A solar flare prediction method based on multi-feature fusion
By employing a multi-feature fusion method, utilizing LVQ neural networks and 3D convolutional neural network models, and combining multiple forecasting factors and flare data, the low confidence problem caused by a single forecasting factor was solved, achieving highly accurate solar flare forecasting.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- YUNNAN OBSERVATORY CHINESE ACADEMY OF SCIENCES
- Filing Date
- 2023-06-25
- Publication Date
- 2026-05-29
AI Technical Summary
Existing methods for predicting solar flares typically use only a single predictor, resulting in low confidence levels in the predictions and making it difficult to accurately predict the occurrence of flares.
A multi-feature fusion method is adopted, which combines LVQ neural network and 3D convolutional neural network models with multiple forecast factors and flare data to construct a dataset for training and prediction. The results of multiple models are used for correction, and finally a high-confidence prediction result is obtained.
It improves the accuracy of solar flare prediction, and allows for the selection of appropriate result models for prediction based on the intended use after a single data input, thus meeting different needs.
Smart Images
Figure CN116956211B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of solar flare prediction technology, specifically relating to a solar flare prediction method based on multi-feature fusion. Background Technology
[0002] A solar flare is a sudden and large-scale energy release process that occurs in a localized area on the surface of the sun. It is a major source of disturbance in the space environment and has a significant impact on the Earth's space environment. Solar flare forecasting is an important component of space weather forecasting.
[0003] Solar flares are among the most violent solar activities and a crucial area of study in space weather forecasting. They are produced by the sudden, large-scale release of localized energy within the solar atmosphere. The eruption of a flare is primarily characterized by a sudden increase in radiation flux from radio waves to X-rays. A large flare, with an area of one two-thousandth of the Sun's surface, can release 4 × 10⁻³ volts of energy in its 10³-second lifetime. 32 The energy of flares causes them to emit intense shortwave radiation, severely disrupting the Earth's lower ionosphere. This causes shortwave radio waves to be strongly absorbed as they pass through, leading to the interruption of shortwave communications. Further ionospheric disturbances disrupt the normal operation of navigation systems. Due to the significant impact of solar flares on space weather, various flare forecasting methods have been extensively studied. Statistical methods for flare forecasting utilize only observed events to predict future conditions, and the statistical patterns of observed solar flares have important guiding significance for flare forecasting.
[0004] The process of solar flare eruption is complex, and the physical mechanism is not yet fully understood. Therefore, most current methods extract forecasting factors from photosphere magnetic field observation data and generate a dataset of whether or not flares occur. From this dataset, a flare prediction model is learned. However, existing prediction methods usually only use one forecasting factor to predict flare occurrence, and the confidence level of the results is open to discussion. To address this, we propose a solar flare prediction method based on multi-feature fusion. Summary of the Invention
[0005] The purpose of this invention is to provide a solar flare prediction method based on multi-feature fusion to solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a solar flare prediction method based on multi-feature fusion, comprising the following steps:
[0007] S1: Acquire solar observation data over a certain period of time, and extract forecast factors and flare eruption data from them. Different forecast factors are labeled with a, b, c... and different flare eruption data are labeled with α, β, γ...
[0008] S2: Construct datasets using different forecast factors and flare eruption data, and each forecast factor needs to be constructed corresponding to each flare eruption data. Different datasets are labeled with (a, α), (a, β), (a, γ)... and each dataset is then divided into training set and test set;
[0009] S3: Use the training set portion of each dataset to train different LVQ neural network models, resulting in multiple trained LVQ neural network models. The resulting LVQ neural network models are labeled using the training set used, i.e., (a, α) model, (a, β) model, (a, γ) model, etc.
[0010] S4: Divide multiple LVQ neural network models into several categories based on the different flare data in the dataset they use. Each category is labeled using flare data, namely, α-class computational model, β-class computational model, γ-class computational model, etc.
[0011] S5: Then, the prediction factors of the test set of different datasets are input into the model corresponding to the category of their flare eruption data, and the result set is obtained. Each type of end data is labeled with flare data, namely α-type result set, β-type result set, γ-type result set, etc.
[0012] S6: Train multiple 3D convolutional neural network models using the flare burst data corresponding to the result set, thereby obtaining result models for different types of flare burst data. Different result models are labeled with flare data, namely α result model, β result model, γ result model, etc.
[0013] S7: Combining the different types of calculation models in S4 with the corresponding result models in S6 forms the prediction model. Different prediction models are labeled using flare data, namely α prediction model, β prediction model, γ prediction model, etc.
[0014] S8: Input the forecast factors into the prediction model of S7 to obtain the prediction results of different solar flares.
[0015] Preferably, the forecasting factors in S1 include McIntoshfen classification data of sunspot groups, sunspot data, data on changes in the maximum horizontal gradient of the solar magnetic field, data on changes in the number of isolated singularities in the solar magnetic field, and data on changes in the length of the solar magnetic field centerline.
[0016] Preferably, the flare data in S1 includes solar A-class flare variation data, solar B-class flare variation data, solar C-class flare variation data, solar M-class flare variation data, and solar X-class flare variation data.
[0017] Preferably, the forecast factor data and flare data in S1 are derived from the longitudinal magnetograph of the entire solar surface and solar activity record data.
[0018] Preferably, in step S2, when constructing a dataset by corresponding the forecast factors with the solar flare eruption data, the time point of the forecast factors should be a certain period of time before the solar flare eruption time point. The specific time interval is determined according to the solar flare time that needs to be predicted in the prediction model.
[0019] Preferably, in step S3, the training of the LVQ neural network model should use the backpropagation method to transmit the error signal back in reverse and modify the weight coefficients of each layer of neurons in turn to reduce the error signal.
[0020] Preferably, the classification of the computational model in S4 is based on the data type of flare, and models trained using the same flare data are classified into the same category.
[0021] Preferably, the time point of the result set in S6 should be a certain period of time before the time point of the solar flare data corresponding to the result set results. The specific time interval is determined according to the solar flare time to be predicted in the prediction model.
[0022] Preferably, the classification standard used to combine the calculation model and the result model in S7 is the flare data type, and those trained using the same flare data are classified into the same category.
[0023] Compared with the prior art, the beneficial effects of the present invention are:
[0024] 1. By using the present invention, the occurrence of solar flares can be predicted by combining multiple predictive factors and computational models, and the results obtained by the computational models can be summarized and corrected by the result models, thereby obtaining results with higher confidence and effectively improving the accuracy of the model in predicting solar flares.
[0025] 2. With the present invention, after a single data input, the present invention can predict the occurrence of solar flares of different levels, thereby enabling users to choose the appropriate result according to different uses. Attached Figure Description
[0026] Figure 1 This is a schematic diagram of the model structure of the present invention;
[0027] Figure 2 This is a schematic diagram of the dataset classification structure process of the present invention;
[0028] Figure 3 This is a schematic diagram of the LVQ neural network model structure of the present invention;
[0029] Figure 4This is a schematic diagram of the result set classification structure of the present invention;
[0030] Figure 5 This is a schematic diagram of the 3D convolutional neural network model structure of the present invention; Detailed Implementation
[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0032] Example 1:
[0033] Please see Figures 1 to 5 This invention provides a technical solution: a solar flare prediction method based on multi-feature fusion, comprising the following steps:
[0034] S1: Acquire solar observation data over a certain period of time, and extract forecast factors and flare eruption data from them. Different forecast factors are labeled with a, b, c... and different flare eruption data are labeled with α, β, γ...
[0035] S2: Construct datasets using different forecast factors and flare eruption data, and each forecast factor needs to be constructed corresponding to each flare eruption data. Different datasets are labeled with (a, α), (a, β), (a, γ)... and each dataset is then divided into training set and test set;
[0036] S3: Use the training set portion of each dataset to train different LVQ neural network models, resulting in multiple trained LVQ neural network models. The resulting LVQ neural network models are labeled using the training set used, i.e., (a, α) model, (a, β) model, (a, γ) model, etc.
[0037] S4: Divide multiple LVQ neural network models into several categories based on the different flare data in the dataset they use. Each category is labeled using flare data, namely, α-class computational model, β-class computational model, γ-class computational model, etc.
[0038] S5: Then, the prediction factors of the test set of different datasets are input into the model corresponding to the category of their flare eruption data, and the result set is obtained. Each type of end data is labeled with flare data, namely α-type result set, β-type result set, γ-type result set, etc.
[0039] S6: Train multiple 3D convolutional neural network models using the flare burst data corresponding to the result set, thereby obtaining result models for different types of flare burst data. Different result models are labeled with flare data, namely α result model, β result model, γ result model, etc.
[0040] S7: Combining the different types of calculation models in S4 with the corresponding result models in S6 forms the prediction model. Different prediction models are labeled using flare data, namely α prediction model, β prediction model, γ prediction model, etc.
[0041] S8: Input the forecast factors into the prediction model of S7 to obtain the prediction results of different solar flares.
[0042] As can be seen from the above description, the present invention has the following beneficial effects: by predicting the occurrence of solar flares using multiple predictive factors and computational models, and by summarizing and correcting the result set obtained by the computational models through the result models, the present invention can obtain results with higher confidence and effectively improve the accuracy of the model in predicting solar flares.
[0043] Furthermore, the forecast factors in S1 include McIntoshfen classification data for sunspot groups, sunspot data, changes in the maximum horizontal gradient of the solar magnetic field, changes in the number of isolated singularities in the solar magnetic field, and changes in the length of the solar magnetic field centerline. The flare data in S1 includes changes in A-class, B-class, C-class, M-class, and X-class solar flares. The forecast factor data and flare data in S1 are from the full-surface longitudinal magnetographs taken by DO / HMI, which can be downloaded from http: / / jsoc.stanford.edu / . Solar activity record data is provided by NOAA, containing information such as the date, flare eruption details, and coordinates for each active region, and can be found at http: / / www.ngdc.noaa.gov. / Downloaded; In S2, when constructing the dataset by mapping the forecast factors to flare eruption data, the time point of the forecast factors should be some time before the flare eruption time point. The specific time interval is determined according to the solar flare time to be predicted in the prediction model; In S3, the error backpropagation method should be used to train the LVQ neural network model, propagating the error signal back in reverse and modifying the weight coefficients of each layer of neurons in turn to reduce the error signal; In S4, the classification basis of the calculation model is the flare data type. Models trained using the same flare data are classified into the same category; In S6, the time point of the result set should be some time before the time point of the flare eruption data corresponding to the result set results. The specific time interval is determined according to the solar flare time to be predicted in the prediction model; In S7, the classification criterion for combining the calculation model and the result model is the flare data type. Models trained using the same flare data are classified into the same category.
[0044] The working principle and usage process of this invention are as follows: Solar observation data over a certain period is acquired, and forecast factors and flare eruption data are extracted from it. Different forecast factors are labeled a, b, c… and different flare eruption data are labeled α, β, γ… A dataset is constructed using different forecast factors and flare eruption data, with each forecast factor corresponding to each flare eruption data point. Different datasets are labeled (a, α), (a, β), (a, γ)… Each dataset is then divided into a training set and a test set. The training set portion of each dataset is used to train different LVQ neural network models, resulting in multiple trained LVQ neural network models. These models are labeled using the training set used, i.e., (a, α) model, (a, β) model, (a, γ) model… The multiple LVQ neural network models are then divided into several categories based on the different flare data in the datasets used. Each category is labeled… Solar flare data are labeled as α-class computational models, α-class computational models, α-class computational models, and so on. Then, prediction factors from the test set of different datasets are input into the models corresponding to the categories of their solar flare eruption data, resulting in result sets. Each type of data is labeled using solar flare data, i.e., α-class result set, β-class result set, γ-class result set, and so on. Multiple 3D convolutional neural network models are trained using the result sets and the corresponding solar flare eruption data to derive result models for different types of solar flare eruption data. Different result models are labeled using solar flare data, i.e., α-result model, β-result model, γ-result model, and so on. Combining different types of computational models with their corresponding result models yields prediction models. Different prediction models are labeled using solar flare data, i.e., α-prediction model, β-prediction model, γ-prediction model, and so on. Finally, prediction factors are input into the prediction model of S7 to obtain prediction results for different solar flare occurrences.
[0045] 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, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus 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 apparatus. 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 apparatus that includes said element.
[0046] The above description is only used to illustrate the technical solution of the present invention and is not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention, as long as they do not depart from the spirit and scope of the technical solution of the present invention, should be covered within the scope of the claims of the present invention.
Claims
1. A solar flare prediction method based on multi-feature fusion, characterized in that, Includes the following steps: S1: Acquire solar observation data over a certain period of time, and extract forecast factors and flare eruption data from them. Different forecast factors are labeled with a, b, c... and different flare eruption data are labeled with α, β, γ... S2: Construct datasets using different forecast factors and flare eruption data, and each forecast factor needs to be constructed corresponding to each flare eruption data. Different datasets are labeled with (a, α), (a, β), (a, γ)... and each dataset is then divided into training set and test set; S3: Use the training set portion of each dataset to train different LVQ neural network models, resulting in multiple trained LVQ neural network models. The resulting LVQ neural network models are labeled using the training set used, i.e., (a, α) model, (a, β) model, (a, γ) model, etc. S4: Divide multiple LVQ neural network models into several categories based on the different flare data in the dataset they use. Each category is labeled using flare data, namely, α-class computational model, β-class computational model, γ-class computational model, etc. S5: Then, the prediction factors of the test set of different datasets are input into the model corresponding to the category of their flare eruption data, and the result set is obtained. Each type of end data is labeled with flare data, namely α-type result set, β-type result set, γ-type result set, etc. S6: Train multiple 3D convolutional neural network models using the flare burst data corresponding to the result set, thereby obtaining result models for different types of flare burst data. Different result models are labeled with flare data, namely α result model, β result model, γ result model, etc. S7: Combining the different types of calculation models in S4 with the corresponding result models in S6 forms the prediction model. Different prediction models are labeled using flare data, namely α prediction model, β prediction model, γ prediction model, etc. S8: Input the forecast factors into the prediction model of S7 to obtain the prediction results of different solar flares.
2. The solar flare prediction method based on multi-feature fusion according to claim 1, characterized in that: The forecasting factors in S1 include McIntoshfen classification data of sunspot groups, sunspot data, data on the variation of the maximum horizontal gradient of the solar magnetic field, data on the variation of the number of isolated singularities in the solar magnetic field, and data on the variation of the length of the solar magnetic field centerline.
3. The solar flare prediction method based on multi-feature fusion according to claim 1, characterized in that: The flare data in S1 includes solar A-class flare variation data, solar B-class flare variation data, solar C-class flare variation data, solar M-class flare variation data, and solar X-class flare variation data.
4. The solar flare prediction method based on multi-feature fusion according to claim 1, characterized in that: The forecast factor data and flare data in S1 are derived from the longitudinal magnetograph of the entire solar surface and solar activity record data.
5. The solar flare prediction method based on multi-feature fusion according to claim 1, characterized in that: In step S2, when constructing a dataset by mapping the forecast factors to solar flare eruption data, the time point of the forecast factors should be some time before the solar flare eruption time point. The specific time interval is determined according to the solar flare time to be predicted in the prediction model.
6. The solar flare prediction method based on multi-feature fusion according to claim 1, characterized in that: In step S3, the LVQ neural network model should be trained using the backpropagation method to transmit the error signal back in reverse and modify the weight coefficients of each layer of neurons in turn to reduce the error signal.
7. The solar flare prediction method based on multi-feature fusion according to claim 1, characterized in that: The classification of computational models in S4 is based on the data type of flares; models trained using the same flare data are classified into the same category.
8. The solar flare prediction method based on multi-feature fusion according to claim 1, characterized in that: In S6, the time point of the result set should be a certain period of time before the time point of the solar flare eruption data corresponding to the result set results. The specific time interval is determined according to the solar flare time that needs to be predicted in the prediction model.
9. The solar flare prediction method based on multi-feature fusion according to claim 1, characterized in that: The classification criterion used to combine the computational model and the result model in S7 is the flare data type; models trained using the same flare data are classified into the same category.