Seismic signal classification model, model training method and earthquake identification method

Abstract

The invention provides a seismic signal classification model, a model training method and a seismic recognition method, and relates to the technical field of seismic monitoring. The seismic signal classification model comprises a feature extraction layer, a feature fusion layer and a decision-making layer which are connected in sequence; the feature extraction layer comprises a time domain encoder, a frequency domain encoder and a feature parameter encoder which are parallel. In the invention, a feature extraction layer of a seismic signal classification model comprises a time domain encoder, a frequency domain encoder and a feature parameter encoder so as to perform feature extraction on multi-modal data of seismic signals, and by integrating time domain parameters, frequency domain parameters and structured feature parameters, the model can more comprehensively characterize the characteristics of the seismic signals, so that the classification accuracy of the seismic signals is improved. And the earthquake identification accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of earthquake monitoring technology, and more specifically, to an earthquake signal classification model, a model training method, and an earthquake identification method. Background Technology

[0002] Microseismic signal identification is a fundamental and crucial technology in the field of earthquake monitoring and research, and it is of great significance for earthquake early warning, seismic activity analysis, and earthquake disaster prevention. With the large-scale construction of seismic stations and technological advancements, seismic networks are able to record more and more microseismic events, making it particularly important to automatically, quickly, and accurately identify different microseismic signals from complex background noise.

[0003] Traditional earthquake event classification methods rely primarily on the experience of seismologists, classifying events through visual analysis of waveform initial polarity, P-wave to S-wave amplitude ratio, and spectral characteristics. However, these methods struggle to quickly process large amounts of data and achieve high-precision classification under complex background noise. Summary of the Invention

[0004] The purpose of this application is to provide an earthquake signal classification model, a model training method, and an earthquake identification method, which combine the time-domain features, frequency-domain features, and structured features of earthquake signals to identify earthquakes and improve the accuracy of earthquake identification.

[0005] In a first aspect, embodiments of this application provide a seismic signal classification model, which includes a feature extraction layer, a feature fusion layer, and a decision layer connected in sequence. The feature extraction layer includes a parallel time-domain encoder, a frequency-domain encoder, and a feature parameter encoder. The time-domain encoder is used to extract features from time-domain parameters to generate a time-domain feature vector. The frequency-domain encoder is used to extract features from frequency-domain parameters to generate a frequency-domain feature vector. The feature parameter encoder is used to extract features from structured feature parameters to generate a statistical feature vector. The feature fusion layer is used to fuse the time-domain feature vector, the frequency-domain feature vector, and the statistical feature vector to generate a fused feature vector. The decision layer is used to obtain the seismic identification result based on the fused feature vector.

[0006] In this embodiment, the feature extraction layer of the seismic signal classification model includes a time-domain encoder, a frequency-domain encoder, and a feature parameter encoder to extract features from the multimodal data of the seismic signal. By integrating time-domain parameters, frequency-domain parameters, and structured feature parameters, the model can more comprehensively characterize the properties of the seismic signal and improve the accuracy of earthquake identification.

[0007] In some embodiments, the temporal encoder includes seven one-dimensional convolutional layers and three long short-term memory layers connected in sequence; the frequency encoder includes seven one-dimensional convolutional layers and one long short-term memory layer connected in sequence; and the feature parameter encoder includes two fully connected layers connected in sequence.

[0008] In this embodiment, the network structure of the encoder differs for different types of parameters, enabling the model to differentiate and target different types of parameters based on their characteristics, thereby improving the accuracy of parameter processing. Furthermore, convolutional layers and long short-term memory layers are used to construct temporal and frequency domain encoders. This type of encoder improves the accuracy of processing temporal and frequency domain parameters through the synergistic effect of local feature extraction and global temporal modeling. Fully connected layers are used to construct a feature parameter encoder. This type of encoder improves the accuracy of processing structured feature parameters through global feature integration and nonlinear mapping.

[0009] Secondly, embodiments of this application provide a training method for an earthquake signal classification model. This method is used to train an earthquake signal classification model according to any embodiment of the first aspect. The method includes: acquiring training data, wherein the training data includes natural earthquake data and non-natural earthquake data; generating a first time-domain sequence sample, a first frequency-domain sequence sample, and a first structured feature parameter sample based on the natural earthquake data; generating a second time-domain sequence sample, a second frequency-domain sequence sample, and a second structured feature parameter sample based on the non-natural earthquake data; inputting the first time-domain sequence sample, the first frequency-domain sequence sample, the first structured feature parameter sample, the second time-domain sequence sample, the second frequency-domain sequence sample, and the second structured feature parameter sample into an initial earthquake signal classification model to obtain predicted values; calculating the loss value between the predicted values ​​and the true values ​​using a loss function; and updating the network parameters in the initial earthquake signal classification model according to the loss value to obtain an earthquake signal classification model.

[0010] In this embodiment, by processing natural and non-natural earthquake data, corresponding multimodal sequence samples are obtained. These multimodal sequence samples are then used to train an initial earthquake signal classification model. The network parameters are updated based on predicted and true values ​​using a loss function, enabling the trained earthquake signal classification model to identify earthquakes based on multimodal data, thereby improving the model's accuracy.

[0011] In some embodiments, calculating the loss between the predicted and actual values ​​using a loss function includes: using the loss function Calculate the loss between the predicted value and the actual value; where, Indicates the number of samples. Represents the true value. This represents the predicted value.

[0012] The embodiments of this application calculate the loss value between the predicted value and the true value using a loss function, and then update the network parameters of the initial seismic signal classification model based on the loss value, thereby improving the accuracy of the seismic signal classification model.

[0013] Thirdly, embodiments of this application provide an earthquake identification method, which includes: acquiring earthquake signal data; generating multimodal data based on the earthquake signal data; wherein the multimodal data includes time-domain parameters, frequency-domain parameters, and structured feature parameters; processing the time-domain parameters, frequency-domain parameters, and structured feature parameters using an earthquake signal classification model of any embodiment of the first aspect to obtain an earthquake identification result; wherein the earthquake signal classification model is trained according to the training method of an earthquake signal classification model of any embodiment of the second aspect.

[0014] In this embodiment, multimodal data is generated based on seismic signal data. Since multimodal data reflects the characteristics of seismic signals from different dimensions, the seismic signal classification model can more comprehensively characterize the characteristics of seismic signals and improve the accuracy of earthquake identification.

[0015] In some embodiments, generating multimodal data based on seismic signal data includes: using a phase picking model to pick phases in the seismic signal data to obtain the arrival time of the P-wave in the seismic event; based on the arrival time, extracting velocity waveform data of a preset duration from the P-wave; simulating the velocity waveform data to obtain acceleration waveform data and displacement waveform data; and generating multimodal data based on the acceleration waveform data, velocity waveform data, and displacement waveform data.

[0016] In this embodiment, the arrival time of the P-wave is accurately identified using a phase picking model. Since this model can accurately determine the arrival time of different phases by analyzing seismic waveform characteristics, it improves the accuracy of multimodal data generated from waveform data of a preset duration corresponding to that arrival time. Furthermore, because P-waves are high-frequency and arrive first, generating multimodal data from P-wave waveform data improves the accuracy and timeliness of the analysis.

[0017] In some embodiments, generating multimodal data based on acceleration waveform data, velocity waveform data, and displacement waveform data includes: generating time-domain parameters and structured feature parameters based on acceleration waveform data, velocity waveform data, and displacement waveform data, respectively; and performing Fourier transforms on acceleration waveform data, velocity waveform data, and displacement waveform data, respectively, to generate frequency-domain parameters.

[0018] In this embodiment, since the waveform data is obtained through the phase picking model and the obtained data is P-wave waveform data, time-domain parameters, frequency-domain parameters and structured feature parameters are generated from the waveform data, which improves the accuracy of multi-modal data and thus improves the accuracy of earthquake identification.

[0019] In some embodiments, the acceleration waveform data includes vertical acceleration waveform data, north-south acceleration waveform data, and east-west acceleration waveform data; the velocity waveform data includes vertical velocity waveform data, north-south velocity waveform data, and east-west velocity waveform data; the displacement waveform data includes vertical displacement waveform data, north-south displacement waveform data, and east-west displacement waveform data; Fourier transforms are performed on the acceleration waveform data, velocity waveform data, and displacement waveform data respectively to generate frequency domain parameters, including: performing Fourier transforms on the vertical acceleration waveform data, north-south acceleration waveform data, east-west acceleration waveform data, vertical velocity waveform data, north-south velocity waveform data, east-west velocity waveform data, vertical displacement waveform data, north-south displacement waveform data, and east-west displacement waveform data respectively to obtain the corresponding Fourier spectra; truncating each Fourier spectrum according to a preset frequency truncation range, and normalizing each truncated spectrum to obtain the corresponding standard parameters; generating frequency domain parameters based on each standard parameter.

[0020] In this embodiment, each type of waveform data includes three components: vertical, north-south, and east-west. By performing a Fourier transform on each component, the time-domain seismic waveform is converted into a frequency-domain representation, thus intuitively reflecting the concentration of seismic wave energy in different frequency bands. Furthermore, normalization processing eliminates amplitude differences between different seismic events, improving the accuracy of frequency domain parameters.

[0021] In some embodiments, an earthquake identification result is obtained by processing time-domain parameters, frequency-domain parameters, and structured feature parameters using an earthquake signal classification model. This includes: extracting features from the time-domain parameters, frequency-domain parameters, and structured feature parameters using the feature extraction layer of the earthquake signal classification model to generate time-domain feature vectors, frequency-domain feature vectors, and statistical feature vectors; generating a fused feature vector based on the time-domain feature vectors, frequency-domain feature vectors, and statistical feature vectors using the feature fusion layer of the earthquake signal classification model; and obtaining the earthquake identification result based on the fused feature vector through the decision layer of the earthquake signal classification model.

[0022] The embodiments of this application integrate time-domain parameters, frequency-domain parameters, and structured feature parameters, enabling the seismic signal classification model to more comprehensively characterize the properties of seismic signals and improve the accuracy of seismic identification.

[0023] Fourthly, embodiments of this application provide a training apparatus for an earthquake signal classification model. The apparatus includes: a first acquisition module for acquiring training data, wherein the training data includes natural earthquake data and non-natural earthquake data; a first generation module for generating a first time-domain sequence sample, a first frequency-domain sequence sample, and a first structured feature parameter sample based on the natural earthquake data; and generating a second time-domain sequence sample, a second frequency-domain sequence sample, and a second structured feature parameter sample based on the non-natural earthquake data; a training module for inputting the first time-domain sequence sample, the first frequency-domain sequence sample, the first structured feature parameter sample, the second time-domain sequence sample, the second frequency-domain sequence sample, and the second structured feature parameter sample into an initial earthquake signal classification model to obtain predicted values; a calculation module for calculating the loss between the predicted values ​​and the true values ​​using a loss function; and an update module for updating the network parameters in the initial signal classification model according to the loss values ​​to obtain an earthquake signal classification model.

[0024] Fifthly, embodiments of this application provide an earthquake identification device, which includes: a second acquisition module for acquiring earthquake signal data; a second generation module for generating multimodal data based on the earthquake signal data; wherein the multimodal data includes time-domain parameters, frequency-domain parameters, and structured feature parameters; and a processing module for processing the time-domain parameters, frequency-domain parameters, and structured feature parameters using an earthquake signal classification model according to any embodiment of the first aspect to obtain an earthquake identification result; wherein the earthquake signal classification model is trained according to the training method of an earthquake signal classification model according to any embodiment of the second aspect.

[0025] In a sixth aspect, embodiments of this application provide an electronic device, including: a processor and a memory, the memory storing machine-readable instructions executable by the processor, the machine-readable instructions being able to perform the method steps of any one of the embodiments of the second aspect, the third aspect, or both when executed by the processor.

[0026] In a seventh aspect, embodiments of this application provide a computer-readable storage medium, comprising: computer program instructions stored on the computer-readable storage medium, wherein the computer program instructions are executed by a processor to perform the method steps of any one of the second aspect, the third aspect, or both of the two aspects.

[0027] Eighthly, embodiments of this application provide a computer program product, including: computer program instructions, which are executed by a processor to perform the method steps of any one of the second aspect, the third aspect, or both.

[0028] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing embodiments of this application. Attached Figure Description

[0029] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a schematic diagram of the structure of a seismic signal classification model provided in an embodiment of this application; Figure 2 A flowchart illustrating a training method for a seismic signal classification model provided in an embodiment of this application; Figure 3 A schematic flowchart illustrating an earthquake identification method provided in an embodiment of this application; Figure 4 A schematic diagram of the structure of a training device for a seismic signal classification model provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an earthquake identification device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the electronic device structure provided in an embodiment of this application. Detailed Implementation

[0031] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0032] It should be noted that all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0033] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0034] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0035] Microseismic signal identification is a fundamental and crucial technology in the field of earthquake monitoring and research, and it is of great significance for earthquake early warning, seismic activity analysis, and earthquake disaster prevention. With the large-scale construction of seismic stations and technological advancements, seismic networks are able to record more and more microseismic events, making it particularly important to automatically, quickly, and accurately identify different microseismic signals from complex background noise.

[0036] Microseismic signals typically include small-magnitude natural and non-natural earthquakes, which may be triggered by human activities such as mine blasting, underground nuclear tests, and the impoundment of large reservoirs. These non-natural seismic events not only increase the complexity of the earthquake catalog but may also induce secondary tectonic earthquakes, introducing more uncertainty into earthquake disaster prevention and control. Therefore, improving the accuracy and efficiency of microseismic signal identification is crucial for earthquake monitoring systems. It can reduce false alarms, improve the accuracy of the earthquake catalog, and support earthquake early warning and emergency response.

[0037] Traditional earthquake event classification methods primarily rely on the experience of seismologists, classifying events through visual analysis of waveform initial polarity, P-wave to S-wave amplitude ratio, and spectral characteristics. However, in recent years, with the increase in the number and density of monitoring stations, the volume of earthquake data has surged dramatically, making manual processing insufficient for practical needs. Against the backdrop of growing demands for big data and automation, these methods have gradually revealed their limitations, such as cumbersome extraction processes, reliance on manual labor, difficulty in rapidly processing large amounts of data, and difficulty in achieving high-precision classification under complex background noise.

[0038] The rapid advancement of information technology has greatly enhanced data analysis capabilities, leading to the increasingly widespread application of machine learning methods in earthquake event monitoring and early warning. Deep learning, as a highly efficient data processing tool, has provided new insights into earthquake signal classification. However, existing deep learning models mostly utilize only single-modal data, failing to comprehensively capture the characteristics of earthquake signals, thus limiting their recognition performance.

[0039] To address the aforementioned issues, this application provides an earthquake signal classification model, a model training method, and an earthquake identification method, which combine the time-domain, frequency-domain, and structural features of earthquake signals to identify earthquakes and improve earthquake identification accuracy.

[0040] Based on the above ideas, this application provides a seismic signal classification model. Figure 1 This is a schematic diagram of the structure of a seismic signal classification model provided in an embodiment of this application, as shown below. Figure 1 As shown, the seismic signal classification model includes a feature extraction layer, a feature fusion layer, and a decision layer connected in sequence. The feature extraction layer includes a parallel time-domain encoder, a frequency-domain encoder, and a feature parameter encoder. The time-domain encoder is used to extract features from time-domain parameters to generate a time-domain feature vector; the frequency-domain encoder is used to extract features from frequency-domain parameters to generate a frequency-domain feature vector; the feature parameter encoder is used to extract features from structured feature parameters to generate a statistical feature vector; the feature fusion layer is used to fuse the time-domain feature vector, the frequency-domain feature vector, and the statistical feature vector to generate a fused feature vector; and the decision layer is used to obtain the seismic identification result based on the fused feature vector.

[0041] Time-domain parameters refer to the original waveform of a seismic signal as it changes over time, including the three components of velocity, acceleration, and displacement. Because time-domain parameters preserve the original temporal evolution information of the seismic signal, they can directly reflect the amplitude, phase, and duration of ground motion.

[0042] The three components refer to the vertical direction, the north-south direction, and the east-west direction.

[0043] In the specific implementation process, the temporal parameters are extracted by a temporal encoder to generate a temporal feature vector.

[0044] Frequency domain parameters refer to the amplitude spectrum, phase spectrum, or power spectrum of time-domain data converted into the frequency domain through Fourier transform. They demonstrate the energy concentration of seismic signals at different frequencies.

[0045] In the specific implementation process, the frequency domain parameters are extracted by the frequency domain encoder to generate frequency domain feature vectors.

[0046] Structured feature parameters refer to statistical and physical features extracted from time-domain or frequency-domain data, such as peak parameters and integral parameters.

[0047] In the specific implementation process, the structured feature parameters are extracted by the feature parameter encoder to generate statistical feature vectors.

[0048] After generating three feature vectors, a feature fusion layer concatenates and fuses them to form a comprehensive multimodal fused feature vector. This vector contains the time-domain features, frequency-domain features, and structured data features of the seismic signal. Finally, the fused feature vector is input into the decision layer for processing to obtain the earthquake identification result.

[0049] In this embodiment, the feature extraction layer of the seismic signal classification model includes a time-domain encoder, a frequency-domain encoder, and a feature parameter encoder to extract features from the multimodal data of the seismic signal. By integrating time-domain parameters, frequency-domain parameters, and structured feature parameters, the model can more comprehensively characterize the properties of the seismic signal and improve the accuracy of earthquake identification.

[0050] In some embodiments, the temporal encoder includes seven one-dimensional convolutional layers and three long short-term memory layers connected in sequence; the frequency encoder includes seven one-dimensional convolutional layers and one long short-term memory layer connected in sequence; and the feature parameter encoder includes two fully connected layers connected in sequence.

[0051] In its implementation, the temporal encoder consists of seven one-dimensional convolutional layers and three long short-term memory layers. Each one-dimensional convolutional layer can use a convolutional kernel of size 2. It should be noted that one-dimensional convolutional layers can also use convolutional kernels of size 3. Furthermore, the kernel sizes of different convolutional layers can be the same or different.

[0052] To further enhance the model's expressive power, a ReLU activation function can be added after each convolutional layer to perform a non-linear transformation on the convolutional result.

[0053] After multi-layer feature extraction by the temporal encoder, the final temporal feature vector is obtained.

[0054] For example, the temporal feature vector can be a 50×1 dimensional feature vector.

[0055] The frequency domain encoder consists of seven one-dimensional convolutional layers and one long short-term memory layer. Each one-dimensional convolutional layer can use a convolutional kernel of size 2. It should be noted that one-dimensional convolutional layers can also use convolutional kernels of size 3. Furthermore, the kernel sizes of different convolutional layers can be the same or different.

[0056] Furthermore, to further enhance the model's expressive power, a ReLU activation function can be added after each convolutional layer in the frequency domain encoder. This allows for a non-linear transformation of the convolutional results using the ReLU activation function, effectively capturing the periodicity and frequency distribution characteristics of the frequency domain parameters.

[0057] After multi-layer feature extraction by the frequency domain encoder, the final frequency domain feature vector is obtained.

[0058] For example, the temporal feature vector can be a 30×1 dimensional feature vector.

[0059] The feature parameter encoder consists of two fully connected layers. These layers extract structured feature parameters to obtain statistical feature vectors.

[0060] For example, the feature parameter encoder can effectively encode structured features such as peak acceleration, peak velocity, and peak displacement to form an 8×1 dimensional feature vector, which concentrates the key statistical information of the seismic signal.

[0061] In this embodiment, the network structure of the encoder differs for different types of parameters, enabling the model to differentiate and target different types of parameters based on their characteristics, thereby improving the accuracy of parameter processing. Furthermore, convolutional layers and long short-term memory layers are used to construct temporal and frequency domain encoders. This type of encoder improves the accuracy of processing temporal and frequency domain parameters through the synergistic effect of local feature extraction and global temporal modeling. Fully connected layers are used to construct a feature parameter encoder. This type of encoder improves the accuracy of processing structured feature parameters through global feature integration and nonlinear mapping.

[0062] In some embodiments, the decision layer includes three fully connected layers connected in sequence as a classification layer for making classification decisions based on fused feature vectors.

[0063] For example, the output of the decision layer is a 2×1 dimensional vector that includes classification scores for two classes of samples (non-natural earthquakes and natural earthquakes).

[0064] Following the decision layer, a torch.max function is connected to extract the classification label with the highest score from the 2×1 dimension vector output by the decision layer, thus obtaining a more intuitive recognition result.

[0065] For example, if the category label is 0, the event is determined to be a non-natural earthquake; if the output label is 1, it is determined to be a natural earthquake.

[0066] Based on the above seismic signal classification model, in order to improve the accuracy and robustness of the seismic signal classification model, it is necessary to train the initial seismic signal classification model to obtain a trained seismic signal classification model for identifying seismic signal data.

[0067] Based on this, embodiments of this application provide a training method for a seismic signal classification model. It should be understood that this method can be applied to terminal devices (also known as electronic devices) and servers; the terminal device can specifically be a smartphone, tablet computer, computer, personal digital assistant (PDA), etc.; the server can specifically be an application server or a web server. To facilitate understanding of the technical solution provided by embodiments of this application, the following uses a server as the execution subject as an example to introduce the application scenario of the seismic signal classification model training method provided by embodiments of this application.

[0068] Figure 2 A flowchart illustrating a training method for a seismic signal classification model provided in this application embodiment is shown below. Figure 2 As shown, the method includes: Step S201: Obtain training data; wherein, the training data includes natural earthquake data and non-natural earthquake data.

[0069] Step S202: Generate a first time-domain sequence sample, a first frequency-domain sequence sample, and a first structured feature parameter sample based on natural earthquake data; generate a second time-domain sequence sample, a second frequency-domain sequence sample, and a second structured feature parameter sample based on non-natural earthquake data.

[0070] Step S203: Input the first time-domain sequence sample, the first frequency-domain sequence sample, the first structured feature parameter sample, the second time-domain sequence sample, the second frequency-domain sequence sample, and the second structured feature parameter sample into the initial seismic signal classification model to obtain the predicted value.

[0071] Step S204: Calculate the loss between the predicted value and the actual value using the loss function.

[0072] Step S205: Update the network parameters in the initial seismic signal classification model according to the loss value to obtain the seismic signal classification model.

[0073] In the specific implementation process, earthquake data from the China Earthquake Networks Center is collected as training data, which includes both natural and non-natural earthquake data.

[0074] Natural earthquake data refers to the recording of vibration signals generated by earthquake events triggered by natural geological activities within the Earth. Examples include plate tectonics, fault rupture, and volcanic earthquakes.

[0075] Non-natural earthquake data are records of ground vibration signals directly or indirectly caused by human activities. Examples include earthquakes induced by artificial blasting, industrial activities, and reservoirs.

[0076] When analyzing earthquake signals, it is necessary to distinguish between natural and non-natural earthquakes in order to avoid misjudgment.

[0077] Before acquiring sample data, the sampling frequency of all seismic waveform records was uniformly adjusted to 100 Hz. Waveform records for a total of 15 seconds were extracted from the P-waves of each seismic event, including the 5 seconds before and 10 seconds after their arrival. A Butterworth high-pass filter was applied to the extracted waveform records to eliminate low-frequency drift, thus obtaining preprocessed training data.

[0078] It should be noted that the duration of the waveform recording can be adjusted adaptively according to the actual situation. For example, waveform recordings of a total of 16 seconds can be extracted, including the 8 seconds before and after the arrival of the P-wave for each seismic event.

[0079] After obtaining the training data, multimodal sequence samples are generated based on the training data. These multimodal sequence samples include time-domain sequence samples, frequency-domain sequence samples, and structured feature parameter samples.

[0080] Since the training data includes both natural and non-natural earthquake data, corresponding multimodal sequence samples are generated for different types of training data.

[0081] That is, a first time-domain sequence sample, a first frequency-domain sequence sample, and a first structured feature parameter sample are generated based on natural earthquake data; a second time-domain sequence sample, a second frequency-domain sequence sample, and a second structured feature parameter sample are generated based on non-natural earthquake data.

[0082] In one implementation, the process of generating multimodal sequence samples is as follows: The time-domain sequence samples consist of 100Hz three-component waveform data within a 15-second time window, including 5 seconds before and 10 seconds after P-wave triggering. The three-component waveform data includes acceleration, velocity, and displacement waveforms, arranged in a 9... 1500-dimensional input modulus time-domain encoder.

[0083] The frequency domain sequence samples are normalized Fourier spectra of three-component waveform data (acceleration, velocity, and displacement) with a frequency range of 0.1 Hz to 20 Hz, and are presented in 9... 200-dimensional input frequency domain encoder.

[0084] The structured feature parameter samples are calculated from the three-component waveform data of acceleration, velocity, and displacement: peak acceleration Pa, peak velocity Pv, peak displacement Pd, peak value of acceleration timesvelocity Pav, peak value of acceleration timesvelocity Pad, peak value of acceleration timesvelocity Pvd, peak value of velocity timesvelocity Pvd, integral of absolute acceleration IAA, integral of absolute velocity IAV, and integral of absolute displacement IAD, with a value of 27. A 1-dimensional input feature parameter encoder. The formula for calculating the structured feature parameters is as follows:

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097]

[0098]

[0099]

[0100]

[0101]

[0102]

[0103]

[0104]

[0105]

[0106]

[0107]

[0108]

[0109]

[0110]

[0111] Where a(t)UD, a(t)NS, and a(t)EW represent the acceleration time histories in the vertical, north-south, and east-west directions, respectively; v(t)UD, v(t)NS, and v(t)EW represent the velocity time histories in the vertical, north-south, and east-west directions, respectively; and d(t)UD, d(t)NS, and d(t)EW represent the displacement time histories in the vertical, north-south, and east-west directions, respectively. tc represents the arrival time of the P-wave, and t represents time.

[0112] After obtaining the multimodal sequence samples, the initial seismic signal classification model is trained based on the multimodal sequence samples. The training process is as follows: Forward propagation: The initial seismic signal classification model receives input data, performs calculations through the network structure of each layer, and finally produces predicted output values. In this process, the flow of data in the network is unidirectional, from the input layer to the output layer.

[0113] During training, the cross-entropy loss function is used to measure the difference between the predicted output and the actual label. The cross-entropy loss function is a commonly used loss function in multi-class classification problems, effectively guiding the model to learn how to reduce prediction error.

[0114] For each sample, the difference between the probability distribution of the predicted value and the probability distribution of the true value is calculated using the loss function, and the overall loss is obtained by summing them using the expected value. The formula for calculating the classification cross-entropy loss function is as follows:

[0115] in, Indicates the number of samples. Represents the true value. This represents the predicted value.

[0116] The embodiments of this application calculate the loss value between the predicted value and the true value using a loss function, and then update the network parameters of the initial seismic signal classification model based on the loss value, thereby improving the accuracy of the seismic signal classification model.

[0117] Backpropagation: Based on the gradient calculated from the loss function, the weights and bias parameters of each layer in the network are updated by backpropagation through the network.

[0118] Parameter update: Based on the gradient calculated by backpropagation and the preset learning rate, the network parameters are updated using the Adam optimizer to minimize the loss function.

[0119] Early stopping strategy: To prevent overfitting and improve the model's generalization ability, the loss on the validation set is continuously monitored during training. If the validation loss does not decrease significantly within 10 consecutive training epochs, the model is considered to have converged, and training is terminated at this point.

[0120] Model validation: After each training cycle, the model is evaluated using an independent validation dataset.

[0121] In this embodiment, by processing natural and non-natural earthquake data, corresponding multimodal sequence samples are obtained. These multimodal sequence samples are then used to train an initial earthquake signal classification model. The network parameters are updated based on predicted and true values ​​using a loss function, enabling the trained earthquake signal classification model to identify earthquakes based on multimodal data, thereby improving the model's accuracy.

[0122] After obtaining a trained seismic signal classification model, the seismic signal data is identified using the seismic signal classification model.

[0123] Based on this, embodiments of this application provide an earthquake identification method. It should be understood that this method can be applied to terminal devices (also known as electronic devices) and servers; wherein the terminal device can specifically be a smartphone, tablet computer, computer, personal digital assistant (PDA), etc.; the server can specifically be an application server or a web server. To facilitate understanding of the technical solution provided by embodiments of this application, the application scenario of the earthquake identification method provided by embodiments of this application will be introduced below using a server as the execution subject as an example.

[0124] Figure 3 This is a flowchart illustrating an earthquake identification method provided in an embodiment of this application, as shown below. Figure 3 As shown, the method includes: Step S301: Obtain seismic signal data.

[0125] Step S302: Generate multimodal data based on seismic signal data; wherein, the multimodal data includes time domain parameters, frequency domain parameters, and structured feature parameters.

[0126] Step S303: Using the seismic signal classification model, the time domain parameters, frequency domain parameters, and structured feature parameters are processed to obtain the earthquake identification result; wherein, the seismic signal classification model is trained according to the training method of the seismic signal classification model in the above embodiment.

[0127] Seismic signal data refers to the data obtained by sensors (such as seismographs and accelerometers) recording surface or underground vibrations caused by seismic events (such as natural earthquakes and non-natural earthquakes).

[0128] In the specific implementation process, after acquiring seismic signal data, multimodal data is generated based on the seismic signal data.

[0129] Multimodal data refers to information generated from multiple different types of data.

[0130] In this embodiment, the multimodal data includes time-domain parameters, frequency-domain parameters, and structured feature parameters.

[0131] As can be seen from the above embodiments, the seismic signal classification model includes a time-domain encoder, a frequency-domain encoder, and a feature parameter encoder. When processing multimodal data using the seismic signal classification model, the time-domain parameters, frequency-domain parameters, and structured feature parameters are respectively input into the corresponding encoders for feature extraction, so that the seismic signal classification model generates earthquake identification results based on the extracted feature vectors.

[0132] In this embodiment, multimodal data is generated based on seismic signal data. Since multimodal data reflects the characteristics of seismic signals from different dimensions, the seismic signal classification model can more comprehensively characterize the characteristics of seismic signals and improve the accuracy of earthquake identification.

[0133] In some embodiments, generating multimodal data based on seismic signal data includes: using a phase picking model to pick phases in the seismic signal data to obtain the arrival time of the P-wave in the seismic event; based on the arrival time, extracting velocity waveform data of a preset duration from the P-wave; simulating the velocity waveform data to obtain acceleration waveform data and displacement waveform data; and generating multimodal data based on the acceleration waveform data, velocity waveform data, and displacement waveform data.

[0134] A phase picking model is an algorithm or neural network model used in seismology to automatically identify and label the arrival time of seismic waves (such as P-waves, S-waves, etc.). This phase picking model can utilize existing commonly used models, such as PhaseNet and EQTransformer; this application does not impose specific limitations on it.

[0135] In the specific implementation process, the acquired seismic signal data is input into the phase picking model to perform phase picking on the seismic signal data and obtain the arrival time of the P-wave in the seismic event.

[0136] Based on the arrival time, capture velocity waveform data for a preset duration. The preset duration is a pre-defined value; for example, capture velocity waveform data for 5 seconds before and 10 seconds after the arrival time of the longitudinal wave. Differentiating the velocity waveform data yields the acceleration waveform data; integrating the velocity waveform data yields the displacement waveform data.

[0137] It should be noted that existing simulation software can also be used to simulate acceleration and displacement waveforms based on velocity waveform data.

[0138] Finally, multimodal data is generated based on the acceleration waveform data, velocity waveform data, and displacement waveform data.

[0139] It should be noted that, in addition to using the phase picking model to pick phases from seismic signal data and obtain the arrival time of the P-wave in a seismic event, traditional algorithms (such as the STA / LTA long-term average ratio) can also be used to obtain the arrival time of the P-wave in a seismic event. The arrival time of the P-wave in a seismic event can also be determined through manual annotation.

[0140] In this embodiment, the arrival time of the P-wave is accurately identified using a phase picking model. Since this model can accurately determine the arrival time of different phases by analyzing seismic waveform characteristics, it improves the accuracy of multimodal data generated from waveform data of a preset duration corresponding to that arrival time. Furthermore, because P-waves are high-frequency and arrive first, generating multimodal data from P-wave waveform data improves the accuracy and timeliness of the analysis.

[0141] In some embodiments, generating multimodal data based on acceleration waveform data, velocity waveform data, and displacement waveform data includes: generating time-domain parameters and structured feature parameters based on acceleration waveform data, velocity waveform data, and displacement waveform data, respectively; and performing Fourier transforms on acceleration waveform data, velocity waveform data, and displacement waveform data, respectively, to generate frequency-domain parameters.

[0142] As can be seen from the above embodiments, each type of waveform data is three-component waveform data. Therefore, acceleration waveform data includes vertical acceleration waveform data, north-south acceleration waveform data, and east-west acceleration waveform data; velocity waveform data includes vertical velocity waveform data, north-south velocity waveform data, and east-west velocity waveform data; and displacement waveform data includes vertical displacement waveform data, north-south displacement waveform data, and east-west displacement waveform data.

[0143] According to the definition of time-domain parameters, time-domain parameters refer to the original waveform of a seismic signal that changes over time, including the three components of velocity, acceleration, and displacement.

[0144] Therefore, after obtaining acceleration waveform data, velocity waveform data, and displacement waveform data, time-domain parameters can be directly generated based on the vertical acceleration waveform data, north-south acceleration waveform data, east-west acceleration waveform data, vertical velocity waveform data, north-south velocity waveform data, east-west velocity waveform data, vertical displacement waveform data, north-south displacement waveform data, and east-west displacement waveform data.

[0145] According to the definition of structured feature parameters, structured feature parameters refer to statistical and physical features extracted from time-domain or frequency-domain data, such as peak parameters and integral parameters.

[0146] Therefore, after obtaining acceleration waveform data, velocity waveform data, and displacement waveform data, the structured feature parameters are calculated based on the above-mentioned formula for calculating structured feature parameters, using the vertical acceleration waveform data, north-south acceleration waveform data, east-west acceleration waveform data, vertical velocity waveform data, north-south velocity waveform data, east-west velocity waveform data, vertical displacement waveform data, north-south displacement waveform data, and east-west displacement waveform data.

[0147] According to the definition of frequency domain parameters, frequency domain parameters refer to the amplitude spectrum, phase spectrum, or power spectrum of time domain data converted into frequency domain data through Fourier transform.

[0148] Therefore, Fourier transforms are performed on the acceleration waveform data, velocity waveform data, and displacement waveform data respectively to generate frequency domain parameters. This includes: performing Fourier transforms on the vertical acceleration waveform data, north-south acceleration waveform data, east-west acceleration waveform data, vertical velocity waveform data, north-south velocity waveform data, east-west velocity waveform data, vertical displacement waveform data, north-south displacement waveform data, and east-west displacement waveform data respectively to obtain the corresponding Fourier spectra; truncating each Fourier spectrum according to a preset frequency truncation range, and normalizing each truncated spectrum to obtain the corresponding standard parameters; and generating frequency domain parameters based on each standard parameter.

[0149] The preset frequency is a pre-defined value. For example, for each Fourier spectrum, a frequency range of 0.1 Hz to 20 Hz is selected. Alternatively, a frequency range of 1 Hz to 10 Hz is selected.

[0150] In this embodiment, since the waveform data is obtained through the phase picking model and the obtained data is P-wave waveform data, time-domain parameters, frequency-domain parameters and structured feature parameters are generated from the waveform data, which improves the accuracy of multi-modal data and thus improves the accuracy of earthquake identification.

[0151] In some embodiments, an earthquake identification result is obtained by processing time-domain parameters, frequency-domain parameters, and structured feature parameters using an earthquake signal classification model. This includes: extracting features from the time-domain parameters, frequency-domain parameters, and structured feature parameters using the feature extraction layer of the earthquake signal classification model to generate time-domain feature vectors, frequency-domain feature vectors, and statistical feature vectors; generating a fused feature vector based on the time-domain feature vectors, frequency-domain feature vectors, and statistical feature vectors using the feature fusion layer of the earthquake signal classification model; and obtaining the earthquake identification result based on the fused feature vector through the decision layer of the earthquake signal classification model.

[0152] In the specific implementation process, the temporal encoder, frequency encoder and feature parameter encoder in the feature extraction layer of the seismic signal classification model are used to extract features of temporal parameters, frequency parameters and structured feature parameters to obtain the corresponding temporal feature vector, frequency feature vector and statistical feature vector.

[0153] After obtaining each feature vector, a fused feature vector is generated based on the time-domain feature vector, frequency-domain feature vector, and statistical feature vector using the feature fusion layer of the seismic signal classification model. The earthquake identification result is then obtained through the decision layer of the seismic signal classification model based on the fused feature vector. To make the identification results more intuitive, the earthquake identification results can be input into the `torch.max` function to obtain the final output category.

[0154] The embodiments of this application integrate time-domain parameters, frequency-domain parameters, and structured feature parameters, enabling the seismic signal classification model to more comprehensively characterize the properties of seismic signals and improve the accuracy of seismic identification.

[0155] Figure 4 This is a schematic diagram of the structure of a training device for a seismic signal classification model provided in an embodiment of this application, as shown below. Figure 4 As shown, the device includes: a first acquisition module 401, a first generation module 402, a training module 403, a calculation module 404, and an update module 405; wherein, The first acquisition module 401 is used to acquire training data, which includes natural earthquake data and non-natural earthquake data. The first generation module 402 is used to generate a first time-domain sequence sample, a first frequency-domain sequence sample, and a first structured feature parameter sample based on the natural earthquake data; and to generate a second time-domain sequence sample, a second frequency-domain sequence sample, and a second structured feature parameter sample based on the non-natural earthquake data. The training module 403 is used to input the first time-domain sequence sample, the first frequency-domain sequence sample, the first structured feature parameter sample, the second time-domain sequence sample, the second frequency-domain sequence sample, and the second structured feature parameter sample into the initial seismic signal classification model to obtain predicted values. The calculation module 404 is used to calculate the loss value between the predicted value and the true value using a loss function. The update module 405 is used to update the network parameters in the initial signal classification model according to the loss value to obtain the seismic signal classification model.

[0156] Based on the above embodiments, the calculation module 404 is specifically used to: utilize the loss function Calculate the loss between the predicted value and the actual value; where, Indicates the number of samples. Represents the true value. This represents the predicted value.

[0157] It should be understood that this device corresponds to the training method embodiment of the seismic signal classification model described above, and is capable of performing the various steps involved in the above method embodiment. The specific functions of this device can be found in the description above, and detailed descriptions are omitted here to avoid repetition. The device includes at least one software functional module that can be stored in memory or embedded in the device's operating system (OS) in the form of software or firmware.

[0158] Figure 5 This is a schematic diagram of the structure of an earthquake identification device provided in an embodiment of this application, as shown below. Figure 5 As shown, the device includes: a second acquisition module 501, a second generation module 502, and a processing module 503; wherein, The second acquisition module 501 is used to acquire seismic signal data; the second generation module 502 is used to generate multimodal data based on the seismic signal data; wherein, the multimodal data includes time-domain parameters, frequency-domain parameters, and structured feature parameters; the processing module 503 is used to process the time-domain parameters, frequency-domain parameters, and structured feature parameters using the seismic signal classification model of any embodiment of the first aspect to obtain seismic identification results; wherein, the seismic signal classification model is trained according to the training method of the seismic signal classification model of any embodiment of the second aspect.

[0159] Based on the above embodiments, the second generation module 502 is used to: use the phase picking model to pick phases of the seismic signal data to obtain the arrival time of the P-wave in the seismic event; based on the arrival time, extract velocity waveform data of a preset duration from the P-wave; simulate the velocity waveform data to obtain acceleration waveform data and displacement waveform data; and generate multimodal data based on the acceleration waveform data, velocity waveform data and displacement waveform data.

[0160] Based on the above embodiments, the second generation module 502 is specifically used to: generate time-domain parameters and structured feature parameters according to acceleration waveform data, velocity waveform data and displacement waveform data respectively; and perform Fourier transform on acceleration waveform data, velocity waveform data and displacement waveform data respectively to generate frequency-domain parameters.

[0161] Based on the above embodiments, the acceleration waveform data includes vertical acceleration waveform data, north-south acceleration waveform data, and east-west acceleration waveform data; the velocity waveform data includes vertical velocity waveform data, north-south velocity waveform data, and east-west velocity waveform data; the displacement waveform data includes vertical displacement waveform data, north-south displacement waveform data, and east-west displacement waveform data; the second generation module 502 is specifically used to: perform Fourier transform on the vertical acceleration waveform data, north-south acceleration waveform data, east-west acceleration waveform data, vertical velocity waveform data, north-south velocity waveform data, east-west velocity waveform data, vertical displacement waveform data, north-south displacement waveform data, and east-west displacement waveform data respectively to obtain the corresponding Fourier spectrum; truncate each Fourier spectrum according to a preset frequency truncation range, and normalize each truncated spectrum to obtain the corresponding standard parameters; generate frequency domain parameters based on each standard parameter.

[0162] Based on the above embodiments, the processing module 503 is specifically used to: extract features from the time-domain parameters, frequency-domain parameters, and structured feature parameters using the feature extraction layer of the seismic signal classification model, and generate time-domain feature vectors, frequency-domain feature vectors, and statistical feature vectors; generate a fused feature vector based on the time-domain feature vectors, frequency-domain feature vectors, and statistical feature vectors using the feature fusion layer of the seismic signal classification model; and obtain the earthquake identification result based on the fused feature vector through the decision layer of the seismic signal classification model.

[0163] It should be understood that this device corresponds to the earthquake identification method embodiment described above and is capable of performing the various steps involved in the above method embodiment. The specific functions of this device can be found in the description above, and detailed descriptions are omitted here to avoid repetition. The device includes at least one software functional module that can be stored in memory or embedded in the device's operating system (OS) in the form of software or firmware.

[0164] Figure 6 This is a schematic diagram of the electronic device structure provided in the embodiments of this application, such as... Figure 6 As shown, the electronic device includes a processor 601, a memory 602, and a bus 603; wherein the processor 601 and the memory 602 communicate with each other via the bus 603. The processor 601 is used to call program instructions in the memory 602 to execute the methods provided in the above-described method embodiments.

[0165] Processor 601 can be an integrated circuit chip with signal processing capabilities. The processor 601 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), 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. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor.

[0166] The memory 602 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable red-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0167] This application provides a computer program product, including: computer program instructions, which are executed by a processor to perform the methods provided in the above-described method embodiments.

[0168] This application provides a computer-readable storage medium, including: computer program instructions stored on the computer-readable storage medium, which, when executed by a processor, perform the methods provided in the above-described method embodiments.

[0169] Computer-readable storage media can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0170] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0171] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0172] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0173] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A seismic signal classification model, characterized in that, The seismic signal classification model includes a feature extraction layer, a feature fusion layer, and a decision layer connected in sequence. The feature extraction layer includes a parallel temporal encoder, a frequency encoder, and a feature parameter encoder; The time-domain encoder is used to extract features from time-domain parameters and generate a time-domain feature vector; the frequency-domain encoder is used to extract features from frequency-domain parameters and generate a frequency-domain feature vector. The feature parameter encoder is used to extract features from structured feature parameters and generate statistical feature vectors; The feature fusion layer is used to fuse the time-domain feature vector, the frequency-domain feature vector, and the statistical feature vector to generate a fused feature vector; The decision layer is used to obtain earthquake identification results based on the fused feature vector.

2. The seismic signal classification model according to claim 1, characterized in that, in, The temporal encoder comprises seven one-dimensional convolutional layers and three long short-term memory layers connected in sequence; the frequency encoder comprises seven one-dimensional convolutional layers and one long short-term memory layer connected in sequence; and the feature parameter encoder comprises two fully connected layers connected in sequence.

3. A training method for a seismic signal classification model, characterized in that, The method is used to train the seismic signal classification model according to any one of claims 1-2; the method includes: Acquire training data; wherein, the training data includes natural earthquake data and non-natural earthquake data; Based on the natural earthquake data, a first time-domain sequence sample, a first frequency-domain sequence sample, and a first structured feature parameter sample are generated; based on the non-natural earthquake data, a second time-domain sequence sample, a second frequency-domain sequence sample, and a second structured feature parameter sample are generated. The first time-domain sequence sample, the first frequency-domain sequence sample, the first structured feature parameter sample, the second time-domain sequence sample, the second frequency-domain sequence sample, and the second structured feature parameter sample are input into the initial seismic signal classification model to obtain the predicted value. The loss function is used to calculate the loss between the predicted value and the actual value. The network parameters in the initial seismic signal classification model are updated based on the loss value to obtain the seismic signal classification model.

4. The method according to claim 1, characterized in that, The calculation of the loss value between the predicted value and the true value using a loss function includes: Using loss function Calculate the loss value between the predicted value and the actual value; where, Indicates the number of samples. This represents the actual value. This represents the predicted value.

5. An earthquake identification method, characterized in that, The method includes: Acquire seismic signal data; Multimodal data is generated based on the seismic signal data; wherein, the multimodal data includes time-domain parameters, frequency-domain parameters, and structured feature parameters; Using the seismic signal classification model according to any one of claims 1-2, the time-domain parameters, the frequency-domain parameters, and the structured feature parameters are processed to obtain the seismic identification result; wherein, the seismic signal classification model is trained using the training method of the seismic signal classification model according to any one of claims 3-4.

6. The method according to claim 5, characterized in that, The generation of multimodal data based on the seismic signal data includes: The phase picking model is used to pick phases in the seismic signal data to obtain the arrival time of the P-wave in the seismic event. Based on the arrival time, extract the velocity waveform data of the longitudinal wave for a preset duration; The velocity waveform data is simulated to obtain acceleration waveform data and displacement waveform data; The multimodal data is generated based on the acceleration waveform data, the velocity waveform data, and the displacement waveform data.

7. The method according to claim 6, characterized in that, in, The step of generating the multimodal data based on the acceleration waveform data, the velocity waveform data, and the displacement waveform data includes: The time-domain parameters and the structured feature parameters are generated based on the acceleration waveform data, the velocity waveform data, and the displacement waveform data, respectively. The frequency domain parameters are generated by performing Fourier transforms on the acceleration waveform data, the velocity waveform data, and the displacement waveform data, respectively.

8. The method according to claim 7, characterized in that, in, The acceleration waveform data includes vertical acceleration waveform data, north-south acceleration waveform data, and east-west acceleration waveform data; the velocity waveform data includes vertical velocity waveform data, north-south velocity waveform data, and east-west velocity waveform data; the displacement waveform data includes vertical displacement waveform data, north-south displacement waveform data, and east-west displacement waveform data. The step of performing Fourier transforms on the acceleration waveform data, the velocity waveform data, and the displacement waveform data respectively to generate the frequency domain parameters includes: Fourier transforms are performed on the vertical acceleration waveform data, the north-south acceleration waveform data, the east-west acceleration waveform data, the vertical velocity waveform data, the north-south velocity waveform data, the east-west velocity waveform data, the vertical displacement waveform data, the north-south displacement waveform data, and the east-west displacement waveform data respectively to obtain the corresponding Fourier spectra. Each Fourier spectrum is truncated according to a preset frequency truncation range, and each truncated spectrum is normalized to obtain the corresponding standard parameters. The frequency domain parameters are generated based on each of the standard parameters.

9. The method according to any one of claims 5-8, characterized in that, The process of using a seismic signal classification model to process the time-domain parameters, frequency-domain parameters, and structured feature parameters to obtain seismic identification results includes: The feature extraction layer of the earthquake signal classification model is used to extract features from the time-domain parameters, the frequency-domain parameters, and the structured feature parameters, respectively, to generate time-domain feature vectors, frequency-domain feature vectors, and statistical feature vectors. Using the feature fusion layer of the seismic signal classification model, a fused feature vector is generated based on the time-domain feature vector, the frequency-domain feature vector, and the statistical feature vector; The earthquake identification result is obtained through the decision layer of the earthquake signal classification model based on the fused feature vector.

10. An electronic device, characterized in that, include: A processor and a memory, the memory storing machine-readable instructions executable by the processor, which, when executed by the processor, perform the method as described in any one of claims 3 to 9.

Images