Medium and low latitude ionosphere space-time prediction method and system during strong magnetic storm

Abstract

The invention discloses a medium-low latitude ionosphere space-time prediction method and system during a strong magnetic storm, and the method comprises the steps: fusing ground-based GNSS and air-based COSMIC occultation ionosphere observation data, employing a multi-mode space-time sequence ionosphere model, integrating the advantages of a time convolution neural network and a Transform model through a cross attention mechanism, constructing a dual-space-time-flow TransTCN-XA prediction model, and carrying out the prediction of the medium-low latitude ionosphere through a time convolution neural network. And the spatial-temporal change of the medium-low latitude ionosphere during the strong magnetic storm period can be effectively captured. By adopting the technical scheme of the invention, higher resolution and more accurate ionosphere forecasting can be provided for the fields of space weather forecasting and navigation positioning, so that the accuracy and reliability of forecasting are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of satellite navigation, and particularly relates to a method and system for spatio-temporal prediction of the mid-low latitude ionosphere during strong magnetic storms. Background Art

[0002] The ionosphere is an important part of the solar-terrestrial space environment. Especially in the mid-low latitude regions, the combined effects of the equatorial electric field, geomagnetic field, neutral wind field and other factors are more obvious than in other regions. The dynamic changes of the mid-low latitude ionosphere are intense and complex, showing significant high-frequency characteristics on spatio-temporal scales. The dynamic changes of the ionosphere directly affect the stable operation of systems such as short-wave communication, satellite navigation, and energy network security. During strong magnetic storms, the solar wind energy strongly perturbs the ionosphere structure through the magnetosphere-ionosphere bidirectional coupling, resulting in phenomena such as total electron content (TEC) anomalies and equatorial plasma bubbles (EPBs), seriously reducing the navigation and positioning accuracy, and even causing signal interruption. The mid-low latitude regions have complex geographical features, gathering 70% of the global population and the core sea and air routes, and facing double challenges due to the significant spatio-temporal heterogeneity of the ionosphere in this region. Existing ionosphere prediction models mostly rely on a single data source. Although ground-based GNSS observations can provide high-precision TEC data, they are limited by the uneven distribution of stations and are difficult to cover the ocean and remote areas; although space-based occultation observations (such as COSMIC satellites) have global detection capabilities, due to the lack of a multi-source data fusion mechanism in traditional models, it is difficult to accurately depict the spatio-temporal evolution. In addition, during strong magnetic storms, the existing methods have insufficient resolution and accuracy in ionosphere modeling, resulting in significant prediction deviations for the abnormal ionosphere structure.

[0003] Current ionosphere prediction models based on deep learning mainly rely on architectures such as convolutional neural networks (CNNs), recurrent neural networks (RNNs) and their variants (such as LSTMs, GRUs), but there are significant defects: CNNs are difficult to model long-term dependencies in time series, RNN-based models are vulnerable to the problem of vanishing gradients, and both usually decouple spatio-temporal features and ignore the spatio-temporal coupling characteristics of ionospheric perturbations. In recent years, Transformer has shown advantages in global dependency modeling with its self-attention mechanism, but its computational complexity is high and it is sensitive to local time dynamics. Temporal convolutional networks (TCNs) can capture long-range temporal dependencies through dilated convolutions, but lack the ability to interact with spatial features. In addition, existing ionosphere prediction models have not fully utilized space-based and ground-based data sources and have not fully explored the cross-modal correlations of the ionosphere. For example, international empirical models (IRI / Nequick) are restricted by the parameterization framework and have insufficient spatio-temporal resolution, making it difficult to analyze the rapidly evolving three-dimensional electron density structure; the global ionosphere map (GIM) lags in response to space weather events, etc. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a spatio-temporal prediction method and system for the mid-latitude and low-latitude ionosphere during strong magnetic storms, which can improve the prediction performance through a multi-level spatio-temporal feature coupling and cross-source interaction mechanism, and combine a gating mechanism to integrate the local time series modeling of TCN and the global context awareness of Transformer to accurately analyze the multi-scale evolution laws of each phase of magnetic storms; by integrating the fine-grained time series modeling of TCN and the global spatial correlation advantages of Transformer, and using space-ground data fusion to significantly improve the spatio-temporal prediction accuracy and physical interpretability of ionospheric disturbances during strong magnetic storms, so as to provide an innovative solution for mid-latitude and low-latitude space weather warning and navigation positioning performance improvement.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A spatio-temporal prediction method for the mid-latitude and low-latitude ionosphere during strong magnetic storms, comprising:

[0007] Step 1: Obtain the original observation files of GNSS observation stations in the mid-latitude and low-latitude regions during strong magnetic storms, COSMIC occultation ionospheric profile data files, and geomagnetic and solar activity indices.

[0008] Step 2: Estimate the ionospheric delay through the original observation files of GNSS observation stations to obtain the ground-based GNSS-based VTEC, and extract the occultation VTEC from the COSMIC ionospheric profile file.

[0009] Step 3: Perform time resampling and spatial Kriging interpolation on the ionospheric VTEC data of each observation system; at the same time, interpolate the geomagnetic and solar activity indices.

[0010] Step 4: Construct a multi-modal ionospheric spatio-temporal prediction model for the TranTCN-XA model during strong magnetic storms.

[0011] Step 5: Divide the data set into a training set, a validation set, and a test set, and use the solar radiation flux F10.7, the geomagnetic activity index Dst, the solar wind SW, and the sunspot number as the model channel features, and VTEC as the label.

[0012] Step 6: Use the training set to train the model to construct the model weight parameters, use the validation set data to perform grid hyperparameter tuning on the model to achieve the best effect, and finally perform spatio-temporal prediction on the test set.

[0013] Preferably, in Step 2, the ionospheric delay is estimated through the observation files of the dual-frequency GNSS receiver, and the total ionospheric electron content on the path of the satellite signal to the ground GNSS station is obtained, which is called the tilted TEC. After projection conversion, the VTEC is obtained, and the calculation formula is as follows:

[0014]

[0015] where ΔDCB = ΔDCB sat + ΔDCB rec is the differential code bias correction value between the satellite and the receiver; f1 and f2 are two carrier frequencies of the GNSS signal; p1 and p2 are dual-frequency pseudorange observations; 40.3×10 16 is the ionospheric delay constant; R E is the radius of the Earth; θ is the zenith angle of the satellite relative to the GNSS station, and 350 is the ionospheric pierce point.

[0016] Preferably, in step three, the cubic Hermite interpolation polynomial PCHIP interpolation is used to interpolate the geomagnetic solar activity index.

[0017] Preferably, in step three, the VTEC in the ground-based GNSS-VTEC and space-based COSMIC occultation ionospheric data is resampled at a 30-minute resolution, and a two-dimensional ionospheric map with a resolution of 4°×2° at 40 degrees north and south latitudes is obtained through Kriging interpolation.

[0018] Preferably, in step four, a TCN network is deployed to capture the global spatial characteristics of the ionosphere in the sequence data, and the Transformer architecture is combined to identify the long-term dependent temporal characteristics of the ionosphere in the sequence; meanwhile, a cross-attention mechanism is introduced to achieve the efficient fusion of spatio-temporal characteristics in the network.

[0019] The present invention also provides a spatio-temporal prediction system for the mid-low latitude ionosphere during strong magnetic storms, including:

[0020] A first processing module for obtaining the original observation file of the mid-low latitude GNSS observation station, the COSMIC occultation ionospheric profile data file, and the geomagnetic solar activity index during strong magnetic storms;

[0021] A second processing module for estimating the ionospheric delay through the original observation file of the GNSS observation station to obtain the VTEC based on ground-based GNSS, and extracting the occultation VTEC from the ionospheric profile file of COSMIC;

[0022] A third processing module for performing time resampling and spatial Kriging interpolation on the ionospheric VTEC data of each observation system; meanwhile, interpolating the geomagnetic solar activity index;

[0023] A fourth processing module for constructing a multi-modal ionospheric spatio-temporal prediction model for the TranTCN-XA model during strong magnetic storms;

[0024] The fifth processing module is used to divide the dataset into a training set, a validation set, and a test set, and use the solar radiation flux F10.7, the geomagnetic activity index Dst, the solar wind SW, and the sunspot number as model channel features, and VTEC as the label.

[0025] The sixth processing module is used to use the training set for model training to construct model weight parameters, use the validation set data to perform grid hyperparameter tuning on the model to achieve the best effect, and finally perform spatio-temporal prediction on the test set.

[0026] Preferably, the second processing module estimates the ionospheric delay through the observation file of the dual-frequency GNSS receiver, calculates the total ionospheric electron content on the path of the satellite signal to the ground GNSS station, which is called the tilted TEC, and obtains the VTEC through projection conversion. The calculation formula is as follows:

[0027]

[0028] where, ΔDCB = ΔDCB sat +ΔDCB rec is the differential code bias correction value of the satellite and the receiver; f1 and f2 are the two carrier frequencies of the GNSS signal; p1 and p2 are the dual-frequency pseudorange observations; 40.3×10 16 is the ionospheric delay constant; R E is the radius of the earth; θ is the zenith angle of the satellite relative to the GNSS station, and 350 is the ionospheric piercing point.

[0029] Preferably, the third processing module uses the cubic Hermite interpolation polynomial PCHIP interpolation to interpolate the geomagnetic and solar activity indices.

[0030] Preferably, the third processing module resamples the VTEC in the ground-based GNSS-VTEC and space-based COSMIC occultation ionospheric data at a 30-minute resolution, and obtains a two-dimensional ionospheric map of 4°×2° at 40 degrees north and south latitudes through Kriging interpolation.

[0031] Preferably, the fourth processing module deploys a TCN network to capture the global spatial characteristics of the ionosphere in the sequence data, combines the Transformer architecture to identify the long-term dependent temporal characteristics of the ionosphere in the sequence; at the same time, a cross-attention mechanism is introduced to achieve the efficient fusion of spatio-temporal characteristics in the network.

[0032] The present invention integrates ground-based GNSS and space-based COSMIC occultation ionospheric observation data. Relying on a deep learning framework, a multi-modal spatio-temporal sequence ionospheric model is adopted. By means of a cross-attention mechanism (CrossAttention), the advantages of a temporal convolutional neural network (TCN) and a Transformer model are integrated to construct a TranTCN-XA prediction model with dual spatio-temporal streams, which can effectively capture the spatio-temporal variations of the mid-latitude and low-latitude ionosphere during strong magnetic storms. Compared with traditional ionospheric models, the present invention can provide higher resolution and more accurate ionospheric forecasts for the fields of space weather forecasting and navigation positioning, thus significantly improving the accuracy and reliability of predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to the provided drawings.

[0034] Figure 1 It is a flow chart of the spatio-temporal prediction method for the mid-latitude and low-latitude ionosphere during strong magnetic storms in the embodiments of the present invention;

[0035] Figure 2 It is the distribution of ground-based GNSS observation stations and space-based COSMIC occultation observations in the mid-latitude and low-latitude regions;

[0036] Figure 3 It is the analysis of the spatial prediction performance of the TranTCN-XA model in a strong magnetic storm event;

[0037] Figure 4 It is the error distribution histogram and correlation scatter distribution of the TranTCN-XA model;

[0038] Figure 5 It is the analysis of the temporal prediction performance of the TranTCN-XA model in the mid-latitude and low-latitude regions under the background of a strong magnetic storm event. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0040] To make the above objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] Embodiment 1:

[0042] As Figure 1 shown, the embodiment of the present invention provides a spatio-temporal prediction method for the mid-low latitude ionosphere during strong geomagnetic storms, including:

[0043] Step 1: Obtain the original observation files of mid-low latitude GNSS observation stations, COSMIC ionospheric observation data files, and geomagnetic and solar activity indices (solar radiation flux F10.7, geomagnetic activity index Dst, solar wind SW, and sunspot number) during strong geomagnetic storms from 2015 to 2023. The purpose of this step is to obtain data sources. The original observation files of mid-low latitude GNSS observation stations are downloaded through the open-access Global International GNSS Service Center (https: / / igs.org / ), the COSMIC ionospheric observation data files are obtained from the National Science Foundation's Center for Atmospheric Research (NSF NCAR: https: / / www.cosmic.ucar.edu / ), and the geomagnetic and solar activity indices (solar radiation flux F10.7, sunspot number, geomagnetic activity index Dst, and solar wind SW) come from the NASA Goddard Space Flight Center's SPDF database (https: / / omniweb.gsfc.nasa.gov / ), with a time resolution of on average per hour. In particular, the solar radiation flux F10.7 is on average per day.

[0044] Step 2: Estimate the ionospheric delay through the original observation files of GNSS observation stations to obtain the ground-based GNSS-based VTEC, and extract the occultation VTEC (RO-VTEC) from the COSMIC ionospheric profile files. The COSMIC occultation VTEC data can be directly extracted from the ionospheric profile files downloaded from the NSF NCAR center. The ground-based GNSS-based VTEC is estimated by observing the ionospheric delay in the observation files of dual-frequency GNSS receivers, as shown in formula (1):

[0045]

[0046] where ΔDCB = ΔDCB sat +ΔDCB rec is the differential code bias correction value between the satellite and the receiver; f1 and f2 are the two carrier frequencies of the GNSS signal (e.g., GPS L1 = 1575.42 MHz, L2 = 1227.60 MHz); p1 and p2 are the dual-frequency pseudorange observations (unit: m); 40.3×10 16 is the ionospheric delay constant (unit: m3 / s 2 );R E is the radius of the Earth; θ is the zenith angle of the satellite relative to the GNSS station, and 350 is the ionospheric pierce point (km).

[0047] Step 3: Resample the ionospheric VTEC data of each observation system in time and perform spatial Kriging interpolation. At the same time, use the cubic Hermite interpolation polynomial (PCHIP) to interpolate the geomagnetic and solar activity indices to unify the spatio-temporal resolution. The resolution of ground-based GNSS-VTEC is 30 seconds, while the time resolution of COSMIC occultation observation data is not fixed. Since the ionosphere usually does not change significantly in a short period of time, in this example, the ground-based GNSS-VTEC is resampled to data with a resolution of 30 minutes, and based on the time of ground-based GNSS-VTEC, the RO-VTEC within 10 minutes before and after each moment is integrated into the mid- and low-latitude ionospheric VTEC data at one moment. Further, a two-dimensional ionospheric map with a longitude and latitude of 4°×2° in the mid- and low-latitudes is obtained through Kriging interpolation. Kriging interpolation needs to meet two assumptions: unbiased and optimal. The unbiasedness requires that the expected value of the predicted value of the interpolation is equal to the true value, as shown in the following formula (2):

[0048]

[0049] where is the predicted value (Kriging interpolation result) of the point x0 to be estimated; z i is the observed value (known value) of the known point x i ; λ i is the weight of the point x i (to be solved by a system of equations); N is the total number of known points participating in the interpolation. This means that there is no systematic bias in the predicted value. Through mathematical derivation, it can be obtained that the weight coefficient needs to satisfy the constraint of formula (3):

[0050]

[0051] where N is the total number of known points; λ i is the weight coefficient, satisfying the constraint condition of unbiasedness.

[0052] The optimality requires that the error variance between the predicted value and the true value is minimized, that is, formula (4), which ensures that the predicted value is as close as possible to the true value and the error is minimized.

[0053]

[0054] The key to the Kriging method lies in using the correlation between points in space, which is described by the semi-variance function. The definition of the semi-variance function is:[[]]

[0055]

[0056] Among them, x i and x j are any two points in space. z ( x i) , z ( x j) is the observed value of point x i mixed x j . γ ( x i , x j) describes the spatial correlation between point x i and x j . The smaller the semi-variance value, the stronger the correlation between points. Equation (6) is used to solve the weight coefficient λ i and the Lagrange multiplier ψ

[0057]

[0058] Among them, γ ( x i , x j) is the semi-variance value of point x i and point x j ; λ i is the weight coefficient. Ψ is the Lagrange multiplier, which is used to introduce the unbiasedness constraint condition.

[0059] The geomagnetic solar activity index adopts the piecewise cubic Hermite interpolation polynomial (PCHIP) interpolation method. To ensure the monotonicity and smoothness of the data, PCHIP constructs a cubic polynomial on each interval [x i , x i+1 , as shown in the following formula:

[0060] P i (x) = y i + d i (x - x i ) + c i (x - x i ) 2 + d i (x - x i ) 3 (7)

[0061] Among them, y i is the data point value, d i is the first derivative at node x i , c i and d i are polynomial coefficients, which are determined by the interpolation conditions and monotonicity constraints.

[0062] Step 4: By deploying the TCN network to capture the global spatial characteristics of the ionosphere in sequence data and combining the Transformer architecture to identify the long-term dependent temporal features of the ionosphere in the sequence, we adopted a parallel processing architecture to improve the efficiency of model training and inference. In addition, this technology also introduced a cross-attention mechanism to achieve efficient fusion of spatio-temporal features in the network, which not only allows the model to consider temporal and spatial relationships simultaneously, but also enables more accurate extraction of the key features of the spatio-temporal sequence data of the ionosphere, thereby enhancing the feature expression ability of the model and achieving more accurate prediction results.

[0063] (1) TCN time feature extraction (dilated causal convolution) is given by equation (8): For the input time series signal The output of the l-th layer of TCN is:

[0064]

[0065] where *d is the causal convolution with dilation rate d, ensuring irreversibility in the time direction. is the convolutional kernel weight, and the receptive field is enlarged by stacking layers (e.g., d = 2 l ). Specifically, capture the multi-scale time dependencies of ionospheric parameters (e.g., sudden disturbances d = 1, diurnal variations d = 24)

[0066] (2) Transformer spatial feature modeling (self-attention): For the spatial grid data (N is the number of spatial positions), the formula is as follows:

[0067]

[0068] where Q s = X s W Q , K s = X s W K , V s = X s W V , and the key point of the formula is to model the global spatial correlation (e.g., the interaction between the equatorial anomaly region and the polar region).

[0069] (3) Spatio-temporal cross-attention fusion: Using a two-way interaction formula, time-to-space (TCN features guiding spatial modeling) is as follows:

[0070]

[0071] Space-to-time (spatial features constraining time evolution) is as follows:

[0072]

[0073] The key point of the formula is to achieve two-way alignment of spatio-temporal features through dynamic weights. For example, model the propagation direction of ionospheric traveling disturbances (TIDs).

[0074] Step 5: Divide the dataset during 15 strong magnetic storms from 2015 to 2023 into a training set, a validation set, and a test set. Among them, the first 13 magnetic storm events are used as the training set, the 14th magnetic storm event is used as the validation set, and the 15th magnetic storm event is used as the test set. Solar radiation flux F10.7, sunspot number, geomagnetic activity index Dst, and solar wind SW are used as model channel features, and VTEC is used as the label.

[0075] Step 6: Use the ionospheric data of 13 super magnetic storm events in the training set for model training, construct model weight parameters, and use the validation set data to perform grid hyperparameter tuning on the model to achieve the best model effect. Finally, perform spatio-temporal prediction on the test set.

[0076] Step 7: Model performance evaluation. Systematically evaluate the model using root mean square error (RMSE), mean absolute error (MAE), and Pearson correlation coefficient (PCC). The expressions of the evaluation metrics are as follows:

[0077]

[0078] where, y i is the true value of VTEC, is the VTEC value predicted by the model.

[0079]

[0080] where, x i and y i are the observed values of two variables respectively. and are the average values of the two variables. PCC is the Pearson correlation coefficient.

[0081] As Figure 2 shown, the distribution of ground-based GNSS observation stations and COSMIC occultation observations in the mid-low latitudes. COSMIC occultation data effectively make up for the lack of GNSS ground observation stations in ocean areas and provide a rich data source for high spatio-temporal resolution ionospheric monitoring.

[0082] As Figure 3As shown, four typical geomagnetic storm moments during DOY307 to DOY311 were selected for the experimental case (DOY307(19UT) = 8 nT, DOY309(19UT) = -163 nT, DOY310(08UT) = -102 nT, DOY311(10UT) = -50 nT). Despite the strong geomagnetic storm interference on the ionosphere, the prediction results of the TranTCN-XA model are highly consistent with the true ionospheric data. From the residual distribution diagram ( Figure 3 (i)-(l)) shown, the prediction error distribution mostly concentrates within 1 TECu, and is similar to the test set error, with most errors within ±1.5 TECu. In particular, as Figure 3 (i) shows, overestimation also occurs during the quiet period of DOY307(19UT) in the low-latitude region. In addition, as Figure 3 (j) shows, at the moment of DOY309(19UT) during the strong geomagnetic storm period. The overestimation in some parts of the low-latitude region and underestimation at the mid-latitude edge of the model are obvious. Most of the errors are within -2 to 2 TECu, and these underestimations and overestimations are generally within ±4 TECu, and very few exceed ±5 TECu. It shows that the model still has excellent prediction ability under the background of strong geomagnetic storms.

[0083] As Figure 4 shown, at the four typical geomagnetic storm moments (DOY307-19UT, DOY309-19UT, DOY310-08UT, DOY311-10UT) in the case. The error distribution histograms (such as Figure 4 (a), (c), (e), (g)) show that the absolute value of the error of more than 80% of the samples ≤ 1 TECu, and the overall distribution range is [-2, +2]; among them, the accuracy of DOY310-08UT is the best (MAE = 0.70, RMSE = 1.02), while the error of DOY309-19UT is the largest (MAE = 1.19, RMSE = 1.66), but still meets the monitoring requirements. The proportion of outliers with an absolute error greater than 5 TECu < 0.5%, and the results of the correlation analysis (such as Figure 4 (b), (d), (f), (h)) show that the PCC = 0.999 at all moments, the normalized absolute error < 0.2, and the data points are closely distributed near the fitting line. In addition, the error distribution of the test set is more approaching normal, and the peak shifts towards 0 TECu, confirming that the model has more stable prediction characteristics in the conventional space environment. This result systematically verifies two aspects of the advantages of the model in complex ionospheric perturbation scenarios: the guarantee of high resolution and high precision, and the generalization ability across different geomagnetic storm levels.

[0084] As Figure 5As shown, the TranTCN-XA model can effectively capture the temporal variation characteristics of the total electron content (VTEC) in the ionosphere of the mid-latitude and low-latitude regions, and the prediction curve is highly consistent with the true value. The overall performance indicators show that the minimum values of the model's MAE and RMSE reach 0.86 TECu and 1.03 TECu respectively, and the PCC can reach as high as 0.999, indicating that it has strong robustness in temporal prediction under extreme space weather events. Specifically, the model performs particularly well during the prediction process, and the prediction curve almost completely coincides with the true value.

[0085] The present invention realizes high-precision spatio-temporal prediction of the mid-latitude and low-latitude ionosphere during strong magnetic storms by fusing ground-based GNSS and space-based COSMIC occultation heterogeneous data and combining the spatio-temporal modeling advantages of deep learning. Compared with traditional methods, its innovative advantages are reflected in the following three aspects:

[0086] (1) Multi-source data fusion enhances spatial resolution: By jointly using the complementary observations of ground-based GNSS-VTEC and space-based occultation RO-VTEC, the spatial coverage limitation of a single data source is broken through, and the spatial resolution of ionospheric modeling is improved to the order of 4°×2° (longitude-latitude grid).

[0087] (2) Dual spatio-temporal flow model improves prediction accuracy: The TranTCN-XA prediction model proposed in the present invention uses a temporal convolutional network to accurately capture the hourly disturbance characteristics of the ionosphere, and realizes mid-latitude and local spatio-temporal coupling modeling with the help of a cross-attention mechanism. During strong magnetic storms, the model achieves high-precision prediction results, with a root mean square error (RMSE) of less than 1.5 TECu, a Pearson correlation coefficient (PCC) as high as 0.999, and a mean absolute error (MAE) of about 1 TECu. It is proved that the TranTCN-XA dual spatio-temporal flow model can significantly improve the prediction accuracy:

[0088] (3) Near-real-time forecasting supports application effectiveness: Provide spatio-temporally continuous high-resolution ionospheric TEC field forecasting products (update cycle ≤ 0.3 hours), which can be synchronously served in engineering scenarios such as space weather disturbance monitoring, satellite navigation ionospheric delay correction, and high-frequency communication link optimization.

[0089] Embodiment 2:

[0090] The embodiment of the present invention also provides a spatio-temporal prediction system for the mid-latitude and low-latitude ionosphere during strong magnetic storms, including:

[0091] A first processing module for obtaining the original observation files of mid-latitude and low-latitude GNSS observation stations, COSMIC occultation ionospheric profile data files, and geomagnetic and solar activity indices during strong magnetic storms;

[0092] The second processing module is used to estimate the ionospheric delay through the original observation files of GNSS observation stations, obtain the ground-based GNSS-based VTEC, and extract the occultation VTEC from the ionospheric profile files of COSMIC;

[0093] The third processing module is used to perform time resampling and spatial Kriging interpolation on the ionospheric VTEC data of each observation system; at the same time, interpolate the geomagnetic and solar activity indices;

[0094] The fourth processing module is used to construct a multimodal ionospheric spatio-temporal prediction model for the TranTCN-XA model during strong magnetic storms;

[0095] The fifth processing module is used to divide the data set into a training set, a validation set, and a test set, and use the solar radiation flux F10.7, the geomagnetic activity index Dst, the solar wind SW, and the sunspot number as model channel features, and VTEC as the label;

[0096] The sixth processing module is used to use the training set for model training to construct model weight parameters, use the validation set data to perform grid hyperparameter tuning on the model to achieve the best effect, and finally perform spatio-temporal prediction on the test set.

[0097] As an implementation manner of the embodiment of the present invention, the second processing module estimates the ionospheric delay through the observation files of the dual-frequency GNSS receiver, calculates the total ionospheric electron content on the path from the satellite signal to the ground GNSS station, which is called the tilted TEC, and obtains the VTEC through projection conversion. The calculation formula is as follows:

[0098]

[0099] where, ΔDCB = ΔDCB sat +ΔDCB rec is the differential code bias correction value between the satellite and the receiver; f1 and f2 are the two carrier frequencies of the GNSS signal; p1 and p2 are the dual-frequency pseudorange observations; 40.3×10 16 is the ionospheric delay constant; R E is the radius of the earth; θ is the zenith angle of the satellite relative to the GNSS station, and 350 is the ionospheric pierce point.

[0100] As an implementation manner of the embodiment of the present invention, the third processing module interpolates the geomagnetic and solar activity indices using the cubic Hermite interpolation polynomial PCHIP.

[0101] As an implementation manner of an embodiment of the present invention, the third processing module resamples the VTEC in the ground-based GNSS-VTEC and space-based COSMIC occultation ionospheric data at a resolution of 30 minutes, and obtains a two-dimensional ionospheric map with a size of 4°×2° at 40 degrees north and south latitudes through Kriging interpolation.

[0102] As an implementation manner of an embodiment of the present invention, the fourth processing module captures the global spatial characteristics of the ionosphere in the sequence data by deploying a TCN network, and combines the Transformer architecture to identify the long-term dependence time series characteristics of the ionosphere in the sequence; at the same time, a cross-attention mechanism is introduced to achieve efficient fusion of spatio-temporal characteristics in the network.

[0103] The above-described embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A spatio-temporal prediction method for the mid-latitude and low-latitude ionosphere during a strong geomagnetic storm, characterized in that, including: Step 1: Obtain the original observation files of mid - low latitude GNSS observation stations, COSMIC occultation ionospheric profile data files, and geomagnetic and solar activity indices during strong magnetic storms; Step 2: Estimate the ionospheric delay through the original observation files of GNSS observation stations to obtain ground - based GNSS - based VTEC, and extract occultation VTEC from the ionospheric profile files of COSMIC; Step 3: Perform time resampling and spatial Kriging interpolation on the ionospheric VTEC data of each observation system; at the same time, interpolate the geomagnetic and solar activity indices; Step 4: Construct a multi - modal ionospheric spatio - temporal prediction model for the TranTCN - XA model during strong magnetic storms; Step 5: Divide the data set into a training set, a validation set, and a test set, and use the solar radiation flux F10.7, geomagnetic activity index Dst, solar wind SW, and sunspot number as model channel features, and VTEC as the label; Step 6: Use the training set for model training to construct model weight parameters, use the validation set data to perform grid - based hyperparameter tuning on the model to achieve the best effect, and finally perform spatio - temporal prediction on the test set.

2. The spatio-temporal prediction method for the mid-low latitude ionosphere during strong geomagnetic storms according to claim 1, wherein, In Step 2, the ionospheric delay is estimated through the observation files of dual - frequency GNSS receivers to obtain the total ionospheric electron content on the path from the satellite signal to the ground GNSS station, which is called tilted TEC. After projection conversion, VTEC is obtained. The calculation formula is as follows: Among them, ΔDCB = ΔDCB sat +ΔDCB rec is the differential code bias correction value between the satellite and the receiver; f1 and f2 are the two carrier frequencies of the GNSS signal; p1 and p2 are the dual-frequency pseudorange observations; 40.3×10 16 is the ionospheric delay constant; R E is the radius of the Earth; θ is the zenith angle of the satellite relative to the GNSS station, and 350 is the ionospheric pierce point.

3. The spatio-temporal prediction method for the mid-low latitude ionosphere during strong geomagnetic storms according to claim 2, characterized in that, In Step 3, the cubic Hermite interpolation polynomial PCHIP interpolation is used to interpolate the geomagnetic and solar activity indices.

4. The spatio-temporal prediction method for the mid-low latitude ionosphere during a strong geomagnetic storm according to claim 3, wherein In Step 3, the VTEC in the ground - based GNSS - VTEC and space - based COSMIC occultation ionospheric data is resampled at a 30 - minute resolution, and a two - dimensional ionospheric map with a resolution of 4°×2° at 40 degrees north and south latitudes is obtained through Kriging interpolation.

5. The spatio-temporal prediction method of the mid-low latitude ionosphere during strong geomagnetic storms according to claim 3, characterized in that In Step 4, a TCN network is deployed to capture the global spatial characteristics of the ionosphere in sequence data, and the Transformer architecture is combined to identify the long - term dependent temporal characteristics of the ionosphere in the sequence; at the same time, a cross - attention mechanism is introduced to achieve efficient fusion of spatio - temporal features in the network.

6. A spatio-temporal prediction system for the mid-latitude and low-latitude ionosphere during strong geomagnetic storms, characterized in that, including: The first processing module is used to obtain the original observation files of mid - low latitude GNSS observation stations, COSMIC occultation ionospheric profile data files, and geomagnetic and solar activity indices during strong magnetic storms; The second processing module is used to estimate the ionospheric delay through the original observation files of GNSS observation stations to obtain ground - based GNSS - based VTEC, and extract occultation VTEC from the ionospheric profile files of COSMIC; The third processing module is used to perform time resampling and spatial Kriging interpolation on the ionospheric VTEC data of each observation system; at the same time, interpolate the geomagnetic and solar activity indices; The fourth processing module is used to construct a multi - modal ionospheric spatio - temporal prediction model for the TranTCN - XA model during strong magnetic storms; The fifth processing module is used to divide the data set into a training set, a validation set, and a test set, and use the solar radiation flux F10.7, the geomagnetic activity index Dst, the solar wind SW, and the sunspot number as model channel features, and VTEC as the label; The sixth processing module is used to use the training set for model training to build model weight parameters, use the validation set data to perform grid hyperparameter tuning on the model to achieve the best effect, and finally perform spatio-temporal prediction on the test set.

7. The spatio-temporal prediction system for the mid-low latitude ionosphere during strong geomagnetic storms according to claim 6, characterized in that, The second processing module estimates the ionospheric delay through the observation file of the dual-frequency GNSS receiver, and obtains the total electron content of the ionosphere on the path from the satellite signal to the ground GNSS station, which is called the inclined TEC. After projection conversion, VTEC is obtained. The calculation formula is as follows: where ΔDCB = ΔDCB sat + ΔDCB rec is the differential code bias correction value between the satellite and the receiver; f1 and f2 are two carrier frequencies of the GNSS signal; p1 and p2 are dual-frequency pseudorange observations; 40.3×10 16 is the ionospheric delay constant; R E is the radius of the Earth; θ is the zenith angle of the satellite relative to the GNSS station, and 350 is the ionospheric pierce point.

8. The spatio-temporal prediction system for the mid-low latitude ionosphere during strong geomagnetic storms according to claim 7, wherein The third processing module uses the cubic Hermite interpolation polynomial PCHIP to interpolate the geomagnetic and solar activity indices.

9. The spatio-temporal prediction system for the mid-low latitude ionosphere during strong geomagnetic storms according to claim 8, wherein, The third processing module resamples the VTEC in the ground-based GNSS-VTEC and space-based COSMIC occultation ionospheric data at a 30-minute resolution, and obtains a two-dimensional ionospheric map of 4°×2° at 40 degrees north and south latitudes through Kriging interpolation.

10. The spatio-temporal prediction system for the mid-low latitude ionosphere during strong geomagnetic storms according to claim 9, characterized in that, The fourth processing module deploys a TCN network to capture the global spatial characteristics of the ionosphere in the sequence data, and combines the Transformer architecture to identify the long-term dependent time series characteristics of the ionosphere in the sequence; at the same time, a cross-attention mechanism is introduced to achieve efficient fusion of spatio-temporal characteristics in the network.

Images