Deep learning-based Zhang Zheng No.1 satellite multi-parameter space-time correlation analysis method

Abstract

The invention discloses a Zhang Zheng No.1 satellite multi-parameter space-time correlation analysis method based on deep learning. The method comprises the following steps: firstly, acquiring and preprocessing ionosphere multi-parameter data, including magnetic field three-component waveform data, electron density Ne and GPS TEC data; then, a time sequence relation of different parameter earthquake precursor anomalies of the ionosphere is established by utilizing a random walk and a yoke theory; and then, on the basis of a CNN-BiLSTM deep learning network, a Swarm B satellite ionized layer multi-parameter seismic anomaly intelligent identification model is constructed. And then, by using a transfer learning technology, migrating the Swarm B satellite model to a ZH-1 (01) satellite, and constructing a ZH-1 (01) multi-parameter seismic anomaly identification model. And finally, seismic anomaly identification and verification are carried out by using Zhang Heng No.1 satellite constellation data, and the model is continuously optimized. According to the invention, time-space correlation analysis is carried out on the multi-source ionosphere parameters through a deep learning method, the accuracy of seismic anomaly recognition can be effectively improved, and a scientific basis is provided for seismic early warning.

Description

Technical Field

[0001] The present invention belongs to the field of earthquake early warning technology, and specifically relates to a Zhangheng-1 satellite multi-parameter spatiotemporal correlation analysis method based on deep learning. This innovative method realizes the intelligent identification and prediction of earthquake precursors by comprehensively utilizing a variety of ionospheric parameters, including magnetic field three-component waveform data, electron density Ne and GPS TEC data. The method integrates advanced deep learning technology, random walk theory and martingale theory to construct a complex and accurate multi-parameter spatiotemporal correlation analysis model. Through in-depth analysis and model migration of Swarm B satellite and Zhangheng-1 satellite data, the present invention not only improves the accuracy of earthquake anomaly identification, but also enhances the generalization ability and real-time performance of the model. The application of this method is of great significance for improving the performance of earthquake early warning systems and reducing the losses caused by earthquake disasters, and represents an important breakthrough in the field of earthquake early warning technology. Background Art

[0002] The background technology section of the present invention is intended to provide readers with a more comprehensive technical background to help them better understand the technical principles and application value of the present invention. It should be noted that this information is not intended to admit or imply that these technologies are already known to those skilled in the art, but to provide a more detailed technical background to help readers better understand the technical value of the present invention.

[0003] Earthquake early warning technology is one of the important directions of earthquake science research. Earthquake is a sudden and destructive natural disaster that has a huge impact on human society and economy. Therefore, accurately predicting the time, location and intensity of earthquakes has always been a major challenge facing the scientific community. In recent years, with the development of satellite observation technology, using satellite observation data to identify and predict earthquake anomalies has become an important research direction. However, there are some limitations in existing technologies. The accuracy of earthquake anomaly identification with a single parameter is not high, and the mutual influence between multiple parameters cannot be fully considered. The spatiotemporal correlation between multiple parameters is difficult to establish effectively, resulting in weak generalization ability of the model. In addition, there is subjectivity in manually extracted features, which may ignore some important features or introduce unnecessary features. These limitations have seriously affected the development and application of earthquake early warning technology, and new methods and technologies are urgently needed to overcome these challenges and improve the accuracy and reliability of earthquake early warning. Summary of the invention

[0004] The purpose of the present invention is to provide a deep learning-based Zhangheng-1 satellite multi-parameter spatiotemporal correlation analysis method to solve the technical problem of low accuracy in earthquake anomaly recognition in the prior art. This method combines deep learning technology with multi-parameter spatiotemporal correlation analysis to achieve intelligent recognition and prediction of earthquake precursors, thereby improving the accuracy and real-time performance of earthquake early warning.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A deep learning-based multi-parameter spatiotemporal correlation analysis method for Zhangheng-1 satellite includes the following steps: (1) Acquire and preprocess ionospheric multi-parameter data, including magnetic field three-component waveform data, Ne and GPS TEC data; (2) Using random walk and martingale theory, the temporal relationship between earthquake precursor anomalies of different ionospheric parameters is established; (3) Construct a Swarm B satellite ionospheric multi-parameter earthquake anomaly intelligent recognition model based on the CNN-BiLSTM deep learning network; (4) Using transfer learning technology, the Swarm B satellite model was transferred to the ZH-1(01) satellite to build a ZH-1(01) multi-parameter earthquake anomaly recognition model; (5) Use the Zhangheng-1 satellite constellation data to identify and verify earthquake anomalies and continuously optimize the model.

[0006] By adopting the above technical solution, the present invention has the following beneficial effects: The present invention automatically extracts features through deep learning methods, avoids the subjectivity of manual feature extraction, and improves the efficiency and accuracy of feature extraction. The transfer learning technology is used to solve the problem of insufficient ZH-1 (01) satellite data and improve the generalization ability of the model. The multi-parameter spatiotemporal correlation analysis is used to make full use of the correlation between different parameters and improve the accuracy of earthquake anomaly identification. Through continuous model verification and optimization, the model can continuously adapt to new data and situations and maintain a high recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0008] Figure 1 A flowchart of a multi-parameter spatiotemporal correlation analysis method provided by an embodiment of the present invention; Figure 2 A schematic diagram of a transfer learning process provided by an embodiment of the present invention; DETAILED DESCRIPTION

[0009] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example

[0010] This embodiment 1 provides a method for analyzing the multi-parameter spatiotemporal correlation of the Zhangheng-1 satellite based on deep learning. Before describing the specific steps, some professional terms used in the present invention are first explained: 1. ZH-1(01) Satellite: refers to ZhangHeng-1 01 Satellite, which is China's first space seismic electromagnetic monitoring satellite used to detect earthquake precursor information.

[0011] 2. Swarm B satellite: One of the Earth magnetic field detection satellites launched by the European Space Agency (ESA). It is part of the Swarm Constellation and is used to study the Earth's magnetic field.

[0012] 3. CNN-BiLSTM: A deep learning network structure that combines the advantages of Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM). CNN is used to extract local features, and BiLSTM is used to capture long-distance dependencies.

[0013] 4. Ne: refers to electron density, which is an important parameter in the ionosphere.

[0014] 5. GPS TEC: Global Positioning System Total Electron Content, an important indicator of the electron content in the ionosphere.

[0015] 6. IBPT Classifier: Improved Binary Particle Swarm Optimization-based Decision Tree Classifier is a machine learning algorithm for data classification.

[0016] 7. Dobrovolsky distance: An empirical formula for estimating the impact range of an earthquake, R = 10^(0.43M), where R is the impact radius (km) and M is the magnitude of the earthquake. This formula was proposed by Russian geophysicist Dobrovolsky.

[0017] Now, let's describe the specific steps of this method in detail: Step 1: Data acquisition and preprocessing Acquire multi-parameter data of the ionosphere, including three-component waveform data of the magnetic field, Ne and GPS TEC data. Data sources include ZH-1(01) high-precision magnetometer, Swarm B satellite vector field magnetometer, ZH-1(01) Langmuir probe, Swarm B satellite Langmuir probe and domestic tectonic environment monitoring network. The time range is selected from Swarm B satellite data from April 2014 to 2021 and ZH-1(01) data from 2019 to 2021, and the spatial range is China and its surrounding areas.

[0018] The magnetic field data were processed by first-order difference and cubic spline curve fitting to eliminate low-frequency magnetic field disturbance signals and obtain trajectory residuals. The Ne data were interpolated to unify the time resolution of ZH-1(01) and Swarm B satellites to 3s.

[0019] Step 2: Establishing the timing relationship The random walk and martingale theory are used to establish the time series relationship of earthquake precursor anomalies of different ionospheric parameters. The lengths of the first and second windows are preset (initial setting is 30 or 45). The random walk probability of each time series data point in the first and second windows is calculated. The ratio of the random walk probability of the two windows is used to analyze the changes before and after the data point. Then, the power martingale value in the martingale theory is used to characterize the continuous changes of the time series data.

[0020] Step 3: Swarm B Satellite Model Construction The Swarm B satellite ionospheric multi-parameter earthquake anomaly intelligent recognition model is constructed based on the CNN-BiLSTM deep learning network. The initial number of CNN network layers is 3, including one-dimensional convolutional layers and pooling layers, which are used to extract local parallel features. The initial number of BiLSTM layers is 3, which is used for feature extraction and sequence learning of long-distance dependent features. The fully connected layer and Softmax classifier are used to obtain the final classification results.

[0021] Step 4: ZH-1(01) satellite model construction like Figure 2As shown in the figure, the Swarm B satellite model is migrated to the ZH-1(01) satellite using transfer learning technology to build a ZH-1(01) multi-parameter earthquake anomaly recognition model. The trained Swarm B satellite CNN-BiLSTM deep learning network (source domain) model parameters are migrated to the ZH-1(01) satellite (target domain) model. The obtained weight parameters are used to initialize the parameters of the remaining network layers, and the model is corrected by adjusting the model parameters such as the loss function and optimizer function.

[0022] Step 5: Model Validation and Optimization The Zhangheng-1 satellite constellation data is used to identify and verify earthquake anomalies, and the model is continuously optimized. The measured parameters of the ZH-1(01) in orbit and ionospheric multi-parameters such as GPS TEC are used to conduct research on the effectiveness of identifying anomalies in real earthquake cases. Based on the feedback from application evaluation, the model parameters are adjusted and the model is optimized, and the process is repeated. Example

[0023] This embodiment 2 further refines the structure of the CNN-BiLSTM deep learning network on the basis of embodiment 1. This embodiment 2 includes the technical content disclosed in embodiment 1, and the technical content of this embodiment 2 that is the same as that of embodiment 1 is not repeated here. The differences between this embodiment 2 and embodiment 1 are described below.

[0024] In this embodiment, the initial number of CNN network layers is increased to 5, and each layer is followed by a maximum pooling layer. The initial number of BiLSTM layers is also increased to 5 to improve the model's ability to capture long-distance dependent features. A dropout layer is added before the fully connected layer to prevent overfitting. This deeper network structure can extract more complex features, which helps to improve the recognition accuracy of the model. Example

[0025] like Figure 2 As shown, the third embodiment further refines the process of transfer learning based on the first embodiment. The third embodiment includes the technical contents disclosed in the first embodiment, and the technical contents of the third embodiment and the first embodiment are not repeated here. The differences between the third embodiment and the first embodiment are described below.

[0026] In this embodiment, the transfer learning process adopts a progressive fine-tuning strategy. First, all layers of the Swarm B satellite model are frozen, and only the newly added fully connected layers are trained. Then, more layers are gradually unfrozen, starting from the last layer, until all layers are trainable. This strategy can better adapt to the data characteristics of the ZH-1(01) satellite while retaining the useful features learned by the Swarm B satellite model.

[0027] The technical features described in the specification using words such as "better", "preferred", "preferably" or "more preferably" are not essential technical features of the present invention. When the technical features are missing in the embodiment, the content protected by the embodiment also falls within the scope of protection of the present invention.

[0028] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based multi-parameter spatiotemporal correlation analysis method for Zhangheng-1 satellite, characterized in that: The following steps are involved: (1) acquiring and preprocessing ionospheric multi-parameter data, wherein the ionospheric multi-parameter data includes magnetic field three-component waveform data, Ne and GPS TEC data; (2) Using random walk and martingale theory, the temporal relationship between earthquake precursor anomalies of different ionospheric parameters is established; (3) Construct a Swarm B satellite ionospheric multi-parameter earthquake anomaly intelligent recognition model based on the CNN-BiLSTM deep learning network; (4) Using transfer learning technology, the Swarm B satellite model was transferred to the ZH-1(01) satellite to build a ZH-1(01) multi-parameter earthquake anomaly recognition model; (5) Use the Zhangheng-1 satellite constellation data to identify and verify earthquake anomalies and continuously optimize the model.

2. The method according to claim 1, characterized in that The pretreatment in step (1) comprises: Perform first-order difference and cubic spline curve fitting on magnetic field data; The Ne data were interpolated to unify the time resolution to 3 s.

3. The method according to claim 1, characterized in that The method for establishing the timing relationship in step (2) includes: Set the length of the front and back windows, the initial length is 30 or 45; Calculate the random walk probability of each time series data point in the first and last two windows; The ratio of the random walk probabilities of the two windows is used to analyze the changes before and after the data points; Use the martingale value in martingale theory to describe the continuous changes of time series data; Determine earthquake anomalies based on the above calculation results; Establish the temporal relationship of multi-parameter pre-earthquake anomalies.

4. The method according to claim 1, characterized in that The CNN-BiLSTM deep learning network in step (3) includes: CNN network layer, including 3-5 one-dimensional convolutional layers and pooling layers, used to extract local parallel features; BiLSTM layer, which contains 3-5 layers of bidirectional long short-term memory network, is used for feature extraction and sequence learning of long-distance dependent features; Fully connected layers and Softmax classifier are used to obtain the final classification results.

5. The method according to claim 1, characterized in that The transfer learning method in step (4) includes: Migrate the trained Swarm B satellite CNN-BiLSTM deep learning network model parameters to the ZH-1(01) satellite model; Use the obtained weight parameters to initialize the parameters of the remaining network layers; Modify the model by adjusting model parameters such as loss function and optimizer function; A progressive fine-tuning strategy is adopted, where all layers are first frozen, only the newly added fully connected layers are trained, and then more layers are gradually unfrozen.

6. The method according to claim 1, characterized in that The seismic anomaly identification adopts an IBPT classifier model, which is an improved binary particle swarm optimization decision tree classifier.

7. The method according to claim 1, characterized in that The time range of the ionospheric multi-parameter data is the Swarm B satellite data from April 2014 to 2021 and the ZH-1(01) data from 2019 to 2021, and the spatial range is China and its surrounding areas.

8. The method according to claim 1, characterized in that The object of earthquake anomaly identification is earthquakes of magnitude 5.5 or above.

9. The method according to claim 1, characterized in that The spatial range of the ionospheric multi-parameter data is a time series within the Dobrovolsky distance range from the epicenter, where the Dobrovolsky distance is calculated by the formula R = 10^(0.43M), R is the impact radius (km), and M is the earthquake magnitude.

Images