In-situ single-station earthquake early warning method, electronic equipment and computer program product

Abstract

The invention provides an in-situ single-station earthquake early warning method, electronic equipment and a computer program product, and relates to the technical field of earthquake early warning. The method comprises the steps of obtaining first acceleration waveform data based on three-component acceleration of ground vibration, and judging whether an earthquake event occurs based on the first acceleration waveform data; if it is determined that the earthquake event occurs, obtaining P wave arrival time based on the first acceleration waveform data; obtaining second acceleration waveform data under the corresponding time window after the arrival of the P wave based on the obtained arrival time of the P wave; and on the basis of the second acceleration waveform data, related parameters predicted under the corresponding time window of the P-wave arrival time are obtained, and the related parameters comprise the epicentral distance, the azimuth angle, the epicentral position, the magnitude and the ground peak acceleration. The method is used for in-situ earthquake early warning earthquake event identification, P wave arrival time pickup, earthquake positioning, earthquake magnitude estimation and ground peak acceleration prediction so as to improve the reliability of in-situ earthquake early warning information.

Description

Technical Field

[0001] The present application relates to the technical field of earthquake early warning, and in particular to an on-site single-station earthquake early warning method, electronic equipment, and computer program product. Background Art

[0002] The existing long-short time average and pooling criterion algorithms used in earthquake event identification and P-wave picking for on-site earthquake early warning mainly rely on single amplitude information, and have low accuracy in earthquake event identification, and there are large errors in picking up the P-wave arrival time. The low earthquake event identification accuracy and large P-wave arrival time errors will seriously affect the reliability of subsequent on-site earthquake early warning information release; the on-site earthquake early warning epicentral distance estimation, magnitude estimation and ground peak acceleration (PGA) prediction are mainly based on prediction equations established using a single characteristic parameter extracted from the P-wave. The single characteristic parameter contains less earthquake information and cannot fully characterize the epicentral distance, magnitude and ground peak acceleration (PGA), which will also lead to relatively low reliability of on-site earthquake early warning information release; At present, in on-site earthquake early warning, earthquake event identification, P-wave arrival time picking, epicentral distance estimation, magnitude estimation and ground peak acceleration (PGA) prediction mainly rely on manual extraction of limited information from P-wave data, and have not achieved artificial intelligence processing. Summary of the invention

[0003] In view of this, the purpose of the embodiments of the present application is to provide a local single-station earthquake early warning method, electronic equipment and computer program product to improve the problems in the prior art that the reliability of local earthquake early warning information release is relatively low and artificial intelligence processing is not realized.

[0004] An embodiment of the present application provides an on-site single-station earthquake early warning method, the method comprising: obtaining first acceleration waveform data based on the three-dimensional acceleration of ground shaking, and judging whether an earthquake event occurs based on the first acceleration waveform data; if it is determined that an earthquake event occurs, obtaining a P-wave arrival time based on the first acceleration waveform data; obtaining second acceleration waveform data in a corresponding time window after the P-wave arrives based on the obtained P-wave arrival time; and obtaining relevant parameters predicted in the corresponding time window of the P-wave arrival time based on the second acceleration waveform data, wherein the relevant parameters include epicentral distance and azimuth.

[0005] In the above implementation process, the acceleration change of the ground in three directions is measured. The acceleration data of these devices (i.e., three-dimensional acceleration) are collected in real time, and the first acceleration waveform data is formed after preprocessing. The first acceleration waveform data is obtained through the three-dimensional acceleration of the ground vibration, and the first acceleration waveform data obtained is analyzed to determine whether an earthquake event will occur. If the analyzed first acceleration waveform data can determine that an earthquake event will occur, then the first acceleration waveform data obtained is used to obtain the P wave arrival time. According to the P wave arrival time, a time window is determined, and the second acceleration waveform data within the time window is obtained from the monitoring device. Using the second acceleration waveform data, by analyzing the characteristics of the wave amplitude, frequency and duration, the relevant parameters of the earthquake are predicted, wherein the relevant parameters include the epicenter distance and azimuth. Feedback information after the early warning is collected, including actual earthquake data and early warning effects, for optimizing the early warning algorithm and improving the accuracy of the early warning. Through the early warning system, this information is quickly transmitted to the areas that may be affected so that emergency response measures can be taken.

[0006] Optionally, the first acceleration waveform data is obtained based on the three-dimensional acceleration of ground vibration, and judging whether an earthquake event occurs based on the first acceleration waveform data, including: performing baseline correction on the three-dimensional acceleration, obtaining the first acceleration waveform data, and converting the first acceleration waveform data into first acceleration spectrum data; wherein the first acceleration spectrum data is used to judge whether an earthquake event occurs.

[0007] In the above implementation process, after collecting the three-dimensional acceleration data, baseline correction is first required. Baseline correction refers to eliminating non-seismic signals in the data, such as environmental noise, equipment errors, etc., to improve the recognition accuracy of seismic signals. The corrected three-dimensional acceleration data is converted into first acceleration waveform data. The time domain signal is converted into a frequency domain signal through Fourier transform, that is, the first acceleration waveform data is further converted into first acceleration spectrum data. By analyzing the first acceleration spectrum data, the characteristic frequency of the seismic signal is identified to determine whether an earthquake event occurs.

[0008] Optionally, the first acceleration waveform data obtained based on the three-dimensional acceleration of ground vibration, and judging whether an earthquake event occurs based on the first acceleration waveform data, also includes: inputting the first acceleration waveform data and the first acceleration spectrum data into a first model to judge whether an earthquake event occurs; wherein the first model is used to capture time domain and spectrum data information of earthquake events and non-earthquake events.

[0009] In the above implementation process, the first acceleration waveform data and the first acceleration spectrum data are provided as input to the first model. Before the first model is put into use, it is necessary to train it with data of historical earthquakes and non-earthquake events. During the training process, the model learns to distinguish the characteristics of earthquake events and non-earthquake events in the time domain and spectrum. The first model analyzes the input waveform data and spectrum data, captures the characteristics of earthquake events and non-earthquake events, and determines the occurrence of earthquake events.

[0010] Optionally, if it is determined that an earthquake event has occurred, the P-wave arrival time is obtained based on the first acceleration waveform data, including: inputting the first acceleration waveform data into a probability model; after the probability model performs feature extraction on the first acceleration waveform data, obtaining a probability density function of the P-wave arrival time; and obtaining the P-wave arrival time based on the probability density function.

[0011] In the above implementation process, the first acceleration waveform data after baseline correction and preprocessing is input into the probability model, and the probability model extracts features from the input first acceleration waveform data. Based on the extracted features, the probability model generates a probability density function of the P wave arrival time, and estimates the P wave arrival time according to the probability density function.

[0012] Optionally, the method of obtaining the predicted relevant parameters in the corresponding time window of the P-wave arrival time based on the second acceleration waveform data includes: inputting the second acceleration waveform data into a second model to obtain the predicted epicentral distance; inputting the second acceleration waveform data into a third model to obtain the predicted azimuth; wherein the second model is used to extract local spatial features of the second acceleration waveform data related to the epicentral distance estimation; the third model is used to extract local spatial features of the second acceleration waveform data related to the azimuth estimation; and obtaining the predicted epicenter position based on the predicted epicentral distance and the azimuth, and the earthquake location calculation rule.

[0013] In the above implementation process, the second acceleration waveform data after the arrival of the P wave is input into the second model and the third model respectively. The second model extracts the local spatial features related to the estimation of the epicentral distance from the second acceleration waveform data to predict the epicentral distance. The third model extracts the local spatial features related to the azimuth estimation from the second acceleration waveform data to predict the azimuth. Based on the predicted epicentral distance and azimuth, and the earthquake location calculation law, the predicted epicenter position is calculated.

[0014] Optionally, the relevant parameters also include: an estimated earthquake magnitude; the relevant parameters predicted in the corresponding time window of the P-wave arrival time are obtained based on the second acceleration waveform data, including: inputting the second acceleration waveform data into a fourth model to obtain the predicted estimated earthquake magnitude; wherein the fourth model is used to extract local spatial features of the second acceleration waveform data related to the earthquake magnitude estimate.

[0015] In the above implementation process, the second acceleration waveform data after the arrival of the P wave is input into the fourth model. The fourth model extracts local spatial features related to the earthquake magnitude estimation value from the second acceleration waveform data to predict the earthquake magnitude.

[0016] Optionally, the related parameters also include: ground peak acceleration; the related parameters predicted in the corresponding time window of the P-wave arrival time are obtained based on the second acceleration waveform data, including: inputting the second acceleration waveform data into a fifth model to obtain the predicted ground peak acceleration; wherein the fifth model is used to extract local spatial features of the second acceleration waveform data related to the ground peak acceleration.

[0017] In the above implementation process, the second acceleration waveform data after the P wave arrives is input into the fifth model. The fifth model extracts local spatial features related to the ground peak acceleration (PGA) from the second acceleration waveform data to predict the ground peak acceleration (PGA).

[0018] Optionally, the method further comprises: if it is determined that an earthquake event occurs, issuing on-site earthquake warning information; wherein the on-site earthquake warning information comprises an earthquake event identification result and the related parameter prediction result.

[0019] In the above implementation process, once an earthquake event is determined to occur through the above model, the earthquake event identification results and the prediction results of all relevant parameters are integrated into local earthquake warning information.

[0020] An embodiment of the present application further provides an electronic device, which includes a memory and a processor, wherein the memory stores program instructions, and when the processor reads and runs the program instructions, the steps in any of the above implementation methods are executed.

[0021] In the above implementation process, the electronic device has sufficient hardware resources, including but not limited to processors, memories, input / output interfaces, etc., to support program operation and data processing. The memory stores program instructions, which include algorithms and models for executing the local single-station earthquake early warning method. The processor is responsible for reading and running the program instructions in the memory to execute each step in the local single-station earthquake early warning method. The program instructions implement functions including data acquisition, baseline correction, feature extraction, model prediction, warning information generation and release, etc.

[0022] An embodiment of the present application further provides a computer program product, which includes a computer program / instructions. When the computer program / instructions are executed by a processor, the steps in any of the above implementation methods are performed.

[0023] In the above implementation process, the functions that the computer program product needs to implement are determined, including the collection, processing, analysis and release of early warning information of earthquake data. According to the results of the demand analysis, the architecture and components of the program are designed so that the program can efficiently execute the various steps in the local single-station earthquake early warning method. The program is decomposed into multiple modules, each of which is responsible for a specific function, such as data acquisition module, data processing module, early warning algorithm module, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 A schematic diagram of an on-site single-station earthquake early warning method provided in an embodiment of the present application;

[0026] Figure 2 An overall schematic diagram of an on-site single-station earthquake early warning method provided in an embodiment of the present application;

[0027] Figure 3 An overall schematic diagram of an on-site single-station earthquake early warning method provided in an embodiment of the present application;

[0028] Figure 4 A schematic diagram of a first model preprocessing provided in an embodiment of the present application;

[0029] Figure 5 A schematic diagram of a first model provided in an embodiment of the present application;

[0030] Figure 6A schematic diagram of a probability model provided in an embodiment of the present application;

[0031] Figure 7 A schematic diagram of a second model provided in an embodiment of the present application;

[0032] Figure 8 A schematic diagram of a third model provided in an embodiment of the present application;

[0033] Fig. 9 A schematic diagram of a fourth model provided in an embodiment of the present application;

[0034] Fig.10 A schematic diagram of a fifth model provided in an embodiment of the present application;

[0035] Fig.11 A schematic diagram of an electronic device provided in an embodiment of the present application.

[0036] Icon: 100 - electronic device; 111 - memory; 112 - storage controller; 113 - processor; 114 - peripheral interface; 115 - input and output unit; 116 - display unit. DETAILED DESCRIPTION

[0037] The technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments of the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the embodiments of the present application.

[0038] See also Figure 1 , Figure 1 A schematic diagram of an on-site single-station earthquake early warning method provided in an embodiment of the present application.

[0039] An embodiment of the present application provides an on-site single-station earthquake early warning method, the method comprising: obtaining first acceleration waveform data based on the three-dimensional acceleration of ground shaking, and judging whether an earthquake event occurs based on the first acceleration waveform data; if it is determined that an earthquake event occurs, obtaining a P-wave arrival time based on the first acceleration waveform data; obtaining second acceleration waveform data in a corresponding time window after the P-wave arrives based on the obtained P-wave arrival time; and obtaining relevant parameters predicted in the corresponding time window of the P-wave arrival time based on the second acceleration waveform data, wherein the relevant parameters include epicentral distance and azimuth.

[0040] In the above implementation process, the acceleration change of the ground in three directions (usually vertical, north-south and east-west) is measured. The acceleration data (i.e., three-dimensional acceleration) of these devices are collected in real time, and the first acceleration waveform data is formed after preprocessing. The first acceleration waveform data is obtained through the three-dimensional acceleration of the ground vibration, and the first acceleration waveform data obtained is analyzed to determine whether an earthquake event will occur. If the analyzed first acceleration waveform data can determine that an earthquake event will occur, then the first acceleration waveform data obtained is used to obtain the P wave arrival time. According to the P wave arrival time, a time window is determined, and the second acceleration waveform data within the time window is obtained from the monitoring device. Using the second acceleration waveform data, by analyzing the characteristics of the wave amplitude, frequency and duration, the relevant parameters of the earthquake are predicted, wherein the relevant parameters include the epicenter distance and azimuth. Feedback information after the early warning is collected, including actual earthquake data and early warning effects, is used to optimize the early warning algorithm and improve the accuracy of the early warning. Through the early warning system, this information is quickly transmitted to the areas that may be affected so that emergency response measures can be taken.

[0041] In one embodiment of the present application, the on-site single-station earthquake early warning method is applied to major projects such as high-speed railways. The overall goal of the earthquake monitoring and early warning system for major projects such as high-speed railways based on artificial intelligence participation is to complete earthquake monitoring and early warning, issue earthquake emergency response orders, and ensure driving safety. The system can quickly estimate the basic earthquake parameters based on the ground motion parameters measured in real time by the seismic station and the initial earthquake information, determine the earthquake impact range and alarm level, and issue earthquake emergency response information to local and non-local railways within the earthquake impact range before the destructive seismic waves arrive, and activate relevant systems to take effective emergency response measures for driving.

[0042] Please combine Figure 1 , see Figure 2 , Figure 2 An overall schematic diagram of the on-site single-station earthquake early warning method provided in an embodiment of the present application.

[0043] Optionally, first acceleration waveform data is obtained based on the three-dimensional acceleration of ground vibration, and whether an earthquake event occurs is determined based on the first acceleration waveform data, including: after baseline correction of the three-dimensional acceleration, the first acceleration waveform data is obtained, and the first acceleration waveform data is converted into first acceleration spectrum data; wherein the first acceleration spectrum data is used to determine whether an earthquake event occurs.

[0044] In the above implementation process, after collecting the three-dimensional acceleration data, baseline correction is first required. Baseline correction refers to eliminating non-seismic signals in the data, such as environmental noise, equipment errors, etc., to improve the recognition accuracy of seismic signals. The corrected three-dimensional acceleration data is converted into first acceleration waveform data. The time domain signal is converted into a frequency domain signal by Fourier transform, that is, the first acceleration waveform data is further converted into first acceleration spectrum data. By analyzing the first acceleration spectrum data, the parameters of the spectrum analysis, such as frequency range, resolution, etc., are set to optimize the characteristic frequency recognition of the seismic signal and determine whether a seismic event occurs. The judgment of a seismic event can be based on specific features in the spectrum data, such as the frequency characteristics of P waves and S waves, or a sudden increase in spectrum energy.

[0045] In one embodiment of the present application, in the identification of earthquake events, the first acceleration waveform data is analyzed, and whether an earthquake event has occurred is determined by deep learning methods such as threshold judgment and pattern recognition. If the waveform data meets the characteristics of an earthquake, such as the rapid rise and fall of the P wave and the long-lasting vibration of the S wave, it is determined to be an earthquake event.

[0046] Optionally, obtaining first acceleration waveform data based on the three-dimensional acceleration of ground vibration, and judging whether an earthquake event occurs based on the first acceleration waveform data, also includes: inputting the first acceleration waveform data and the first acceleration spectrum data into a first model to judge whether an earthquake event occurs; wherein the first model is used to capture time domain and spectrum data information of earthquake events and non-earthquake events.

[0047] In the above implementation process, the three-dimensional acceleration data is collected and baseline corrected to obtain the first acceleration waveform data, and the corrected first acceleration waveform data is Fourier transformed to obtain the first acceleration spectrum data. The first acceleration waveform data and the first acceleration spectrum data are provided as input to the first model. Before the first model is put into use, it is necessary to train it with data from historical earthquakes and non-earthquake events. During the training process, the model learns to distinguish the characteristics of earthquake events and non-earthquake events in the time domain and spectrum. The first model analyzes the input waveform data and spectrum data, captures the characteristics of earthquake events and non-earthquake events, and determines the occurrence of earthquake events.

[0048] In one embodiment of the present application, actual earthquake data and feedback on early warning effects are collected to evaluate the performance of the first model. Model parameters are adjusted or the model is retrained based on the feedback to improve its accuracy and robustness. The first model is regularly updated to incorporate new earthquake event data and the latest seismological research results. The hardware and software environment of the model is maintained to ensure stable operation.

[0049] Optionally, if an earthquake event is determined to have occurred, the P-wave arrival time is obtained based on the first acceleration waveform data, including: inputting the first acceleration waveform data into a probability model; after the probability model performs feature extraction on the first acceleration waveform data, obtaining a probability density function of the P-wave arrival time; and obtaining the P-wave arrival time based on the probability density function.

[0050] In the above implementation process, the first acceleration waveform data that has been baseline corrected and preprocessed is input into the probability model, and the probability model extracts features from the input first acceleration waveform data. Feature extraction includes but is not limited to the amplitude, frequency, time points of the peaks and troughs, the starting point of the waveform, etc. Based on the extracted features, the probability model generates a probability density function of the P wave arrival time. This probability density function describes the possibility of the P wave arriving at different time points, and can be a normal distribution, a uniform distribution, or other statistical distribution suitable for describing the P wave arrival time. According to the probability density function, the P wave arrival time is estimated.

[0051] In one embodiment of the present application, the P wave arrival time is estimated based on the probability density function. This may be done by finding the maximum point of the probability density function or calculating the expected value as the estimated arrival time of the P wave. Preferably, the maximum point is selected as the P wave arrival time.

[0052] Optionally, based on the second acceleration waveform data, relevant parameters predicted in the corresponding time window of the P-wave arrival time are obtained, including: inputting the second acceleration waveform data into a second model to obtain a predicted epicentral distance; inputting the second acceleration waveform data into a third model to obtain a predicted azimuth; wherein the second model is used to extract local spatial features of the second acceleration waveform data related to the epicentral distance estimation; the third model is used to extract local spatial features of the second acceleration waveform data related to the azimuth estimation; and the predicted epicenter position is obtained based on the predicted epicentral distance and azimuth, and the earthquake location calculation rule.

[0053] In the above implementation process, the second acceleration waveform data after the arrival of the P wave is input into the second model and the third model respectively. The second model extracts local spatial features related to the estimation of the epicentral distance from the second acceleration waveform data to predict the epicentral distance. The third model extracts local spatial features related to the azimuth estimation from the second acceleration waveform data to predict the azimuth. Based on the predicted epicentral distance and azimuth, and the earthquake location calculation rules, the predicted epicenter position is calculated. These features may include the amplitude, frequency, arrival time difference, phase of the waveform, etc.

[0054] Optionally, the relevant parameters also include: an estimated earthquake magnitude; based on the second acceleration waveform data, obtaining relevant parameters predicted in the corresponding time window of the P-wave arrival time, including: inputting the second acceleration waveform data into a fourth model to obtain a predicted estimated earthquake magnitude; wherein the fourth model is used to extract local spatial features of the second acceleration waveform data related to the earthquake magnitude estimation.

[0055] In the above implementation process, the second acceleration waveform data after the P wave arrives is input into the fourth model. The fourth model extracts local spatial features related to the earthquake magnitude estimation value from the second acceleration waveform data to predict the earthquake magnitude. These features may include the amplitude, frequency, energy distribution, duration of the waveform, etc.

[0056] Optionally, the relevant parameters also include: ground peak acceleration; based on the second acceleration waveform data, obtaining relevant parameters predicted under the corresponding time window of the P-wave arrival time, including: inputting the second acceleration waveform data into the fifth model to obtain the predicted ground peak acceleration; wherein the fifth model is used to extract local spatial features related to the second acceleration waveform data and the ground peak acceleration.

[0057] In the above implementation process, the second acceleration waveform data after the P wave arrives is input into the fifth model. The fifth model extracts local spatial features related to the ground peak acceleration (PGA) from the second acceleration waveform data to predict the ground peak acceleration (PGA). These features may include the amplitude, frequency, peak time, energy distribution, etc. of the waveform.

[0058] Optionally, the method further includes: if it is determined that an earthquake event has occurred, issuing on-site earthquake warning information; wherein the on-site earthquake warning information includes earthquake event identification results and related parameter prediction results.

[0059] In the above implementation process, once an earthquake event is determined to occur through the above model, the earthquake event identification results and the prediction results of all relevant parameters (such as epicenter distance, azimuth, magnitude, ground peak acceleration, etc.) are integrated into local earthquake warning information.

[0060] In one embodiment of the present application, the warning information is formatted into a format that is easy to understand and operate, such as a text message, chart or map. The information contains all key data, such as the warning level, the area that may be affected, the recommended actions, etc. The warning information is released through appropriate communication channels, which may include mobile phone text messages, radio, television, the Internet, dedicated warning systems, etc., so that the warning information can be quickly and widely disseminated to the areas that may be affected.

[0061] Please combine Figure 2 , see Figure 3 , Figure 3 An overall schematic diagram of an on-site single-station earthquake early warning method provided in an embodiment of the present application.

[0062] In one embodiment of the present application, first, the three-dimensional acceleration waveform data recorded in real time by the seismic station is used as the input of the multimodal deep learning on-site earthquake early warning earthquake event identification model (HRSeisNet model, i.e., the first model), and the HRSeisNet model is used to determine whether it is an earthquake event. If it is determined to be an earthquake event, the three-dimensional acceleration waveform data recorded in real time by the seismic station is used as the input of the on-site earthquake early warning P-wave picking model (HRPNet model, i.e., the probability model) based on the Unet network, and the HRPNet model is used to obtain the P-wave arrival time; then, based on the P-wave arrival time obtained by the HRPNet model, the three-dimensional acceleration after the P-wave arrival recorded in real time by the seismic station is used as the convolutional neural network on-site earthquake early warning epicenter distance estimation model (HRDisNet model, i.e., the second model ), the azimuth estimation model of the local earthquake early warning police based on convolutional neural network (HRAziNet model, i.e. the third model), the magnitude estimation model of the local earthquake early warning based on convolutional neural network and Transformer (HRMagNet model, i.e. the fourth model) and the ground peak acceleration (PGA) prediction model of the local earthquake early warning based on long short-term memory network (HRPGANet model, i.e. the fifth model) are input, the epicentral distance predicted by the HRDisNet model, the azimuth predicted by the HRAziNet model, the magnitude predicted by the HRMagNet model and the ground peak acceleration (PGA) predicted by the HRPGANet model under the corresponding time window of P wave arrival can be obtained, and the predicted epicentral position can be obtained by using the earthquake location formula based on the epicentral distance and azimuth and the predicted epicentral distance and the predicted azimuth.

[0063] Specifically, see Figure 4 , Figure 4 A schematic diagram of a first model preprocessing provided in an embodiment of the present application.

[0064] In one embodiment of the present application, the real-time three-dimensional acceleration waveform (a) of the ground vibration is obtained according to the monitoring of the seismic station. UD 、a EW 、a NS ). Real-time three-dimensional acceleration waveform (a UD 、a EW 、a NS ) to perform baseline correction and obtain the three-dimensional acceleration waveform after baseline correction (A UD , A EW , A NS ); then the baseline-corrected three-dimensional acceleration waveform (A UD , A EW , ANS ) to perform fast Fourier transform (FFT) to obtain the three-dimensional acceleration waveform (A) after baseline correction. UD , A EW , A NS ) corresponds to the three-dimensional spectrum diagram (Spec UD Spec EW Spec NS ).

[0065] Please combine Figure 4 , see Figure 5 , Figure 5 A schematic diagram of a first model provided in an embodiment of the present application.

[0066] Will Figure 4 The three-dimensional acceleration waveform (A) obtained after baseline correction of data preprocessing UD , A EW , A NS ) and three-dimensional spectrum (Spec UD Spec EW Spec NS ) as the input of the HRSeisNet model. The HRSeisNet model mainly consists of a time domain encoder, a spectrum encoder and a prediction module. The three-dimensional acceleration waveform after baseline correction (A UD , A EW , A NS ) is used as the input of the time domain encoder of the HRSeisNet model. The shape of the input tensor is a matrix of (N, 3), where N represents the number of sampling points used as the input of the time domain encoder and 3 represents the three-dimensional spectrum. UD Spec EW Spec NS ) is used as the input of the spectrum encoder of the HRSeisNet model. The shape of its input tensor is (2, N, 3), where 2 represents the 2D spectrum graph, N represents the number of sampling points of the 2D spectrum graph, and 3 represents the three-dimensional direction.

[0067] The time domain encoder mainly consists of 3 1D convolutional layers, 3 1D maximum pooling layers, 2 BiLSTM layers, and 2 Dropout layers. The 1D convolutional layer performs convolution operations on the input through sliding convolution kernels, allowing the 1D convolutional layer to capture local spatial features in the input data that are relevant to the recognition of earthquake events for on-site earthquake warning. The number of convolution kernels in each convolutional layer is 128, 64, and 32, respectively, the size of the convolution kernel is 3, the moving step is 1, and each convolutional layer uses the rectified linear unit activation function (ReLU function). The type of padding in each convolutional layer is "same", and the weight initialization type is "Henormal". Each convolutional layer is followed by a maximum pooling layer to extract important spatial features and reduce the complexity of model calculation. The size of the pooling kernel and the moving step in each pooling layer are both 2, and the type of padding is "same". The BiLSTM layer better captures the contextual information of each time step in the sequence by processing the input sequence in two directions (forward and backward), thereby extracting the corresponding time series features related to the recognition of earthquake events for on-site earthquake warning. Each BiLSTM layer has 128 and 64 units, respectively. In the first BiLSTM layer, the "return sequences" parameter is set to "True", which means that the first BiLSTM layer returns the entire sequence after each round of iteration, which is helpful for processing time series data or text tasks that require understanding the context of the entire sequence. In the second BiLSTM layer, the "return sequences" parameter is set to "False", which means that the second BiLSTM layer only returns the last output of the sequence after the last round of iterations, because returning only the final output can reduce the amount of calculation. In order to reduce overfitting, the second pooling layer and the second BiLSTM layer are followed by a Dropout layer, and the parameter in the Dropout layer is set to 0.2. Through the time domain encoder, the time domain features related to the identification of earthquake events for on-site earthquake warning are output.

[0068] The spectral encoder mainly consists of 2 1D convolutional layers, 2 1D max pooling layers, 2 BiLSTM layers, and 2 Dropout layers. The number of convolution kernels in each convolutional layer is 64 and 32 respectively, the convolution kernel size is 3, the moving step is 1, and each convolutional layer uses the ReLU activation function. The padding type in each convolutional layer is "same", and the weight initialization type is "He normal". Each convolutional layer is followed by a max pooling layer. The pooling kernel size and moving step in each pooling layer are both 2, and the padding type is "same". Each BiLSTM layer has 64 and 32 units respectively. Each BiLSTM layer has 64 and 32 units respectively. In the first BiLSTM layer, the "return sequences" parameter is set to "True"; in the second BiLSTM layer, the "return sequences" parameter is set to "False". In order to reduce overfitting, the second pooling layer and the second BiLSTM layer are followed by a Dropout layer respectively, and the parameter in the Dropout layer is set to 0.2. Through the spectrum encoder, the spectrum characteristics related to the identification of earthquake events for on-site earthquake warning are output.

[0069] Then, the HRSeisNet model uses the Concatenate layer to concatenate the time domain features extracted by the time domain encoder related to the identification of earthquake events for on-site earthquake warning and the spectral features extracted by the spectral encoder related to the identification of earthquake events for on-site earthquake warning to achieve feature fusion. The result of feature fusion is then input into the prediction module. The prediction module consists of three fully connected layers, and the number of neurons in each fully connected layer is 64, 32, and 2 respectively. The first two fully connected layers use the ReLU activation function; the last fully connected layer uses the sigmoid activation function to determine whether it is an earthquake event or a non-earthquake event. If it is determined to be an earthquake event, the following steps are executed.

[0070] Optionally, regarding the training and verification of the HRSeisNet model. During the training of the HRSeisNet model, the Adam optimizer was used, the initial learning rate was set to 0.001, and the loss function was the cross entropy loss function. The HRSeisNet model was trained based on the TensorFlow2.3 framework. During the training process, in order to optimize the learning rate, the "ReduceLROnPlateau" recall function was used, which is mainly used to monitor the accuracy of the validation set. In this study, when the accuracy of the validation set no longer increases after 5 iterations, the current learning rate is multiplied by 0.1 as the new learning rate in the next iterative training process. At the same time, in order to reduce model overfitting, the "EarlyStopping" recall function is also used to monitor the accuracy of the validation set, that is: when the accuracy of the validation set no longer increases after 10 iterations, the model will stop training and save the model with the highest accuracy of the current validation set as the final trained HRSeisNet model.

[0071] See also Figure 6 , Figure 6 A schematic diagram of a probability model provided in an embodiment of the present application.

[0072] If it is judged to be an earthquake event, the three-dimensional acceleration waveform (A UD , A EW , A NS ) as the input of the HRPNet model. The shape of the input tensor of the HRPNet model is a matrix of (N, 3), where N represents the three-dimensional acceleration waveform (A) as the input of the HRPNet model. UD , A EW , A NS) is the number of sampling points, 3 represents the three-way. The HRPNet model consists of 5 encoders and 4 decoders, and the feature flow between encoders and decoders is realized through jump connections. The encoder is closer to the input than the decoder, and the low-dimensional feature map it extracts contains more original waveform information. The jump connection retains this information as much as possible. The role of the encoder is feature extraction in the downsampling process. Downsampling extracts more abstract high-level information features related to the P-wave picking of earthquake early warning in the field by reducing the size of the feature map. For the first four encoders, each encoder contains 2 1D convolutional layers and 1D maximum pooling layers. The size of the convolution kernel in the convolution layer of each encoder is 3, and the number of convolution kernels is 16, 32, 64 and 128 respectively. The moving step of the convolution kernel is 1. Each convolution layer uses L2 regularization with a parameter of 0.003 to reduce model overfitting. Each convolution layer uses the ReLU activation function. The size and moving step of the pooling kernel of the maximum pooling layer in each encoder are both 2. At the same time, the padding type of the convolution layer and pooling layer in each encoder is "same". The fifth encoder contains two 1D convolutional layers, and the number of convolution kernels in each convolutional layer is 256.

[0073] The role of the decoder is to extract features during the upsampling process. Upsampling restores the image spatial structure and captures the detailed information features related to the on-site earthquake warning P wave picking by enlarging the size of the feature map. Each decoder consists of an upsampling layer, a splicing layer, and two convolutional layers. The upsampling layer performs upsampling with a parameter of 2; the splicing layer is used to integrate multiple feature maps extracted by the previous layer of the jump connection; the two convolutional layers have the same role as the encoder. At the same time, the size of the convolution kernel in the convolutional layer of each decoder is 3, and the number of convolution kernels is 128, 64, 32, and 16, respectively. The moving step of the convolution kernel is 1. Each convolutional layer uses L2 regularization with a parameter of 0.003 to reduce model overfitting. Each convolutional layer uses the ReLU activation function.

[0074] Finally, the feature map output by the decoder is integrated with the probability density function of the P wave through a single-layer convolution structure. The size of the convolution kernel is 1, the number of convolution kernels is 1, and the step size is 1. The softmax activation function is used to output the probability density function of the P wave arrival time. The point with the largest probability is the P wave arrival time.

[0075] See also Figure 7 , Figure 7 A schematic diagram of a second model provided in an embodiment of the present application.

[0076] The three-dimensional acceleration waveform (A) of the corresponding time window after the P wave arrives is obtained. UD , A EW , A NS) is used as the input of the HRDisNet model. The shape of the input tensor is (N,3), where N represents the number of sampling points used as the input of the HRDisNet model after the P wave arrives, and 3 represents the three directions. The HRDisNet model mainly consists of 3 convolution modules, 1 flattening layer, 2 fully connected modules and a fully connected layer. Each convolution module consists of 1 1D convolution layer, 1 normalization layer and 1 1D maximum pooling layer; each fully connected layer consists of 1 fully connected layer and 1 Dropout layer. The 1D convolution layer is used to automatically extract the local spatial features of the input data related to the estimation of the epicenter distance of the on-site earthquake warning. The convolution kernel slides on the input data and generates feature maps by calculating the dot product and applying the activation function. These feature maps not only capture the changes in the acceleration data in the time dimension, but also implicitly contain the relationship between different directions. The number of convolution kernels in the convolution layers of the three convolution modules is 32, 64, and 128, respectively. The size of the convolution kernel is 3, and the moving step of the convolution kernel is 1. Each convolution layer uses the ReLU activation function to provide nonlinear mapping. A normalization layer is used after each convolution layer to accelerate training and improve the stability of model training. The internal covariate shift is reduced by normalizing the output of each layer. A 1D maximum pooling layer is used after each normalization layer. The size of the pooling kernel in the pooling layer is 2. Inserting the pooling layer can reduce the dimension of the feature map, thereby reducing the amount of calculation and reducing model overfitting. The convolution module finally flattens the output of the last pooling layer into a one-dimensional vector through the flattening layer and inputs it to the fully connected module. The number of neurons in the fully connected layers of the two fully connected modules is 128 and 64 respectively. And the ReLU activation function is used. The last fully connected layer has 1 neuron and uses a linear activation function to output the predicted epicenter distance. At the same time, in order to reduce model overfitting, L2 regularization is used in each convolutional layer and the fully connected layer in the fully connected module, and the dropout rate of the Dropout layer in each fully connected module is set to 0.3.

[0077] See also Figure 8 , Figure 8 A schematic diagram of a third model provided in an embodiment of the present application.

[0078] The three-dimensional acceleration waveform (A) of the corresponding time window after the P wave arrives is obtained. UD , A EW , A NS) is used as the input of the HRAziNet model. The shape of the input tensor is (N,3), where N represents the number of sampling points used as the input of the HRAziNet model after the P wave arrives, and 3 represents the three directions. The HRAziNet model mainly consists of 2 convolution modules, 1 flattening layer, and 3 fully connected modules. Each convolution module consists of a 1D convolution layer, an attention mechanism, a normalization layer, and a 1D maximum pooling layer. The 1D convolution layer is used to automatically extract the local spatial features of the input data related to the prediction of the local earthquake warning police position angle. The convolution kernel slides on the input data, and generates feature maps by calculating the dot product and applying the activation function. These feature maps not only capture the changes in the acceleration data in the time dimension, but also implicitly contain the relationship between different directions. The number of convolution kernels in the convolution layers in the two convolution modules is 128 and 256 respectively, the size of the convolution kernel is 3, the moving step of the convolution kernel is 2, and each convolution layer uses the ReLU activation function to provide nonlinear mapping. The attention mechanism is used after each convolution layer. The attention mechanism allows the model to dynamically focus on the important parts of the data when processing the input data. Adding the attention mechanism after the convolutional layer can further enhance the model's ability to capture key features, thereby improving the performance of the model. The normalization layer is used after the attention mechanism. A 1D maximum pooling layer is used after each normalization layer. The size of the pooling kernel in the pooling layer is 2. Inserting the pooling layer can reduce the dimension of the feature map, thereby reducing the amount of calculation and reducing model overfitting. The convolution module finally flattens the output of the last pooling layer into a one-dimensional vector through the flattening layer and inputs it to the fully connected layer. The number of neurons in the three fully connected layers is 32, 16, and 1 respectively. The first two fully connected layers use the ReLU activation function. The last fully connected layer has 1 neuron and uses a linear activation function to output the predicted azimuth. At the same time, in order to reduce model overfitting, each convolutional layer and the first two fully connected layers use L2 regularization.

[0079] Optionally, the predicted epicenter position can be obtained by using an earthquake location formula based on epicenter distance and azimuth according to the predicted epicenter distance obtained by the HRDisNet model and the predicted azimuth obtained by the HRAziNet model. This formula refers to the "circle-circle" location method in earthquake location, also known as the "intersection circle method". This method uses the P-wave arrival time recorded by two or more seismic monitoring stations, combined with the propagation speed of seismic waves, to calculate the distance from the epicenter to each monitoring station (epicenter distance), and then determines the location of the epicenter based on these distances and the azimuth of the monitoring station.

[0080] See also Fig. 9 , Fig. 9 A schematic diagram of a fourth model provided in an embodiment of the present application.

[0081] The HRMagNet model combines convolutional neural networks and Transformers. The advantage of the HRMagNet model is that it takes into account the ability of convolutional neural networks to extract local spatial features from input data and the ability of Transformer networks to solve long-term dependency problems. At the same time, the advantage of the Transformer network is the attention mechanism, which can guide the network to extract the most meaningful features from the input data, and through training, the network gives higher weights to important information in the input data and lower weights to other samples.

[0082] The three-dimensional acceleration waveform (A) of the corresponding time window after the P wave arrives is obtained. UD , A EW , A NS ) is used as the input of the HRMagNet model. The shape of the input tensor is (N,3), where N represents the number of sampling points after the P wave arrives as the input of the HRMagNet model, and 3 represents the three-dimensional. Five convolution blocks are used to convolution the three-dimensional acceleration waveform (A UD , A EW , A NS) for feature extraction. Each convolutional block consists of a 1D convolutional layer, a ReLU activation function, and a 1D max pooling layer. In each convolutional block, each convolutional layer has 256 convolutional kernels of size 3 and the moving step of the convolutional kernel is 3, and the type of convolutional layer parameter initialization is "Henormal". At the same time, the size of the pooling kernel in the 1D max pooling layer is 4, and the moving step is 2. The padding type in the 1D convolutional layer and the 1D max pooling layer is "same". The output results of the 5 convolutional blocks are used as the input of 5 Transformer encoders. Each Transformer encoder consists of two normalization layers, a multi-head attention mechanism, and two fully connected layers. A normalization layer is used before the multi-head attention mechanism. Then, after the multi-head attention mechanism, 1 normalization layer and 2 fully connected layers are connected in sequence. The two fully connected layers have 512 and 256 neurons respectively. Residual connection is implemented after the multi-head attention mechanism and the fully connected layer. The residual path is added between the input sequence and the output of the multi-head attention mechanism. Its purpose is to enhance the learning ability of the network by mixing the extracted low-order and high-order features. At the same time, the final output of the Transformer encoder is obtained by adding the output of the fully connected layer and the input of the last normalization layer. Finally, a Dropout layer is connected after the Transformer encoder, and the dropout rate of the Dropout layer is set to 0.5. For the output results of the Dropout layer, we used the global average pooling layer and the global maximum pooling layer for dimensionality reduction, which played a role in reducing overfitting to a certain extent. The global average pooling layer focuses more on extracting the overall information features related to the magnitude estimation of the on-site earthquake warning, while the global maximum pooling layer focuses more on extracting the most significant information features related to the magnitude estimation of the on-site earthquake warning. Then, the overall information features and the most significant information features of the output of the global average pooling layer and the global maximum pooling layer are spliced ​​by the splicing layer and input into two fully connected layers. Among them, the first fully connected layer has 256 neurons and uses the ReLU activation function. The last fully connected layer has one neuron and uses a linear activation function to output the magnitude estimate.

[0083] See also Fig.10 , Fig.10 A schematic diagram of a fifth model provided in an embodiment of the present application.

[0084] The three-dimensional acceleration waveform (A) of the corresponding time window after the P wave arrives is obtained. UD , A EW , A NS) is used as the input of the HRPGANet model. The shape of the input tensor is (N,3), where N represents the number of sampling points used as the input of the HRPGANet model after the P wave arrives, and 3 represents the three-dimensionality. The HRPGANet model mainly consists of 4 LSTM modules, 1 flattening layer, and 2 fully connected layers. Each LSTM module consists of 1 LSTM layer, an attention mechanism, and a Dropout layer. The number of units in the LSTM layer in the 4 LSTM modules is 256, 128, 64, and 32, respectively. At the same time, in each LSTM layer, the "return sequences" parameter is set to "True", and the LSTM layer returns the entire sequence after each round of iteration, which helps to process tasks such as time series data or text that require understanding of the entire sequence context, and then extract time series features related to the prediction of the peak ground acceleration (PGA) for on-site earthquake warning. In each LSTM module, an attention mechanism is added after the LSTM layer. By introducing the attention mechanism, the LSTM module can adaptively adjust the attention to different parts of the input sequence, better capture the important information in the sequence, and reduce the attention to irrelevant information, which can improve the model's ability to model time series and improve prediction performance. In addition, the attention mechanism can clearly show the degree of attention paid by the model to different parts when processing the input sequence by calculating the attention weight. This helps to understand the decision-making process of the model and improve the interpretability of the model. In order to reduce the overfitting of the model, in each LSTM module, the attention mechanism is followed by a Dropout layer, and the dropout rate in the Dropout layer is set to 0.2. The output of the fourth LSTM module is flattened into a 1-dimensional feature vector using a flattening layer and used as the input of two fully connected layers. The number of neurons in the two fully connected layers is 100 and 1 respectively. The first fully connected layer uses the ReLU activation function; the second fully connected layer uses the linear activation function to output the predicted peak ground acceleration (PGA).

[0085] See also Fig.11 , Fig.11 A block diagram of an electronic device provided in an embodiment of the present application.

[0086] An embodiment of the present application further provides an electronic device, which includes a memory and a processor, wherein program instructions are stored in the memory, and when the processor reads and runs the program instructions, the steps in any of the above implementation methods are executed.

[0087] In the above implementation process, the electronic device has sufficient hardware resources, including but not limited to processors, memories, input / output interfaces, etc., to support program operation and data processing. The memory stores program instructions, which include algorithms and models for executing the local single-station earthquake early warning method. The processor is responsible for reading and running the program instructions in the memory to execute the various steps in the local single-station earthquake early warning method. The program instructions implement functions including data acquisition, baseline correction, feature extraction, model prediction, warning information generation and release. It should be understood that the electronic device can be a personal computer (PC), a tablet computer, a smart phone, a personal digital assistant (PDA) and other electronic devices with logical computing functions.

[0088] The electronic device 100 may include a memory 111, a storage controller 112, a processor 113, a peripheral interface 114, an input and output unit 115, and a display unit 116. A person skilled in the art may understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the electronic device 100. For example, the electronic device 100 may further include Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0089] The memory 111, storage controller 112, processor 113, peripheral interface 114, input / output unit 115 and display unit 116 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The processor 113 is used to execute the executable module stored in the memory.

[0090] The memory 111 may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc. The memory 111 is used to store programs, and the processor 113 executes the program after receiving the execution instruction. The method executed by the electronic device 100 defined by the process disclosed in any embodiment of the present application can be applied to the processor 113, or implemented by the processor 113.

[0091] The processor 113 may be an integrated circuit chip with signal processing capability. The processor 113 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may 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 gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application may be implemented or executed. The general-purpose processor may be a microprocessor or any conventional processor.

[0092] The above-mentioned peripheral interface 114 couples various input / output devices to the processor 113 and the memory 111. In some embodiments, the peripheral interface 114, the processor 113 and the storage controller 112 can be implemented in a single chip. In some other examples, they can be implemented by separate chips respectively. The above-mentioned input and output unit 115 is used to provide input data to the user. The input and output unit 115 can be, but not limited to, a mouse and a keyboard. The above-mentioned display unit 116 provides an interactive interface (such as a user operation interface) between the electronic device 100 and the user or is used to display image data for user reference. In this embodiment, the display unit can be a liquid crystal display or a touch display. If it is a touch display, it can be a capacitive touch screen or a resistive touch screen that supports single-point and multi-point touch operations. Supporting single-point and multi-point touch operations means that the touch display can sense touch operations generated simultaneously from one or more positions on the touch display, and the sensed touch operations are handed over to the processor for calculation and processing. In an embodiment of the present application, the display unit 116 can display information related to the on-site earthquake early warning.

[0093] In several embodiments provided in the present application, it should be understood that the disclosed device can also be implemented in other ways. The device embodiments described above are merely schematic, for example, the block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices according to the multiple embodiments of the present application. In this regard, each box in the block diagram can represent a module, a program segment or a part of a code, and a part of a module, a program segment or a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram, and the combination of the block diagrams, can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0094] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0095] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0096] The above are only specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

[0097] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

Claims

1. A method for earthquake early warning at a single local station, characterized in that: The method comprises: obtaining first acceleration waveform data based on the three-dimensional acceleration of the ground vibration, and determining whether an earthquake event occurs based on the first acceleration waveform data; If it is determined that an earthquake event occurs, obtaining a P wave arrival time based on the first acceleration waveform data; Based on the acquired P wave arrival time, obtaining second acceleration waveform data in a corresponding time window after the P wave arrives; Based on the second acceleration waveform data, relevant parameters predicted in a corresponding time window of the P-wave arrival time are obtained, wherein the relevant parameters include epicentral distance and azimuth.

2. The method according to claim 1, characterized in that: The method of obtaining the first acceleration waveform data based on the three-dimensional acceleration of the ground vibration and judging whether an earthquake event occurs based on the first acceleration waveform data includes: After baseline correction is performed on the three-dimensional acceleration, the first acceleration waveform data is obtained, and the first acceleration waveform data is converted into first acceleration spectrum data; wherein the first acceleration spectrum data is used to determine whether an earthquake event occurs.

3. The method according to claim 2, characterized in that The method further comprises: obtaining first acceleration waveform data based on the three-dimensional acceleration of the ground vibration, and judging whether an earthquake event occurs based on the first acceleration waveform data; The first acceleration waveform data and the first acceleration spectrum data are input into a first model to determine whether an earthquake event occurs; wherein the first model is used to capture the time domain and spectrum data information of earthquake events and non-earthquake events.

4. The method according to claim 1, characterized in that If it is determined that an earthquake event occurs, obtaining the P wave arrival time based on the first acceleration waveform data includes: inputting the first acceleration waveform data into a probability model; After the probability model performs feature extraction on the first acceleration waveform data, a probability density function of the P wave arrival time is obtained; Based on the probability density function, the P wave arrival time is obtained.

5. The method according to claim 1, characterized in that The step of obtaining the relevant parameters predicted in the corresponding time window of the P wave arrival time based on the second acceleration waveform data includes: Inputting the second acceleration waveform data into a second model to obtain the predicted epicentral distance; inputting the second acceleration waveform data into a third model to obtain the predicted azimuth; Wherein, the second model is used to extract the local spatial features related to the second acceleration waveform data and the epicenter distance estimation; the third model is used to extract the local spatial features related to the second acceleration waveform data and the azimuth estimation; The predicted epicenter position is obtained based on the predicted epicenter distance and the azimuth, and an earthquake location calculation rule.

6. The method according to claim 1, characterized in that in, The related parameters also include: an estimated value of earthquake magnitude; the related parameters predicted under the corresponding time window of the P wave arrival time obtained based on the second acceleration waveform data include: The second acceleration waveform data is input into a fourth model to obtain a predicted earthquake magnitude estimate; wherein the fourth model is used to extract local spatial features of the second acceleration waveform data related to the earthquake magnitude estimate.

7. The method according to claim 1, characterized in that in, The related parameters also include: ground peak acceleration; the related parameters predicted under the corresponding time window of the P wave arrival time obtained based on the second acceleration waveform data include: The second acceleration waveform data is input into a fifth model to obtain the predicted ground peak acceleration; wherein the fifth model is used to extract local spatial features of the second acceleration waveform data related to the ground peak acceleration.

8. The method according to claim 1, characterized in that: The method further comprises: If an earthquake event is determined to have occurred, local earthquake warning information will be issued; The on-site earthquake warning information includes earthquake event identification results and related parameter prediction results.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the memory stores program instructions, and when the processor runs the program instructions, the steps in the method according to any one of claims 1 to 8 are executed.

10. A computer program product, characterized in that The computer program product comprises a computer program / instruction, and when the computer program / instruction is executed by a processor, the steps in the method according to any one of claims 1 to 8 are implemented.

Images