An earthquake prediction method and system based on spatiotemporal attention mechanism

Abstract

The present invention provides an earthquake prediction method and system based on a spatiotemporal attention mechanism, comprising: placing a plurality of piezoelectric seismic sensors in a detection area to acquire seismic wave detection data; acquiring surface temperature data, performing a one-dimensional wavelet transform and a two-dimensional wavelet transform on the surface temperature data, and obtaining pre-earthquake temperature data using an eddy current algorithm and a Recurrent Stochastic Transformation (RST) algorithm; constructing an earthquake prediction model using a doorway recurrent unit and a spatiotemporal attention mechanism, and training the earthquake prediction model based on the seismic wave detection data and pre-earthquake temperature data; using the trained earthquake prediction model, outputting a seismic wave value at a specific time based on the pre-earthquake temperature value calculated at the corresponding time, and determining whether an earthquake has occurred based on the magnitude of the seismic wave value. The pre-earthquake temperature calculated by the present invention effectively eliminates the influence of external factors and enables the model to learn the changing patterns of the data and their correlations at the same location and time, thereby achieving accurate prediction of seismic waves.

Description

Technical Field

[0001] The present invention relates to the field of earthquake exploration technology, and in particular to an earthquake prediction method and system based on a spatiotemporal attention mechanism. Background Art

[0002] Earthquakes are natural phenomena in which underground rocks break and shift, releasing huge amounts of pent-up energy in the form of seismic waves. This energy release is often accompanied by changes in the temperature of the rock interface, leading to abnormal ground temperature. Therefore, abnormal changes in underground temperature can effectively predict the occurrence of earthquakes. However, earthquake predictions based solely on calculating temperature changes before an earthquake are often inaccurate. On the one hand, temperature changes are only used as a reference for judging the occurrence of an earthquake, not the direct cause of an earthquake, nor the main basis for judging the occurrence of an earthquake. On the other hand, temperature changes are often affected by other factors such as human activities or surface cover, resulting in inaccurate earthquake prediction results. Therefore, there is an urgent need to analyze the relationship between temperature changes before an earthquake and the main factors of an earthquake in order to improve the accuracy of earthquake prediction. Summary of the Invention

[0003] In view of this, the present invention proposes an earthquake prediction method and system based on a spatiotemporal attention mechanism to solve the problems existing in the above-mentioned prior art.

[0004] On the one hand, to achieve the above-mentioned purpose, the present invention proposes an earthquake prediction method based on a spatiotemporal attention mechanism, comprising:

[0005] Arranging a plurality of piezoelectric seismic sensors in the detection area, and acquiring seismic wave detection data based on the piezoelectric seismic sensors;

[0006] Obtaining surface temperature data based on meteorological station observation data in the detection area, performing one-dimensional wavelet transform and two-dimensional wavelet transform on the surface temperature data to obtain the in-situ temperature of the detection area, using the vorticity algorithm to obtain the in-situ vorticity based on the in-situ temperature, and using the RST algorithm to obtain pre-earthquake temperature value data;

[0007] An earthquake prediction model is constructed using a doorway recurrent unit and a spatiotemporal attention mechanism, a dataset is constructed based on the seismic wave detection data and pre-earthquake temperature value data, and the earthquake prediction model is trained using the dataset;

[0008] The trained earthquake prediction model is used to predict the seismic wave value at the corresponding moment based on the pre-earthquake temperature value calculated at a specific time, and whether an earthquake will occur is determined based on the size of the seismic wave value.

[0009] Alternatively, the method for obtaining the in-situ vorticity using the vorticity algorithm is as follows:

[0010]

[0011] Where, is the central pixel of the detection area at a certain moment and position, and W(x,y) is the size difference between the central pixel and the four neighboring pixels.

[0012] Optionally, the process of obtaining pre-earthquake temperature data using the RST algorithm includes:

[0013] Obtaining the measured surface temperature and the average surface temperature of the detection area at a certain time in previous years based on the observation data of the meteorological station, and obtaining the DN value of the detection area at that time based on the difference between the measured surface temperature and the average surface temperature;

[0014] The pre-earthquake temperature value data is obtained based on the DN value, and the mean and standard deviation corresponding to the DN value.

[0015] Optionally, the process of training the earthquake prediction model using a data set includes:

[0016] Constructing the change curves of the seismic wave detection data and the pre-earthquake temperature value data over time according to the data set;

[0017] Obtaining seismic wave values ​​in the X-axis, Y-axis, and Z-axis directions at a certain location based on the seismic wave detection data, obtaining pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions at the same location based on the pre-seismic temperature value data, and splicing the seismic wave values ​​and pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions to construct an earthquake characteristic vector;

[0018] A relational dimension encoder is constructed using a spatiotemporal attention mechanism, which inputs the change curve of the seismic wave detection data and the pre-earthquake temperature value data as well as the earthquake feature vector to generate potential feature variables; and a graph attention mechanism is used to encode the earthquake feature vector to obtain input variables;

[0019] Using the potential feature vector and the input variable as the hidden state and input variable in a gated recurrent unit, and cyclically updating the potential feature variable to obtain an optimal potential feature variable;

[0020] A variational autoencoder based on a gated recurrent unit obtains seismic wave variation characteristics according to a variation curve of the seismic wave detection data;

[0021] The potential characteristic variables, the seismic wave change characteristics and the pre-earthquake temperature value at the prediction time are input into the autovariation decoder to obtain the seismic wave value at the prediction time.

[0022] Optionally, the process of determining whether an earthquake has occurred based on the magnitude of the seismic wave value includes:

[0023] A minimum seismic wave value at the time an earthquake occurs in the detection area is constructed based on the seismic wave detection data. When the seismic wave value output by the earthquake prediction model reaches the minimum seismic wave value, it is determined that an earthquake is about to occur at the predicted time.

[0024] On the other hand, to achieve the above-mentioned purpose, the present invention proposes an earthquake prediction system based on a spatiotemporal attention mechanism, comprising a seismic wave detection module, a pre-earthquake temperature calculation module, a seismic wave prediction module, and an earthquake judgment module;

[0025] The seismic wave detection module is used to obtain seismic wave detection data of the detection area;

[0026] The pre-earthquake temperature calculation module is used to obtain pre-earthquake temperature value data of the detection area;

[0027] The seismic wave prediction module constructs an earthquake prediction model based on the seismic wave detection data and the pre-earthquake temperature value data acquired at different times, using a doorway recurrent unit and a spatiotemporal attention mechanism to predict the seismic wave value at a future time;

[0028] The earthquake judgment module is used to judge whether an earthquake will occur based on the magnitude of the earthquake wave value at a future moment.

[0029] Optionally, the seismic wave detection module includes several piezoelectric seismic sensors, which use piezoelectric single crystal chips as piezoelectric plates. When an earthquake occurs, the piezoelectric plates are deformed due to the vibration, and the mechanical energy of the vibration is converted into electrical energy based on the positive piezoelectric effect, thereby obtaining seismic electrical signals; the seismic wave detection module obtains vibration signals of different detection areas through several piezoelectric seismic sensors, and performs statistics on the vibration signals to generate seismic wave detection data.

[0030] Optionally, the pre-earthquake temperature calculation module obtains meteorological station observation data in the detection area, obtains surface temperature data based on the meteorological station observation data, performs one-dimensional wavelet transform and two-dimensional wavelet transform on the surface temperature data, obtains the in-situ temperature of the detection area, uses the vorticity algorithm to obtain the in-situ vorticity based on the in-situ temperature, and uses the RST algorithm to obtain the pre-earthquake temperature value data.

[0031] Optionally, the earthquake prediction module constructs curves showing changes of the seismic wave detection data and the pre-earthquake temperature value data over time respectively;

[0032] Obtaining seismic wave values ​​in the X-axis, Y-axis, and Z-axis directions at a certain location based on the seismic wave detection data, obtaining pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions at the same location based on the pre-seismic temperature value data, and splicing the seismic wave values ​​and pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions to construct an earthquake characteristic vector;

[0033] A relational dimension encoder is constructed using a spatiotemporal attention mechanism, which inputs the change curve of the seismic wave detection data and the pre-earthquake temperature value data as well as the earthquake feature vector to generate potential feature variables; and a graph attention mechanism is used to encode the earthquake feature vector to obtain input variables;

[0034] Using the potential feature vector and the input variable as the hidden state and input variable in a gated recurrent unit, and cyclically updating the potential feature variable to obtain an optimal potential feature variable;

[0035] A variational autoencoder based on a gated recurrent unit obtains seismic wave variation characteristics according to a variation curve of the seismic wave detection data;

[0036] The potential characteristic variables, the seismic wave change characteristics and the pre-earthquake temperature value at the prediction time are input into the autovariation decoder to obtain the seismic wave value at the prediction time.

[0037] Optionally, the earthquake judgment module constructs a minimum seismic wave value at the time an earthquake occurs in the detection area based on the seismic wave detection data. When the seismic wave value output by the earthquake prediction model reaches the minimum seismic wave value, it is judged that an earthquake is about to occur at the predicted time.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention uses the wavelet transform method to eliminate influencing factors such as human activities and surface cover, obtains abnormal information in the spatial domain through the vorticity algorithm, and adopts the RST algorithm to extract thermal anomaly information in the time domain, realizing the extraction of temperature under the joint space-time. At the same time, the spatiotemporal attention mechanism is used as the encoder of the gated recurrent unit, so that the model learns the respective change laws of seismic waves and pre-earthquake temperature through the change curves of their respective changes and their functional correlation at the same location and time, thereby realizing the accurate prediction of seismic waves in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0041] Figure 1 Flowchart of an earthquake prediction method based on spatiotemporal attention mechanism in an embodiment of the present invention;

[0042] Figure 2 This is a structural diagram of an earthquake prediction system based on the spatiotemporal attention mechanism in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, unless there is a conflict, the embodiments of the present disclosure and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0044] Example 1

[0045] This embodiment proposes an earthquake prediction method based on spatiotemporal attention mechanism, such as Figure 1 Shown, including:

[0046] Arranging a plurality of piezoelectric seismic sensors in the detection area, and acquiring seismic wave detection data based on the piezoelectric seismic sensors;

[0047] Obtaining surface temperature data based on meteorological station observation data in the detection area, performing one-dimensional wavelet transform and two-dimensional wavelet transform on the surface temperature data to obtain the in-situ temperature of the detection area, using the vorticity algorithm to obtain the in-situ vorticity based on the in-situ temperature, and using the RST algorithm to obtain pre-earthquake temperature value data;

[0048] An earthquake prediction model is constructed using a doorway recurrent unit and a spatiotemporal attention mechanism, a dataset is constructed based on the seismic wave detection data and pre-earthquake temperature value data, and the earthquake prediction model is trained using the dataset;

[0049] The trained earthquake prediction model is used to predict the seismic wave value at the corresponding moment based on the pre-earthquake temperature value calculated at a specific time, and whether an earthquake will occur is determined based on the size of the seismic wave value.

[0050] As a preferred embodiment, the piezoelectric seismic sensor described in this embodiment uses piezoelectric single crystal material as the piezoelectric piece. When vibration occurs, the piezoelectric material will be deformed under force. Due to the positive piezoelectric effect of the piezoelectric material, when the piezoelectric material is deformed, the mechanical energy will be converted into electrical energy. Then, by collecting the electrical signal on the piezoelectric piece, the seismic electrical signal can be obtained. By statistically analyzing the seismic electrical signals detected at different positions, the seismic wave detection data of the area can be obtained.

[0051] As a preferred embodiment, in processing the surface temperature data, a one-dimensional wavelet transform is used to decompose the surface temperature data to eliminate the influence of solar activity, geographical location, and surface cover. Based on the above processing results, a two-dimensional wavelet transform is used to eliminate the influence of non-tectonic factors such as atmospheric activity and human activities in space, thereby extracting the required in situ temperature;

[0052] After extracting the local temperature, the vorticity algorithm is used to detect the location information of the thermal anomaly signal by calculating the size difference of the thermal signal at a certain location relative to the thermal signals of the surrounding locations at the same time.

[0053]

[0054] In the formula is the central pixel at a certain moment in the detection area, and W(x,y) is the size difference between the central pixel and the four neighboring pixels.

[0055] Finally, the RST algorithm is used to calculate the pre-earthquake temperature value information:

[0056]

[0057] Among them, R represents the thermal anomaly information before the earthquake, V(r i ,t) represents the position r i The DN value of the remote sensing image at time t, or other index values ​​calculated from the DN value. v and σ v is r i V(r i ,t)’s mean and standard deviation.

[0058] RST based on the average of the same period in previous years:

[0059] V(r i ,t)=ΔLST(r i ,t)=LST(r i ,t)-LST(r i ,d)

[0060] Among them, LST(r i ,d) is the mean of a series of effective values ​​over the same period of the previous year, LST(r i ,t) is r i The surface temperature value at time t is used to eliminate the multi-year background and highlight the differences.

[0061] It can be understood that this embodiment uses the wavelet transform method to eliminate influencing factors such as human activities and surface cover, then obtains abnormal information in the spatial domain through the vorticity algorithm, and finally uses the RST algorithm to extract thermal anomaly information in the time domain, achieving time and space combination, and then extracting and analyzing thermal anomalies, which facilitates the later model to learn the spatiotemporal evolution laws of seismic waves and temperature.

[0062] During the model construction process, this embodiment first constructs the time-varying curves of the seismic wave detection data and pre-seismic temperature value data obtained in different time periods. At the same time, the earthquake characteristics that need to be learned by the model are composed of six vectors, which are the seismic wave detection data and pre-seismic temperature value data in the X-axis, Y-axis, and Z-axis directions at a certain position in the area at a certain moment. To construct the characteristic vector, this embodiment obtains the seismic wave values ​​in the X-axis, Y-axis, and Z-axis directions at a certain position based on the seismic wave detection data, and obtains the pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions at the same position based on the pre-seismic temperature value data. The seismic wave values ​​and pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions are spliced ​​respectively to complete the construction of the earthquake characteristic vector. Through the characteristic vector, the model can learn the relationship between temperature and seismic waves at each position, and then realize the prediction of seismic wave values ​​in a specific time period in the future.

[0063] The spatiotemporal attention mechanism is used as the relational dimension encoder of the model. The seismic wave data, the change curve of the pre-earthquake temperature value, and the earthquake feature vector constructed above are input to generate potential feature variables. The earthquake feature vector is then encoded using the graph attention mechanism to obtain the input variables.

[0064] Furthermore, in the spatiotemporal attention mechanism, seismic wave data, the change curve of pre-earthquake temperature values, and earthquake feature vectors first enter the spatial attention mechanism, where a graph convolutional neural network is stacked and a gating mechanism is used to fuse graph convolution features to obtain a first latent feature. Subsequently, the temporal latent feature enters the temporal attention mechanism to obtain a second latent feature, wherein the temporal attention mechanism does not have a graph convolution and a gating mechanism. Finally, a global temporal pooling function is used to pool the time dimension of the second latent feature to obtain the latent feature variable.

[0065] Using the potential feature vector and the input variable as the hidden state and input variable in the gated neural network, and cyclically updating the potential feature variable to obtain the optimal potential feature variable;

[0066] A variational autoencoder based on a gated recurrent unit obtains seismic wave variation characteristics according to a variation curve of the seismic wave detection data;

[0067] The potential characteristic variables, the seismic wave change characteristics and the pre-earthquake temperature value at the prediction time are input into the autovariation decoder to obtain the seismic wave value at the prediction time.

[0068] Finally, the minimum seismic wave value at the time of earthquake occurrence in the detection area is constructed based on the historical seismic wave detection data. When the seismic wave value output by the earthquake prediction model reaches the minimum seismic wave value, it is judged that an earthquake is about to occur at the predicted time.

[0069] This embodiment uses the spatiotemporal attention mechanism as the encoder of the gated recurrent unit, allowing the model to learn the respective changing patterns of earthquake waves and pre-earthquake temperature, as well as their interaction at the same location and time, through the changing curves of earthquake waves and pre-earthquake temperature, thereby achieving accurate prediction of earthquake waves at future times.

[0070] Example 2

[0071] This embodiment proposes an earthquake prediction system based on spatiotemporal attention mechanism, such as Figure 2 As shown, it includes a seismic wave detection module, a pre-earthquake temperature calculation module, a seismic wave prediction module, and an earthquake judgment module;

[0072] The seismic wave detection module is used to obtain seismic wave detection data of the detection area;

[0073] The pre-earthquake temperature calculation module is used to obtain pre-earthquake temperature value data of the detection area;

[0074] The seismic wave prediction module constructs an earthquake prediction model based on the seismic wave detection data and the pre-earthquake temperature value data acquired at different times, using a doorway recurrent unit and a spatiotemporal attention mechanism to predict the seismic wave value at a future time;

[0075] The earthquake judgment module is used to judge whether an earthquake will occur based on the magnitude of the earthquake wave value at a future moment.

[0076] As a preferred embodiment, the seismic wave detection module includes several piezoelectric seismic sensors, which use piezoelectric single crystal chips as piezoelectric sheets. When an earthquake occurs, the piezoelectric sheets are deformed due to the vibration, and the mechanical energy of the vibration is converted into electrical energy based on the positive piezoelectric effect, thereby obtaining seismic electrical signals; the seismic wave detection module obtains vibration signals of different detection areas through several piezoelectric seismic sensors, and performs statistics on the vibration signals to generate seismic wave detection data.

[0077] As a preferred embodiment, the pre-earthquake temperature calculation module obtains meteorological station observation data in the detection area, obtains surface temperature data based on the meteorological station observation data, performs one-dimensional wavelet transform and two-dimensional wavelet transform on the surface temperature data, obtains the in-situ temperature of the detection area, uses the vorticity algorithm to obtain the in-situ vorticity based on the in-situ temperature, and uses the RST algorithm to obtain the pre-earthquake temperature value data.

[0078] As a preferred embodiment, the earthquake prediction module constructs the change curves of the seismic wave detection data and the pre-earthquake temperature value data over time respectively;

[0079] Obtaining seismic wave values ​​in the X-axis, Y-axis, and Z-axis directions at a certain location based on the seismic wave detection data, obtaining pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions at the same location based on the pre-seismic temperature value data, and splicing the seismic wave values ​​and pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions to construct an earthquake characteristic vector;

[0080] A relational dimension encoder is constructed using a spatiotemporal attention mechanism, which inputs the change curve of the seismic wave detection data and the pre-earthquake temperature value data as well as the earthquake feature vector to generate potential feature variables; and a graph attention mechanism is used to encode the earthquake feature vector to obtain input variables;

[0081] Using the potential feature vector and the input variable as the hidden state and input variable in a gated recurrent unit, and cyclically updating the potential feature variable to obtain an optimal potential feature variable;

[0082] A variational autoencoder based on a gated recurrent unit obtains seismic wave variation characteristics according to a variation curve of the seismic wave detection data;

[0083] The potential characteristic variables, the seismic wave change characteristics and the pre-earthquake temperature value at the prediction time are input into the autovariation decoder to obtain the seismic wave value at the prediction time.

[0084] As a preferred embodiment, the earthquake judgment module constructs a minimum seismic wave value at the time when an earthquake occurs in the detection area based on the seismic wave detection data. When the seismic wave value output by the earthquake prediction model reaches the minimum seismic wave value, it is judged that an earthquake is about to occur at the predicted moment.

[0085] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0086] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. They should all be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. An earthquake prediction method based on spatiotemporal attention mechanism, characterized in that: The following steps are involved: Arranging a plurality of piezoelectric seismic sensors in the detection area, and acquiring seismic wave detection data based on the piezoelectric seismic sensors; Obtaining surface temperature data based on meteorological station observation data in the detection area, performing one-dimensional and two-dimensional wavelet transforms on the surface temperature data to obtain the in-situ temperature of the detection area, using a vorticity algorithm to obtain in-situ vorticity based on the in-situ temperature, and using a RST algorithm to obtain pre-earthquake temperature value data; An earthquake prediction model is constructed using a doorway recurrent unit and a spatiotemporal attention mechanism, a dataset is constructed based on the seismic wave detection data and pre-earthquake temperature value data, and the earthquake prediction model is trained using the dataset; Using the trained earthquake prediction model, the seismic wave value at the corresponding moment is predicted based on the pre-earthquake temperature value calculated at a specific time, and the magnitude of the seismic wave value is used to determine whether an earthquake has occurred. The process of training the earthquake prediction model using the data set includes: Constructing the change curves of the seismic wave detection data and the pre-earthquake temperature value data over time according to the data set; Obtaining seismic wave values ​​in the X-axis, Y-axis, and Z-axis directions at a certain location based on the seismic wave detection data, obtaining pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions at the same location based on the pre-seismic temperature value data, and splicing the seismic wave values ​​and pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions to construct an earthquake characteristic vector; A relational dimension encoder is constructed using a spatiotemporal attention mechanism, which inputs the change curve of the seismic wave detection data and the pre-earthquake temperature value data as well as the earthquake feature vector to generate potential feature variables; and a graph attention mechanism is used to encode the earthquake feature vector to obtain input variables; Using the potential feature variable and the input variable as the hidden state and input variable in a gated recurrent unit, and performing a cyclic update on the potential feature variable to obtain an optimal potential feature variable; A variational autoencoder based on a gated recurrent unit obtains seismic wave variation characteristics according to a variation curve of the seismic wave detection data; The potential characteristic variables, the seismic wave change characteristics and the pre-earthquake temperature value at the prediction time are input into the variational autoencoder to obtain the seismic wave value at the prediction time.

2. The earthquake prediction method based on spatiotemporal attention mechanism according to claim 1, characterized in that: The method of obtaining the in-situ vorticity using the vorticity algorithm is as follows: Where, is the central pixel of the detection area at a certain position at a certain moment, is the size difference between the center pixel and the four neighboring pixels.

3. The earthquake prediction method based on spatiotemporal attention mechanism according to claim 1, characterized in that: The process of obtaining pre-earthquake temperature data using the RST algorithm includes: Obtaining the measured surface temperature and the average surface temperature of the detection area at a certain time in previous years based on the observation data of the meteorological station, and obtaining the DN value of the detection area at that time based on the difference between the measured surface temperature and the average surface temperature; The pre-earthquake temperature value data is obtained based on the DN value, and the mean and standard deviation corresponding to the DN value.

4. The earthquake prediction method based on spatiotemporal attention mechanism according to claim 1, characterized in that: The process of determining whether an earthquake has occurred based on the magnitude of the seismic wave value includes: A minimum seismic wave value at the time an earthquake occurs in the detection area is constructed based on the seismic wave detection data. When the seismic wave value output by the earthquake prediction model reaches the minimum seismic wave value, it is determined that an earthquake is about to occur at the predicted time.

5. An earthquake prediction system based on spatiotemporal attention mechanism, characterized in that: It includes seismic wave detection module, pre-earthquake temperature calculation module, seismic wave prediction module and earthquake judgment module; The seismic wave detection module is used to obtain seismic wave detection data of the detection area; The pre-earthquake temperature calculation module is used to obtain pre-earthquake temperature value data of the detection area; The seismic wave prediction module constructs an earthquake prediction model based on the seismic wave detection data and the pre-earthquake temperature value data acquired at different times, using a doorway recurrent unit and a spatiotemporal attention mechanism to predict the seismic wave value at a future time; The earthquake judgment module is used to judge whether an earthquake will occur based on the magnitude of the earthquake wave value at a future moment; The earthquake prediction module constructs the change curves of the seismic wave detection data and the pre-earthquake temperature value data over time respectively; Obtaining seismic wave values ​​in the X-axis, Y-axis, and Z-axis directions at a certain location based on the seismic wave detection data, obtaining pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions at the same location based on the pre-seismic temperature value data, and splicing the seismic wave values ​​and pre-seismic temperature values ​​in the X-axis, Y-axis, and Z-axis directions to construct an earthquake characteristic vector; A relational dimension encoder is constructed using a spatiotemporal attention mechanism, which inputs the change curve of the seismic wave detection data and the pre-earthquake temperature value data as well as the earthquake feature vector to generate potential feature variables; And use the graph attention mechanism to encode the earthquake feature vector to obtain the input variable; Using the potential feature variable and the input variable as the hidden state and input variable in a gated recurrent unit, and performing a cyclic update on the potential feature variable to obtain an optimal potential feature variable; A variational autoencoder based on a gated recurrent unit obtains seismic wave variation characteristics according to a variation curve of the seismic wave detection data; The potential characteristic variables, the seismic wave change characteristics and the pre-earthquake temperature value at the prediction time are input into the variational autoencoder to obtain the seismic wave value at the prediction time.

6. The earthquake prediction system based on spatiotemporal attention mechanism according to claim 5, characterized in that: The seismic wave detection module includes a plurality of piezoelectric seismic sensors, each of which uses a piezoelectric single crystal chip as a piezoelectric plate. When an earthquake occurs, the piezoelectric plate deforms due to the vibration, and the mechanical energy of the vibration is converted into electrical energy based on the direct piezoelectric effect, thereby obtaining a seismic electrical signal. The seismic wave detection module obtains vibration signals from different detection areas through a plurality of piezoelectric seismic sensors, and performs statistics on the vibration signals to generate seismic wave detection data.

7. The earthquake prediction system based on spatiotemporal attention mechanism according to claim 5, characterized in that: The pre-earthquake temperature calculation module obtains meteorological station observation data in the detection area, obtains surface temperature data based on the meteorological station observation data, performs one-dimensional wavelet transform and two-dimensional wavelet transform on the surface temperature data, obtains the in-situ temperature of the detection area, uses the vorticity algorithm to obtain the in-situ vorticity based on the in-situ temperature, and uses the RST algorithm to obtain pre-earthquake temperature value data.

8. The earthquake prediction system based on spatiotemporal attention mechanism according to claim 5, characterized in that: The earthquake judgment module constructs a minimum seismic wave value at the time when an earthquake occurs in the detection area based on the seismic wave detection data. When the seismic wave value output by the earthquake prediction model reaches the minimum seismic wave value, it is judged that an earthquake is about to occur at the predicted time.

Images