor; The feature set is input into the trained continuous noise recognition model, and the prediction results at each time point on the seismic wave are output. The predicted results and characteristic values ​​at each time point on the seismic wave are weighted and summed point by point to obtain the noise value at each time point; Calculate the judgment value at each time point, and identify the time points where the noise value on the seismic wave is greater than the judgment value as noise points; The training of the continuous noise recognition model includes: Construct a training sample feature set; A hybrid autoencoder-attention model is constructed. Temporal features of the feature set are extracted by sliding convolutional layers. The temporal features are input into the reconstruction branch and the classification branch in parallel. The reconstruction branch reconstructs the signal through a deconvolutional layer. The classification branch generates a dynamic weight matrix by introducing an attention module. Weights are set for the reconstruction branch and the classification branch respectively, and the dynamic weight matrix is ​​calculated. The features are recalibrated by matrix multiplication using the dynamic weight matrix. The noise prediction result is output after passing through a fully connected layer. Construct a bi-branch dynamic loss function, iterate until the loss function converges, and output the trained continuous noise recognition model.

2. The method for automatic identification of seismic wave continuity noise according to claim 1, characterized in that, The training of the continuous noise recognition model also includes; For the reconstruction branch, calculate the mean square error of the reconstructed signal and the original signal multiplied by the adaptive factor. Losses from reconstruction , This represents the difference between the reconstructed signal and the original signal value; For the classification branch, the classification loss is obtained by calculating the binary cross-entropy between the noisy prediction results and the true results. ; Reconstruction losses and classification loss The two-branch dynamic loss function is obtained by weighting them separately. ; The iteration process brings the loss function to its minimum value, and then outputs the prediction results for each point on the seismic wave at this point.

3. The method for automatic identification of seismic wave continuity noise according to claim 1, characterized in that, The method further includes: The mean, standard deviation, maximum value, minimum value, and gradient mean are used as new scale features. The new scale features are calculated within a time window, and the new scale features and feature values ​​at the same moment are combined to form a comprehensive feature set.

4. The method for automatic identification of seismic wave continuity noise according to claim 1, characterized in that, Calculate the decision value at each time point, including: Based on the model prediction results Calculate the adaptive decision value for each point on the seismic wave. .

5. An automatic seismic wave continuity noise identification system, characterized in that, include: The first main module is used to acquire the seismic wave data to be identified; The second main module is used to calculate the characteristic values ​​of the seismic wave data to be identified at each time point based on the constructed seismic wave continuity noise index function, forming a feature set; the constructed seismic wave continuity noise index function is: ,in, Let be the eigenvalue at time t. For logarithmically corrected difference operators, For local information entropy, For local standard deviation, This is an adaptive amplification factor; The third main module is used to input the feature set into the trained continuous noise recognition model and output the prediction results for each time point on the seismic wave. The training of the continuous noise recognition model includes: constructing a training sample feature set; constructing a hybrid autoencoder-attention model, extracting temporal features from the feature set through sliding convolutional layers, inputting the temporal features into a reconstruction branch and a classification branch in parallel, reconstructing the signal through a deconvolutional layer in the reconstruction branch, generating a dynamic weight matrix in the classification branch by introducing an attention module, setting weights for the reconstruction branch and the classification branch respectively and calculating the dynamic weight matrix, recalibrating the features using the dynamic weight matrix through matrix multiplication, and outputting the noise prediction results through a fully connected layer; constructing a dual-branch dynamic loss function, iterating until the loss function converges, and outputting the trained continuous noise recognition model. The fourth main module is used to weight and sum the prediction results and characteristic values ​​at each time point on the seismic wave to obtain the noise value at each time point; The fifth main module is used to calculate the judgment value at each time point and identify the time points where the noise value on the seismic wave is greater than the judgment value as noise points.

6. The automatic seismic wave continuity noise identification system according to claim 5, characterized in that, The third main module also includes: The first submodule is used to construct the feature set of training samples; The second submodule is used to construct a hybrid autoencoder-attention model. It extracts temporal features of the feature set by sliding convolutional layers. The temporal features are input into the reconstruction branch and the classification branch in parallel. The reconstruction branch reconstructs the signal through a deconvolutional layer. The classification branch generates a dynamic weight matrix by introducing an attention module. Weights are set for the reconstruction branch and the classification branch respectively, and the dynamic weight matrix is ​​calculated. The features are recalibrated by matrix multiplication using the dynamic weight matrix. The noise prediction result is output after passing through a fully connected layer. The third submodule is used to construct a two-branch dynamic loss function, iterate until the loss function converges, and output the trained continuous noise recognition model.

7. An automatic seismic wave continuity noise identification device, characterized in that, The method includes a memory and a processor, wherein the memory stores program instructions that are executed by the processor, and the processor invokes the program instructions to execute the automatic identification method for seismic wave continuity noise as described in any one of claims 1 to 4.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to execute the automatic seismic wave continuity noise identification method according to any one of claims 1 to 4.

Images