Prediction method and prediction device for coronal mass ejection event, and storage medium

Abstract

The invention discloses a prediction method and a prediction device for a coronal mass ejection event, and a storage medium. The method comprises the following steps: acquiring a coronal mass ejection CME physical parameter and a CME image; performing feature extraction on the CME physical parameters to obtain target parameter feature vectors; performing feature extraction on the CME image to obtain a target image feature vector; performing feature fusion on the target parameter feature vector and the target image feature vector based on an attention mechanism to obtain a fused feature vector; and predicting arrival time according to the fused feature vector. Therefore, according to the method, the attention mechanism is introduced, the target parameter feature vector obtained based on the CME physical parameters and the target image feature vector obtained based on the CME image are subjected to feature fusion, redundant information between the CME physical parameters and the CME image can be effectively deleted, and fusion of complementary information is realized; the CME arrival time can be quickly and accurately predicted.

Description

Technical Field

[0001] The present invention relates to the field of corona detection technology, and in particular to a method for predicting a coronal mass ejection event, a device for predicting a coronal mass ejection event, and a computer-readable storage medium. Background Art

[0002] Coronal Mass Ejection (CME) is a strong solar eruption phenomenon that can cause drastic changes in the Earth's near-Earth space environment and have a huge impact on normal human life. In order to effectively reduce or prevent the damage that CME may cause, it is urgent to predict the arrival time of CME on Earth in a fast and accurate way.

[0003] In order to accurately predict the arrival time of CME, the relevant technology uses empirical models, drag models, and magnetohydrodynamic models based on CME physical parameter data to predict the time when CME will arrive on Earth. However, this prediction method is largely limited by physical theories based on prior assumptions or expert knowledge, and fails to fully explore the information contained in the data itself, reducing the accuracy of the prediction. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems in the related art to a certain extent. To this end, the first object of the present invention is to propose a prediction method for coronal mass ejection events, which can effectively delete the redundant information between the CME physical parameters and the CME image by introducing an attention mechanism, and realize the fusion of complementary information, so as to quickly and accurately predict the arrival time of the CME.

[0005] A second object of the present invention is to provide a prediction device for coronal mass ejection events.

[0006] A third object of the present invention is to provide a computer-readable storage medium.

[0007] To achieve the above-mentioned purpose, a first embodiment of the present invention proposes a method for predicting a coronal mass ejection event, the method comprising: obtaining physical parameters and CME images of coronal mass ejections (CMEs); performing feature extraction on the physical parameters of the CMEs to obtain a target parameter feature vector; performing feature extraction on the CME images to obtain a target image feature vector; performing feature fusion on the target parameter feature vector and the target image feature vector based on an attention mechanism to obtain a fused feature vector; and predicting an arrival time based on the fused feature vector.

[0008] According to the prediction method of the coronal mass ejection event of the embodiment of the present invention, firstly, the physical parameters and CME images of the coronal mass ejection CME are obtained, and the feature extraction of the CME physical parameters is performed to obtain the target parameter feature vector, and the feature extraction of the CME image is performed to obtain the target image feature vector, and the target parameter feature vector and the target image feature vector are feature fused based on the attention mechanism to obtain the fused feature vector; and the arrival time is predicted based on the fused feature vector. Therefore, by introducing the attention mechanism, the method performs feature fusion on the target parameter feature vector obtained based on the CME physical parameters and the target image feature vector obtained based on the CME image, which can effectively delete the redundant information between the CME physical parameters and the CME image, realize the fusion of complementary information, and realize the rapid and accurate prediction of the CME arrival time.

[0009] In addition, the method for predicting a coronal mass ejection event according to the above embodiment of the present invention may also have the following additional technical features:

[0010] According to one embodiment of the present invention, feature extraction is performed on CME physical parameters to obtain a target parameter feature vector, including: feature mapping the CME physical parameters to form a plurality of first-layer feature vectors; dimensionality enhancement of the plurality of first-layer feature vectors to form a second-layer feature vector; nonlinear enhancement of the second-layer feature vector to form a third-layer feature vector; and addition and normalization of the plurality of first-layer feature vectors and the third-layer feature vector to form a target parameter feature vector.

[0011] According to one embodiment of the present invention, feature extraction is performed on a CME image to obtain a target image feature vector, including: using an improved ResNet-18 network to extract features from the CME image to form a target image feature vector.

[0012] According to one embodiment of the present invention, the improved ResNet-18 network includes: an input layer, a residual block, a skip connection, a flatten layer and an output layer.

[0013] According to one embodiment of the present invention, feature fusion is performed on a target parameter feature vector and a target image feature vector based on an attention mechanism, including: performing a self-attention mechanism operation on the target parameter feature vector and the target image feature vector to obtain an enhanced parameter feature vector and an enhanced image feature vector respectively; performing a cross-attention mechanism operation on the enhanced parameter feature vector and the enhanced image feature vector to obtain a fused feature vector.

[0014] According to one embodiment of the present invention, a self-attention mechanism operation is performed on a target parameter feature vector and a target image feature vector respectively to obtain an enhanced parameter feature vector and an enhanced image feature vector respectively, including: determining a first query vector, a first key vector and a first value vector; performing a matrix product of the first query vector and the first key vector to generate a first similarity matrix; applying a Softmax function to normalize the first similarity matrix to generate a first attention weight; applying the first attention weight to the first value vector and performing a weighted sum to obtain an enhanced parameter feature vector and an enhanced image feature vector.

[0015] According to one embodiment of the present invention, a cross-attention mechanism operation is performed on an enhanced parameter feature vector and an enhanced image feature vector to obtain a fused feature vector, including: determining a second query vector based on the enhanced parameter feature vector, and determining a second key vector and a second value vector based on the enhanced image feature vector; performing a matrix product between the second query vector and the second key vector to generate a second similarity matrix; applying a Softmax function to normalize the second similarity matrix to generate a second attention weight; applying the second attention weight to the second value vector and performing a weighted sum to obtain a fused feature vector.

[0016] According to one embodiment of the present invention, the method for predicting a coronal mass ejection event further includes: evaluating the predicted arrival time according to a mean absolute error and a root mean square error.

[0017] To achieve the above-mentioned purpose, the second embodiment of the present invention proposes a prediction device for coronal mass ejection events, which includes: a first acquisition module, used to acquire physical parameters and CME images of coronal mass ejections (CME); a second acquisition module, used to perform feature extraction on the CME physical parameters to obtain a target parameter feature vector; a third acquisition module, used to perform feature extraction on the CME image to obtain a target image feature vector; a fusion module, used to perform feature fusion on the target parameter feature vector and the target image feature vector based on an attention mechanism to obtain a fused feature vector; and a prediction module, used to predict the arrival time according to the fused feature vector.

[0018] According to the prediction device of the coronal mass ejection event of the embodiment of the present invention, the physical parameters and CME images of the coronal mass ejection CME are obtained by the first acquisition module, the feature extraction of the CME physical parameters is performed by the second acquisition module to obtain the target parameter feature vector, the feature extraction of the CME image is performed by the third acquisition module to obtain the target image feature vector, the feature fusion of the target parameter feature vector and the target image feature vector is performed by the fusion module based on the attention mechanism to obtain the fused feature vector, and the prediction module predicts the arrival time according to the fused feature vector. Therefore, the device can effectively delete the redundant information between the CME physical parameters and the CME image by introducing the attention mechanism to fuse the target parameter feature vector obtained based on the CME physical parameters and the target image feature vector obtained based on the CME image, and realize the fusion of complementary information, so as to realize the rapid and accurate prediction of the CME arrival time.

[0019] To achieve the above objectives, a third aspect of the present invention provides a computer-readable storage medium having stored thereon a prediction program for a coronal mass ejection event, which, when executed by a processor, implements the above-mentioned method for predicting a coronal mass ejection event.

[0020] According to the computer-readable storage medium of an embodiment of the present invention, the above-mentioned method for predicting a coronal mass ejection event is implemented when the prediction program for a coronal mass ejection event is executed by a processor. Based on the above-mentioned method for predicting a coronal mass ejection event, redundant information between CME physical parameters and CME images is effectively deleted, and complementary information is fused to achieve rapid and accurate prediction of the CME arrival time.

[0021] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of a method for predicting a coronal mass ejection event according to an embodiment of the present invention;

[0023] Figure 2 A schematic diagram of the overall framework of a prediction system model for coronal mass ejection events according to an embodiment of the present invention;

[0024] Figure 3 is a structural schematic diagram of a parameter feature extraction module according to an embodiment of the present invention;

[0025] Figure 4 A flowchart of a method for predicting a coronal mass ejection event according to a specific embodiment of the present invention;

[0026] Figure 5FIG. 4 is a connection diagram of a device for predicting coronal mass ejection events according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0028] The following describes a method for predicting a coronal mass ejection event, a device for predicting a coronal mass ejection event, and a computer-readable storage medium according to embodiments of the present invention with reference to the accompanying drawings.

[0029] As one of the strongest phenomena in the Sun-Earth system, coronal mass ejections may interfere with the Earth's space environment and cause disturbances in the geomagnetic field. Therefore, it is crucial to quickly and accurately predict the arrival time of CMEs on Earth to reduce their impact on the near-Earth space environment and human daily life. In order to accurately predict the arrival time of CMEs, the following two technical solutions are mainly used:

[0030] Solution 1: Calculate the physical parameters of CME by analyzing CME images. These physical parameters make full use of human prior knowledge and usually include the most critical attributes of CME. Then, the arrival time of CME is predicted based on the CME physical parameters. However, the physical parameters of CME are obtained based on the detection and tracking of CME. Researchers only calculate the pre-set physical parameters based on CME images, which introduces a certain degree of subjectivity. Therefore, the physical parameters of CME may not be comprehensive enough and may miss some critical but imperceptible image information, resulting in unsatisfactory prediction results.

[0031] Solution 2: CME image data is used as input, and features are directly extracted from CME images through convolutional neural networks to predict the arrival time of CME. Although this technical solution can fully obtain CME-related information, convolutional neural networks are easily affected by image noise, causing them to over-focus on unimportant background areas in the image, thereby reducing performance.

[0032] In view of the shortcomings of the CME arrival time prediction methods in the related arts, the present invention proposes a prediction method for coronal mass ejection events. The method is based on a target parameter feature vector determined by the CME physical parameters and a target image feature vector determined by the CME image. Then, by introducing an attention mechanism, the redundant information between the target parameter feature vector and the target image feature vector is effectively deleted, and the fusion of complementary information is achieved, thereby quickly and accurately predicting the CME arrival time.

[0033] Figure 1 The figure is a flow chart of a method for predicting a coronal mass ejection event according to an embodiment of the present invention.

[0034] like Figure 1 As shown, the method for predicting a coronal mass ejection event according to an embodiment of the present invention may include the following steps:

[0035] S1, obtain the physical parameters and CME images of coronal mass ejection CME;

[0036] S2, feature extraction of CME physical parameters to obtain target parameter feature vector;

[0037] S3, extracting features from the CME image to obtain a target image feature vector;

[0038] S4, based on the attention mechanism, feature fusion is performed on the target parameter feature vector and the target image feature vector to obtain a fused feature vector;

[0039] S5, predict the arrival time based on the fused feature vector.

[0040] Specifically, the prediction system model is as follows Figure 2 As shown, it includes a parameter feature extraction module 1, an image feature extraction module 2 and a feature fusion module 3.

[0041] The parameter feature extraction module 10 extracts features of the CME physical parameters to obtain a target parameter feature vector based on the CME physical parameters. The CME physical parameters are obtained by fitting the original CME observation image through professional knowledge, including parameters such as the initial velocity, acceleration and opening angle of the CME, which are used to reflect the motion characteristics of the CME.

[0042] The image feature extraction module 20 extracts features from the CME image to obtain a target image feature vector. Specifically, the target image feature vector can be automatically extracted directly from the CME image using an artificial intelligence method, and comprehensive CME-related information can be obtained, including details that may not be perceived by the human visual system.

[0043] The feature fusion module 30 performs feature fusion on the target parameter feature vector and the target image feature vector based on the attention mechanism to obtain a fused feature vector, which not only eliminates the redundancy between the CME physical parameter data and the CME image data, but also realizes the complementary fusion of information.

[0044] Then, the CME arrival time is predicted based on the fused feature vector to improve the prediction speed and accuracy of the CME arrival time. A preset relationship between the fused feature vector and the CME arrival time can be preset, such as a preset relationship expression, a preset mapping table, etc., without specific limitation. In the CME arrival time prediction process, the CME arrival time is predicted based on the determined fused feature vector and the preset relationship.

[0045] This embodiment extracts features from both CME physical parameters and CME images, and fuses these features by introducing an attention mechanism, which can simultaneously capture human prior knowledge and CME-related image information to maximize the use of all available data and achieve rapid and accurate prediction of CME arrival time.

[0046] In one embodiment of the present invention, feature extraction is performed on CME physical parameters to obtain a target parameter feature vector, including: feature mapping the CME physical parameters to form a plurality of first-layer feature vectors; dimensionality enhancement of the plurality of first-layer feature vectors to form a second-layer feature vector; nonlinear enhancement of the second-layer feature vector to form a third-layer feature vector; and addition and normalization of the plurality of first-layer feature vectors and the third-layer feature vector to form a target parameter feature vector.

[0047] Specifically, CME physical parameters cover key CME attribute information and human prior knowledge, and thus play a vital role in the prediction task of CME arrival time. In order to achieve automatic feature extraction and avoid manually specifying the function form, the parameter feature extraction module of this embodiment is used to extract the CME physical parameters based on the neural network. Figure 2 As shown, it includes four learnable neural layers. Assuming that there are 12 CME physical parameters, the specific feature extraction process is as follows:

[0048] In the first layer, the 12 CME physical parameters are mapped into 12 512-dimensional feature vectors to obtain multiple first-layer feature vectors.

[0049] In the second layer, the 512-dimensional feature vector is upgraded to 2048 dimensions to optimize the encoding of high-level features and obtain the second-layer feature vector. Since each feature has a unique physical meaning, direct fusion may lead to the loss of diversity between different physical parameters. To this end, this embodiment uses feature alignment to map multiple first-layer feature vectors to a high-dimensional feature space to promote feature fusion. In addition, selecting a high-dimensional rise can also enhance the expressive power of the model, better capture the complex patterns and relationships in the input data, and deeply study various feature combinations to obtain more complex feature representations.

[0050] In the third layer, the second layer feature vector is reduced from 2048 dimensions to 512 dimensions to obtain the third layer feature vector, aiming to reduce the number of parameters, enhance the nonlinear ability of the model, avoid overfitting of the model, and improve the training speed and generalization ability of the model.

[0051] In the fourth layer, the 512-dimensional first-layer feature vector output by the first layer is added to the 512-dimensional third-layer feature vector output by the third layer, and normalized to obtain the target parameter feature vector. This residual connection helps to deal with the problem of gradient disappearance and gradient explosion, making the model easier to train. In addition, normalizing each feature vector dimension helps ensure that the input has a similar distribution in different feature dimensions, further improving the convergence speed of the model and enhancing the robustness of the model to the input.

[0052] In one embodiment of the present invention, feature extraction is performed on a CME image to obtain a target image feature vector, including: using an improved ResNet-18 network to extract features from the CME image to form a target image feature vector.

[0053] In one embodiment of the present invention, the improved ResNet-18 network includes: an input layer, a residual block, a skip connection, a flatten layer and an output layer.

[0054] Specifically, with the rapid development of deep learning, convolutional neural networks have begun to be widely used in the field of CME arrival time prediction. Convolutional neural networks can directly extract features from CME images to achieve comprehensive acquisition of CME-related information, including details that are difficult to capture in CME physical parameters. In view of the excellent performance of convolutional neural networks, this embodiment uses the classic and efficient convolutional neural network model ResNet-18 as the basic network, and improves the ResNet-18 network to form a new convolutional neural network, namely the improved ResNet-18 network, to construct an image feature extraction module to extract the target image feature vector from the CME image.

[0055] In terms of the basic network, the improved ResNet-18 network includes an input layer conv1 and four convolutional layers conv2, conv3, conv4, and conv5, with input channels of 3, 64, 64, 128, and 256, respectively. In order to adapt to the feature extraction task of CME image data, we adjusted the original ResNet-18 network: the last pooling layer and fully connected layer of the original network were deleted, a flatten layer was added, and finally a 512-dimensional image feature vector was output.

[0056] In the process of feature extraction of CME images, the image feature extraction module receives the CME image as input and finally generates the target image feature vector through multiple convolution operations. The image feature extraction module is composed of a network layer composed of multiple convolution modules. Each network layer is responsible for detecting image features of a specific scale and adopts a shortcut operation to dynamically fuse the features of each convolution layer. By stacking network layers layer by layer and using shortcuts to fuse the features of different convolution layers, the target image feature vector generated by the image feature extraction module can fully and detailedly reflect the features of the CME image.

[0057] In one embodiment of the present invention, feature fusion is performed on the target parameter feature vector and the target image feature vector based on the attention mechanism, including: performing self-attention mechanism operations on the target parameter feature vector and the target image feature vector to obtain an enhanced parameter feature vector and an enhanced image feature vector respectively; performing cross-attention mechanism operations on the enhanced parameter feature vector and the enhanced image feature vector to obtain a fused feature vector.

[0058] That is to say, a self-attention mechanism operation is performed on the target parameter feature vector to obtain an enhanced parameter feature vector, and a self-attention mechanism operation is performed on the target image feature vector to obtain an enhanced image feature vector. Then, a cross-attention mechanism operation is performed on the enhanced parameter feature vector and the enhanced image feature vector to obtain a fused feature vector, thereby fully fusing the target parameter feature vector and the target image feature vector, so that while extracting the CME physical parameters and CME image features, the key and comprehensive information of the CME is captured to improve the prediction accuracy of the CME arrival time.

[0059] In one embodiment of the present invention, a self-attention mechanism operation is performed on a target parameter feature vector and a target image feature vector respectively to obtain an enhanced parameter feature vector and an enhanced image feature vector respectively, including: determining a first query vector, a first key vector and a first value vector; performing a matrix product of the first query vector and the first key vector to generate a first similarity matrix; applying a Softmax function to normalize the first similarity matrix to generate a first attention weight; applying the first attention weight to the first value vector and performing a weighted sum to obtain an enhanced parameter feature vector and an enhanced image feature vector.

[0060] Specifically, the calculation formula of the attention mechanism is as follows:

[0061]

[0062] Among them, Q is the query vector Query, which is used to represent the specific elements in the input data that need to be paid attention to; K is the key vector Key, which is used to represent the characteristics of the input data; V is the value vector Value, which is used to represent the specific value of each element in the input data; k Represents the dimension of the key vector. In order to optimize network efficiency, d is usually k Set to a relatively small value.

[0063] Taking the self-attention mechanism operation on the target parameter feature vector to obtain the enhanced parameter feature vector as an example, the specific operation is as follows:

[0064] First, after determining the first query vector Q1, the first key vector K1 and the first value vector V1 based on the target parameter feature vector, in order to simplify the computational complexity, the first key vector K1 is set to be equal to the first value vector V1, and a first similarity matrix is ​​obtained by performing matrix product of the first query vector Q1 and the first key vector K1. The first similarity matrix reflects the similarity between the elements in the first query vector Q1 and the elements in the input data.

[0065] Then, the first similarity matrix is ​​normalized by applying the Softmax function to generate the first attention weights. These attention weights reflect the relative importance of each element in the input data, giving higher weights to key elements and lower weights to less important elements, thereby improving the recognition of related features.

[0066] Finally, the first attention weight is applied to the first value vector V1 and weighted sum is performed to obtain an enhanced parameter feature vector, which realizes the fusion of the contribution of each element in the input data, thereby obtaining an enhanced feature vector, namely, the enhanced parameter feature vector.

[0067] Similarly, the self-attention mechanism is operated on the target image feature vector to obtain the enhanced image feature vector. The specific process is not described in detail.

[0068] In addition, in the process of obtaining the enhanced parameter feature vector based on the target parameter feature vector, the first query vector Q1, the first key vector K1 and the first value vector V1 are all equal to the target parameter feature vector, which helps the model focus on the key CME physical parameter components. In the process of obtaining the enhanced image feature vector based on the target image feature vector, the first query vector Q1, the first key vector K1 and the first value vector V1 are set as the target image feature vector of the CME image, which enables the convolutional neural network model to highlight the key image areas.

[0069] This embodiment achieves targeted enhancement of the features of CME physical parameters and CME images through the application of the self-attention mechanism, thereby better capturing the correlation between the two.

[0070] In one embodiment of the present invention, a cross-attention mechanism operation is performed on the enhanced parameter feature vector and the enhanced image feature vector to obtain a fused feature vector, including: determining a second query vector based on the enhanced parameter feature vector, and determining a second key vector and a second value vector based on the enhanced image feature vector; performing matrix product between the second query vector and the second key vector to generate a second similarity matrix; applying a Softmax function to normalize the second similarity matrix to generate a second attention weight; applying the second attention weight to the second value vector and performing weighted summation to obtain a fused feature vector.

[0071] That is, the second query vector Q2 is determined according to the enhanced parameter feature vector, the second key vector K2 and the second value vector V2 are determined according to the enhanced image feature vector, and then the second query vector Q2 is matrix-producted with the second value vector V2 to obtain the second similarity matrix, and the second similarity matrix is ​​normalized by the Softmax function to obtain the second attention weight, and the second attention weight obtained is determined based on the feature correlation between the CME physical parameters and the CME image, so that the model can focus on the CME-related area, thereby ignoring the irrelevant parts. Finally, the second attention weight is applied to the second value vector V2, and a weighted sum is performed to calculate the fused feature vector, which successfully highlights the significant features and omits unimportant details.

[0072] This embodiment adopts the cross-attention mechanism, introduces the prior knowledge of CME physical parameters, further optimizes the association between the features of CME physical parameters and the features of CME images, and maximizes the use of complete CME-related image information. Therefore, the prediction model based on the cross-attention mechanism can capture the key and comprehensive information of CME through the attention mechanism while extracting the features of CME physical parameters and CME images.

[0073] In one embodiment of the present invention, the method for predicting a coronal mass ejection event further includes: evaluating the predicted arrival time according to the mean absolute error and the root mean square error, so as to improve the accuracy of the time prediction.

[0074] As a specific embodiment of the present application, the prediction system model of the coronal mass ejection event is as follows: Figure 2As shown, it includes parameter feature extraction module 1, image feature extraction module 2 and feature fusion module 3. Among them, parameter feature extraction module 1 has good nonlinear transformation ability to extract features of CME physical parameters and can effectively extract key CME information; image feature extraction module 2 is designed based on the widely used ResNet-18 model with excellent performance as the basic network to extract features of CME images; feature fusion module 3 uses attention mechanism to complete the feature fusion of CME physical parameter data and CME image data for time prediction. The prediction method of the coronal mass ejection event is as follows Figure 4 As shown, the following steps may be specifically included:

[0075] S101, obtaining physical parameters and CME images of coronal mass ejections (CMEs), wherein step S102 is performed on the physical parameters of CMEs, and step S109 is performed on the CME images.

[0076] S102, performing feature mapping on the physical parameters of the CME to form a plurality of first-layer feature vectors.

[0077] S103, performing dimension enhancement on multiple first-layer feature vectors to form second-layer feature vectors.

[0078] S104, performing nonlinear enhancement on the second layer feature vectors to form third layer feature vectors.

[0079] S105, adding the plurality of first-layer feature vectors and the third-layer feature vectors and performing normalization processing to form a target parameter feature vector.

[0080] S106, performing a self-attention mechanism operation on the target parameter feature vector to obtain an enhanced parameter feature vector.

[0081] S107 , performing a matrix product on the first query vector and the first key vector to generate a first similarity matrix.

[0082] S108, determine a second query vector according to the enhanced parameter feature vector. Execute step S112.

[0083] S109, using an improved ResNet-18 network to extract features from the CME image to form a target image feature vector.

[0084] S110, performing self-attention mechanism operations on the target image feature vectors respectively to obtain enhanced image feature vectors.

[0085] S111, determining a second key vector and a second value vector according to the enhanced image feature vector.

[0086] S112, performing a matrix product on the second query vector and the second key vector to generate a second similarity matrix.

[0087] S113, applying a Softmax function to the second similarity matrix for normalization to generate a second attention weight.

[0088] S114, applying the second attention weight to the second value vector and performing weighted summation to obtain a fused feature vector.

[0089] S115, predicting the arrival time according to the fused feature vector.

[0090] This coronal mass ejection event prediction method adopts a method of combining CME physical parameter data and CME image data in the CME arrival time prediction to achieve a comprehensive and in-depth understanding of CME and improve the accuracy of CME arrival time prediction.

[0091] In summary, according to the prediction method of the coronal mass ejection event of the embodiment of the present invention, firstly, the physical parameters and CME images of the coronal mass ejection CME are obtained, and the feature extraction of the CME physical parameters is performed to obtain the target parameter feature vector, and the feature extraction of the CME image is performed to obtain the target image feature vector, and the target parameter feature vector and the target image feature vector are feature fused based on the attention mechanism to obtain the fused feature vector; and the arrival time is predicted according to the fused feature vector. Therefore, by introducing the attention mechanism, the method performs feature fusion on the target parameter feature vector obtained based on the CME physical parameters and the target image feature vector obtained based on the CME image, which can effectively delete the redundant information between the CME physical parameters and the CME image, realize the fusion of complementary information, and realize the rapid and accurate prediction of the CME arrival time.

[0092] Corresponding to the above embodiment, the present invention also proposes a prediction device for coronal mass ejection events.

[0093] like Figure 5 As shown, the device for predicting a coronal mass ejection event according to the embodiment of the present invention may include: a first acquisition module 10 , a second acquisition module 20 , a third acquisition module 30 , a fusion module 40 and a prediction module 50 .

[0094] Among them, the first acquisition module 10 is used to obtain the physical parameters and CME images of the coronal mass ejection CME. The second acquisition module 20 is used to extract the features of the CME physical parameters to obtain the target parameter feature vector. The third acquisition module 30 is used to extract the features of the CME image to obtain the target image feature vector. The fusion module 40 is used to perform feature fusion on the target parameter feature vector and the target image feature vector based on the attention mechanism to obtain a fused feature vector. The prediction module 50 is used to predict the arrival time according to the fused feature vector.

[0095] According to one embodiment of the present invention, the second acquisition module 20 performs feature extraction on the CME physical parameters to obtain the target parameter feature vector, which is specifically used for: performing feature mapping on the CME physical parameters to form multiple first-layer feature vectors; performing dimensionality enhancement on the multiple first-layer feature vectors to form second-layer feature vectors; performing nonlinear enhancement on the second-layer feature vectors to form third-layer feature vectors; adding the multiple first-layer feature vectors and the third-layer feature vectors and performing normalization processing to form the target parameter feature vector.

[0096] According to one embodiment of the present invention, the third acquisition module 30 performs feature extraction on the CME image to obtain a target image feature vector, and is specifically used to: use an improved ResNet-18 network to extract features from the CME image to form a target image feature vector.

[0097] According to one embodiment of the present invention, the improved ResNet-18 network includes: an input layer, a residual block, a skip connection, a flatten layer and an output layer.

[0098] According to one embodiment of the present invention, the fusion module 40 performs feature fusion on the target parameter feature vector and the target image feature vector based on the attention mechanism, and is specifically used for: performing self-attention mechanism operations on the target parameter feature vector and the target image feature vector respectively to obtain an enhanced parameter feature vector and an enhanced image feature vector respectively; performing cross-attention mechanism operations on the enhanced parameter feature vector and the enhanced image feature vector to obtain a fused feature vector.

[0099] According to one embodiment of the present invention, the fusion module 40 performs a self-attention mechanism operation on the target parameter feature vector and the target image feature vector, respectively, to obtain an enhanced parameter feature vector and an enhanced image feature vector, respectively, which are specifically used to: determine a first query vector, a first key vector and a first value vector; perform a matrix product based on the first query vector and the first key vector to generate a first similarity matrix; apply a Softmax function to normalize the first similarity matrix to generate a first attention weight; apply the first attention weight to the first value vector and perform a weighted sum to obtain an enhanced parameter feature vector and an enhanced image feature vector.

[0100] According to one embodiment of the present invention, the fusion module 40 performs a cross-attention mechanism operation on the enhanced parameter feature vector and the enhanced image feature vector to obtain a fused feature vector, which is specifically used for: determining a second query vector based on the enhanced parameter feature vector, and determining a second key vector and a second value vector based on the enhanced image feature vector; performing a matrix product between the second query vector and the second key vector to generate a second similarity matrix; applying a Softmax function to normalize the second similarity matrix to generate a second attention weight; applying the second attention weight to the second value vector and performing a weighted sum to obtain a fused feature vector.

[0101] According to one embodiment of the present invention, the prediction module 50 is further configured to: evaluate the predicted arrival time according to the mean absolute error and the root mean square error.

[0102] It should be noted that for details not disclosed in the device for predicting a coronal mass ejection event in an embodiment of the present invention, please refer to the details disclosed in the method for predicting a coronal mass ejection event in the above-mentioned embodiment of the present invention, and no further details will be given here.

[0103] According to the prediction device of the coronal mass ejection event of the embodiment of the present invention, the physical parameters and CME images of the coronal mass ejection CME are obtained by the first acquisition module, the feature extraction of the CME physical parameters is performed by the second acquisition module to obtain the target parameter feature vector, the feature extraction of the CME image is performed by the third acquisition module to obtain the target image feature vector, the feature fusion of the target parameter feature vector and the target image feature vector is performed by the fusion module based on the attention mechanism to obtain the fused feature vector, and the prediction module predicts the arrival time according to the fused feature vector. Therefore, the device can effectively delete the redundant information between the CME physical parameters and the CME image by introducing the attention mechanism to fuse the target parameter feature vector obtained based on the CME physical parameters and the target image feature vector obtained based on the CME image, and realize the fusion of complementary information, so as to realize the rapid and accurate prediction of the CME arrival time.

[0104] Corresponding to the above embodiment, the present invention also proposes a computer-readable storage medium.

[0105] According to the computer-readable storage medium of an embodiment of the present invention, the above-mentioned method for predicting a coronal mass ejection event is implemented when the prediction program for a coronal mass ejection event is executed by a processor. Based on the above-mentioned method for predicting a coronal mass ejection event, redundant information between CME physical parameters and CME images is effectively deleted, and complementary information is fused to achieve rapid and accurate prediction of the CME arrival time.

[0106] According to the computer-readable storage medium of an embodiment of the present invention, when the prediction program of the coronal mass ejection event is executed by the processor, the above-mentioned prediction method of the coronal mass ejection event is implemented. Based on the above-mentioned prediction method of the coronal mass ejection event,

[0107] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0108] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0109] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0110] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0111] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0112] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for predicting a coronal mass ejection event, characterized in that: The method comprises: Obtain CME physical parameters and CME images; Performing feature extraction on the CME physical parameters to obtain a target parameter feature vector; Performing feature extraction on the CME image to obtain a target image feature vector; Based on the attention mechanism, feature fusion is performed on the target parameter feature vector and the target image feature vector to obtain a fused feature vector; The arrival time is predicted according to the fused feature vector.

2. The method for predicting a coronal mass ejection event according to claim 1, wherein: Feature extraction is performed on the CME physical parameters to obtain a target parameter feature vector, including: Performing feature mapping on the CME physical parameters to form a plurality of first-layer feature vectors; Performing dimension enhancement on the plurality of first-layer feature vectors to form second-layer feature vectors; Performing nonlinear enhancement on the second-layer feature vectors to form a third-layer feature vector; A plurality of the first-layer feature vectors and the third-layer feature vectors are added and normalized to form the target parameter feature vector.

3. The method for predicting a coronal mass ejection event according to claim 2, wherein: Performing feature extraction on the CME image to obtain a target image feature vector includes: An improved ResNet-18 network is used to extract features of the CME image to form the target image feature vector.

4. The method for predicting a coronal mass ejection event according to claim 3, wherein: The improved ResNet-18 network includes: an input layer, a residual block, a skip connection, a flatten layer and an output layer.

5. The method for predicting a coronal mass ejection event according to claim 3, characterized in that: Performing feature fusion on the target parameter feature vector and the target image feature vector based on an attention mechanism includes: Performing a self-attention mechanism operation on the target parameter feature vector and the target image feature vector respectively to obtain an enhanced parameter feature vector and an enhanced image feature vector respectively; A cross-attention mechanism operation is performed on the enhanced parameter feature vector and the enhanced image feature vector to obtain the fused feature vector.

6. The method for predicting a coronal mass ejection event according to claim 5, characterized in that: The target parameter feature vector and the target image feature vector are respectively subjected to a self-attention mechanism operation to obtain an enhanced parameter feature vector and an enhanced image feature vector, respectively, including: determining a first query vector, a first key vector, and a first value vector; Generate a first similarity matrix by performing a matrix product of the first query vector and the first key vector; Applying a Softmax function to normalize the first similarity matrix to generate a first attention weight; The first attention weight is applied to the first value vector and weighted sum is performed to obtain the enhanced parameter feature vector and the enhanced image feature vector.

7. The method for predicting a coronal mass ejection event according to claim 5, characterized in that: Performing a cross attention mechanism operation on the enhanced parameter feature vector and the enhanced image feature vector to obtain the fused feature vector includes: Determine a second query vector according to the enhanced parameter feature vector, and determine a second key vector and a second value vector according to the enhanced image feature vector; Generate a second similarity matrix by performing a matrix product of the second query vector and the second key vector; Applying a Softmax function to normalize the second similarity matrix to generate a second attention weight; The second attention weight is applied to the second value vector and weighted sum is performed to obtain the fused feature vector.

8. The method for predicting a coronal mass ejection event according to claim 1, wherein: The method further comprises: The predicted arrival times were evaluated based on mean absolute error and root mean square error.

9. A device for predicting a coronal mass ejection event, characterized in that: The device comprises: The first acquisition module is used to obtain the physical parameters and CME images of the coronal mass ejection CME; A second acquisition module is used to extract features of the CME physical parameters to obtain a target parameter feature vector; A third acquisition module is used to extract features from the CME image to obtain a target image feature vector; A fusion module, used for performing feature fusion on the target parameter feature vector and the target image feature vector based on an attention mechanism to obtain a fused feature vector; A prediction module is used to predict the arrival time according to the fused feature vector.

10. A computer-readable storage medium, characterized in that: A prediction program for coronal mass ejection events is stored thereon, and when the prediction program for coronal mass ejection events is executed by a processor, a prediction method for coronal mass ejection events according to any one of claims 1-9 is implemented.

Images