Solar flare spectral line inversion method

Abstract

The invention discloses a solar flare spectral line inversion method, which belongs to the crossing field of astronomical observation and artificial intelligence, and comprises the following steps: simulating a solar flare event data set according to RADYN, training a constructed full-connection network after preprocessing, carrying out simulation test by using simulated spectral data, carrying out actual test by using actual spectral data in sequence, and obtaining a solar flare spectral line inversion result; according to the solar flare spectrum inversion model, height distribution data of physical parameters can be obtained through inversion according to actually measured spectrum data of a solar flare event, and solar flare spectral line inversion is completed. According to the method, the full-connection neural network is used for inversion of the solar flares, rapid inversion of physical parameters of the solar chromosphere layer can be achieved, data processing time is greatly shortened, meanwhile, high-dimensional spectral data are effectively processed, complex features are automatically learned and extracted, and the nonlinear relation between the spectral data and the physical parameters is captured; and high-precision parameter inversion is realized.

Description

Technical Field

[0001] The present invention relates to a solar flare spectral line inversion method, which belongs to the intersection of astronomical observation and artificial intelligence, and is specifically applied to solar physics research and forecast analysis of solar eruption activities. Background Art

[0002] Solar eruptions release enormous amounts of energy, which heat different layers of the solar atmosphere through particle beam bombardment, heat conduction, thermal radiation and other mechanisms, causing dramatic changes in the originally tranquil solar atmosphere. These changes are mainly manifested in significant changes in parameters such as plasma number density, temperature, and movement speed. The changes in these parameters are closely related to the generation and evolution of solar eruption activities. As with magnetic field measurements, we cannot directly measure these physical parameters. However, changes in these parameters will be directly reflected in the observed spectrum, and we can obtain these characteristic parameters through spectral inversion. Therefore, accurate inversion and measurement of these physical parameters are of great scientific significance and application value for a deep understanding of the eruption mechanism of solar activity and improving the accuracy of solar eruption forecasts.

[0003] Although inversion methods based on physical models have been used for spectral inversion, these methods often rely on complex physical assumptions and approximations, with high computational costs. Sometimes the assumptions and approximations do not hold, resulting in erroneous inversions. In particular, the Xihe provides high-resolution spectra from the solar photosphere to the chromosphere. These spectral lines are mainly optically thick and cannot be achieved with simple inversion models. Machine learning methods perform fast and efficient inversions based on large-scale statistical data, showing broad application prospects. With the improvement of computing power and the development of deep learning technology, it has become possible to use artificial intelligence to invert spectral lines of solar flare activities.

[0004] The present invention applies deep learning methods to the study of solar flare spectral line inversion, aiming to build a model that uses Hα spectral data to quickly and accurately invert the physical parameters of the solar chromosphere (including particle number density, temperature and plasma motion speed) by building a suitable fully connected neural network model. At the same time, the model is applied to invert the high-resolution Hα spectral data observed by China's "Xihe" solar exploration satellite (CHASE) during solar flare outbreaks, and the plasma temperature, density and radial velocity field of the observed flare area are quickly and real-time given. The model is expected to analyze the dynamic evolution of solar flare outbreaks observed by "Xihe" in real time, and provide physical data support for flare outbreak prediction. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a solar flare spectral line inversion method, which can invert the physical parameter data of the solar flare event based on the actual spectral data obtained during the solar flare outbreak, thereby solving the problem that it is very difficult to directly measure the physical parameters of the solar flare event.

[0006] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a solar flare spectral line inversion method, comprising the following steps:

[0007] S1: According to the RADYN numerical simulation algorithm, a simulated solar flare event data set is obtained;

[0008] S2: preprocessing the simulated solar flare event data set to obtain preprocessed spectral data and smoothed physical parameter height distribution data;

[0009] S3: Establish a fully connected network and use the preprocessed spectral data and smoothed physical parameter height distribution data for training to obtain the flare spectrum inversion model;

[0010] S4: According to the RADYN numerical simulation algorithm, simulated spectral data is obtained, and the simulated spectral data is input into the flare spectrum inversion model for simulation test, and it is judged whether the error index is less than the set error threshold. If so, it goes to S5, otherwise, it returns to S3 and retrains to obtain the flare spectrum inversion model;

[0011] S5: According to the solar exploration satellite, actual spectrum data is obtained, and the actual spectrum data is input into the flare spectrum inversion model for actual testing, and it is determined whether the physical parameter height distribution data output by the flare spectrum inversion model is consistent with the actual physical parameter height distribution data. If so, enter S6, otherwise, return to S3 and retrain to obtain the flare spectrum inversion model;

[0012] S6: Input the measured spectral data of the solar flare event into the flare spectrum inversion model for processing, obtain the height distribution data of the physical parameters of the solar flare event, and complete the inversion of the solar flare spectrum line.

[0013] The beneficial effects of the present invention are as follows: the present invention adopts multiple hidden layers to form a fully connected network, and is trained by respectively using simulated spectral data and actual spectral data generated by the RADYN simulation algorithm, which can fully mine the implicit information in the spectral data and complete the rapid inversion of physical parameters during solar flares. It has high computational efficiency while ensuring high accuracy. Through the accuracy and high efficiency of solar activity forecasts, the potential impact of solar flares on human society can be reduced.

[0014] Further: the specific steps of S2 are as follows:

[0015] S201: Acquire original spectral data and physical parameter height distribution data in a simulated solar flare event data set;

[0016] S202: performing point spread function convolution on the original spectral data to obtain spectral data with a resolution consistent with the measured spectral data;

[0017] S203: performing interpolation and normalization processing within a set wavelength range according to the spectral data having a resolution consistent with the measured spectral data, to obtain pre-processed spectral data;

[0018] S204: performing interpolation processing according to the physical parameter height distribution data to obtain smooth physical parameter height distribution data.

[0019] The beneficial effect of the above further scheme is that preprocessing the original spectral data and physical parameter height distribution data in the simulated solar flare event data set can improve the data processing efficiency of the fully connected network and the accuracy of the inversion results.

[0020] Further: the fully connected network comprises: an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a fifth hidden layer and an output layer connected in sequence; the first hidden layer, the second hidden layer, the third hidden layer, the fourth hidden layer and the fifth hidden layer are all connected in sequence to an activation layer and a discard layer;

[0021] The input layer is used to receive 64-dimensional spectral data;

[0022] The first hidden layer is used to map the 64-dimensional spectral data to a 128-dimensional space through a linear transformation and perform feature extraction to obtain 128-dimensional data and a first feature;

[0023] The second hidden layer is used to map the 128-dimensional data to a 256-dimensional space through a linear transformation and perform feature extraction to obtain the 128-dimensional data and the second feature;

[0024] The third hidden layer is used to map the 256-dimensional data to a 512-dimensional space through a linear transformation, and perform feature extraction to obtain the 512-dimensional data and the third feature;

[0025] The fourth hidden layer is used to map the 512-dimensional data to a 1024-dimensional space through a linear transformation to obtain 1024-dimensional data, and integrate the first feature, the second feature and the third feature to obtain an integrated feature;

[0026] The fifth hidden layer is used to compress the 1024-dimensional data into a 512-dimensional space through a linear transformation to obtain 512-dimensional spectral data;

[0027] The output layer is used to compress the 512-dimensional data into a 400-dimensional space to obtain 400-dimensional output data, wherein the 400-dimensional output data is the inverted physical parameter height distribution data;

[0028] The activation layer is used to introduce nonlinear characteristics by suppressing negative values ​​and retaining positive values;

[0029] The dropout layer is used to randomly drop some neurons.

[0030] The beneficial effects of the above further scheme are as follows: the present invention uses a fully connected neural network to perform solar flare inversion, which has a simple structure and strong versatility, and can effectively process high-dimensional spectral data. Through multiple hidden layers and a large number of neurons, it can automatically learn and extract complex features in the input data, capture the nonlinear relationship between spectral data and physical parameters, and achieve high-precision parameter inversion.

[0031] Further: the specific steps of S3 are as follows:

[0032] S301: Establishing a fully connected network, wherein the fully connected network includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a fifth hidden layer and an output layer;

[0033] S302: inputting the preprocessed spectral data into a fully connected network, and outputting inverted physical parameter height distribution data;

[0034] S303: Calculating loss values ​​of the inverted physical parameter height distribution data and the smoothed physical parameter height distribution data according to the Huber loss function;

[0035] S304: According to the loss value, the gradient of the loss value with respect to the weight of each layer is calculated using the back propagation algorithm;

[0036] S305: Using the Adam optimizer to update the model parameters of the fully connected network according to the gradient of the loss value to the weights of each layer;

[0037] S306: Determine whether the number of iterations reaches the set number of iterations. If so, complete the training of the fully connected network to obtain the flare spectrum inversion model. Otherwise, return to S302.

[0038] The beneficial effects of the above further scheme are as follows: the present invention uses the Huber loss function to perform loss value analysis, which can reduce the impact of noise and extreme abnormal data in spectral data on training. At the same time, it can robustly process a large amount of complex high-dimensional spectral data to ensure the accuracy and stability of physical parameter inversion.

[0039] Further: the specific steps of S4 are as follows:

[0040] S401: According to the RADYN numerical simulation algorithm, a simulated solar flare event data set is obtained, and simulated spectral data and simulated physical parameter height distribution data are randomly selected;

[0041] S402: preprocessing the simulated spectrum data, and inputting it into the flare spectrum inversion model, and outputting the inverted physical parameter height distribution data;

[0042] S403: Calculating an error index according to the inverted physical parameter height distribution data and the simulated physical parameter height distribution data;

[0043] S404: Determine whether the error index is less than the set error threshold. If so, complete the simulation test of the flare spectrum inversion model and enter S5. Otherwise, optimize the model parameters of the flare spectrum inversion model, adjust the preprocessed spectral data and the smoothed physical parameter height distribution data, and return to S3 to retrain and obtain the flare spectrum inversion model.

[0044] The beneficial effect of the above further scheme is that by simulating the solar flare event data set through RADYN, the flare spectrum inversion model can be tested, the performance of the model on the simulated data can be evaluated, and optimization can be performed to improve its performance and inversion accuracy.

[0045] Further: the specific steps of S5 are as follows:

[0046] S501: Based on the solar exploration satellite, historical spectrum data is obtained, and the historical data is selected as actual spectrum data;

[0047] S502: pre-processing the actual spectrum data, and inputting it into the flare spectrum inversion model, and outputting the inverted physical parameter height distribution data;

[0048] S503: Compare the inverted physical parameter height distribution data with the known physical parameter height distribution data, and determine whether they are consistent. If so, complete the actual test of the flare spectrum inversion model and enter S6. Otherwise, adjust the preprocessed spectral data and the smoothed physical parameter height distribution data, return to S3, and retrain to obtain the flare spectrum inversion model.

[0049] The beneficial effect of the above further scheme is that by using actual spectral data to test the flare spectrum inversion model, its accuracy when used for measured spectral data can be ensured, and the deficiencies of the model can be further discovered.

[0050] Further: the specific steps of S6 are as follows:

[0051] S601: Acquire measured spectrum data of solar flare events;

[0052] S602: Input the measured spectrum data into the flare spectrum inversion model, and output the inverted physical parameter height distribution data;

[0053] S603: Construct a three-dimensional stereogram of the flare area based on the inverted physical parameter height distribution data, and complete the inversion of the solar flare spectrum.

[0054] The beneficial effects of the above further scheme are: the present invention constructs and trains a fully connected network to obtain a flare spectrum inversion model, which can process measured spectral data, quickly and accurately invert the height distribution data of physical parameters in solar flare events, and further establish a three-dimensional stereogram to facilitate the analysis and research of solar flares. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of the solar flare spectral line inversion method of the present invention;

[0056] Figure 2 Data from the RADYN simulated solar flare event dataset;

[0057] Figure 3 A diagram of a fully connected network structure constructed for the present invention;

[0058] Figure 4 The results of model tests for simulating solar flares before eruption;

[0059] Figure 5 Results of model tests for simulations of solar flare explosions;

[0060] Figure 6 The results of the model test for the actual test before the solar flare eruption;

[0061] Figure 7 Results of model tests for actual tests in solar flare eruptions;

[0062] Figure 8 This is an image of the X1.0 solar flare event that occurred on October 2, 2022 at 19:53-20:34;

[0063] Fig. 9 This is the height distribution characteristic diagram of the inverted particle number density in the flare region in the photosphere and chromosphere;

[0064] Fig.10 This is the characteristic diagram of the height distribution of the temperature inverted in the flare area in the photosphere and chromosphere;

[0065] Fig.11 This is a characteristic diagram of the height distribution of the inverted plasma motion velocity in the flare region in the photosphere and chromosphere. DETAILED DESCRIPTION

[0066] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0067] like Figure 1 FIG. 1 is a flow chart of the solar flare spectrum inversion method of the present invention, comprising the following steps:

[0068] S1: According to the RADYN numerical simulation algorithm, a simulated solar flare event data set is obtained;

[0069] S2: preprocessing the simulated solar flare event data set to obtain preprocessed spectral data and smoothed physical parameter height distribution data;

[0070] S3: Establish a fully connected network and use the preprocessed spectral data and smoothed physical parameter height distribution data for training to obtain the flare spectrum inversion model;

[0071] S4: According to the RADYN numerical simulation algorithm, simulated spectral data is obtained, and the simulated spectral data is input into the flare spectrum inversion model for simulation test, and it is judged whether the error index is less than the set error threshold. If so, it goes to S5, otherwise, it returns to S3 and retrains to obtain the flare spectrum inversion model;

[0072] S5: According to the solar exploration satellite, actual spectrum data is obtained, and the actual spectrum data is input into the flare spectrum inversion model for actual testing, and it is determined whether the physical parameter height distribution data output by the flare spectrum inversion model is consistent with the actual physical parameter height distribution data. If so, enter S6, otherwise, return to S3 and retrain to obtain the flare spectrum inversion model;

[0073] S6: Input the measured spectral data of the solar flare event into the flare spectrum inversion model for processing, obtain the height distribution data of the physical parameters of the solar flare event, and complete the inversion of the solar flare spectrum line.

[0074] In one embodiment of the present invention, RADYN is a numerical simulation algorithm based on the radiation fluid dynamics method, and its full name is Radiative Hydrodynamics Code. It was originally developed by Carlsson et al. and applied to helioseismology research. Subsequently, this model was further applied to calculate the flare response problem in the solar chromosphere. The Hα spectral line is one of the brightest emission lines in the solar chromosphere, which can provide high-contrast images of solar active regions and is very suitable for observing and analyzing solar flares. By constructing the relationship between the Hα spectral line and the changes in these physical parameters during the solar flare, indirect measurement of the physical state of the chromosphere during the solar flare can be achieved.

[0075] In one embodiment of the present invention, in S1, according to the RADYN numerical simulation algorithm, a simulated solar flare event data set is obtained; wherein, the simulated solar flare event data set includes 79 simulated solar flare events with different parameter configurations, such as spectral index, cutoff energy, total energy and other parameters, each simulated solar flare event lasts for 50 seconds, the data recording interval is 0.1 second, and a total of 39,500 data are generated, each data contains Hα spectral data and height distribution data of corresponding physical parameters, such as particle number density, temperature, plasma movement speed and other physical parameters.

[0076] like Figure 2 Shown are data from the RADYN simulated solar flare event dataset; Figure 2 The upper left picture is the RADYN spectrum data, and the horizontal axis Wavelength is the wavelength, and the unit is The vertical axis Intensity is the normalized intensity; the upper right picture is the RADYN particle number density distribution data, the horizontal axis Height is the height, the unit is km, and the vertical axis Density is the particle number density, the unit is / cm 3 ; The lower left picture is the RADYN temperature distribution data, the horizontal axis Height is the height, the unit is km, the vertical axis Temperature is the temperature, the unit is K; the lower right picture is the RADYN plasma motion velocity distribution data, the horizontal axis Height is the height, the unit is km, the vertical axis Velocity is the plasma motion velocity, the unit is km / s.

[0077] In one embodiment of the present invention, the specific steps of S2 are as follows:

[0078] S201: Acquire original spectral data and physical parameter height distribution data in a simulated solar flare event data set;

[0079] S202: performing point spread function convolution on the original spectral data to obtain spectral data with a resolution consistent with the measured spectral data, so that it can simulate the resolution of the spectrometer of a solar exploration satellite, such as CHASE;

[0080] S203: Based on the spectral data whose resolution is consistent with the measured spectral data, interpolation and normalization are performed within a set wavelength range to obtain pre-processed spectral data; wherein the set wavelength range is selected is the length symbol, angstrom; the pre-processed spectral data matches the CHASE "Xihe" data and has 64 wavelength points;

[0081] S204: Perform interpolation processing on the physical parameter height distribution data to obtain smooth physical parameter height distribution data, wherein the physical parameter height distribution data includes physical parameters such as particle number density, temperature, and plasma movement speed. 400 equally spaced points within the height range of 0-2000km are selected to generate smooth physical parameter height distribution data.

[0082] In one embodiment of the present invention, Figure 3 As shown in FIG. 1 , it is a fully connected network structure diagram constructed by the present invention, where n [0] is the input layer, n [1] is the first hidden layer, n [2] is the second hidden layer, n [3] is the third hidden layer, n [4] is the fourth hidden layer, n [5] is the fifth hidden layer, n [6] is the output layer, Figure 2 In the figure, 64 represents 64-dimensional data, 128 represents 128-dimensional data, 256 represents 256-dimensional data, 512 represents 512-dimensional data, 1024 represents 1024-dimensional data, and 400 represents 400-dimensional data. The fully connected network includes: an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a fifth hidden layer, and an output layer connected in sequence; the first hidden layer, the second hidden layer, the third hidden layer, the fourth hidden layer, and the fifth hidden layer are all connected in sequence to an activation layer and a dropout layer;

[0083] The input layer is used to receive 64-dimensional spectral data. Each input data point represents the Hα spectral intensity observed at a specific wavelength. The main task of the input layer is to introduce spectral data into the fully connected network and provide raw information for subsequent layers.

[0084] The first hidden layer is used to map the 64-dimensional spectral data to a 128-dimensional space through linear transformation and perform feature extraction to obtain 128-dimensional data and the first feature. The linear transformation increases the dimension of the data and expands the feature space, which can better identify potential complex patterns.

[0085] The second hidden layer is used to map the 128-dimensional data to a 256-dimensional space through linear transformation and perform feature extraction to obtain 128-dimensional data and the second feature. The dimension of the data is increased through linear transformation, further expanding the feature space.

[0086] The third hidden layer is used to map the 256-dimensional data to a 512-dimensional space through linear transformation and perform feature extraction to obtain 512-dimensional data and the third feature. The linear transformation feature extraction is continued to learn more complex and abstract representations from the spectral data.

[0087] The fourth hidden layer is used to map the 512-dimensional data to a 1024-dimensional space through a linear transformation to obtain 1024-dimensional data, and integrate the first feature, the second feature, and the third feature to obtain an integrated feature;

[0088] The fifth hidden layer is used to compress the 1024-dimensional data into a 512-dimensional space through linear transformation to obtain 512-dimensional spectral data, compressing the dimension to 512 dimensions to prepare for the output layer. The fifth hidden layer compresses the features to ensure that the output layer can obtain the most representative features;

[0089] The output layer is used to compress the 512-dimensional data into a 400-dimensional space to obtain 400-dimensional output data, where the 400-dimensional output data is the inverted physical parameter height distribution data. The output layer of the model finally compresses the 512-dimensional data into 400 dimensions, including the height distribution of physical parameters such as particle number density, temperature, and plasma motion speed, which directly corresponds to the inverted physical parameter distribution, reflecting the final interpretation ability of the fully connected network for the input spectral data;

[0090] The activation layer is used to introduce nonlinear characteristics by suppressing negative values ​​and retaining positive values. The activation layer is used after each hidden layer to improve the expressiveness of the fully connected network, which helps to improve the training speed and model performance.

[0091] The dropout layer is used to randomly drop some neurons. The dropout rate is initially set to 20%. Adding a dropout layer after each hidden layer can prevent overfitting of the fully connected network, making the fully connected network more robust during training, avoiding overfitting of training data, and improving the generalization ability on actual data.

[0092] In one embodiment of the present invention, the specific steps of S3 are as follows:

[0093] S301: Establish a fully connected network, where the fully connected network includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a fifth hidden layer and an output layer;

[0094] S302: inputting the preprocessed spectral data into a fully connected network, and outputting inverted physical parameter height distribution data;

[0095] S303: Calculating loss values ​​of the inverted physical parameter height distribution data and the smoothed physical parameter height distribution data according to the Huber loss function;

[0096] S304: According to the loss value, the gradient of the loss value with respect to the weight of each layer is calculated using the back propagation algorithm;

[0097] S305: Using the Adam optimizer to update the model parameters of the fully connected network according to the gradient of the loss value to the weights of each layer;

[0098] S306: Determine whether the number of iterations reaches the set number of iterations. If so, complete the training of the fully connected network to obtain the flare spectrum inversion model. Otherwise, return to S302.

[0099] Among them, the Huber loss function has the advantages of both mean square error MSE and mean absolute error MAE. By introducing a threshold δ, it is expressed as MSE when the prediction error is small, and converted to MAE when the error is large, so that it has stronger robustness to outliers. In the solar flare spectral line inversion method of the present invention, the Huber loss function is selected, which can effectively reduce the adverse effects of extreme abnormal data on model training while ensuring that the fully connected network performs high-precision inversion on normal data. Since the inversion of the solar flare Hα spectral line involves a large amount of complex high-dimensional spectral data, which may contain noise or extreme conditions, the Huber loss function can be more robust than MSE when processing such data through its smooth loss curve, avoiding the problem of overfitting or model performance degradation that may be caused by the simple use of MSE. Therefore, the use of the Huber loss function in the present invention helps to improve the generalization ability of the model, ensure the accuracy and stability of the inversion results under different conditions, and enhance the ability to accurately invert physical parameters during solar flares.

[0100] The Adam optimizer combines the advantages of momentum and adaptive learning rate, and can efficiently find the optimal solution in a high-dimensional parameter space. The present invention uses the Adam optimizer to improve the convergence speed and stability of training. The Adam optimizer can set the initial learning rate and use a learning rate decay mechanism.

[0101] During the training process, you can improve model performance by adjusting hyperparameters such as learning rate, batch size, dropout rate, etc.

[0102] In one embodiment of the present invention, the specific steps of S4 are as follows:

[0103] S401: According to the RADYN numerical simulation algorithm, a simulated solar flare event data set is obtained, and simulated spectral data and simulated physical parameter height distribution data are randomly selected, wherein the simulated spectral data include spectral data before the flare eruption, spectral data during the eruption, and spectral data at different development stages, so as to comprehensively cover different physical states.

[0104] S402: preprocessing the simulated spectrum data and inputting it into the flare spectrum inversion model, outputting the inverted physical parameter height distribution data, including physical parameters such as particle number density, temperature and plasma motion speed;

[0105] S403: Calculate an error index based on the inverted physical parameter height distribution data and the simulated physical parameter height distribution data, focusing on the physical parameter distribution differences at various heights;

[0106] S404: Determine whether the error index is less than the set error threshold. If so, the flare spectrum inversion model has excellent performance. Complete the simulation test of the flare spectrum inversion model and enter S5. Otherwise, optimize the model parameters of the flare spectrum inversion model, adjust the preprocessed spectrum data and the smoothed physical parameter height distribution data, and return to S3 to retrain the flare spectrum inversion model. Among them, the mean square error can be used as an error index, and the possible deficiencies of the model can be identified by analyzing the weights and activations of the flare spectrum inversion model layer by layer. The performance of the flare spectrum inversion model can be improved by adjusting the model hyperparameters, improving the data preprocessing method, and adjusting the diversity of the training set.

[0107] like Figure 4 As shown in the figure, it is the model test result of the simulation test before the solar flare eruption. Figure 5 The following are the results of the model test for the simulation test of the solar flare. The upper left figure is the RADYN simulation spectrum, and the horizontal axis Wavelength is the wavelength, and the unit is The vertical axis Intensity is the normalized intensity; the upper right figure is the particle number density distribution diagram, the horizontal axis Height is the height, the unit is km, and the vertical axis Density is the particle number density, the unit is / cm 3; The lower left figures are all temperature distribution diagrams, with the horizontal axis Height being the height in km and the vertical axis Temp being the temperature in K; the lower right figures are all plasma movement velocity distribution diagrams, with the horizontal axis Height being the height in km and the vertical axis Velocity being the plasma movement velocity in km / s; among them, the blue line Real is the physical parameter distribution in the RADYN simulation data, and the orange line Inversed is the physical parameter distribution inverted by the model.

[0108] In one embodiment of the present invention, the specific steps of S5 are as follows:

[0109] S501: Based on the solar exploration satellite, historical spectrum data is obtained, and the data is selected as actual spectrum data, wherein historical Hα spectrum data is obtained from China's "Xihe" solar exploration satellite;

[0110] S502: Preprocessing the actual spectrum data, and inputting it into the flare spectrum inversion model, and outputting the inverted physical parameter height distribution data;

[0111] S503: Compare the inverted physical parameter height distribution data with the known physical parameter height distribution data, and determine whether they are consistent. If so, the flare spectrum inversion model has excellent performance, complete the actual test of the flare spectrum inversion model, and enter S6; otherwise, adjust the preprocessed spectrum data and the smoothed physical parameter height distribution data, return to S3, and retrain to obtain the flare spectrum inversion model; wherein, the performance of the flare spectrum inversion model can be improved by adjusting the model hyperparameters, improving the data preprocessing method, and adjusting the diversity of the training set.

[0112] like Figure 6 As shown in Figure 2, the model test results of the actual test before the solar flare eruption are as follows: Figure 7 The following are the actual test results of the model in the solar flare. The upper left figure is the actual spectrum of CHASE, and the horizontal axis Wavelength is the wavelength in units of The vertical axis Intensity is the intensity, the unit is erg / s·cm 2 ; The upper right figure is the inverted particle number density distribution diagram, the horizontal axis Height is the height, the unit is km, the vertical axis Density is the particle number density, the unit is / cm 3 ; The lower left figures are all inverted temperature distribution maps, with the horizontal axis Height being the height in km and the vertical axis Temp being the temperature in K; the lower right figures are all inverted plasma motion velocity distribution maps, with the horizontal axis Height being the height in km and the vertical axis Velocity being the plasma motion velocity in km / s.

[0113] In one embodiment of the present invention, a representative solar flare event is selected as the research object, and a X1.0 solar flare event occurring at 19:53-20:34 on October 2, 2022 is selected, such as Figure 8 As shown, the horizontal coordinate Helioprojective Longitude (Solar-X) is the longitude in arc seconds, and the vertical coordinate Helioprojective Lantitude (Solar-Y) is the latitude in arc seconds; obtain the measured data, focusing on the peak moment of the flare and the spectral changes before and after the flare occurs; the specific steps of S6 are as follows:

[0114] S601: Acquire measured spectral data of a solar flare event; wherein, extract Hα spectral data during the solar flare event, covering observation data of the entire flare region;

[0115] S602: Input the measured spectral data into the flare spectrum inversion model, and output the inverted physical parameter height distribution data, including the distribution of physical parameters such as particle number density, temperature and plasma movement speed, which can reflect the physical changes in different regions of the solar atmosphere during the solar flare eruption.

[0116] S603: Based on the inverted physical parameter height distribution data, a three-dimensional stereogram of the flare area is constructed to display the spatial distribution characteristics of the physical parameters in the photosphere and chromosphere, provide an intuitive display of physical phenomena, facilitate the understanding of the flare mechanism and improve the forecasting ability, and complete the inversion of the solar flare spectrum.

[0117] like Fig. 9 As shown, this is the height distribution characteristic diagram of the inverted particle number density in the photosphere and chromosphere in the flare area. The unit of particle number density is / cm3. Ten height slices of 400km, 800km, 1200km, 1300km, 1400km, 1500km, 1600km, 1700km, 1800km and 1900km are selected.

[0118] like Fig.10 As shown, this is a characteristic diagram of the height distribution of the inverted temperature in the photosphere and chromosphere in the flare area. The unit of temperature is K, and ten height slices of 400km, 800km, 1200km, 1300km, 1400km, 1500km, 1600km, 1700km, 1800km and 1900km are selected.

[0119] like Fig.11As shown in the figure, this is the height distribution characteristic diagram of the inverted plasma motion speed in the photosphere and chromosphere in the flare area. The unit of plasma motion speed is km / s, and ten height slices of 400km, 800km, 1200km, 1300km, 1400km, 1500km, 1600km, 1700km, 1800km and 1900km are selected.

[0120] The beneficial effects of the present invention are as follows: the present invention adopts multiple hidden layers to form a fully connected network, and performs training and testing on simulated spectral data and actual spectral data generated by the RADYN simulation algorithm, which can fully mine the implicit information in the spectral data and complete the rapid inversion of physical parameters during solar flares. It has high computational efficiency and ensures high accuracy. Through the accuracy and high efficiency of solar activity forecasts, it can reduce the potential impact of solar flares on human society.

Claims

1. A solar flare spectral line inversion method, characterized in that: The following steps are involved: S1: According to the RADYN numerical simulation algorithm, a simulated solar flare event data set is obtained; S2: preprocessing the simulated solar flare event data set to obtain preprocessed spectral data and smoothed physical parameter height distribution data; S3: Establish a fully connected network and use the preprocessed spectral data and smoothed physical parameter height distribution data for training to obtain the flare spectrum inversion model; S4: According to the RADYN numerical simulation algorithm, simulated spectral data is obtained, and the simulated spectral data is input into the flare spectrum inversion model for simulation test, and it is judged whether the error index is less than the set error threshold. If so, it goes to S5, otherwise, it returns to S3 and retrains to obtain the flare spectrum inversion model; S5: According to the solar exploration satellite, actual spectrum data is obtained, and the actual spectrum data is input into the flare spectrum inversion model for actual testing, and it is determined whether the physical parameter height distribution data output by the flare spectrum inversion model is consistent with the actual physical parameter height distribution data. If so, enter S6, otherwise, return to S3 and retrain to obtain the flare spectrum inversion model; S6: Input the measured spectral data of the solar flare event into the flare spectrum inversion model for processing, obtain the height distribution data of the physical parameters of the solar flare event, and complete the inversion of the solar flare spectrum line.

2. The solar flare spectrum inversion method according to claim 1, characterized in that: The specific steps of S2 are as follows: S201: Acquire original spectral data and physical parameter height distribution data in a simulated solar flare event data set; S202: performing point spread function convolution on the original spectral data to obtain spectral data with a resolution consistent with the measured spectral data; S203: performing interpolation and normalization processing within a set wavelength range according to the spectral data having a resolution consistent with the measured spectral data, to obtain pre-processed spectral data; S204: performing interpolation processing according to the physical parameter height distribution data to obtain smooth physical parameter height distribution data.

3. The solar flare spectrum inversion method according to claim 1, characterized in that: The fully connected network comprises: an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a fifth hidden layer and an output layer connected in sequence; the first hidden layer, the second hidden layer, the third hidden layer, the fourth hidden layer and the fifth hidden layer are all connected in sequence to an activation layer and a discard layer; The input layer is used to receive 64-dimensional spectral data; The first hidden layer is used to map the 64-dimensional spectral data to a 128-dimensional space through a linear transformation and perform feature extraction to obtain 128-dimensional data and a first feature; The second hidden layer is used to map the 128-dimensional data to a 256-dimensional space through a linear transformation and perform feature extraction to obtain the 128-dimensional data and the second feature; The third hidden layer is used to map the 256-dimensional data to a 512-dimensional space through a linear transformation, and perform feature extraction to obtain the 512-dimensional data and the third feature; The fourth hidden layer is used to map the 512-dimensional data to a 1024-dimensional space through a linear transformation to obtain 1024-dimensional data, and integrate the first feature, the second feature and the third feature to obtain an integrated feature; The fifth hidden layer is used to compress the 1024-dimensional data into a 512-dimensional space through a linear transformation to obtain 512-dimensional spectral data; The output layer is used to compress the 512-dimensional data into a 400-dimensional space to obtain 400-dimensional output data, wherein the 400-dimensional output data is the inverted physical parameter height distribution data; The activation layer is used to introduce nonlinear characteristics by suppressing negative values ​​and retaining positive values; The dropout layer is used to randomly drop some neurons.

4. The solar flare spectrum inversion method according to claim 1, characterized in that: The specific steps of S3 are as follows: S301: Establishing a fully connected network, wherein the fully connected network includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a fifth hidden layer and an output layer; S302: inputting the preprocessed spectral data into a fully connected network, and outputting inverted physical parameter height distribution data; S303: Calculating loss values ​​of the inverted physical parameter height distribution data and the smoothed physical parameter height distribution data according to the Huber loss function; S304: According to the loss value, the gradient of the loss value with respect to the weight of each layer is calculated using the back propagation algorithm; S305: Using the Adam optimizer to update the model parameters of the fully connected network according to the gradient of the loss value to the weights of each layer; S306: Determine whether the number of iterations reaches the set number of iterations. If so, complete the training of the fully connected network to obtain the flare spectrum inversion model. Otherwise, return to S302.

5. The solar flare spectrum inversion method according to claim 1, characterized in that: The specific steps of S4 are as follows: S401: According to the RADYN numerical simulation algorithm, a simulated solar flare event data set is obtained, and simulated spectral data and simulated physical parameter height distribution data are randomly selected; S402: pre-processing the simulated spectrum data, and inputting it into the flare spectrum inversion model, and outputting the inverted physical parameter height distribution data; S403: Calculating an error index according to the inverted physical parameter height distribution data and the simulated physical parameter height distribution data; S404: Determine whether the error index is less than the set error threshold. If so, complete the simulation test of the flare spectrum inversion model and enter S5. Otherwise, optimize the model parameters of the flare spectrum inversion model, adjust the preprocessed spectral data and the smoothed physical parameter height distribution data, and return to S3 to retrain and obtain the flare spectrum inversion model.

6. The solar flare spectrum inversion method according to claim 1, characterized in that: The specific steps of S5 are as follows: S501: Based on the solar exploration satellite, historical spectrum data is obtained, and the historical data is selected as actual spectrum data; S502: Preprocessing the actual spectrum data, and inputting it into the flare spectrum inversion model, and outputting the inverted physical parameter height distribution data; S503: Compare the inverted physical parameter height distribution data with the known physical parameter height distribution data, and determine whether they are consistent. If so, complete the actual test of the flare spectrum inversion model and enter S6. Otherwise, adjust the preprocessed spectral data and the smoothed physical parameter height distribution data, return to S3, and retrain to obtain the flare spectrum inversion model.

7. The solar flare spectrum inversion method according to claim 1, characterized in that: The specific steps of S6 are as follows: S601: Acquire measured spectrum data of solar flare events; S602: Input the measured spectrum data into the flare spectrum inversion model, and output the inverted physical parameter height distribution data; S603: Construct a three-dimensional stereogram of the flare area based on the inverted physical parameter height distribution data to complete the inversion of the solar flare spectrum.

Images