Geomagnetic storm forecasting method and system based on multi-source coronal continuous observation image fusion
By fusing CME and coronal hole features through a multi-source adaptive 3D convolutional neural network, the uncertainty and insufficient accuracy of existing geomagnetic storm forecasting methods are solved, and efficient and accurate geomagnetic storm forecasting is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-12
- Publication Date
- 2026-04-14
AI Technical Summary
Existing geomagnetic storm forecasting methods rely on human experience and limited parameters, making it difficult to accurately predict the synergistic or competitive effects of CME and coronal holes. Furthermore, the lack of deep fusion of multi-source coronal observation images leads to forecast uncertainty and insufficient accuracy.
A multi-source adaptive 3D convolutional neural network (MSA-3DCNN) and a cross-source feature alignment mechanism are used to extract spatiotemporal features from multi-source corona observation images. CME and coronal hole features are fused through a deep learning model and combined with a multi-task learning framework for geomagnetic storm forecasting.
It has achieved automation, accuracy, and comprehensiveness in geomagnetic storm forecasting, improved forecasting efficiency and accuracy, met real-time operational needs, and is directly applicable to space weather operational platforms.
Smart Images
Figure CN121860937A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of space weather monitoring and forecasting technology, specifically to a geomagnetic storm forecasting method and system based on the fusion of multi-source continuous corona observation images. Background Technology
[0002] Geomagnetic storms are violent, global disturbances in Earth's magnetosphere, primarily caused by massive energy injections from the Sun lasting from several hours to several days. Their two main driving forces are: 1) coronal mass ejection (CME) events, particularly interplanetary CMEs carrying southward magnetic fields that reconnect with Earth's magnetosphere; and 2) the sustained impact of coronal interaction regions (CIRs) generated by high-speed coronal hole flows on Earth's magnetosphere. Strong geomagnetic storms can severely threaten the on-orbit safety of spacecraft, disrupt radio communications and navigation systems, and even paralyze ground-based power grids.
[0003] Existing geomagnetic storm prediction methods mainly rely on post-hoc identification of solar activity and empirical model extrapolation: 1. Empirical models based on CME parameters. When a CME is observed by a coronagraph (such as SOHO / LASCO), forecasters or automated systems measure its sky-plane velocity, angular width, source region location, and other parameters, combining these with empirical models to predict its arrival time at Earth and the intensity of the geomagnetic storm it may trigger (usually measured by the Dst or Kp index). These methods heavily rely on accurate measurements of the initial CME parameters and struggle to precisely determine the direction of its internal magnetic field (especially the crucial southward component), leading to significant forecast uncertainty.
[0004] 2. Statistical Relationships Based on Coronal Hole Characteristics. This approach identifies coronal holes by analyzing extreme ultraviolet (EUV) images, measuring their area, location, and other characteristics, and establishing empirical relationships between these characteristics and subsequent geomagnetic activity levels based on historical statistical data. While this method has some predictive capability for recurring geomagnetic storms, it cannot accurately predict non-recurring mega-magnetic storms caused by coronal magnetic fields (CMEs), and its prediction accuracy for the timing and intensity of geomagnetic storms is limited.
[0005] 3. Numerical models. Models such as the WSA-Enlil model can simulate the propagation of solar wind parameters in interplanetary space. However, these models are computationally expensive, and their accuracy is limited by initial conditions and model physics, making them difficult to meet the needs of real-time operational forecasting.
[0006] In summary, existing technologies have the following shortcomings: First, the forecast information of CME and coronal holes is not effectively integrated, and they are often processed separately, ignoring their synergistic or competitive effects; second, they heavily rely on limited manually extracted parameters (such as velocity and area), failing to fully utilize the rich spatiotemporal evolution information contained in the original images; and third, the forecast models are mostly linear or empirical models, making it difficult to capture the complex nonlinear relationships in the solar wind-magnetospheric coupling system. Therefore, it is urgent to develop a new method that can automatically, efficiently, and deeply integrate multi-source coronal observation images for accurate geomagnetic storm forecasting. Summary of the Invention
[0007] To achieve the objective of this invention, this application provides a geomagnetic storm prediction method based on the fusion of continuous multi-source corona observation images, comprising: Step S1: Acquire continuous time series of coronal mass ejection and coronal hole observation images; Step S2: Using the first deep learning model, extract the dynamic feature vector of coronal mass ejection from the sequence of observed images; Step S3: Extract coronal hole evolution feature vectors from the coronal hole observation image sequence using a second deep learning model; Step S4: Fuse the dynamic feature vector of the coronal mass ejection and the feature vector of the coronal hole evolution to obtain a comprehensive feature vector; Step S5: Input the comprehensive feature vector into the geomagnetic storm forecast model to obtain the geomagnetic storm forecast results for a specified time period in the future; the geomagnetic storm forecast results include the probability of geomagnetic storm occurrence or the predicted value of geomagnetic index determined based on the geomagnetic index threshold.
[0008] In one specific embodiment, step S1 includes: Step S11: Acquire a sequence of coronal mass ejection (CME) observation images and a sequence of coronal hole observation images from at least two different observation satellites or instruments; Step S12: Preprocess the acquired image sequence, including image alignment, noise reduction, pixel value normalization, and region of interest extraction.
[0009] In one specific embodiment, in step S2, the first deep learning model is an improved multi-source adaptive 3D convolutional neural network used to extract spatiotemporal features from the coronal mass ejection observation image sequence; the multi-source adaptive 3D convolutional neural network processes multi-source data through a multi-source adaptive convolutional kernel mechanism, and the convolutional kernel weights are obtained by modulating the shared basic convolutional kernel through the source feature embedding vector according to the following formula: in, For the first Adaptive convolution kernels for each data source To share the basic convolutional kernel, For the modulation matrix, For the first Embedding features of a data source This represents element-wise multiplication. These are learnable parameters.
[0010] In one specific embodiment, in step S2, the first deep learning model further includes a cross-source feature alignment and fusion mechanism to fuse feature maps from different data sources according to the following formula: in, According to the source Interpolation alignment operations performed on the spatial-temporal resolution parameters. These are adaptive weights computed via an attention mechanism, where K represents the number of feature maps from different data sources. Indicates the spatial-temporal resolution parameter. This represents the corona feature map from the k-th data source.
[0011] In one specific embodiment, in step S3, the second deep learning model is a multi-source adaptive three-dimensional convolutional neural network, and when training the second deep learning model to extract coronal hole evolution features, a morphology-aware loss function is introduced, determined according to the following formula: in, This is a morphological feature extraction function. For the first reconstructed by the decoder Coronal hole image, Indicates the first observation obtained Image of a coronal hole.
[0012] In one specific embodiment, step S4 includes: Step S41: Determine the attention weight of the coronal mass ejection dynamic features on the coronal hole evolution features, and obtain the modulated coronal hole evolution feature vector; Step S42: Determine the attention weight of coronal hole evolution features on coronal mass ejection dynamic features, and obtain the modulated coronal mass ejection dynamic feature vector. Step S43: The coronal hole evolution feature vector and the coronal mass ejection dynamic feature vector are concatenated, and a comprehensive feature vector is obtained through a multilayer perceptron.
[0013] In one specific embodiment, in step S5, the geomagnetic storm forecasting model is trained using a multi-task learning framework, and the total loss function of the geomagnetic storm forecasting model is determined by combining classification loss, regression loss, morphological perception loss, and physical constraint loss.
[0014] In one specific embodiment, the geomagnetic storm forecast result in step S5 includes: Classification results: The probability of geomagnetic storms of different levels occurring within the next 1 to 3 days is predicted; the levels are classified according to thresholds of the Dst or Kp index; and / or, Regression results: Predict the Dst index or Kp index value at a specific future time point.
[0015] To achieve the same inventive objective, this invention also provides a geomagnetic storm forecasting system based on the fusion of multi-source continuous coronal observation images, comprising: Image acquisition module: used to acquire continuous time series of observation images of coronal mass ejections and coronal holes; CME Feature Extraction Module: Used to extract dynamic feature vectors of coronal mass ejections from a sequence of observed images of coronal mass ejections using a first deep learning model; Coronal hole feature extraction module: used to extract coronal hole evolution feature vectors from coronal hole observation image sequences using a second deep learning model; Feature fusion module: used to fuse the dynamic feature vector of the coronal mass ejection and the feature vector of the coronal hole evolution to obtain a comprehensive feature vector; Result acquisition module: used to input the comprehensive feature vector into the geomagnetic storm forecast model to obtain the geomagnetic storm forecast results for a specified time period in the future; the geomagnetic storm forecast results include the probability of geomagnetic storm occurrence or the predicted value of geomagnetic index determined based on the geomagnetic index threshold.
[0016] In one specific embodiment, in the CME feature extraction module, the first deep learning model is an improved multi-source adaptive 3D convolutional neural network, used to extract spatiotemporal features from the coronal mass ejection observation image sequence; the multi-source adaptive 3D convolutional neural network processes multi-source data through a multi-source adaptive convolutional kernel mechanism, and the convolutional kernel weights are obtained by modulating a shared basic convolutional kernel through a source feature embedding vector according to the following formula: in, For the first Adaptive convolution kernels for each data source To share the basic convolutional kernel, For the modulation matrix, For the first Embedding features of a data source This represents element-wise multiplication. These are learnable parameters.
[0017] The beneficial effects of the above technical solution are as follows: (1) End-to-end automation of the forecast process improves efficiency and objectivity. This invention realizes full-process automation from input of raw multi-source images to output of geomagnetic storm forecast products, overcomes the heavy reliance of existing methods on manual experience interpretation and parameter extraction, significantly improves forecast efficiency, and ensures the consistency and objectivity of forecast results.
[0018] (2) Deep fusion of multi-source heterogeneous image information enhances forecast comprehensiveness. By introducing an improved multi-source adaptive 3D convolutional network (MSA-3DCNN) and a cross-source feature alignment mechanism, this invention achieves for the first time effective fusion of CME and coronal hole images from different satellites, instruments, and resolutions at the feature level. The model can simultaneously capture the evolutionary information of transient CME events and persistent coronal hole structures, and understand their synergistic or competitive effects, thereby providing a more comprehensive assessment of the driving sources of geomagnetic storms.
[0019] (3) Enhanced spatiotemporal dynamic feature extraction capability, enabling the discovery of deeper predictive factors. The network structures such as MSA-3DCNN adopted can directly and automatically learn the propagation dynamics features of CME (such as three-dimensional velocity field and expansion acceleration) and the morphological evolution laws of coronal holes (such as area change rate and boundary stability) from image sequences, capturing subtle spatiotemporal patterns that cannot be described by traditional artificial parameters, thereby discovering deeper features with greater predictive value.
[0020] (4) The model exhibits enhanced physical interpretability and generalization ability. By introducing a bidirectional cross-modal attention mechanism, the model can dynamically learn the relative importance of CME features and coronal hole features in the forecast, thereby enhancing its understanding of complex solar activity scenarios. Simultaneously, the introduction of morphological and physical relationship constraints into the loss function guides the model's learning process to conform to known physical laws, which not only improves forecast accuracy, especially for strong geomagnetic storms, but also enhances the model's interpretability and generalization ability.
[0021] (5) It has great potential for business applications and meets the needs of real-time forecasting. The method has a clear modular design. Once the model is trained, the computational efficiency in the online prediction stage is high, and it can quickly process real-time data streams, meeting the strict timeliness requirements of space weather business forecasting. It has the potential to be directly applied to business platforms. Attached Figure Description
[0022] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0023] Figure 1 A flowchart illustrating a geomagnetic storm prediction method based on the fusion of continuous multi-source corona observation images, provided as an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a geomagnetic storm forecasting system based on the fusion of continuous observation images of the multi-source corona, as provided in one embodiment of the present invention. Detailed Implementation
[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0025] Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar symbols denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the invention, and should not be construed as limiting the invention.
[0026] Example 1 One embodiment of the present invention provides a geomagnetic storm prediction method based on the fusion of multi-source continuous coronal observation images, referring to... Figure 1 As shown, it includes: Step S1: Acquire continuous time series of coronal mass ejection and coronal hole observation images; In a specific embodiment of the present invention, step S1 includes: Step S11: Acquire a sequence of coronal mass ejection (CME) observation images and a sequence of coronal hole observation images from at least two different observation satellites or instruments; Step S12: Preprocess the acquired image sequence, including image alignment, noise reduction, pixel value normalization, and region of interest extraction.
[0027] Specifically, continuous time-series corona observation image data are acquired from solar observation satellites (such as SDO, SOHO, and STEREO). The data includes: Coronal mass ejection (CME) image sequence: mainly from continuous observation images from coronagraphs (such as LASCO / C2, C3), used to capture the burst and propagation process of CMEs.
[0028] Coronal hole (CH) image sequence: mainly from continuous observation images from extreme ultraviolet telescopes (such as SDO / AIA193Å), used to monitor changes in the area, shape and location of coronal holes.
[0029] The acquired images are preprocessed, including alignment, noise reduction, normalization, and region of interest (ROI) extraction.
[0030] Download the C2 and C3 band coronagraph image sequence with a temporal resolution of 30 minutes within the most recent 24 hours from the LASCO instrument database on the SOHO satellite. Download the 193 Å extreme ultraviolet image sequence for the same time period from the AIA instrument database on the SDO satellite. All images were preprocessed, including image alignment based on solar center coordinates, median filtering for noise reduction, pixel value normalization to the [0,1] interval, and uniform cropping of a 512x512 pixel region of interest centered on the heliocenter.
[0031] Step S2: Using the first deep learning model, extract the dynamic feature vector of coronal mass ejection from the sequence of observed images; In a specific embodiment of the present invention, in step S2, the first deep learning model is an improved multi-source adaptive three-dimensional convolutional neural network, used to extract spatiotemporal features from the coronal mass ejection observation image sequence; the multi-source adaptive three-dimensional convolutional neural network processes multi-source data through a multi-source adaptive convolutional kernel mechanism, and the convolutional kernel weights are obtained by modulating the shared basic convolutional kernel through the source feature embedding vector according to the following formula: in, For the first Adaptive convolution kernels for each data source To share the basic convolutional kernel, For the modulation matrix, For the first Embedding features of a data source This represents element-wise multiplication. These are learnable parameters.
[0032] In a specific embodiment of the present invention, in step S2, the first deep learning model further includes a cross-source feature alignment and fusion mechanism to fuse feature maps from different data sources according to the following formula: in, According to the source Interpolation alignment operations performed on the spatial-temporal resolution parameters. These are adaptive weights computed via an attention mechanism, where K represents the number of feature maps from different data sources. Indicates the spatial-temporal resolution parameter. This represents the corona feature map from the k-th data source.
[0033] Image sequences from LASCO C2 and C3 are used as two data sources and input into the MSA-3DCNN network. The network first embeds vectors based on the characteristics of the two data sources (e.g., embedding e_c2 at C2 resolution and e_c3 at C3 resolution), and then modulates a shared basic convolutional kernel using a modulation formula to generate source-specific convolutional kernels. After extracting features separately, a cross-source attention fusion formula is used to obtain the fused CME dynamic feature vector f_CME (dimension 1024). The AIA 193Å image sequence is input into another structurally similar MSA-3DCNN branch (which can share some of the underlying weights with the CME branch) to extract the coronal hole evolution feature vector f_CH (dimension 512). When training this branch, an additional morphological awareness loss L_shape is introduced to constrain the image reconstructed by the decoder to maintain consistency with the original image in morphological features (such as area and roundness).
[0034] f_CME and f_CH are fused using a bidirectional cross-modal attention mechanism. First, the attention weight β_{CME→CH} for CME→CH is calculated to obtain the enhanced coronal hole feature f_CH'; similarly, the attention weight for CH→CME is calculated to obtain f_CME'. Finally, the two are concatenated and passed through an MLP to output the final comprehensive feature vector f_fused (dimension 1536).
[0035] Step S3: Extract coronal hole evolution feature vectors from the coronal hole observation image sequence using a second deep learning model; In a specific embodiment of the present invention, in step S3, the second deep learning model is a multi-source adaptive three-dimensional convolutional neural network, and when training the second deep learning model to extract coronal hole evolution features, a morphology-aware loss function is introduced, which is determined according to the following formula: in, For morphological feature extraction functions, For the first reconstructed by the decoder Coronal hole image, Indicates the first observation obtained Image of a coronal hole.
[0036] Step S4: Fuse the dynamic feature vector of the coronal mass ejection and the feature vector of the coronal hole evolution to obtain a comprehensive feature vector; In a specific embodiment of the present invention, step S4 includes: Step S41: Determine the attention weight of the coronal mass ejection dynamic features on the coronal hole evolution features, and obtain the modulated coronal hole evolution feature vector; Step S42: Determine the attention weight of coronal hole evolution features on coronal mass ejection dynamic features, and obtain the modulated coronal mass ejection dynamic feature vector. Step S43: The coronal hole evolution feature vector and the coronal mass ejection dynamic feature vector are concatenated, and a comprehensive feature vector is obtained through a multilayer perceptron.
[0037] Step S5: Input the comprehensive feature vector into the geomagnetic storm forecast model to obtain the geomagnetic storm forecast results for a specified time period in the future; the geomagnetic storm forecast results include the probability of geomagnetic storm occurrence or the predicted value of geomagnetic index determined based on the geomagnetic index threshold; In a specific embodiment of the present invention, in step S5, the geomagnetic storm forecasting model is trained using a multi-task learning framework, and the total loss function of the geomagnetic storm forecasting model is determined by combining classification loss, regression loss, morphological perception loss and physical constraint loss.
[0038] In a specific embodiment of the present invention, the geomagnetic storm forecast result in step S5 includes: Classification results: The probability of geomagnetic storms of different levels occurring within the next 1 to 3 days is predicted; the levels are classified according to thresholds of the Dst or Kp index; and / or, Regression results: Predict the Dst index or Kp index value at a specific future time point.
[0039] The world's largest Kp index for the next 24 hours after obtaining the corresponding time period from the geomagnetic index data center.
[0040] Define the label: If the maximum Kp ≥ 6, the label is "Geomagnetic storm" (Category 1); otherwise, it is "No geomagnetic storm" (Category 0).
[0041] Associate the f_fused corresponding to each image sequence sample with the binary classification label to construct the training dataset.
[0042] Construct a multi-task learning model. The input layer has 1536 nodes. The model has two output heads: Classification Header: After passing through two fully connected layers (768 and 256 nodes), the Softmax function is used to output the probabilities of the two classes.
[0043] Regression Head: After passing through two fully connected layers (768 and 256 nodes), it outputs a scalar that predicts the maximum Kp exponent value for the next 24 hours.
[0044] Loss function: The total loss is a weighted sum: L_total = λ1 * L_cls + λ2 * L_reg + λ3 * L_shape + λ4 * L_phy. Where L_phy is the physical constraint loss, for example, the constrained prediction Kp value is positively correlated with the CME velocity v_CME estimated from the image. We set λ1=1.0, λ2=0.5, λ3=0.2, λ4=0.1.
[0045] The model is trained using historical data from the past 10 years (e.g., 2012-2022) as the training set and employing the Adam optimizer.
[0046] A geomagnetic storm warning is issued based on the geomagnetic storm forecast results. The latest SOHO and SDO image data are acquired in real time, and steps S1 to S2 are repeated to obtain the current f_fused value. f_fused is then input into a pre-trained model. The model outputs the probability P of a "major geomagnetic storm". If P exceeds a preset threshold (e.g., 0.7), a warning message stating "a major geomagnetic storm may occur within the next 24 hours (Kp≥6)" is automatically generated and issued.
[0047] Example 2 One embodiment of the present invention provides a geomagnetic storm prediction system based on the fusion of multi-source continuous coronal observation images, referring to... Figure 2 As shown, it includes: Image acquisition module 10: used to acquire continuous time series observation images of coronal mass ejections and coronal holes; CME Feature Extraction Module 20: Used to extract the dynamic feature vector of the coronal mass ejection from the observation image sequence of the coronal mass ejection using a first deep learning model; Coronal hole feature extraction module 30: used to extract coronal hole evolution feature vectors from the coronal hole observation image sequence using a second deep learning model; Feature fusion module 40: used to fuse the dynamic feature vector of the coronal mass ejection and the feature vector of the coronal hole evolution to obtain a comprehensive feature vector; Result acquisition module 50: used to input the comprehensive feature vector into the geomagnetic storm forecast model to obtain the geomagnetic storm forecast results for a specified time period in the future; the geomagnetic storm forecast results are based on the geomagnetic index threshold to determine the probability of geomagnetic storm occurrence or the predicted value of the geomagnetic index; Alarm output module 60: Used to issue geomagnetic storm alarms based on the geomagnetic storm forecast results.
[0048] In a specific embodiment of the present invention, in the CME feature extraction module 20, the first deep learning model is an improved multi-source adaptive three-dimensional convolutional neural network to extract spatiotemporal features from image sequences; the multi-source adaptive three-dimensional convolutional neural network processes multi-source data through a source adaptive convolutional kernel mechanism, and the convolutional kernel weights are obtained by modulating the shared basic convolutional kernels through the source feature embedding vector according to the following formula: in, For the first Adaptive convolution kernels for each data source To share the basic convolutional kernel, For the modulation matrix, For the first Embedding features of a data source This represents element-wise multiplication. These are learnable parameters.
[0049] Data source: Coronagraph images using both SOHO / LASCO (Sun-Earth connection view) and STEREO-A / COR (lateral view).
[0050] Feature embedding: Define a feature embedding vector e_k for each data source, which contains information such as: spatial resolution, the heliocentric latitude and longitude of the satellite (to reflect the difference in viewpoint), and noise level estimate.
[0051] Feature Alignment and Fusion: The feature maps Y_LASCO and Y_STEREO of LASCO and STEREO differ significantly in spatial scale and viewpoint. Using a learnable interpolation function, Resize, they are aligned to a unified spatial-temporal grid based on their resolution parameter θ_k. Then, a cross-source attention mechanism is used to calculate the weights α_LASCO and α_STEREO, and the weighted sum is obtained to obtain the fused CME features. This allows the model to comprehensively utilize frontal and side viewpoint information, more accurately determining the 3D propagation direction of the CME and whether it is facing the Earth.
[0052] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
[0053] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the functions specified in one or more boxes. Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the invention. Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.
[0054] The methods and apparatus provided by the present invention have been described in detail above. Specific examples have been used 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 methods and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
[0055] In the description of this specification, references to terms such as "an embodiment," "some embodiments," "example," "specific example," or "a specific embodiment" or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A geomagnetic storm forecasting method based on the fusion of continuous multi-source corona observation images, characterized in that, include: Step S1: Acquire continuous time series of coronal mass ejection and coronal hole observation images; Step S2: Using the first deep learning model, extract the dynamic feature vector of coronal mass ejection from the sequence of observed images; Step S3: Extract coronal hole evolution feature vectors from the coronal hole observation image sequence using a second deep learning model; Step S4: Fuse the dynamic feature vector of the coronal mass ejection and the feature vector of the coronal hole evolution to obtain a comprehensive feature vector; Step S5: Input the comprehensive feature vector into the geomagnetic storm forecast model to obtain the geomagnetic storm forecast results for a specified time period in the future; the geomagnetic storm forecast results include the probability of geomagnetic storm occurrence or the predicted value of geomagnetic index determined based on the geomagnetic index threshold.
2. The geomagnetic storm forecasting method based on multi-source continuous coronal observation image fusion according to claim 1, characterized in that, Step S1 includes: Step S11: Acquire a sequence of coronal mass ejection (CME) observation images and a sequence of coronal hole observation images from at least two different observation satellites or instruments; Step S12: Preprocess the acquired image sequence, including image alignment, noise reduction, pixel value normalization, and region of interest extraction.
3. The geomagnetic storm forecasting method based on multi-source continuous coronal observation image fusion according to claim 1, characterized in that, In step S2, the first deep learning model is an improved multi-source adaptive three-dimensional convolutional neural network, used to extract spatiotemporal features from the coronal mass ejection observation image sequence. The multi-source adaptive 3D convolutional neural network processes multi-source data through a multi-source adaptive convolutional kernel mechanism. The convolutional kernel weights are obtained by modulating a shared basic convolutional kernel with source feature embedding vectors according to the following formula: in, For the first Adaptive convolution kernels for each data source To share the basic convolutional kernel, For the modulation matrix, For the first Embedding features of a data source This represents element-wise multiplication. These are learnable parameters.
4. The geomagnetic storm forecasting method based on multi-source continuous coronal observation image fusion according to claim 1, characterized in that, In step S2, the first deep learning model further includes a cross-source feature alignment and fusion mechanism to fuse feature maps from different data sources according to the following formula: in, According to the source Interpolation alignment operations performed on the spatial-temporal resolution parameters. These are adaptive weights computed via an attention mechanism, where K represents the number of feature maps from different data sources. Indicates the spatial-temporal resolution parameter. This represents the corona feature map from the k-th data source.
5. The geomagnetic storm forecasting method based on multi-source continuous coronal observation image fusion according to claim 1, characterized in that, In step S3, the second deep learning model is a multi-source adaptive 3D convolutional neural network, and when training the second deep learning model to extract coronal hole evolution features, a morphology-aware loss function is introduced, which is determined according to the following formula: in, This is a morphological feature extraction function. For the first reconstructed by the decoder Coronal hole image, Indicates the first observation obtained Image of a coronal hole.
6. The geomagnetic storm forecasting method based on multi-source continuous coronal observation image fusion according to claim 1, characterized in that, Step S4 includes: Step S41: Determine the attention weight of the coronal mass ejection dynamic features on the coronal hole evolution features, and obtain the modulated coronal hole evolution feature vector; Step S42: Determine the attention weight of coronal hole evolution features on coronal mass ejection dynamic features, and obtain the modulated coronal mass ejection dynamic feature vector. Step S43: The coronal hole evolution feature vector and the coronal mass ejection dynamic feature vector are concatenated, and a comprehensive feature vector is obtained through a multilayer perceptron.
7. The geomagnetic storm forecasting method based on multi-source continuous coronal observation image fusion according to claim 1, characterized in that, In step S5, the geomagnetic storm forecasting model is trained using a multi-task learning framework, and the total loss function of the geomagnetic storm forecasting model is determined by combining classification loss, regression loss, morphological perception loss, and physical constraint loss.
8. The geomagnetic storm forecasting method based on multi-source continuous coronal observation image fusion according to claim 1, characterized in that, The geomagnetic storm forecast results mentioned in step S5 include: Classification results: The probability of geomagnetic storms of different levels occurring within the next 1 to 3 days is predicted; the levels are classified according to thresholds of the Dst or Kp index; and / or, Regression results: Predict the Dst index or Kp index value at a specific future time point.
9. A geomagnetic storm forecasting system based on the fusion of continuous multi-source corona observation images, characterized in that, include: Image acquisition module: used to acquire continuous time series of observation images of coronal mass ejections and coronal holes; CME Feature Extraction Module: Used to extract dynamic feature vectors of coronal mass ejections from a sequence of observed images of coronal mass ejections using a first deep learning model; Coronal hole feature extraction module: used to extract coronal hole evolution feature vectors from coronal hole observation image sequences using a second deep learning model; Feature fusion module: used to fuse the dynamic feature vector of the coronal mass ejection and the feature vector of the coronal hole evolution to obtain a comprehensive feature vector; Result acquisition module: used to input the comprehensive feature vector into the geomagnetic storm forecast model to obtain the geomagnetic storm forecast results for a specified time period in the future; the geomagnetic storm forecast results include the probability of geomagnetic storm occurrence or the predicted value of geomagnetic index determined based on the geomagnetic index threshold.
10. The geomagnetic storm forecasting system based on multi-source continuous coronal observation image fusion according to claim 9, characterized in that, In the CME feature extraction module, the first deep learning model is an improved multi-source adaptive three-dimensional convolutional neural network, used to extract spatiotemporal features from the coronal mass ejection observation image sequence. The multi-source adaptive 3D convolutional neural network processes multi-source data through a multi-source adaptive convolutional kernel mechanism. The convolutional kernel weights are obtained by modulating a shared basic convolutional kernel with source feature embedding vectors according to the following formula: in, For the first Adaptive convolution kernels for each data source To share the basic convolutional kernel, For the modulation matrix, For the first Embedding features of a data source This represents element-wise multiplication. These are learnable parameters.