Lightweight in-situ seismic oscillation parameter prediction model construction method and earthquake early warning system

Abstract

The invention relates to the technical field of earthquake prediction, and particularly provides a lightweight in-situ seismic oscillation parameter prediction model construction method and an earthquake early warning system. The lightweight in-situ seismic oscillation parameter prediction model construction method comprises the steps of training an initial prediction model according to acceleration waveform data when a seismic P wave arrives, and obtaining a pre-trained prediction model; taking the pre-trained prediction model as a teacher model, and migrating knowledge of the teacher model to a student model through distillation; and constructing a lightweight in-situ seismic oscillation parameter prediction model based on the student model. According to the method, knowledge in the pre-training prediction model is migrated to the student model through distillation, so that the complexity and the calculation amount of the student model can be reduced; and the lightweight in-situ seismic oscillation parameter prediction model is constructed based on the student model, so that the in-situ seismic oscillation parameter prediction efficiency can be ensured, the demand of the model for hardware resources can be reduced, and the deployable scene of the model can be expanded.

Description

Technical Field

[0001] The present application relates to the technical field of earthquake prediction, and in particular to a method for constructing a lightweight on-site seismic parameter prediction model and an earthquake early warning system. Background Art

[0002] The prediction of local ground motion parameters is an important task in earthquake early warning. Its purpose is to predict the local ground motion parameters using the P-wave signals recorded by the stations after an earthquake occurs, so as to quickly issue an alarm and effectively reduce the casualties and economic losses caused by the earthquake disaster.

[0003] Existing models for predicting local earthquake parameters usually predict local earthquake parameters based on multiple features extracted manually from P waves; or use a multi-layer neural network structure to extract feature data from the original earthquake waveform, and predict local earthquake parameters based on the extracted feature data. Among them, the workload of manually extracting features from P waves is large and time-consuming, which will affect the rapidity of alarm issuance. The use of a multi-layer neural network structure to extract feature data from the original earthquake waveform requires a large amount of memory, which limits the deployment environment of the corresponding prediction model. Summary of the invention

[0004] In view of this, the purpose of the embodiments of the present application is to provide a lightweight on-site seismic parameter prediction model construction method and an earthquake early warning system to solve the technical problems of low parameter prediction efficiency or limited deployment scenarios of existing on-site seismic parameter prediction models.

[0005] In a first aspect, an embodiment of the present application provides a method for constructing a lightweight on-site ground motion parameter prediction model, the method comprising:

[0006] The initial prediction model is trained according to the acceleration waveform data when the earthquake P wave arrives to obtain a pre-trained prediction model;

[0007] Using the pre-trained prediction model as a teacher model, and transferring the knowledge of the teacher model to the student model through distillation;

[0008] A lightweight on-site earthquake parameter prediction model is constructed based on the student model.

[0009] In the above implementation process, the lightweight on-site seismic parameter prediction model construction method obtains a pre-trained prediction model by training the initial prediction model according to the acceleration waveform data when the earthquake P wave arrives; the pre-trained prediction model is used as a teacher model, and the knowledge of the teacher model is transferred to the student model through distillation; and a lightweight on-site seismic parameter prediction model is constructed based on the student model. The lightweight on-site seismic parameter prediction model construction method transfers the knowledge in the pre-trained prediction model to the student model through distillation, which can reduce the complexity and calculation amount of the student model; and then the lightweight on-site seismic parameter prediction model is constructed based on the student model, which can reduce the model's demand for hardware resources while ensuring the efficiency of on-site seismic parameter prediction, and expand the model's deployable scenarios. The technical problems of low parameter prediction efficiency or limited deployment scenarios of existing on-site seismic parameter prediction models are solved.

[0010] In addition, compared with the existing on-site seismic parameter prediction model constructed based on a multi-layer neural network structure, the lightweight on-site seismic parameter prediction model constructed based on the lightweight on-site seismic parameter prediction model construction method provided in this application is simpler and can further improve the running speed of the model, thereby improving the efficiency of on-site seismic parameter prediction and obtaining better earthquake warning effects.

[0011] Optionally, in an embodiment of the present application, the initial prediction model is trained according to the acceleration waveform data when the earthquake P wave arrives to obtain a pre-trained prediction model, including: integrating the acceleration waveform data to obtain velocity waveform data; integrating the velocity waveform data to obtain displacement waveform data; calculating the local seismic motion parameters when the earthquake P wave arrives according to the acceleration waveform data, the velocity waveform data and the displacement waveform data; training the initial prediction model according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the local seismic motion parameters to obtain the pre-trained prediction model.

[0012] In the above implementation process, by integrating the acceleration waveform data, the velocity waveform data can be obtained; by integrating the velocity waveform data, the displacement waveform data can be obtained. Compared with the existing local earthquake parameter prediction equation based on a single characteristic parameter, more characteristic parameters can be obtained based on the acceleration waveform data, the velocity waveform data and the displacement waveform data, which reduces the uncertainty of the local earthquake parameter prediction results obtained based on the constructed lightweight local earthquake parameter prediction model, that is, improves the accuracy of the obtained local earthquake parameter prediction results. Therefore, the lightweight local earthquake parameter prediction model constructed based on the lightweight local earthquake parameter prediction model construction method provided in the present application can improve the efficiency and accuracy of earthquake early warning.

[0013] Optionally, in an embodiment of the present application, the pre-trained prediction model includes a trained first feature extraction module and a first fully connected module; the using the pre-trained prediction model as a teacher model and migrating the knowledge of the teacher model to the student model through distillation comprises: using the pre-trained prediction model as the teacher model and migrating the first feature extraction module to the student model through distillation; wherein the student model includes the first feature extraction module and the second fully connected module;

[0014] The method of constructing a lightweight local seismic parameter prediction model based on the student model includes: inputting the acceleration waveform data, the velocity waveform data and the displacement waveform data into the student model to obtain the prediction parameters output by the student model; adjusting the parameters of the second fully-connected module in the student model according to the local seismic parameters and the prediction parameters, until the similarity between the prediction parameters output by the adjusted student model and the local seismic parameters reaches a preset similarity threshold; and constructing the lightweight local seismic parameter prediction model based on the adjusted student model.

[0015] In the above implementation process, the first feature extraction module is migrated to the student model through distillation, and the parameters of the second fully connected module in the student model are adjusted through acceleration waveform data, velocity waveform data, displacement waveform data and local seismic motion parameters; the accuracy of local seismic motion parameter prediction can be further improved while reducing the complexity and calculation amount of the constructed lightweight local seismic motion parameter prediction model and improving the efficiency of local seismic motion parameter prediction.

[0016] Optionally, in an embodiment of the present application, the first fully connected module includes a linear activation function layer; the second fully connected module includes a nonlinear activation function layer and a linear activation function layer; the lightweight on-site seismic parameter prediction model is constructed based on the adjusted student model, including: pruning the nonlinear activation function layer in the adjusted student model until the sparsity of the student model after the pruning operation reaches a preset sparsity threshold; and determining the student model after the pruning operation as the lightweight on-site seismic parameter prediction model.

[0017] In the above implementation process, the second fully connected module includes a nonlinear activation function layer and a linear activation function layer; by pruning the nonlinear activation function layer in the adjusted student model, it is possible to further reduce the complexity and computational complexity of the constructed lightweight ground seismic parameter prediction model while ensuring the accuracy of the ground seismic parameter prediction of the constructed model, thereby improving the efficiency of the ground seismic parameter prediction.

[0018] Optionally, in an embodiment of the present application, the first feature extraction module includes: a convolution layer and a pooling layer; the initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the local seismic parameters to obtain the pre-trained prediction model, including: based on the convolution layer, feature extraction is performed on the acceleration waveform data, the velocity waveform data and the displacement waveform data to obtain feature extraction data; based on the pooling layer, pooling operation is performed on the feature extraction data to obtain reduced dimension feature data; based on the first fully connected module, the initial model prediction parameters are output according to the reduced dimension feature data; according to the initial model prediction parameters and the local seismic parameters, the internal parameters of the initial prediction model are adjusted to obtain the pre-trained prediction model.

[0019] In the above implementation process, the convolution layer can be used to obtain feature extraction data from acceleration waveform data, velocity waveform data, and displacement waveform data; and the pooling layer can be used to select the maximum value in the local area to reduce the length of the input sequence, thereby extracting key feature data and improving the robustness of the pre-trained prediction model obtained through training.

[0020] Optionally, in an embodiment of the present application, the feature extraction data includes acceleration feature data, velocity feature data and displacement feature data; the first feature extraction module further includes: a random deactivation layer; before performing a pooling operation on the feature extraction data based on the pooling layer to obtain the reduced dimension feature data, the initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the local ground motion parameter to obtain the pre-trained prediction model, and further includes: based on the random deactivation layer, randomly determining valid feature data in the feature extraction data; the valid feature data includes at least one of the acceleration feature data, the velocity feature data and the displacement feature data;

[0021] The performing a pooling operation on the feature extraction data based on the pooling layer to obtain the dimension-reduced feature data includes: performing a pooling operation on the effective feature data based on the pooling layer to obtain the dimension-reduced feature data.

[0022] In the above implementation process, the random dropout layer can be used to randomly determine the effective feature data in the feature extraction data, that is, based on the random dropout layer, part of the feature extraction data can be randomly discarded during the training process; thereby reducing the dependence of the pre-trained prediction model on specific feature data, reducing the possibility of overfitting of the pre-trained prediction model, and improving the generalization ability of the pre-trained prediction model obtained through training.

[0023] Optionally, in an embodiment of the present application, the first feature extraction module further includes: a bidirectional gating unit layer; before the first fully connected module outputs the initial model prediction parameters according to the reduced dimension feature data, the initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the ground seismic parameters to obtain the pre-trained prediction model, and further includes: acquiring feature representation data of the reduced dimension feature data based on the bidirectional gating unit layer;

[0024] The outputting of initial model prediction parameters based on the first fully-connected module according to the dimension reduction feature data includes: outputting the initial model prediction parameters based on the first fully-connected module according to the feature representation data.

[0025] In the above implementation process, the bidirectional gated unit layer can help the pre-trained prediction model to better capture the long-term dependencies in the sequence, so as to improve the performance of the pre-trained prediction model obtained through training.

[0026] Optionally, in an embodiment of the present application, the first feature extraction module includes: a convolution layer, a random inactivation layer, a pooling layer, and a bidirectional gating unit layer; the initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data, and the local ground motion parameter to obtain the pre-trained prediction model, including: based on the convolution layer, feature extraction is performed on the acceleration waveform data, the velocity waveform data, and the displacement waveform data to obtain feature extraction data; the feature extraction data includes: acceleration feature data, velocity feature data, and displacement feature data; based on the random inactivation layer, the initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data, and the local ground motion parameter to obtain the pre-trained prediction model; the feature extraction data includes: acceleration feature data, velocity feature data, and displacement feature data; based on the random inactivation layer, the initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data, and the local ground motion parameter, to obtain the pre-trained prediction model; the pre-trained prediction model is trained according to the acceleration waveform data, the velocity waveform data, and the displacement waveform data ... layer, randomly determining effective feature data in the feature extraction data; the effective feature data includes at least one of the acceleration feature data, the velocity feature data and the displacement feature data; based on the pooling layer, performing a pooling operation on the effective feature data to obtain reduced-dimensional feature data; based on the bidirectional gating unit layer, obtaining feature representation data of the reduced-dimensional feature data; based on the first fully connected module, outputting initial model prediction parameters according to the feature representation data; according to the initial model prediction parameters and the local seismic parameters, adjusting the internal parameters of the initial prediction model to obtain the pre-trained prediction model.

[0027] Optionally, in an embodiment of the present application, the calculating the ground ground motion parameters when the earthquake P wave arrives according to the acceleration waveform data, the velocity waveform data and the displacement waveform data comprises: filtering the acceleration waveform data, the velocity waveform data and the displacement waveform data respectively to obtain the acceleration waveform processing data, the velocity waveform processing data and the displacement waveform processing data after filtering; calculating the ground ground motion parameters when the earthquake P wave arrives according to the acceleration waveform processing data, the velocity waveform processing data and the displacement waveform processing data;

[0028] Wherein, the initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the local seismic parameters to obtain the pre-trained prediction model, including: the initial prediction model is trained according to the acceleration waveform processing data, the velocity waveform processing data, the displacement waveform processing data and the local seismic parameters to obtain the pre-trained prediction model.

[0029] In the above implementation process, by filtering the acceleration waveform data, velocity waveform data and displacement waveform data, the noise data in the acceleration waveform data, velocity waveform data and displacement waveform data can be removed, thereby improving the data quality of the acceleration waveform processing data, velocity waveform processing data and displacement waveform processing data after filtering.

[0030] Optionally, in an embodiment of the present application, before the initial prediction model is trained according to the acceleration waveform data when the earthquake P wave arrives to obtain the pre-trained prediction model, the method also includes: automatically picking up the monitored vertical acceleration data when the P wave arrives, and obtaining the acceleration waveform data when the earthquake P wave arrives based on the monitored three-dimensional acceleration waveform data.

[0031] In a second aspect, an embodiment of the present application provides an earthquake early warning system, the system comprising: a server and an early warning unit;

[0032] The server is configured to predict the local earthquake parameters based on the lightweight local earthquake parameter prediction model constructed by the method described in any one of the first aspects above;

[0033] The early warning unit is configured to perform earthquake early warning according to the local earthquake motion parameters.

[0034] In a third aspect, an embodiment of the present application further provides an electronic device; the electronic device includes:

[0035] Memory;

[0036] processor;

[0037] The memory stores a computer program executable by the processor. When the computer program is executed by the processor, the method for constructing a lightweight on-site seismic parameter prediction model as described in any one of the first aspects is performed.

[0038] In a fourth aspect, an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements a method for constructing a lightweight on-site seismic parameter prediction model as described in any one of the first aspects above.

[0039] In a fifth aspect, an embodiment of the present application further provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the method for constructing a lightweight on-site seismic parameter prediction model as described in any one of the first aspects is executed.

[0040] In a sixth aspect, an embodiment of the present application provides a lightweight on-site ground motion parameter prediction model construction device, the device comprising:

[0041] A training module is used to train the initial prediction model according to the acceleration waveform data when the earthquake P wave arrives, so as to obtain a pre-trained prediction model;

[0042] A distillation module, used to use the pre-trained prediction model as a teacher model and transfer the knowledge of the teacher model to the student model through distillation;

[0043] A model building module is used to build a lightweight on-site earthquake parameter prediction model based on the student model.

[0044] The beneficial effects of the present application include at least: the method for constructing a lightweight on-site seismic parameter prediction model obtains a pre-trained prediction model by training the initial prediction model according to the acceleration waveform data when the earthquake P wave arrives; the pre-trained prediction model is used as a teacher model, and the knowledge of the teacher model is transferred to the student model through distillation; and a lightweight on-site seismic parameter prediction model is constructed based on the student model. The method for constructing a lightweight on-site seismic parameter prediction model transfers the knowledge in the pre-trained prediction model to the student model through distillation, which can reduce the complexity and amount of calculation of the student model; and then the lightweight on-site seismic parameter prediction model is constructed based on the student model, which can reduce the model's demand for hardware resources while ensuring the efficiency of on-site seismic parameter prediction, thereby expanding the model's deployable scenarios. The technical problems of low parameter prediction efficiency or limited deployment scenarios of existing on-site seismic parameter prediction models are solved.

[0045] In addition, compared with the existing on-site seismic parameter prediction model constructed based on a multi-layer neural network structure, the lightweight on-site seismic parameter prediction model constructed based on the lightweight on-site seismic parameter prediction model construction method provided in this application is simpler and can further improve the running speed of the model, thereby improving the efficiency of on-site seismic parameter prediction and obtaining better earthquake warning effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] 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.

[0047] Figure 1 A schematic diagram of a flow chart of a method for constructing a lightweight on-site ground motion parameter prediction model provided in an embodiment of the present application;

[0048] Figure 2 A schematic diagram of a process for obtaining a pre-trained prediction model provided in an embodiment of the present application;

[0049] Figure 3 A flowchart of a method for constructing a lightweight on-site ground motion parameter prediction model based on a student model provided in an embodiment of the present application;

[0050] Figure 4A schematic diagram of the structure of a pre-trained prediction model provided in an embodiment of the present application;

[0051] Figure 5 A schematic diagram of the structure of a lightweight on-site ground motion parameter prediction model provided in an embodiment of the present application;

[0052] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] The following embodiments of the technical solution of the present application are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present application, and are therefore only used as examples, and cannot be used to limit the scope of protection of the present application.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0055] In the description of the embodiments of the present application, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0056] See also Figure 1 The flowchart of a method for constructing a lightweight on-site earthquake parameter prediction model provided by an embodiment of the present application is shown. The method for constructing a lightweight on-site earthquake parameter prediction model may include the following steps:

[0057] S101, training an initial prediction model according to acceleration waveform data when the earthquake P wave arrives to obtain a pre-trained prediction model;

[0058] S102, using the pre-trained prediction model as a teacher model, and transferring the knowledge of the teacher model to the student model through distillation;

[0059] S103. Constructing a lightweight on-site earthquake parameter prediction model based on the student model.

[0060] In step 101, the seismic P wave is also called the seismic longitudinal wave, and the vibration direction of the medium particles propagating it is consistent with the propagation direction of the wave. The acceleration waveform data may include three-dimensional acceleration waveform data (including vertical acceleration a ud , east-west acceleration a ew and the north-south acceleration ans The initial prediction model may be an untrained neural network model, such as a convolutional neural network model or a recurrent neural network model.

[0061] In step 102, the knowledge of the teacher model can be transferred to the student model by online distillation or offline distillation. Exemplarily, when the initial prediction model is implemented by an untrained convolutional neural network model, the pre-trained prediction model can include a trained feature extraction module and a fully connected module; the feature extraction module in the teacher model can be transferred to the student model as knowledge.

[0062] Among them, in step 103, the student model after knowledge transfer can be used as a lightweight local earthquake parameter prediction model; or some parameters in the student model can be fine-tuned based on the acceleration waveform data and the local earthquake parameters corresponding to the acceleration waveform data, and the fine-tuned model can be used as a lightweight local earthquake parameter prediction model. After the knowledge in the pre-trained prediction model is transferred to the student model through distillation, since the student model has the knowledge in the pre-trained prediction model and the network structure is simpler; therefore, a lightweight local earthquake parameter prediction model that can realize local earthquake parameter prediction and has low complexity and calculation amount and low hardware resource requirements can be constructed based on the student model.

[0063] It can be seen that the lightweight on-site earthquake parameter prediction model construction method provided by the embodiment of the present application can reduce the complexity and calculation amount of the student model by transferring the knowledge in the pre-trained prediction model to the student model through distillation; and then constructing a lightweight on-site earthquake parameter prediction model based on the student model can reduce the model's demand for hardware resources while ensuring the efficiency of on-site earthquake parameter prediction, and expand the model's deployment scenario. The technical problem of low parameter prediction efficiency or limited deployment scenario of the existing on-site earthquake parameter prediction model is solved. In addition, compared with the existing on-site earthquake parameter prediction model constructed based on the multi-layer neural network structure, the lightweight on-site earthquake parameter prediction model constructed based on the lightweight on-site earthquake parameter prediction model construction method provided by the present application is simpler, and the running speed of the model can be further improved to improve the efficiency of on-site earthquake parameter prediction and obtain better earthquake early warning effect. And because the lightweight on-site earthquake parameter prediction model constructed based on the lightweight on-site earthquake parameter prediction model construction method provided by the present application is simpler, it can be effectively operated under limited resource conditions. Therefore, the lightweight on-site seismic parameter prediction model constructed based on the lightweight on-site seismic parameter prediction model construction method provided in this application can be deployed in mobile applications, Internet of Things devices, edge computing and other resource-constrained environments, balancing the performance and resource consumption of the prediction model.

[0064] Please refer to Figure 2 , Figure 2 A flowchart of a method for obtaining a pre-trained prediction model provided in an embodiment of the present application.

[0065] In some optional embodiments, S101, according to the acceleration waveform data when the earthquake P wave arrives, the initial prediction model is trained to obtain a pre-trained prediction model, including: S1011, integrating the acceleration waveform data to obtain velocity waveform data; S1012, integrating the velocity waveform data to obtain displacement waveform data; S1013, calculating the local seismic motion parameters when the earthquake P wave arrives according to the acceleration waveform data, the velocity waveform data and the displacement waveform data; S1014, according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the local seismic motion parameters, the initial prediction model is trained to obtain the pre-trained prediction model.

[0066] Among them, the acceleration waveform data may include: vertical acceleration data, east-west acceleration data and north-south acceleration data; the velocity waveform data may include: vertical velocity data, east-west velocity data and north-south velocity data; the displacement waveform data may include: vertical displacement data, east-west displacement data and north-south displacement data. It is possible to deploy stations in the early warning target area and use the time difference between the first arrival wave and the subsequent destructive seismic wave to carry out early warning. And the prediction of local seismic parameters is realized based on the lightweight local seismic parameter prediction model constructed by the method provided in the present application. The local seismic parameters can be adjusted according to the actual earthquake early warning scenario. For example, the local seismic parameters can include at least one of high-speed rail early warning parameters, gas early warning parameters, nuclear power early warning parameters, instrument intensity parameters and aftershock rescue parameters. It can be based on Calculation of ground motion parameters PGA for high-speed railway earthquake early warning GT ; a ew,gt represents the acceleration in the east-west direction after 0.05-5Hz bandpass filtering, a ns,gt Represents the north-south acceleration after 0.05-5Hz bandpass filtering. Calculate the local ground motion parameters SI for gas earthquake early warning; S v,ew represents the velocity response spectrum in the east-west direction, S v,ns represents the velocity response spectrum in the north-south direction, and ξ represents the damping ratio. Calculate the ground motion parameters CAV for nuclear power earthquake early warning; t max Indicates the duration of the earthquake history. Calculation of PGA ground motion parameters for China instrumental earthquake intensity I and PGV I ; a ud,I 、a ew,I and a ns,I They represent the vertical acceleration, east-west acceleration, and north-south acceleration after 0.1-10Hz bandpass filtering, respectively. ud,I 、v ew,I and v ns,I They represent the vertical velocity, east-west velocity, and north-south velocity after 0.1-10Hz bandpass filtering. Calculate the on-site peak seismic motion parameter PI for aftershock rescue. By integrating the acceleration waveform data, the velocity waveform data can be obtained; by integrating the velocity waveform data, the displacement waveform number can be obtained. Compared with the existing on-site seismic motion parameter prediction equation based on a single characteristic parameter, more characteristic parameters can be obtained based on the acceleration waveform data, velocity waveform data and displacement waveform data, reducing the uncertainty of the on-site seismic motion parameter prediction results obtained based on the constructed lightweight on-site seismic motion parameter prediction model. Therefore, the lightweight on-site seismic motion parameter prediction model constructed based on the lightweight on-site seismic motion parameter prediction model construction method provided in the present application can improve the efficiency and accuracy of earthquake early warning.

[0067] Please refer to Figure 3 , Figure 3 A flowchart of an implementation method for constructing a lightweight on-site seismic parameter prediction model based on a student model is provided in an embodiment of the present application.

[0068] Please refer to Figure 4 , Figure 4 A schematic diagram of the structure of a pre-trained prediction model provided in an embodiment of the present application. Figure 4What is shown is a structural diagram of a pre-trained prediction model, that is, a teacher model. In some optional embodiments, the pre-trained prediction model includes a trained first feature extraction module and a first fully connected module; S102, using the pre-trained prediction model as a teacher model, and migrating the knowledge of the teacher model to the student model through distillation, including: using the pre-trained prediction model as the teacher model, and migrating the first feature extraction module to the student model through distillation; wherein the student model includes the first feature extraction module and the second fully connected module; S103, constructing a lightweight local seismic parameter prediction model based on the student model, including: S1031, inputting the acceleration waveform data, the velocity waveform data and the displacement waveform data into the student model to obtain the prediction parameters output by the student model; S1032, adjusting the parameters of the second fully connected module in the student model according to the local seismic parameters and the prediction parameters, until the similarity between the prediction parameters output by the adjusted student model and the local seismic parameters reaches a preset similarity threshold; S1033, constructing the lightweight local seismic parameter prediction model based on the adjusted student model.

[0069] Among them, Figure 4 As shown, the first feature extraction module may include feature processing layers such as convolutional layers and pooling layers. In the process of constructing a lightweight local seismic parameter prediction model based on the student model, the parameters of the first feature extraction module in the student model are no longer adjusted; that is, the first feature extraction module does not participate in the training process of the student model, but provides feature processing data for the student model; in this way, the parameter prediction accuracy of the constructed lightweight local seismic parameter prediction model can be ensured while reducing the difficulty of model training and the complexity of the model. The first feature extraction module is migrated to the student model through distillation, and the parameters of the second fully connected module in the student model are adjusted through acceleration waveform data, velocity waveform data, displacement waveform data and local seismic parameters; while reducing the complexity and calculation amount of the constructed lightweight local seismic parameter prediction model and improving the local seismic parameter prediction efficiency, the accuracy of local seismic parameter prediction can be further improved.

[0070] Please refer to Figure 5 , Figure 5 A schematic diagram of the structure of a lightweight on-site ground motion parameter prediction model provided in an embodiment of the present application. In some optional embodiments, the first fully connected module includes a linear activation function layer; the second fully connected module includes a nonlinear activation function layer and a linear activation function layer (such as Figure 5As shown); S1033, constructing the lightweight on-site seismic parameter prediction model based on the adjusted student model, including: pruning the nonlinear activation function layer in the adjusted student model until the sparsity of the student model after the pruning operation reaches a preset sparsity threshold; determining the student model after the pruning operation as the lightweight on-site seismic parameter prediction model.

[0071] Among them, the nonlinear activation function layer may include 16 neurons (or other reasonable values), and may be specifically implemented by a ReLu activation function layer; the linear activation function layer may include 1 neuron. The degree of pruning of the pruning operation can be controlled by setting the final sparsity. Exemplarily, when the initial sparsity is 0.5, the final sparsity can be set to 0.8 or other reasonable values. The time range of the pruning operation can also be controlled by setting the start step (which can be 0) and the end step (which can be the total number of steps in the training process). The second fully connected module includes a nonlinear activation function layer and a linear activation function layer; by performing a pruning operation on the nonlinear activation function layer in the adjusted student model, the complexity and computational complexity of the constructed lightweight on-site seismic parameter prediction model can be further reduced while ensuring the accuracy of the on-site seismic parameter prediction of the constructed model, thereby improving the efficiency of the on-site seismic parameter prediction.

[0072] In some optional embodiments, the first feature extraction module includes: a convolution layer and a pooling layer; S1014, training the initial prediction model according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the local seismic parameters to obtain the pre-trained prediction model, including: extracting features from the acceleration waveform data, the velocity waveform data and the displacement waveform data based on the convolution layer to obtain feature extraction data; performing pooling operation on the feature extraction data based on the pooling layer to obtain reduced-dimensional feature data; outputting initial model prediction parameters based on the reduced-dimensional feature data based on the first fully-connected module; adjusting internal parameters of the initial prediction model according to the initial model prediction parameters and the local seismic parameters to obtain the pre-trained prediction model.

[0073] Among them, the convolution layer can be implemented by a one-dimensional convolution layer, which can be based on Implement the corresponding convolution operation; X i+j represents the elements in the input sequence (acceleration waveform data, velocity waveform data, and displacement waveform data), i represents the position in the output sequence, k represents the size of the convolution kernel, K j represents the weight of the convolution kernel, Y iRepresents the output result of the convolution layer. The number of convolution kernels in the one-dimensional convolution layer can be specifically 16, the size of the convolution kernel can be 3, and the step size can be 1. The pooling layer can be specifically implemented by a one-dimensional maximum pooling layer, and the pooling window size of the one-dimensional maximum pooling layer can be 4. By setting the pre-training similarity threshold, when the similarity between the prediction parameters output by the adjusted initial prediction model and the local seismic parameters reaches the preset similarity threshold, the adjusted initial prediction model is determined as the pre-training prediction model. The convolution layer can be used to obtain feature extraction data from acceleration waveform data, velocity waveform data, and displacement waveform data; and the pooling layer can be used to select the maximum value in the local area to reduce the length of the input sequence, thereby extracting key feature data and improving the robustness of the pre-trained prediction model obtained through training.

[0074] In some optional embodiments, such as Figure 4 As shown, the first feature extraction module may also include: a random deactivation layer. The feature extraction data may include acceleration feature data, velocity feature data and displacement feature data; before the pooling operation is performed on the feature extraction data based on the pooling layer to obtain the reduced dimension feature data, S1014, the initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the local seismic parameters to obtain the pre-trained prediction model, and further includes: based on the random deactivation layer, randomly determining the valid feature data in the feature extraction data; the valid feature data includes at least one of the acceleration feature data, the velocity feature data and the displacement feature data;

[0075] The performing a pooling operation on the feature extraction data based on the pooling layer to obtain the dimension-reduced feature data includes: performing a pooling operation on the effective feature data based on the pooling layer to obtain the dimension-reduced feature data.

[0076] Among them, acceleration feature data includes feature data obtained by extracting features from acceleration waveform data based on a convolution layer; velocity feature data includes feature data obtained by extracting features from velocity waveform data based on a convolution layer; displacement feature data includes feature data obtained by extracting features from displacement waveform data based on a convolution layer. The drop rate of the random dropout layer (Dropout layer) can be 0.2 or other reasonable values. The random dropout layer can randomly determine valid feature data in the feature extraction data, that is, based on the random dropout layer, part of the feature extraction data can be randomly discarded during the training process; thereby reducing the dependence of the pre-trained prediction model on specific feature data, reducing the possibility of overfitting of the pre-trained prediction model, and improving the generalization ability of the pre-trained prediction model obtained through training.

[0077] In some optional embodiments, such as Figure 4 As shown, the first feature extraction module may further include: a bidirectional gating unit layer. Before the first fully connected module outputs the initial model prediction parameters according to the reduced dimension feature data, S1014, the initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the ground seismic parameters to obtain the pre-trained prediction model, and further includes: acquiring feature representation data of the reduced dimension feature data based on the bidirectional gating unit layer;

[0078] The outputting of initial model prediction parameters based on the first fully-connected module according to the dimension reduction feature data includes: outputting the initial model prediction parameters based on the first fully-connected module according to the feature representation data.

[0079] The output dimension of the bidirectional gated unit layer (BIGRU layer) can be 64 or other reasonable values. The bidirectional gated unit layer can help the pre-trained prediction model better capture the long-term dependencies in the sequence, so as to improve the performance of the pre-trained prediction model obtained through training.

[0080] In some optional embodiments, S1013, calculating the local seismic parameters when the earthquake P wave arrives according to the acceleration waveform data, the velocity waveform data and the displacement waveform data, including: filtering the acceleration waveform data, the velocity waveform data and the displacement waveform data respectively to obtain the acceleration waveform processing data, the velocity waveform processing data and the displacement waveform processing data after filtering; calculating the local seismic parameters when the earthquake P wave arrives according to the acceleration waveform processing data, the velocity waveform processing data and the displacement waveform processing data; wherein, S1014, training the initial prediction model according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the local seismic parameters to obtain the pre-trained prediction model, including: training the initial prediction model according to the acceleration waveform processing data, the velocity waveform processing data, the displacement waveform processing data and the local seismic parameters to obtain the pre-trained prediction model.

[0081] Among them, the filtering processing of the acceleration waveform data, the velocity waveform data and the displacement waveform data can be realized based on a Butterworth high-pass filter (specifically, a fourth-order Butterworth high-pass filter) or a Chebyshev filter. Exemplarily, the vertical acceleration data, the east-west acceleration data and the north-south acceleration data can be filtered based on a Butterworth high-pass filter to obtain vertical acceleration processing data, east-west acceleration processing data and north-south acceleration processing data. By filtering the acceleration waveform data, the velocity waveform data and the displacement waveform data, the noise data in the acceleration waveform data, the velocity waveform data and the displacement waveform data can be removed, and the data quality of the acceleration waveform processing data, the velocity waveform processing data and the displacement waveform processing data obtained after filtering processing can be improved.

[0082] In some optional embodiments, at S101, the initial prediction model is trained according to the acceleration waveform data when the earthquake P wave arrives to obtain the pre-trained prediction model, and the method also includes: automatically picking up the monitored vertical acceleration data when the P wave arrives, and obtaining the acceleration waveform data when the earthquake P wave arrives according to the monitored three-dimensional acceleration waveform data.

[0083] Among them, the monitored vertical acceleration data can be automatically picked up at the time of P wave arrival based on the long short time average (STA / LTA) and Akaike criterion (AIC) algorithms to obtain the acceleration waveform data when the earthquake P wave arrives. Before the monitored vertical acceleration data are automatically picked up at the time of P wave arrival, the monitored three-dimensional acceleration waveform data can be baseline corrected to improve the accuracy of the automatic picking results of the P wave arrival, thereby improving the accuracy of the lightweight on-site seismic motion parameter prediction of the constructed model. By automatically picking up the monitored vertical acceleration data at the time of P wave arrival to obtain the acceleration waveform data when the earthquake P wave arrives, the consistency of the input data of the model in the training process and the input data in the application process can be improved to improve the accuracy of the lightweight on-site seismic motion parameter prediction of the constructed model.

[0084] The embodiment of the present application also provides an earthquake early warning system, the system comprising: a server and an early warning unit;

[0085] The server is configured to predict the local earthquake parameters based on the lightweight local earthquake parameter prediction model constructed by the method described in any one of the first aspects above;

[0086] The early warning unit is configured to perform earthquake early warning according to the local earthquake motion parameters.

[0087] The server may be implemented by a microcontroller, a single chip microcomputer or a programmable logic control unit, etc. The specific implementation of the earthquake early warning system may also refer to the relevant description above.

[0088] An embodiment of the present application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the lightweight on-site ground motion parameter prediction model construction method as described in any one of the first aspects above.

[0089] Please refer to Figure 6 , Figure 6 A schematic diagram of the structure of an electronic device 200 provided in an embodiment of the present application. The electronic device 200 comprises: a memory 202 and a processor 201; the memory 202 stores a computer program executable by the processor 201, and when the computer program is executed by the processor 201, the method for constructing a lightweight on-site ground motion parameter prediction model as described in any one of the first aspects is executed.

[0090] The memory 202 and the processor 201 can be interconnected and communicate with each other via a communication bus 203 and / or other forms of connection mechanisms (not shown). The memory 202 stores a computer program executable by the processor 201. When the computer program is executed by the processor 201, the method for constructing a lightweight on-site ground motion parameter prediction model as described in the first aspect above is executed.

[0091] The embodiment of the present application also provides a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by the processor 201, the method for constructing a lightweight on-site ground motion parameter prediction model as described in the first aspect above is executed.

[0092] Among them, the storage medium 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.

[0093] In several embodiments provided by the embodiments of the present application, it should be understood that the disclosed device / system and method can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and a part of the module, program segment or 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 than 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 / or flowchart, and the combination of boxes in the block diagram and / or flowchart 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 each embodiment of the present application can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part.

[0095] The above description is only an optional implementation manner of the embodiments of the present application, but the protection scope of the embodiments of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed in the embodiments of the present application, which should be covered within the protection scope of the embodiments of the present application.

Claims

1. A lightweight method for constructing a local ground motion parameter prediction model, characterized in that: The method comprises: The initial prediction model is trained according to the acceleration waveform data when the earthquake P wave arrives to obtain a pre-trained prediction model; Using the pre-trained prediction model as a teacher model, and transferring the knowledge of the teacher model to the student model through distillation; A lightweight on-site earthquake parameter prediction model is constructed based on the student model.

2. The method according to claim 1, characterized in that The initial prediction model is trained according to the acceleration waveform data when the earthquake P wave arrives to obtain a pre-trained prediction model, including: Performing integration processing on the acceleration waveform data to obtain velocity waveform data; Performing integration processing on the velocity waveform data to obtain displacement waveform data; Calculating the ground motion parameters when the earthquake P wave arrives based on the acceleration waveform data, the velocity waveform data and the displacement waveform data; The initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the ground seismic parameters to obtain the pre-trained prediction model.

3. The method according to claim 2, characterized in that in, The pre-trained prediction model includes a trained first feature extraction module and a first fully connected module; the pre-trained prediction model is used as a teacher model, and the knowledge of the teacher model is transferred to the student model through distillation, including: Using the pre-trained prediction model as the teacher model, migrating the first feature extraction module to the student model through distillation; wherein the student model includes the first feature extraction module and a second fully connected module; The method of constructing a lightweight on-site ground motion parameter prediction model based on the student model includes: Inputting the acceleration waveform data, the velocity waveform data and the displacement waveform data into the student model to obtain prediction parameters output by the student model; According to the local ground motion parameters and the prediction parameters, adjusting the parameters of the second fully connected module in the student model until the similarity between the prediction parameters output by the adjusted student model and the local ground motion parameters reaches a preset similarity threshold; The lightweight on-site earthquake parameter prediction model is constructed based on the adjusted student model.

4. The method according to claim 3, characterized in that in, The first fully connected module includes a linear activation function layer; the second fully connected module includes a nonlinear activation function layer and a linear activation function layer; the lightweight on-site ground motion parameter prediction model is constructed based on the adjusted student model, including: Performing a pruning operation on the nonlinear activation function layer in the adjusted student model until the sparsity of the student model after the pruning operation reaches a preset sparsity threshold; The student model after the pruning operation is determined as the lightweight on-site seismic parameter prediction model.

5. The method according to claim 3, characterized in that: in, The first feature extraction module includes: a convolution layer and a pooling layer; the initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the ground seismic parameters to obtain the pre-trained prediction model, including: Performing feature extraction on the acceleration waveform data, the velocity waveform data, and the displacement waveform data based on the convolution layer to obtain feature extraction data; Performing a pooling operation on the feature extraction data based on the pooling layer to obtain dimension-reduced feature data; Outputting initial model prediction parameters based on the first fully connected module according to the dimension reduction feature data; According to the initial model prediction parameters and the ground seismic parameters, the internal parameters of the initial prediction model are adjusted to obtain the pre-trained prediction model.

6. The method according to claim 5, characterized in that in, The feature extraction data includes acceleration feature data, velocity feature data and displacement feature data; The first feature extraction module further includes: a random deactivation layer; before performing a pooling operation on the feature extraction data based on the pooling layer to obtain the reduced dimension feature data, training the initial prediction model according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the ground seismic parameters to obtain the pre-trained prediction model, and further includes: Based on the random deactivation layer, randomly determining effective feature data in the feature extraction data; the effective feature data includes at least one of the acceleration feature data, the velocity feature data and the displacement feature data; The performing a pooling operation on the feature extraction data based on the pooling layer to obtain dimension-reduced feature data includes: A pooling operation is performed on the effective feature data based on the pooling layer to obtain the reduced-dimensional feature data.

7. The method according to claim 5, characterized in that in, The first feature extraction module further includes: a bidirectional gating unit layer; before the first fully connected module outputs the initial model prediction parameters according to the dimension reduction feature data, the initial prediction model is trained according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the ground seismic parameters to obtain the pre-trained prediction model, and further includes: Acquire feature representation data of the dimension-reduced feature data based on the bidirectional gating unit layer; Outputting initial model prediction parameters based on the first fully connected module according to the dimension reduction feature data includes: Based on the first fully connected module and according to the feature representation data, the initial model prediction parameters are output.

8. The method according to claim 2, characterized in that: The calculating, based on the acceleration waveform data, the velocity waveform data and the displacement waveform data, the ground motion parameters when the earthquake P wave arrives comprises: Performing filtering processing on the acceleration waveform data, the velocity waveform data and the displacement waveform data respectively to obtain filtered acceleration waveform processing data, velocity waveform processing data and displacement waveform processing data; Calculating the ground motion parameters when the earthquake P wave arrives based on the acceleration waveform processing data, the velocity waveform processing data and the displacement waveform processing data; The step of training the initial prediction model according to the acceleration waveform data, the velocity waveform data, the displacement waveform data and the ground seismic parameters to obtain the pre-trained prediction model includes: The initial prediction model is trained according to the acceleration waveform processing data, the velocity waveform processing data, the displacement waveform processing data and the local ground motion parameters to obtain the pre-trained prediction model.

9. An earthquake early warning system, characterized in that: The system comprises: a server and an early warning unit; The server is configured to predict the local earthquake parameters based on the lightweight local earthquake parameter prediction model constructed by the method according to any one of claims 1 to 8; The early warning unit is configured to perform earthquake early warning according to the local earthquake motion parameters.

10. An electronic device, characterized in that: The electronic device comprises: Memory; processor; The memory stores a computer program executable by the processor, and when the computer program is executed by the processor, the method according to any one of claims 1 to 8 is performed.

Images