Earthquake early warning method and system based on multi-modal data fusion

Abstract

The invention discloses an earthquake early warning method and system based on multi-modal data fusion, and relates to the technical field of earthquake early warning. The method comprises the following specific steps: acquiring multi-modal earthquake monitoring data of a target area in real time; preprocessing the multi-modal seismic monitoring data, and performing feature extraction by using a feature extraction model to obtain vector feature representation of each data source; constructing a multi-modal data fusion network to perform data fusion on the vector feature representations of the data sources to obtain fusion features; inputting the fusion features into a pre-constructed Bayesian network model, and reasoning through the Bayesian network model to obtain seismic probability distribution of the target area; and according to the earthquake probability distribution of the target area obtained by reasoning, setting a grading early warning rule, and triggering earthquake early warning in combination with the grading early warning rule. Earthquake early warning is carried out based on fusion of multi-source heterogeneous data, and the robustness and accuracy of earthquake early warning are improved.

Description

Technical Field

[0001] This invention relates to the field of earthquake early warning technology, and in particular to an earthquake early warning method and system based on multimodal data fusion. Background Technology

[0002] Earthquakes, as the most destructive natural disaster on Earth, have always been a key area of ​​focus and breakthrough for the scientific and engineering communities in terms of prediction and early warning technologies. Although the time window provided by earthquake early warning is usually short, only a few seconds to tens of seconds, this period is crucial for taking emergency measures, protecting lives, and reducing property damage. Traditional earthquake early warning mainly relies on the first-arrival wave information recorded by seismic networks to estimate the magnitude and intensity within seconds after the earthquake and issue warnings to potentially affected areas. However, the complexity of geological structures and environmental noise can lead to attenuation and interference of seismic wave signals, and single seismic wave data has the problem of early warning blind spots, resulting in the risk of false alarms or missed alarms. Multimodal data fusion, by introducing heterogeneous observation methods, can effectively compensate for the above-mentioned shortcomings. Therefore, for those skilled in the art, how to conduct earthquake early warning based on multimodal data fusion is an urgent problem to be solved. Summary of the Invention

[0003] The purpose of this invention is to provide an earthquake early warning method and system based on multimodal data fusion to solve the problems mentioned in the background art and improve the robustness, accuracy and timeliness of earthquake early warning.

[0004] To achieve the above objectives, the present invention provides the following solution: an earthquake early warning method based on multimodal data fusion, the specific steps of which include the following:

[0005] Real-time acquisition of multimodal seismic monitoring data for the target area;

[0006] After preprocessing the multimodal earthquake monitoring data, feature extraction is performed using a feature extraction model to obtain vector feature representations of each data source;

[0007] A multimodal data fusion network is constructed to fuse the vector feature representations of various data sources to obtain fused features;

[0008] The fused features are input into a pre-constructed Bayesian network model, and the earthquake probability distribution of the target area is obtained through inference by the Bayesian network model.

[0009] Based on the earthquake probability distribution of the target area obtained through reasoning, a graded early warning rule is set, and an earthquake early warning is triggered in combination with the graded early warning rule.

[0010] Preferably, the multimodal earthquake monitoring data includes meteorological data, seismic station observation data, and geological data; the seismic station observation data includes earthquake waveform records, source parameters, and station information.

[0011] Preferably, the preprocessing of the multimodal seismic monitoring data includes: data cleaning, noise reduction, spatiotemporal registration, and standardization; for the seismic station observation data, extracting its peak acceleration, peak velocity, peak displacement, and spectrum, and identifying and removing abnormal or distorted observation data by training a support vector machine classifier; unifying the timestamps for the multimodal seismic monitoring data, aligning data at different scales, and completing spatiotemporal registration.

[0012] Preferably, different feature extraction models are used for feature extraction of earthquake monitoring data of different modalities. The specific steps are as follows: using a convolutional neural network to extract spatial features from the earthquake station observation data, using a long short-term memory network to extract temporal features from the meteorological data, and using a fully connected neural network to extract features from the geological data.

[0013] Preferably, the multimodal data fusion network is a Transformer-based multimodal data fusion network; the Transformer-based multimodal data fusion network includes a first encoder, a second encoder, a third encoder, and a Transformer decoder; meteorological data is input into the first encoder, seismic station observation data is input into the second encoder, and geological data is input into the third encoder; the first encoder, the second encoder, and the third encoder are all composed of multiple identical layers stacked together, each layer containing a multi-head self-attention layer connected by residuals and a feedforward neural network, and the generated fusion features are input into the Transformer decoder to obtain the final fusion result.

[0014] Preferably, the method further includes inputting the fused features into a convolutional neural network model, outputting earthquake early warning results, and performing a weighted average of the outputs of the convolutional neural network model and the Bayesian network model to obtain the final earthquake early warning probability prediction.

[0015] Preferably, after an earthquake event, its fused features and earthquake damage results are incorporated into the dataset, and the conditional probability table of the Bayesian network model is retrained and dynamically updated.

[0016] Preferably, the Bayesian network model is constructed by determining the dependencies and conditional probability distributions between nodes based on the nodes and edges.

[0017] On the other hand, an earthquake early warning system based on multimodal data fusion is provided, including a multimodal data acquisition module, a feature extraction module, a data fusion module, an inference module, and an early warning module; wherein,

[0018] The multimodal data acquisition module is used to acquire multimodal seismic monitoring data of the target area in real time;

[0019] The feature extraction module is used to preprocess the multimodal earthquake monitoring data and then extract features using a feature extraction model to obtain vector feature representations of each data source.

[0020] The data fusion module is used to construct a multimodal data fusion network to fuse the vector feature representations of each data source to obtain fused features;

[0021] The inference module is used to input the fused features into a pre-constructed Bayesian network model and obtain the earthquake probability distribution of the target area through inference using the Bayesian network model.

[0022] The early warning module is used to set graded early warning rules based on the earthquake probability distribution of the target area obtained through reasoning, and to trigger earthquake early warning in combination with the graded early warning rules.

[0023] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0024] (1) By integrating seismic station data and meteorological data from multiple sources, the spatial features extracted by CNN can capture the waveform propagation differences of stations around the epicenter, and the temporal features extracted by LSTM can uncover abnormal temporal fluctuations in meteorological data. The fusion of the two can avoid the limitations of a single data source, and the complementary multimodal features can reduce information blind spots and improve the robustness and accuracy of earthquake early warning.

[0025] (2) Bayesian networks quantify the probability of earthquake occurrence corresponding to different fusion features through conditional probability tables and posterior probability calculations, rather than simple binary classification judgments. They use probabilistic reasoning to reduce decision-making errors and weighted average the outputs of convolutional neural network models and Bayesian network models to obtain the final earthquake early warning probability prediction, which further improves the accuracy of earthquake early warning. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart of the method of the present invention;

[0028] Figure 2 This is a system structure diagram of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] The purpose of this invention is to provide an earthquake early warning method based on multimodal data fusion, such as... Figure 1 As shown, the specific steps include the following:

[0031] S1. Real-time acquisition of multimodal seismic monitoring data of the target area;

[0032] S2. After preprocessing the multimodal seismic monitoring data, feature extraction is performed using a feature extraction model to obtain vector feature representations of each data source;

[0033] S3. Construct a multimodal data fusion network to fuse the vector feature representations of each data source to obtain fused features;

[0034] S4. Input the fused features into the pre-constructed Bayesian network model, and obtain the earthquake probability distribution of the target area through inference by the Bayesian network model;

[0035] S5. Based on the earthquake probability distribution of the target area obtained through reasoning, set graded early warning rules, and trigger earthquake early warnings in combination with the graded early warning rules.

[0036] Furthermore, multimodal earthquake monitoring data includes meteorological data, seismic station observation data, and geological data; seismic station observation data includes earthquake waveform records, source parameters, and station information.

[0037] Furthermore, the preprocessing of multimodal earthquake monitoring data includes: data cleaning, denoising, spatiotemporal registration, and standardization; for seismic station observation data, peak acceleration, peak velocity, peak displacement, and spectrum are extracted, and abnormal or distorted observation data are identified and removed by training a support vector machine classifier; for multimodal earthquake monitoring data, a unified timestamp is applied, data at different scales are aligned, and spatiotemporal registration is completed.

[0038] In S2, different feature extraction models are used for earthquake monitoring data of different modalities. The specific steps are as follows: convolutional neural networks are used to extract spatial features from earthquake station observation data, long short-term memory networks are used to extract temporal features from meteorological data, and fully connected neural networks are used to extract features from geological data. In the multimodal earthquake early warning scheme, CNN, LSTM, and fully connected neural networks (FCN) are used for feature extraction to accurately match the characteristics of earthquake station observation data (spatial correlation), meteorological data (temporal dependence), and geological data (static / structured attributes). This is the optimal choice for accurately matching data characteristics. The three types of networks each perform their respective functions, and finally, feature complementarity is achieved through a fusion network, providing high-quality input for subsequent Bayesian network inference.

[0039] In S3, the multimodal data fusion network is a Transformer-based multimodal data fusion network, which includes a first encoder, a second encoder, a third encoder, and a Transformer decoder. Meteorological data is input into the first encoder, seismic station observation data is input into the second encoder, and geological data is input into the third encoder. The first encoder, second encoder, and third encoder are all composed of multiple identical layers stacked together. Each layer contains a multi-head self-attention layer connected by residuals and a feedforward neural network. The self-attention mechanism processes the relationships between features learned from sequential data, captures the complementary information relationships between multi-source heterogeneous data, calculates the attention scores between different modalities, and performs information weighted integration to obtain attention weights, thereby generating fused features. The generated fused features are input into the Transformer decoder to obtain the final fusion result.

[0040] Furthermore, in S4, the dependencies and conditional probability distributions between nodes are determined based on nodes and edges. A Bayesian network model is constructed, and after obtaining real-time fused features, the Bayesian network initiates the inference process. The core of this process is to update the posterior probability of the target node with new evidence. After inference is completed, the posterior probability of the target node is output. Based on the probability distribution obtained from the inference, combined with preset hierarchical early warning rules, the conversion from probability to early warning is achieved. After an earthquake event, its fused features and earthquake damage results are incorporated into the dataset, and the conditional probability table of the Bayesian network model is retrained and dynamically updated.

[0041] Furthermore, this involves inputting fused features into a convolutional neural network (CNN) model to output earthquake early warning results. The outputs of the CNN and Bayesian network models are then weighted and averaged to obtain the final earthquake early warning probability prediction. The CNN model employs a lightweight architecture, with the output layer using the sigmoid activation function to directly output the earthquake occurrence probability. If the difference between the probability output by the Bayesian network and the CNN model exceeds 20%, the output is temporarily suspended, and the inference is repeated after supplementing new evidence. If the inference confidence level is less than 70%, the result is deemed unreliable and retained only as monitoring data, not used in early warning decisions.

[0042] The system compares the probability distribution of the Bayesian network output with the hierarchical early warning rules in real time, automatically determines the early warning level, and issues early warnings only to high-risk sub-regions based on the epicenter probability distribution to avoid notification of the entire region. Different emergency measures are matched to different early warning levels and are automatically triggered by the system or confirmed by manual confirmation. After the seismic waves arrive, the early warning is lifted according to the actual seismic situation (no earthquake / magnitude lower than expected).

[0043] On the other hand, an earthquake early warning system based on multimodal data fusion is provided, such as... Figure 2 As shown, it includes a multimodal data acquisition module, a feature extraction module, a data fusion module, an inference module, and an early warning module; among which,

[0044] The multimodal data acquisition module is used to acquire multimodal seismic monitoring data of the target area in real time.

[0045] The feature extraction module is used to extract features from multimodal seismic monitoring data after preprocessing, and obtain vector feature representations of each data source.

[0046] The data fusion module is used to construct a multimodal data fusion network to fuse the vector feature representations of various data sources to obtain fused features;

[0047] The inference module is used to input fused features into a pre-built Bayesian network model and obtain the earthquake probability distribution of the target area through inference from the Bayesian network model.

[0048] The early warning module is used to set graded early warning rules based on the earthquake probability distribution of the target area obtained through reasoning, and to trigger earthquake early warnings in combination with the graded early warning rules.

[0049] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for earthquake early warning based on multi-modal data fusion, characterized in that, The specific steps include the following: Real-time acquisition of multi-modal seismic monitoring data of the target area; After preprocessing the multi-modal seismic monitoring data, feature extraction is performed using a feature extraction model to obtain vector feature representations of each data source; A multi-modal data fusion network is constructed to fuse the vector feature representations of each data source to obtain fusion features; The fusion features are input into a pre-constructed Bayesian network model, and the target area seismic probability distribution is obtained by reasoning through the Bayesian network model; According to the target area seismic probability distribution obtained by reasoning, a hierarchical warning rule is set, and the hierarchical warning rule is combined to trigger a seismic warning.

2. The earthquake early warning method based on multi-modal data fusion according to claim 1, characterized in that, The multi-modal seismic monitoring data includes meteorological data, seismic station observation data, and geological data; the seismic station observation data includes seismic waveform records, source parameters, and station information.

3. The earthquake early warning method based on multi-modal data fusion according to claim 2, characterized in that, The preprocessing of the multi-modal seismic monitoring data includes data cleaning, denoising, time-space registration, and standardization processing; for the seismic station observation data, the peak acceleration, peak velocity, peak displacement, and spectrum are extracted, and the support vector machine classifier is trained to identify and eliminate abnormal or distorted observation data; for the multi-modal seismic monitoring data, the timestamps are unified, the different scale data are aligned, and the time-space registration is completed.

4. The earthquake early warning method based on multi-modal data fusion according to claim 2, characterized in that, Different feature extraction models are used for feature extraction of different modal seismic monitoring data, and the specific steps are as follows: a convolutional neural network is used to extract spatial features from the seismic station observation data, a long short-term memory network is used to extract time series features from the meteorological data, and a fully connected neural network is used to extract features from the geological data.

5. The earthquake early warning method based on multi-modal data fusion according to claim 2, characterized in that, The multi-modal data fusion network is a Transformer-based multi-modal data fusion network; the Transformer-based multi-modal data fusion network includes a first encoder, a second encoder, a third encoder, and a Transformer decoder; meteorological data is input into the first encoder, seismic station observation data is input into the second encoder, and geological data is input into the third encoder; the first encoder, the second encoder, and the third encoder are stacked by multiple identical layers, each layer contains a multi-head self-attention layer and a feedforward neural network connected in residual, and the generated fusion features are input into the Transformer decoder to obtain the final fusion result.

6. The earthquake early warning method based on multi-modal data fusion according to claim 1, characterized in that, The fusion features are also input into a convolutional neural network model to output a seismic warning result, and the outputs of the convolutional neural network model and the Bayesian network model are weighted and averaged to obtain a final seismic warning probability prediction.

7. The earthquake early warning method based on multi-modal data fusion according to claim 1, characterized in that, After a seismic event, its fusion features and damage results are included in the data set, and the conditional probability table of the Bayesian network model is retrained for dynamic updating. 8.The earthquake early warning method based on multi-modal data fusion of claim 1, wherein, The Bayesian network model is constructed according to the dependency relationship and conditional probability distribution between nodes and edges.

9. A multi-modal data fusion based earthquake early warning system characterized in that, It includes a multi-modal data acquisition module, a feature extraction module, a data fusion module, an inference module, and a warning module; wherein, The multi-modal data acquisition module is configured to acquire multi-modal seismic monitoring data of a target region in real time. The feature extraction module is configured to perform feature extraction on the multi-modal seismic monitoring data after preprocessing, using a feature extraction model, to obtain vector feature representations of each data source. The data fusion module is configured to construct a multi-modal data fusion network to perform data fusion on the vector feature representations of each data source to obtain fusion features. The inference module is configured to input the fusion features into a pre-constructed Bayesian network model, and obtain a target region seismic probability distribution through Bayesian network inference. The early warning module is configured to set a hierarchical early warning rule according to the target region seismic probability distribution obtained through inference, and trigger a seismic early warning in combination with the hierarchical early warning rule.

Images