Adaptive ionospheric prediction method and system based on magnetic coordinate system and physical constraint

Abstract

This application provides an adaptive ionospheric prediction method based on magnetic coordinates and physical constraints, relating to the technical field of ionospheric prediction. The method includes: dividing the prediction area into magnetic coordinate grids; acquiring historical ionospheric and meteorological data for each magnetic coordinate grid and performing normalization processing; constructing sample data based on the normalized ionospheric and meteorological data; building a prediction model based on a CNNGRU module and a Transformer module, and training the model using the sample data; inputting the current data of each magnetic coordinate grid into the trained prediction model to generate ionospheric prediction results for each magnetic coordinate grid; aggregating the ionospheric prediction results of each magnetic coordinate grid and outputting them. By calculating magnetic coordinates and adding ionospheric physical constraints, the prediction model constructed using the CNNGRU and Transformer modules significantly improves the accuracy and stability of global predictions.

Description

Technical Field

[0001] This application relates to the technical field of ionospheric prediction, and in particular to an adaptive ionospheric prediction method and system based on magnetic coordinate system and physical constraints. Background Technology

[0002] The ionosphere is the ionized atmosphere at an altitude of 60-1000 km above the Earth. It is formed by the ionization of neutral atmospheric molecules / atoms due to solar extreme ultraviolet radiation, X-ray radiation, and bombardment by high-energy particles, and contains a large number of free electrons and ions. Its core characteristics include refraction, delay, scattering, and scintillation of radio waves, making it a major source of error and interference for satellite navigation (GNSS), shortwave communication, and radar detection. Its state is driven by solar activity (F10.7, sunspot number), geomagnetic activity (Kp, Dst index), diurnal / seasonal / latitude, and coupling with the lower atmosphere, exhibiting strong nonlinearity, multi-scale, and abrupt changes.

[0003] Current ionospheric prediction primarily employs physical models (such as IRI and NeQuick) and single machine learning models. Physical models, relying on empirical formulas, are ill-suited to sudden space weather events; while prediction methods based on deep learning such as LSTM and Transformer can capture complex nonlinear relationships, they suffer from significant drawbacks.

[0004] 1. Spatial heterogeneity is ignored: Traditional models show significant differences in prediction performance across different regions of the world (equator / polar / mid-latitude) and cannot adaptively adjust; 2. Discontinuous results: The lack of spatial smoothing constraints on discrete prediction points leads to boundary jumps in the global TEC distribution map. Summary of the Invention

[0005] Therefore, it is necessary to provide an adaptive ionospheric prediction method and system based on magnetic coordinates and physical constraints to address the above-mentioned technical problems. By calculating magnetic coordinates and adding ionospheric physical constraints, the ionospheric prediction model constructed using the CNNGRU module and the Transformer module is used to predict the ionospheric state, which significantly improves the accuracy and stability of global prediction.

[0006] Firstly, this application provides an adaptive ionospheric prediction method based on a magnetic coordinate system and physical constraints, including: Select the prediction region and divide the prediction region into magnetic coordinate grids; Historical ionospheric and meteorological data for each magnetic coordinate grid are obtained and normalized separately. Sample data are then constructed based on the normalized ionospheric and meteorological data. An ionospheric prediction model is constructed based on the CNNGRU module and the Transformer module. The ionospheric prediction model is trained using sample data to generate a target ionospheric prediction model. The current ionospheric data and meteorological data of each magnetic coordinate grid are input into the target ionospheric prediction model to generate the ionospheric prediction results for each magnetic coordinate grid. The ionospheric prediction results of each magnetic coordinate grid are then aggregated and output.

[0007] In one embodiment, dividing the prediction region into a magnetic coordinate grid includes: The prediction area is divided according to latitude and longitude. A regular grid is generated, and the center coordinates of each grid are determined. Calculate the magnetic latitude (MLAT) and magnetic local time (MLT) of the current grid based on the center coordinates of the grid, and generate the magnetic coordinates of the current grid.

[0008] In one embodiment, training an ionospheric prediction model using sample data includes: The sample data is input into the ionospheric prediction model, and the features of the ionospheric data and meteorological data are extracted by the CNNGRU module and then spliced ​​to construct time-series features. The temporal features are input into the Transformer module to generate ionospheric predictions. A loss function is constructed based on the predicted and actual values ​​of the ionosphere. The ionosphere prediction model is trained using the loss function, and the parameters of the ionosphere prediction model are adjusted until the loss function is minimized.

[0009] In one embodiment, the CNNGRU module is used to extract features from ionospheric data and meteorological data respectively, and then concatenated to construct time-series features, including: The features of ionospheric data and meteorological data are extracted separately using a CNN neural network, and the first spliced ​​feature is output. The first concatenated feature is input into a bidirectional GRU structure to generate two sets of hidden state vectors. The two sets of hidden state vectors are then merged to form a temporal feature.

[0010] In one embodiment, the loss function is formulated as follows:

[0011] in, L Here, RMSE is the loss function optimized for model training. N The total number of grid points, This represents the actual TEC value for the corresponding grid point. This represents the predicted TEC value for the corresponding grid point.

[0012] In one embodiment, an optimizer is used to adjust the parameters of the ionospheric prediction model:

[0013] in, These are the model parameters from the previous time step; The corrected first moment; The corrected second moment; The learning rate; It is the numerical stability constant; This is the weight decay coefficient.

[0014] In one embodiment, before inputting the time-series features into the Transformer module to generate ionospheric predictions, the method further includes: Add ionospheric physical constraints to time series features:

[0015]

[0016] in, Features of the output after adding ionospheric physical constraints; This refers to the weight of this layer; For bias; The F10.7 index change rate over 24 hours; The trend of the Kp index; It is the sigmoid activation function.

[0017] Secondly, this application also provides an adaptive ionospheric prediction system based on a magnetic coordinate system and physical constraints, comprising: The magnetic coordinate grid division module is used to select the prediction area and divide the prediction area into magnetic coordinate grids; The sample data construction module is used to acquire historical ionospheric and meteorological data for each magnetic coordinate grid, and to normalize the ionospheric and meteorological data respectively, and to construct sample data based on the normalized ionospheric and meteorological data. The model building module is used to build an ionospheric prediction model based on the CNNGRU module and the Transformer module, train the ionospheric prediction model using sample data, and generate a target ionospheric prediction model. The results output module is used to input the current ionospheric data and meteorological data of each magnetic coordinate grid into the target ionospheric prediction model, generate the ionospheric prediction results of each magnetic coordinate grid, aggregate the ionospheric prediction results of each magnetic coordinate grid and output them.

[0018] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps: Select the prediction region and divide the prediction region into magnetic coordinate grids; Historical ionospheric and meteorological data for each magnetic coordinate grid are obtained and normalized separately. Sample data are then constructed based on the normalized ionospheric and meteorological data. An ionospheric prediction model is constructed based on the CNNGRU module and the Transformer module. The ionospheric prediction model is trained using sample data to generate a target ionospheric prediction model. The current ionospheric data and meteorological data of each magnetic coordinate grid are input into the target ionospheric prediction model to generate the ionospheric prediction results for each magnetic coordinate grid. The ionospheric prediction results of each magnetic coordinate grid are then aggregated and output.

[0019] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps: Select the prediction region and divide the prediction region into magnetic coordinate grids; Historical ionospheric and meteorological data for each magnetic coordinate grid are obtained and normalized separately. Sample data are then constructed based on the normalized ionospheric and meteorological data. An ionospheric prediction model is constructed based on the CNNGRU module and the Transformer module. The ionospheric prediction model is trained using sample data to generate a target ionospheric prediction model. The current ionospheric data and meteorological data of each magnetic coordinate grid are input into the target ionospheric prediction model to generate the ionospheric prediction results for each magnetic coordinate grid. The ionospheric prediction results of each magnetic coordinate grid are then aggregated and output.

[0020] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps: Select the prediction region and divide the prediction region into magnetic coordinate grids; Historical ionospheric and meteorological data for each magnetic coordinate grid are obtained and normalized separately. Sample data are then constructed based on the normalized ionospheric and meteorological data. An ionospheric prediction model is constructed based on the CNNGRU module and the Transformer module. The ionospheric prediction model is trained using sample data to generate a target ionospheric prediction model. The current ionospheric data and meteorological data of each magnetic coordinate grid are input into the target ionospheric prediction model to generate the ionospheric prediction results for each magnetic coordinate grid. The ionospheric prediction results of each magnetic coordinate grid are then aggregated and output.

[0021] This application employs the aforementioned adaptive ionospheric prediction method and system based on magnetic coordinate system and physical constraints, which has the following beneficial effects: 1. This application introduces a bidirectional GRU module to explicitly model ionospheric state transitions, including using a forward GRU to learn the evolution from history to the future; using a backward GRU to learn the dependency from the future to the past and fusing the features output by the bidirectional GRU module, using the short / medium period features output by the GRU, and using the Transformer to focus on long-term dependencies, which helps to improve the accuracy of prediction. 2. This paper uses a magnetic coordinate system to directly correlate the particle sedimentation region (|MLAT|>60°) and the equatorial anomaly region (|MLAT|≤15°), which helps to reduce prediction bias; 3. Ionospheric prediction must follow space physics rules. For example, the TEC in the equatorial region at night is controlled by plasma bubbles, while ionospheric physical constraints can effectively suppress physical paradoxes through dynamic gating. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating an adaptive ionospheric prediction method based on magnetic coordinates and physical constraints in one embodiment. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0024] Firstly, referring to Figure 1 This application provides an adaptive ionospheric prediction method based on magnetic coordinate system and physical constraints, including: S100, Select the prediction area and divide the prediction area into magnetic coordinate grids.

[0025] In one embodiment, dividing the prediction region into magnetic coordinate grids includes: dividing the prediction region into latitude and longitude grids. A regular grid is generated, and the center coordinates of each grid are generated. Based on the center coordinates of the grid, the magnetic latitude (MLAT) and magnetic local time (MLT) of the current grid are calculated to generate the magnetic coordinates of the current grid.

[0026] Specifically, the prediction area is divided according to latitude and longitude. A regular grid, each grid The center coordinates are:

[0027]

[0028] in: The starting latitude and longitude; The center coordinates of each grid; Latitude and longitude step size ( ); For grid index ( ), This represents the upper limit of the longitude direction index. This represents the upper limit of the latitude direction index.

[0029] Calculate magnetic latitude (MLAT) and magnetic local time (MLT):

[0030] (Northern Hemisphere) (Northern Hemisphere)

[0031] in, and These are the geomagnetic polar coordinates of the target point (calculated from the International Geomagnetic Reference Field (IGRF) model), and UTC is Coordinated Universal Time.

[0032] Accordingly, the center coordinates of the magnetic grid are ,in, (Magnetic Latitude Index) (Magnetic local time index).

[0033] Among them, the magnetic latitude grid step size magnetic local time step for:

[0034] The same geographical latitude (60°N) may correspond to MLAT=55° (mid-latitude) or 75° (aurotic zone), leading to prediction bias. This paper uses a magnetic coordinate system (MLAT / MLT) to directly correlate particle deposition areas (|MLAT|>60°) and equatorial anomaly areas (|MLAT|≤15°), which helps to prevent prediction bias caused by the above situations.

[0035] S200: Obtain historical ionospheric and meteorological data for each magnetic coordinate grid, and normalize the ionospheric and meteorological data respectively. Construct sample data based on the normalized ionospheric and meteorological data.

[0036] The ionospheric data described in this application includes the ionospheric map exchange format file IONEX, with a temporal resolution of 2 hours and a spatial resolution of 5°×2.5°, denoted as... ,in, T This represents the total number of time steps. H The latitude range is -90° to 90°. W The longitude range is -180° to 180°. The meteorological data described in this application includes space weather data OMNI, from which [data is extracted]. Five indices; among them, The level of global geomagnetic activity; The planetary amplitude index; The main correlation strength of the geomagnetic storm; The solar radio flux is 10.7 cm; Sunspot is the relative number of sunspots, denoted as . .

[0037] The ionospheric TEC data and space meteorological data at each grid point were normalized as follows:

[0038] In the formula, Training statistics for TEC; These are statistics for 5 meteorological indices; t For time indexing; TEC normalization processing; This is the result after normalizing the meteorological index.

[0039] The sample construction (sliding window) is as follows:

[0040] In the formula, Input features for TEC As input characteristics for space weather, To predict the target, set according to a 2-hour resolution. τ =12 (24-hour historical data).

[0041] The S300 constructs an ionospheric prediction model based on the CNNGRU and Transformer modules, trains the ionospheric prediction model using sample data, and generates a target ionospheric prediction model.

[0042] Using only the self-attention mechanism of CNN-Transformer, it is difficult to distinguish between short-term sudden disturbances in the ionosphere (such as geomagnetic storms), medium-term diurnal variations (equatorial bimodal structure), and long-term trends (solar activity cycles).

[0043] The introduction of a bidirectional GRU module allows for the modeling of ionospheric state transitions. This includes using a forward GRU to learn historical evolution (e.g., a gradual decrease in the Dst exponent predicts geomagnetic storms) and a reverse GRU to learn future-to-historical dependencies (e.g., the recovery pattern of TEC after a geomagnetic storm). Furthermore, the features output by the bidirectional GRU module are fused, utilizing short / medium-term features from the GRU output and focusing on long-term dependencies using a Transformer, which helps improve the accuracy of predictions.

[0044] In one embodiment, training an ionospheric prediction model using sample data includes: inputting sample data into the ionospheric prediction model; using a CNNGRU module to extract features from ionospheric data and meteorological data respectively and concatenating them to construct time-series features; inputting the time-series features into a Transformer module to generate ionospheric prediction values; constructing a loss function based on the predicted ionospheric values ​​and the actual ionospheric values; using the loss function to train the ionospheric prediction model; and adjusting the parameters of the ionospheric prediction model until the loss function is minimized.

[0045] In one embodiment, the CNNGRU module is used to extract features from ionospheric data and meteorological data respectively and concatenate them to construct time-series features, including: using a CNN neural network to extract features from ionospheric data and meteorological data respectively, and outputting a first concatenated feature; inputting the first concatenated feature into a bidirectional GRU structure to generate two sets of hidden state vectors, and merging the two sets of hidden state vectors to form time-series features.

[0046] Before inputting the temporal features into the Transformer module to generate ionospheric predictions, the following steps are also included: Add ionospheric physical constraints to time series features:

[0047]

[0048] in, Features of the output after adding ionospheric physical constraints; This refers to the weight of this layer; For bias; The F10.7 index change rate over 24 hours; The trend of the Kp index; sigmoid activation function Ionospheric prediction must follow space physics rules (such as the TEC in the equatorial region at night being controlled by plasma bubbles), while ionospheric physical constraints can effectively suppress physical paradoxes through dynamic gating (such as paradoxes like TEC=50TECU in the polar region during quiet periods).

[0049] The ionospheric prediction model of this application consists of four core components:

[0050] In the formula, C is the CNN spatial feature extractor. For feature concatenation operation, R is the GRU temporal modeler, T is the Transformer global dependency modeler, and D is the decoder. The TEC grid result is predicted at time t.

[0051] The spatial pattern of CNN spatial feature extractors in processing TEC grid data is as follows:

[0052] In this layer, Conv2D is a convolutional layer, and MaxPool is a pooling layer. The feature representation after convolution. This is the feature representation after pooling. This represents the feature representation after convolution and pooling.

[0053] Feature splicing is as follows:

[0054] in, This is the output after splicing. The spatial feature dimension of the output. 5 represents the spatial weather characteristics, and 5 represents the latitude of the weather index.

[0055] The GRU temporal modeler employs a bidirectional GRU architecture, consisting of two independent GRUs: one processes the input sequence forward (from front to back), while the other processes the sequence backward (from back to front). These two GRUs generate hidden states separately, and their outputs are ultimately merged to form a richer feature representation.

[0056] The bidirectional GRU expression is:

[0057] in, The timing characteristics of the GRU output. Let be the hidden state vector at time t.

[0058] The expression for the ionospheric physical constraint layer IPC is:

[0059]

[0060] in, Features of the output after adding ionospheric physical constraints; This refers to the weight of this layer; For bias; The F10.7 index change rate over 24 hours; The trend of the Kp index (rising or falling, represented by 1 or -1); It is the sigmoid activation function.

[0061] The location encoding P of the Transformer global dependency modeler is:

[0062] Where i is the dimension index; pos is the time position; Encoding the position of the sine wave; This is the positional encoding for cosines.

[0063] Transformer layer position encoding injection is as follows:

[0064] Where P is the position encoding matrix; Z is the encoded feature.

[0065] The Transformer block structure is as follows:

[0066] TransformerBlock is:

[0067] in, A For multi-headed attention output; B These are intermediate features after residual connection and normalization; C This is the output of the feedforward neural network; This is the final output of the Transformerblock; FNN is the feedforward feature network, and LayerNorm is the layer normalization function.

[0068] MultiHeadAttention is represented as:

[0069] in, Q The element currently being monitored; K For all elements being queried; V The actual information for each element.

[0070] Global feature extraction is as follows:

[0071] in, The feature vector of the last time step; This is the output global feature vector.

[0072] The decoder (D) is:

[0073] in, To decode the feature vector; W Weights for each layer; b For each layer, offset; For the predicted TEC grid.

[0074] In one embodiment, the loss function is formulated as follows:

[0075] in, L Here, RMSE is the loss function optimized for model training. N The total number of grid points, This represents the actual TEC value for the corresponding grid point. This represents the predicted TEC value for the corresponding grid point.

[0076] In one embodiment, an optimizer is used to adjust the parameters of the ionospheric prediction model:

[0077] in, These are the model parameters from the previous time step; The corrected first moment; The corrected second moment; The learning rate; It is the numerical stability constant; This is the weight decay coefficient.

[0078] Sliding window length during training phase Adjusted dynamically based on geomagnetic activity:

[0079] in, This represents the absolute value of the 3-hour change in the Dst index.

[0080] S400 inputs the current ionospheric data and meteorological data of each magnetic coordinate grid into the target ionospheric prediction model, generates the ionospheric prediction results for each magnetic coordinate grid, aggregates the ionospheric prediction results of each magnetic coordinate grid, and outputs them.

[0081] The final predicted values ​​of each grid obtained Reorganized into a two-dimensional matrix according to magnetic coordinates, this matrix contains These elements constitute the preliminary forecast map: Define the global forecast map matrix It consists of all grid predictions, and its elements satisfy:

[0082] Each element in the matrix is ​​determined by its magnetic grid index position: The elements in the first row correspond to all grids where k=0:

[0083] The last row of elements corresponds to All grids:

[0084] The main diagonal elements satisfy k=l:

[0085] The matrix generated above This is a preliminary global ionospheric TEC forecast map, but there is a problem of grid boundary discontinuity. Therefore, spatial smoothing constraints are applied to the grid: to eliminate grid jumps, bilinear interpolation is used for cross-grid smoothing.

[0086] Specifically, define the target high-resolution mesh. satisfy:

[0087]

[0088] in, This is the resolution enhancement factor. For any target point... Its smoothing value is derived by interpolation from the four most recent original grid points:

[0089] in, This is the smoothing result for the corresponding target point; These are the predicted values ​​of the four nearest original grid points to the target point; To smooth out the weights, bilinear interpolation is used for calculation.

[0090] By traversing all target points, a smoothed global forecast map matrix is ​​obtained, and a global forecast map is constructed:

[0091] The dimension of this matrix is Spatial resolution is improved compared to the original mesh. times.

[0092] Where p is the target grid index of the magnetic latitude ( ), where q is the target grid index for the local magnetic time ( ) Finally, a global ionospheric forecast map is generated, and the resulting smoothed forecast matrix is ​​used. Convert to geographic coordinates :

[0093] Data range: longitude ,latitude Time attribute: Forecast validity period Forecast lead time ( Hour).

[0094] The forecast maps in this application can be represented by multiple coordinates, including dual-coordinate forecast maps: (1) Geographic coordinate layer: Displays the geographic distribution of TEC prediction values ​​(in line with user habits); (2) Magnetic coordinate layer: superimposed magnetic coordinate grid (such as the boundary of the equatorial anomaly zone with MLAT=±15°), or color-coded magnetic coordinate information (such as blue representing the magnetic equator); (3) Explainable layer: Mark the magnetic coordinate sensitive areas during geomagnetic storms (such as MLAT=65°-75° in the aurora zone) to improve decision support capabilities.

[0095] In one embodiment, this application also provides an adaptive ionospheric prediction system based on a magnetic coordinate system and physical constraints, comprising: The magnetic coordinate grid division module is used to select the prediction area and divide the prediction area into magnetic coordinate grids; The sample data construction module is used to acquire historical ionospheric and meteorological data for each magnetic coordinate grid, and to normalize the ionospheric and meteorological data respectively, and to construct sample data based on the normalized ionospheric and meteorological data. The model building module is used to build an ionospheric prediction model based on the CNNGRU module and the Transformer module, train the ionospheric prediction model using sample data, and generate a target ionospheric prediction model. The results output module is used to input the current ionospheric data and meteorological data of each magnetic coordinate grid into the target ionospheric prediction model, generate the ionospheric prediction results of each magnetic coordinate grid, aggregate the ionospheric prediction results of each magnetic coordinate grid and output them.

[0096] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0097] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0098] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0099] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0100] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0101] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0102] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An adaptive ionospheric prediction method based on magnetic coordinate system and physical constraints, characterized in that, include: Select the prediction region and divide the prediction region into magnetic coordinate grids; Historical ionospheric and meteorological data for each magnetic coordinate grid are obtained and normalized separately. Sample data are then constructed based on the normalized ionospheric and meteorological data. An ionospheric prediction model is constructed based on the CNNGRU module and the Transformer module. The ionospheric prediction model is trained using sample data to generate a target ionospheric prediction model. The current ionospheric data and meteorological data of each magnetic coordinate grid are input into the target ionospheric prediction model to generate the ionospheric prediction results for each magnetic coordinate grid. The ionospheric prediction results of each magnetic coordinate grid are then aggregated and output.

2. The method according to claim 1, characterized in that, Divide the prediction region into magnetic coordinate grids, including: The prediction area is divided according to latitude and longitude. A regular grid is generated, and the center coordinates of each grid are determined. Calculate the magnetic latitude (MLAT) and magnetic local time (MLT) of the current grid based on the center coordinates of the grid, and generate the magnetic coordinates of the current grid.

3. The method according to claim 1, characterized in that, Training an ionospheric prediction model using sample data includes: The sample data is input into the ionospheric prediction model, and the features of the ionospheric data and meteorological data are extracted by the CNNGRU module and then spliced ​​to construct time-series features. The temporal features are input into the Transformer module to generate ionospheric predictions. A loss function is constructed based on the predicted and actual values ​​of the ionosphere. The ionosphere prediction model is trained using the loss function, and the parameters of the ionosphere prediction model are adjusted until the loss function is minimized.

4. The method according to claim 3, characterized in that, The CNNGRU module was used to extract features from ionospheric and meteorological data respectively, and then concatenated to construct time-series features, including: The features of ionospheric data and meteorological data are extracted separately using a CNN neural network, and the first spliced ​​feature is output. The first concatenated feature is input into a bidirectional GRU structure to generate two sets of hidden state vectors. The two sets of hidden state vectors are then merged to form a temporal feature.

5. The method according to claim 3, characterized in that, The formula for the loss function is: in, L Here, RMSE is the loss function optimized for model training. N The total number of grid points, This represents the actual TEC value for the corresponding grid point. This represents the predicted TEC value for the corresponding grid point.

6. The method according to claim 3, characterized in that, An optimizer was used to adjust the parameters of the ionospheric prediction model: in, These are the model parameters from the previous time step; The corrected first moment; The corrected second moment; The learning rate; It is the numerical stability constant; This is the weight decay coefficient.

7. The method according to claim 3, characterized in that, Before inputting the temporal features into the Transformer module to generate ionospheric predictions, the following steps are also included: Add ionospheric physical constraints to time series features: in, Features of the output after adding ionospheric physical constraints; This refers to the weight of this layer; For bias; The F10.7 index change rate over 24 hours; The trend of the Kp index; It is the sigmoid activation function.

8. An adaptive ionospheric prediction system based on magnetic coordinate system and physical constraints, characterized in that, include: The magnetic coordinate grid division module is used to select the prediction area and divide the prediction area into magnetic coordinate grids; The sample data construction module is used to acquire historical ionospheric and meteorological data for each magnetic coordinate grid, and to normalize the ionospheric and meteorological data respectively, and to construct sample data based on the normalized ionospheric and meteorological data. The model building module is used to build an ionospheric prediction model based on the CNNGRU module and the Transformer module, train the ionospheric prediction model using sample data, and generate a target ionospheric prediction model. The results output module is used to input the current ionospheric data and meteorological data of each magnetic coordinate grid into the target ionospheric prediction model, generate the ionospheric prediction results of each magnetic coordinate grid, aggregate the ionospheric prediction results of each magnetic coordinate grid and output them.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images