A Dst index calculation method and system integrating neutral atom imaging data and machine learning
By fusing neutral atomic imaging data and machine learning technology, combining residual networks and multi-layer perceptron models, the accurate prediction of the Dst index is achieved, the spatial coverage and temporal resolution limitations of traditional methods are solved, and efficient global geomagnetic storm monitoring and early warning services are provided.
Patent Information
- Application Number
- CN202411769870.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-04
- Publication Date
- 2025-06-24
- Estimated Expiration
- 2044-12-04
AI Technical Summary
The traditional Dst index calculation method has limitations such as limited spatial coverage, low temporal resolution and difficulty in capturing the overall dynamics of the magnetosphere, making it difficult to meet the global needs of geomagnetic storm monitoring and space weather warning.
Using a method of fusing neutral atomic imaging data and machine learning, data is collected through a satellite high-energy neutral atomic imager, combined with residual network and multi-layer perceptron model, image features are extracted and satellite position data are multimodal fusion, realizing nonlinear regression prediction of the Dst index.
It significantly improves the calculation accuracy and real-time performance of the Dst index, breaks through the spatial and time limitations of traditional methods, and provides a new high-resolution solution for global geomagnetic storm monitoring and space weather warning.
Smart Images
Figure CN119810580B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of space physics and geomagnetic index calculation, and particularly relates to a method and system for calculating the Dst index by fusing energetic neutral atom imaging data and machine learning. Background Art
[0002] Geomagnetic storms are the direct products of the interaction between the solar wind and the Earth's magnetosphere, which can trigger global magnetic field disturbances. Their main manifestation is the drastic change in the intensity of the Earth's magnetic field, which has a significant impact on modern technological systems of mankind, such as communication, navigation, satellite operation, and power transmission networks. In order to measure the intensity and dynamic evolution of geomagnetic storms, scientists have proposed a variety of geomagnetic indices, among which the Dst index is one of the most important criteria. The Dst index is obtained by ground magnetometers in the low-latitude region, reflecting the global magnetic field changes caused by the Earth's ring current, and can better characterize the time evolution process of geomagnetic storm intensity.
[0003] Although traditional Dst index calculation methods meet the needs of scientific research and space weather monitoring to a certain extent, they also have some significant limitations: (1) Limited spatial coverage: The distribution of ground magnetometers is restricted by geographical limitations, resulting in limitations in spatial coverage. (2) Low time resolution: Ground magnetometers usually record data once an hour, and this resolution is insufficient to cope with rapidly developing space weather events. (3) Inadequate capture of the overall dynamics of the magnetosphere: Ground observations can only provide local information of the geomagnetic field, and it is difficult to comprehensively reflect the ring current structure and its dynamic evolution in the magnetosphere.
[0004] In order to overcome the above limitations, energetic neutral atom (ENA) imaging technology has become an important tool for studying the ring current and geomagnetic storms. ENA imaging can remotely sense the distribution and intensity of the ring current from a global perspective by detecting neutral atoms generated after charge exchange between energetic ions in the ring current and neutral gas. This technology avoids the spatial limitations of ground observations and provides a new perspective for magnetospheric physics research. Existing studies have shown that there is a strong correlation between ENA flux and the Dst index, especially during the main phase and recovery phase of magnetic storms. However, existing methods mostly rely on theoretical models and empirical formulas to calculate this correlation, and it is difficult to capture complex non-linear relationships. Summary of the Invention
[0005] The purpose of the present invention is to overcome the defects of the prior art and propose a method and system for calculating the Dst index by fusing energetic neutral atom imaging data and machine learning.
[0006] In view of this, the present invention proposes a method for calculating the Dst index by fusing energetic neutral atom imaging data and machine learning, which is implemented based on a spaceborne energetic neutral atom imager and includes:
[0007] Receive the collected neutral atom image data of six energy bands and pre-process them in combination with the satellite position data;
[0008] Input the preprocessed neutral atom image data and satellite position data into the pre-established and trained index prediction model, and output the Dst index at the current moment;
[0009] The index prediction model combines a residual network and a multi-layer perceptron, and includes: an image feature extraction module, a feature fusion module, and a regression prediction module; wherein,
[0010] The image feature extraction module is used to extract image features and capture distribution information related to the Dst index in the neutral atom image;
[0011] The feature fusion module is used to perform multimodal fusion of image features and satellite position data;
[0012] The regression prediction module is used to perform nonlinear regression on the fused features and output the Dst index at the current moment.
[0013] Preferably, the six energy bands include: 10keV, 10-16keV, 16-27keV, 27-39keV, 39-50keV and 50-60keV, including neutral particles of all energies.
[0014] Preferably, the satellite position data includes: the longitude, latitude and distance from the center of the Earth of the satellite.
[0015] Preferably, the preprocessing includes: noise filtering and coordinate conversion and encoding; wherein,
[0016] Noise filtering: used to remove data from neutral atom image data whose flux does not meet the conditions;
[0017] Coordinate conversion and encoding: used to convert satellite positions from rectangular coordinates to spherical coordinates, and use sine-cosine encoding for magnetic latitude and longitude.
[0018] Preferably, the image feature extraction module adopts a ResNet34 network and outputs a 1×512 feature vector.
[0019] Preferably, the processing process of the feature fusion module includes:
[0020] The 6×20×20 tensor of image features and the 1×5 vector of satellite position data are concatenated to obtain a 1×517 fused feature.
[0021] Preferably, the regression prediction module uses an MLP model to perform nonlinear regression on the fusion features and outputs the Dst index at the current moment.
[0022] Preferably, the method further includes a training step of an exponential prediction model, including:
[0023] Collect the neutral atom image data of six energy segments in a specified year, the corresponding satellite position data, and the Dst index in the corresponding time period;
[0024] Preprocess the neutral atom image data in combination with the satellite position data; label the Dst index of each hour to the neutral atom image data in the corresponding time period to achieve time alignment;
[0025] Using the cross-validation method, divide the data into a training set and a test set, use the mean square error as the loss function, and the Adam optimization algorithm for training. The learning rate is dynamically adjusted to accelerate convergence until the training requirements are met, obtaining a trained exponential prediction model, and using the root mean square error and the correlation coefficient to evaluate the trained exponential prediction model.
[0026] Preferably, the method further includes: combining the Dst index at the current moment with other space weather observation indicators to provide multi-modal space weather monitoring and warning services; the other space weather observation indicators include: solar wind parameters and SYM-H index.
[0027] On the other hand, the present invention proposes a Dst index calculation system that fuses neutral atom imaging data and machine learning, implemented based on an on-board high-energy neutral atom imager, including:
[0028] A receiving and preprocessing module, configured to receive the collected neutral atom image data of six energy segments and preprocess them in combination with the satellite position data;
[0029] A prediction output module, configured to input the preprocessed neutral atom image data and satellite position data into a pre-established and trained exponential prediction model, and output the Dst index at the current moment;
[0030] The exponential prediction model combines a residual network and a multi-layer perceptron, including: an image feature extraction module, a feature fusion module, and a regression prediction module; where
[0031] The image feature extraction module is configured to extract image features and capture the distribution information related to the Dst index in the neutral atom image;
[0032] The feature fusion module is configured to perform multi-modal fusion of the image features and the satellite position data;
[0033] The regression prediction module is configured to perform non-linear regression on the fused features and output the Dst index at the current moment.
[0034] Compared with the prior art, the advantages of the present invention are as follows:
[0035] The present invention innovatively combines neutral atom imaging technology with deep learning methods, and for the first time realizes the confirmation of the Dst index based on ENA data, breaking through the limitations of spatial coverage and time resolution in traditional geomagnetic index calculation. By designing a model that combines a residual network (ResNet) and a multi-layer perceptron (MLP), fusing ENA images with satellite position information, accurately capturing the non-linear relationship between the Dst index and the ring current, significantly improving the calculation accuracy and real-time performance. The model achieves excellent performance with a correlation coefficient of 0.87 and a root mean square error of 18.96 nT on the test set, providing a new solution with a global perspective and high resolution for geomagnetic storm monitoring and space weather warning, and having broad scientific research and practical application value. Brief Description of the Drawings
[0036] Figure 1 is a flowchart of the method for calculating the Dst index by fusing neutral atom imaging data and machine learning in the present invention;
[0037] Figure 2 is the comparison result of predicting the Dst index and the actual value by using the method of the present invention. Detailed Embodiments
[0038] In recent years, with the rapid development of artificial intelligence and machine learning technologies, their applications in the field of space weather have gradually increased. For example, machine learning models have been successfully applied to model the relationship between solar wind parameters and the Dst index, and significant progress has been made. However, it is the first time to introduce machine learning technology into the correlation analysis between neutral atom imaging data and the Dst index. Neutral atom imaging data has the characteristics of high-dimensional and multi-modal, which poses challenges to traditional analysis methods, while deep learning technology provides the possibility to process such complex data.
[0039] The present invention combines high-energy neutral atom imaging data and machine learning technology to propose a new method for confirming the Dst index. This method uses deep learning models (such as multi-layer perceptrons and residual networks) to extract features from neutral atom images, and at the same time combines relevant information on satellite position and magnetic field perturbation to construct an accurate and efficient Dst index calculation model. This method breaks through the spatial and time limitations of traditional Dst index calculation, provides a new technical means for global geomagnetic storm monitoring and research, and significantly improves the accuracy and resolution of space weather forecasting.
[0040] The present invention proposes a new method based on neutral atom imaging data, which models the relationship between high-energy neutral atom flux and the Dst index through machine learning technology to achieve accurate confirmation and prediction of the Dst index. This method can greatly improve the accuracy and efficiency of geomagnetic storm monitoring and space weather forecasting.
[0041] 1. Data acquisition and processing
[0042] · Use the High Energy Neutral Atom Imager (HENA) of a space exploration satellite (such as the IMAGE satellite) to collect neutral atom observation data from 2000 to 2005. The data includes neutral atom images in six energy bands, satellite position data (longitude, latitude, and geocentric distance), and the Dst index at the corresponding time. The six energy bands are for the IMAGE satellite data used in the calculation process. According to the instrument design, neutral atom data in six energy bands are provided, including <10keV, 10 - 16keV, 16 - 27keV, 27 - 39keV, 39 - 50keV, 50 - 60keV. The main purpose is to include all detected particles. If different satellite data are applied, the energy division varies according to the instrument design.
[0043] · Preprocess the observation data, including denoising, coordinate transformation, and data alignment. Convert the satellite position to spherical coordinates and perform sine - cosine encoding on the angular information to enhance the model's recognition ability of position features.
[0044] 2. Machine learning model construction
[0045] · Construct a deep - learning model that combines a Multi - Layer Perceptron (MLP) and a Residual Network (ResNet).
[0046] · The model input includes neutral atom image data in six energy bands (each band is a 20×20 pixel matrix) and satellite position features; image features are extracted by ResNet, and position features are fused with image features in the fully - connected layer.
[0047] · The model output is the Dst index at the current moment.
[0048] 3. Model training and validation
[0049] · Use the cross - validation method to divide the data into a training set and a test set to ensure that the model has good generalization ability.
[0050] · Use the mean square error (MSE) as the loss function, and the correlation coefficient (R) and root mean square error (RMSE) as performance evaluation indicators.
[0051] · After optimization, the RMSE of the model on the test set reaches 18.96 nT, and the correlation coefficient is 0.87, indicating that it can accurately predict the Dst index.
[0052] 4. Model application and optimization
[0053] · The method of the present invention can process neutral atom imaging data in real time for Dst index confirmation during space weather events (such as geomagnetic storms).
[0054] It can be further extended to the calculation of geomagnetic indices with higher time resolution (such as SYM-H index), improving space weather forecasting capabilities.
[0055] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.
[0056] Embodiment 1
[0057] As Figure 1 shown, Embodiment 1 of the present invention proposes a method for calculating Dst index by fusing neutral atom imaging data and machine learning, including the following steps:
[0058] 1 Data collection and preprocessing
[0059] 1.1 Data collection
[0060] · Use a satellite equipped with a high-energy neutral atom imager (HENA) (such as the IMAGE satellite) to collect observation data from 2000 to 2005.
[0061] · The collected content includes:
[0062] ENA image data in six energy segments (each image has a resolution of 20×20 pixels);
[0063] Satellite position data (magnetic latitude, longitude, and geocentric distance);
[0064] The Dst index for the corresponding time period, used as the label for model supervised learning.
[0065] 1.2 Data preprocessing
[0066] · Noise filtering: Eliminate data with extremely high or low neutral atom flux, and set filtering conditions (such as total flux > 40, geocentric distance > 3 Earth radii).
[0067] · Coordinate transformation and encoding: Convert the satellite position from Cartesian coordinates to spherical coordinates, and use sine-cosine encoding for magnetic latitude and longitude to enhance the model's ability to process periodic data.
[0068] · Time alignment: Label the hourly Dst index to the ENA data for the corresponding time period to ensure the consistency of input and output data.
[0069] 2 Model design and training
[0070] 2.1 Model architecture
[0071] The present invention designs a deep learning model combining a Residual Network (ResNet) and a Multi-Layer Perceptron (MLP), specifically including the following modules:
[0072] · Image Feature Extraction Module
[0073] Input: ENA images of six energy segments (superimposed as a 6×20×20 tensor).
[0074] Use the ResNet34 network to extract features from the ENA images, and output a feature vector of size 1×512 to capture the distribution information related to the Dst index in the neutral atom image.
[0075] · Feature Fusion Module
[0076] Concatenate the image features extracted by ResNet with the satellite position information (a 1×5 vector) into a 1×517 input feature vector for multi-modal data fusion.
[0077] · Regression Prediction Module
[0078] Use the MLP model to perform non-linear regression on the fused features and output the Dst index at the current moment.
[0079] 2.2 Model Training
[0080] · Use the cross-validation method to divide the data into a training set (80%) and a test set (20%).
[0081] · And perform five-fold cross-validation in the training set.
[0082] · Adopt the mean squared error (MSE) as the loss function and the Adam optimization algorithm for training, and dynamically adjust the learning rate to accelerate convergence.
[0083] · Evaluate the model performance through the root mean squared error (RMSE) and the correlation coefficient (R).
[0084] 3 Model Validation and Performance Evaluation
[0085] 3.1 Validation Set Testing
[0086] · The model achieved good performance on the test set, with the correlation coefficient (R) reaching 0.87 and the root mean squared error (RMSE) being 18.96 nT.
[0087] · The results show that the model can accurately capture the changing trend of the Dst index, especially during magnetic storms with large negative Dst values.
[0088] 4 Real-Time Processing and Application
[0089] 4.1 Real-Time Data Processing
[0090] · Deploy the model to the ground data center, receive the ENA data sent by the satellite in real time, preprocess it and then input it into the model to quickly output the estimated Dst index value.
[0091] 4.2 System Integration
[0092] · Combine the output Dst index with other space weather observation indicators (such as solar wind parameters, SYM-H index) to provide multi-modal space weather monitoring and early warning services.
[0093] Example 2
[0094] Example 2 of the present invention proposes a Dst index calculation system that integrates neutral atom imaging data and machine learning. It is implemented based on an on-board high-energy neutral atom imager and adopts the method of Example 1. The system includes:
[0095] A receiving and preprocessing module, which is used to receive the neutral atom image data of six energy segments collected, and preprocess it in combination with satellite position data;
[0096] A prediction and output module, which is used to input the preprocessed neutral atom image data and satellite position data into a pre-established and trained index prediction model to output the Dst index at the current moment;
[0097] The index prediction model combines a residual network and a multi-layer perceptron, and includes: an image feature extraction module, a feature fusion module and a regression prediction module; among them,
[0098] The image feature extraction module is used to extract image features and capture the distribution information related to the Dst index in the neutral atom image;
[0099] The feature fusion module is used to perform multi-modal fusion of the image features and satellite position data;
[0100] The regression prediction module is used to perform non-linear regression on the fused features and output the Dst index at the current moment.
[0101] As Figure 2 shown is the comparison result of predicting the Dst index with the actual value using the method of the present invention.
[0102] The present invention proposes a new method based on neutral atom imaging data. By using machine learning technology to model the relationship between high-energy neutral atom flux and Dst index, accurate confirmation and prediction of the Dst index are realized. This method can greatly improve the accuracy and efficiency of geomagnetic storm monitoring and space weather forecasting.
[0103] Based on neutral atom imaging data, the present invention effectively realizes the estimation and confirmation of the Dst index through a deep learning model, providing a new technical means for geomagnetic storm monitoring and space weather forecasting.
[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the embodiments, those of ordinary skill in the art should understand that any modification or equivalent replacement of the technical solutions of the present invention does not depart from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.
Claims
1. A Dst index calculation method that integrates neutral atom imaging data and machine learning, based on a satellite-borne high-energy neutral atom imager, includes: Receive the collected neutral atom image data of six energy bands and perform pre-processing in combination with the satellite position data; Input the preprocessed neutral atom image data and satellite position data into the pre-established and trained index prediction model, and output the Dst index at the current moment; The index prediction model combines a residual network and a multi-layer perceptron, and includes: an image feature extraction module, a feature fusion module, and a regression prediction module; wherein, The image feature extraction module is used to extract image features and capture distribution information related to the Dst index in the neutral atom image; The feature fusion module is used to perform multimodal fusion of image features and satellite position data; The regression prediction module is used to perform nonlinear regression on the fused features and output the Dst index at the current moment; The image feature extraction module uses a ResNet34 network to output a 1×512 feature vector; The processing process of the feature fusion module includes: The 6×20×20 tensor of image features and the 1×5 vector of satellite position data are concatenated to obtain a 1×517 fused feature; The regression prediction module uses the MLP model to perform nonlinear regression on the fusion features and outputs the Dst index at the current moment; The method also includes a step of training an exponential prediction model, including: Collect neutral atom image data of six energy bands in a specified year, the corresponding satellite position data and the Dst index of the corresponding time period; The neutral atom image data is preprocessed in combination with the satellite position data; the hourly Dst index is marked on the neutral atom image data of the corresponding time period to achieve time alignment.
2. The Dst index calculation method integrating neutral atom imaging data and machine learning according to claim 1, characterized in that: The six energy bands include: 10keV, 10-16keV, 16-27keV, 27-39keV, 39-50keV and 50-60keV, including neutral particles of all energies.
3. The Dst index calculation method integrating neutral atom imaging data and machine learning according to claim 1, characterized in that: The satellite position data includes: the longitude, latitude and distance from the center of the earth of the satellite.
4. The Dst index calculation method integrating neutral atom imaging data and machine learning according to claim 2, characterized in that: The preprocessing includes: noise filtering and coordinate conversion and encoding; wherein, Noise filtering: used to remove data from neutral atom image data whose flux does not meet the conditions; Coordinate conversion and encoding: used to convert satellite positions from rectangular coordinates to spherical coordinates, and use sine-cosine encoding for magnetic latitude and longitude.
5. The Dst index calculation method integrating neutral atom imaging data and machine learning according to claim 1, characterized in that: The training step also includes: The cross-validation method is used to divide the data into training set and test set. The mean square error is used as the loss function. The Adam optimization algorithm is used for training. The learning rate is dynamically adjusted to accelerate convergence until the training requirements are met. The trained exponential prediction model is obtained, and the root mean square error and correlation coefficient are used to evaluate the trained exponential prediction model.
6. The Dst index calculation method integrating neutral atom imaging data and machine learning according to claim 1, characterized in that: The method also includes: combining the Dst index at the current moment with other space weather observation indicators to provide multi-modal space weather monitoring and early warning services; the other space weather observation indicators include: solar wind parameters and SYM-H index.
7. A system based on the Dst index calculation method of integrating neutral atom imaging data and machine learning according to claim 1, implemented based on a satellite-borne high-energy neutral atom imager, comprising: A receiving preprocessing module is used to receive the collected neutral atom image data of six energy bands and perform preprocessing in combination with the satellite position data; and The prediction output module is used to input the pre-processed neutral atom image data and satellite position data into the pre-established and trained index prediction model, and output the Dst index at the current moment; The index prediction model combines a residual network and a multi-layer perceptron, and includes: an image feature extraction module, a feature fusion module, and a regression prediction module; wherein, The image feature extraction module is used to extract image features and capture distribution information related to the Dst index in the neutral atom image; The feature fusion module is used to perform multimodal fusion of image features and satellite position data; The regression prediction module is used to perform nonlinear regression on the fused features and output the Dst index at the current moment.