Ionized layer three-dimensional electron density chromatography method based on space-time coding neural network

Abstract

The invention discloses an ionized layer three-dimensional electron density chromatography method based on a space-time coding neural network. The method comprises the following steps: acquiring electron density profile data and driving parameters; aligning the electron density profile data and the driving parameters according to a space-time grid to obtain an electron density data set; the electron density data set is input into an implicit neural network for training, an ionosphere three-dimensional electron density model is obtained, and the implicit neural network is obtained through construction based on space-time coding and a multi-layer perceptron algorithm; and simulating or predicting the three-dimensional electron density value of the ionized layer by using the three-dimensional electron density model of the ionized layer. By fusing the multi-source occultation electron density data and the solar geomagnetic driving factor, high-precision and high-temporal-spatial-resolution ionosphere tomography, namely ionosphere three-dimensional electron density reconstruction, can be realized, and quasi-real-time high-precision ionosphere monitoring can be realized.

Description

Technical Field

[0001] This invention belongs to the field of space environment modeling and artificial intelligence technology, and in particular relates to a three-dimensional electron density tomography method for the ionosphere based on a spatiotemporal coding neural network. Background Technology

[0002] The ionosphere, a key component of Earth's atmosphere, extends to altitudes of approximately 60 to 1000 kilometers above the Earth's surface. This region is formed by the ionization of Earth's upper atmosphere under the influence of high-energy solar radiation and cosmic rays, and is primarily composed of charged particles, neutral atoms, and molecules. The ionosphere has a decisive impact on critical infrastructure such as modern communications and navigation. For example, signals from Global Navigation Satellite Systems (GNSS) experience significant delays and refraction effects when traversing this region, leading to positioning errors. Therefore, accurately establishing the three-dimensional electron density distribution of the ionosphere is not only a core requirement for improving GNSS positioning accuracy but also an indispensable key link in deeply understanding space weather phenomena, ensuring the reliability of communication systems, and predicting navigation and positioning errors. However, the inherent high-dynamic characteristics and spatial heterogeneity of the ionosphere present significant challenges to the accurate reconstruction of its three-dimensional structure.

[0003] Ionospheric tomography originates from CT imaging in the medical field. It inverts the three-dimensional distribution of ionospheric electron density by measuring the total electron content of the signal between a satellite and a ground receiving station and its projection relationship. However, due to the uneven spatial distribution of ground GNSS receiving stations and satellite orbits, significant blind spots exist in ray path coverage (such as in ocean and polar regions). Traditional tomographic methods are generally divided into voxel-based and function-based methods. Voxel models, due to their simplicity and efficiency, are currently the main method for ionospheric 3D reconstruction. Voxel-based ionospheric tomography divides the ionosphere into regular voxel grids (typically with a resolution of 20-50 km), resulting in reconstruction results limited by the preset grid resolution. Furthermore, due to insufficient rays and uneven geometric line-of-sight distribution, it faces a highly ill-posed problem (multiple solutions to the equations) when constructing the tomographic projection matrix. Regularization assumptions (such as smoothing constraints or Tikhonov regularization) alleviate this ill-posed problem to some extent; however, the actual structure of the ionosphere often contains steep gradients (such as electron density ridges in equatorial anomalies), and excessively strong smoothing constraints will smooth out these key features. The International Reference Ionospheric Model (IRI), as the most widely used empirical model, constructs a parameterized function library of latitude-local time-solar activity based on a large amount of historical observation data. However, the IRI model has fundamental defects: it lags in responding to ionospheric disturbances caused by sudden events such as solar flares and geomagnetic storms; and the IRI uses analytical functions (such as Chapman functions or parabolic profiles) to describe the vertical distribution of electron density, making it difficult to accurately represent multi-peak structures or small-scale irregularities.

[0004] With the rapid development of artificial intelligence technology, deep learning methods have been gradually introduced into the field of ionospheric modeling. For example, using low-Earth orbit occultation data and combining it with backpropagation neural networks (BP) to construct a high-precision initial electron density field significantly improves the accuracy and efficiency of subsequent reconstruction. Furthermore, tomography methods based on wavelet neural networks have also shown good results. These deep learning models effectively overcome the limitations of traditional methods in handling complex nonlinear relationships in the ionosphere, greatly improving the accuracy and adaptability of modeling. However, existing methods generally rely on discrete grid representations, making it difficult to accurately characterize the continuous variation characteristics and fine structure of the ionospheric electron density field, easily leading to information loss and limited accuracy, which urgently needs breakthroughs. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a three-dimensional electron density tomography method for the ionosphere based on a spatiotemporal coding neural network, which can improve the spatiotemporal resolution and accuracy of ionospheric three-dimensional reconstruction and enhance its adaptability to sparse data.

[0006] To achieve the above objectives, this invention provides a method for three-dimensional electron density tomography of the ionosphere based on a spatiotemporal encoded neural network, comprising:

[0007] Acquire electron density profile data and driving parameters;

[0008] The electron density profile data and driving parameters are aligned according to a spatiotemporal grid to obtain an electron density dataset.

[0009] The electron density dataset is input into an implicit neural network for training to obtain a three-dimensional electron density model of the ionosphere. The implicit neural network is constructed based on spatiotemporal coding and multilayer perceptron algorithms.

[0010] The three-dimensional electron density value of the ionosphere is simulated or predicted using the aforementioned three-dimensional electron density model of the ionosphere.

[0011] Optionally, before aligning the electron density profile data and driving parameters according to the spatiotemporal grid, the method further includes: performing profile quality control using Chapman-α to remove electron density profile data that does not conform to the Chapman-α function and to remove driving data that satisfies extreme space background events.

[0012] Optionally, the method for using Chapman-α to perform profile quality control and remove electron density profile data that do not conform to the Chapman-α function is as follows:

[0013]

[0014] in, The electron density observed at height h, ion_N m It is the peak electron density, ion_h mion_H is the peak height, and ion_H is the elevation.

[0015] Optionally, the electron density dataset includes: spatial coordinates, timestamps, and driving parameters.

[0016] Optionally, before inputting the electron density dataset into the spatiotemporally encoded implicit neural network for training, the method further includes: using the XGBoost model to analyze the nonlinear influence weights of the driving parameters on the electron density in the electron density profile data, quantifying the parameter contribution through SHAP values, and obtaining the target driving parameters.

[0017] Optionally, the three-dimensional electron density model of the ionosphere includes: a spatiotemporal coding module and a multilayer perceptron module;

[0018] The spatiotemporal coding module is used to perform spatiotemporal coding on the electron density dataset;

[0019] The multilayer perceptron module is used to receive high-dimensional feature vectors after spatiotemporal encoding and learn their evolution patterns or features.

[0020] Optionally, the spatiotemporal coding module includes: a time coding unit and a spatial coding unit;

[0021] The time encoding unit is used to perform time encoding on the electron density dataset by combining different periodic changes;

[0022] The spatial coding unit is used to spatially code the electron density dataset using a Gaussian random Fourier transform mapping.

[0023] Optionally, the multilayer perceptron module includes: an input layer, a hidden layer, and an output layer;

[0024] The input layer is used to input the encoded coordinate features and the encoded target driving parameters;

[0025] The hidden layer is used to receive the ionospheric electron density information of the input layer and extract the spatiotemporal evolution characteristics of the ionospheric electron density.

[0026] The output layer is used to receive the spatiotemporal evolution characteristics of the ionospheric electron density output from the last hidden layer, and finally predict the three-dimensional electron density value of the target ionosphere.

[0027] Compared with the prior art, the present invention has the following advantages and technical effects:

[0028] The ionospheric three-dimensional electron density modeling technology based on implicit neural representation method can represent the ionospheric electron density field as a continuous neural function, breaking through the spatial resolution limitations of traditional grid-based modeling methods. It can achieve arbitrarily accurate three-dimensional reconstruction of the ionosphere, providing a powerful tool for the refined study of the ionosphere.

[0029] This invention can achieve high-precision, high-spatiotemporal-resolution ionospheric tomography by fusing multi-source occultation electron density data and solar geomagnetic driving factors, enabling near real-time, high-precision ionospheric monitoring.

[0030] This invention is highly adaptable to the format and distribution of input data, and can fully utilize various types of observation data (such as GNSS TEC data, altimeter data, etc.) for joint modeling. Through the reasonable application of data augmentation technology and optimization algorithms, it improves the model's learning ability and generalization ability to different data characteristics, making it suitable for various complex ionospheric environments and different application scenarios. It provides more accurate and efficient technical support for fields such as ionospheric monitoring, space weather research, and satellite communication and navigation. Attached Figure Description

[0031] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0032] Figure 1 This is a flowchart of a three-dimensional electron density tomography method for the ionosphere based on a spatiotemporal coding neural network according to an embodiment of the present invention;

[0033] Figure 2 This is a flowchart of the importance analysis of solar / geomagnetic driving parameters based on the XGBoost model according to an embodiment of the present invention;

[0034] Figure 3 This is a schematic diagram of the structure of the implicit neural representation method network according to an embodiment of the present invention;

[0035] Figure 4 These are scatter plots analyzing the test set correlation of the model under different extreme space environment backgrounds according to embodiments of the present invention. Among them, (a) is a scatter plot analyzing the test set correlation of the model under a weak geomagnetic storm environment, (b) is a scatter plot analyzing the test set correlation of the model under a large geomagnetic storm environment, and (c) is a scatter plot analyzing the test set correlation of the model under a weak geomagnetic storm environment.

[0036] Figure 5These are comparison diagrams of the electron density profile of the model and the actual electron density of the ionospheric vertical sounder according to embodiments of the present invention. Among them, (a) is a comparison diagram of the electron density profile output by the model at 2:30 UTC on year 102, (b) is a comparison diagram of the electron density profile output by the model at 2:30 UTC on year 104, (c) is a comparison diagram of the electron density profile output by the model at 2:30 UTC on year 106, (d) is a comparison diagram of the electron density profile output by the model at 2:30 UTC on year 108, (e) is a comparison diagram of the electron density profile output by the model at 2:30 UTC on year 110, (f) is a comparison diagram of the electron density profile output by the model at 2:30 UTC on year 112 ...

[0037] Figure 6 These are visualizations of the three-dimensional electron density modeling and reconstruction results of the ionosphere based on the implicit neural representation method according to an embodiment of the present invention. Among them, (a) is the three-dimensional reconstruction visualization result of the example region at DOY102, 2025, at 2:30 UTC; (b) is the three-dimensional reconstruction visualization result of the example region at DOY104, 2025, at 2:30 UTC; (c) is the three-dimensional reconstruction visualization result of the example region at DOY106, 2025, at 2:30 UTC; (d) is the three-dimensional reconstruction visualization result of the example region at DOY108, 2025, at 2:30 UTC; (e) is the three-dimensional reconstruction visualization result of the example region at DOY110, 2025, at 2:30 UTC; and (f) is the three-dimensional reconstruction visualization result of the example region at DOY112, 2025, at 2:30 UTC. Detailed Implementation

[0038] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0039] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0040] This embodiment proposes a three-dimensional electron density tomography method for the ionosphere based on a spatiotemporal encoded neural network, such as... Figure 1 As shown, the specific steps include:

[0041] Acquire electron density profile data and driving parameters;

[0042] The electron density profile data and driving parameters are aligned according to the spatiotemporal grid to obtain the electron density dataset;

[0043] The electron density dataset is input into an implicit neural network for training to obtain a three-dimensional electron density model of the ionosphere. The implicit neural network is constructed based on spatiotemporal coding and multilayer perceptron algorithms.

[0044] The three-dimensional electron density of the ionosphere is simulated or predicted using a three-dimensional electron density model of the ionosphere.

[0045] Specifically, S1, acquire publicly available electron density profile data of FORMOSAT-7 / COSMIC-2 and SwarmA / B / C low-orbit occultations, and perform data preprocessing.

[0046] S2. Dataset creation and partitioning: The dataset is divided into a training set of 70%, a validation set of 15%, and a test set of 15% for training, validation, and testing the model.

[0047] S3. Use the XGBoost model to analyze the nonlinear influence weights of solar / geomagnetic parameters on electron density, and quantify the parameter contribution using SHAP values.

[0048] S4. Construct an implicit neural representation neural network model for 3D reconstruction of the ionosphere based on the principles of spatiotemporal coding and multilayer perceptron (MLP) algorithm.

[0049] S5. Model Training and Optimization: The model is trained using the created dataset. The trained implicit neural representation model is used to reconstruct the three-dimensional ionosphere. The input is the spatial region coordinates, time, and the corresponding solar and geomagnetic driving factors. The output is the corresponding electron density value.

[0050] S6. Verify and quality control the output three-dimensional electron density. Compare the output results with the actual observed values, and calculate the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). 2 The accuracy of model predictions is quantified by indicators such as relative error of foF2 and absolute error of hmF2.

[0051] More specifically, in S1, publicly available radio occultation (RO) electron density profile data from FORMOSAT-7 / COSMIC-2 and SwarmA / B / C satellites from October 2019 to May 2025 are acquired. In this example, the ionosphere of a specific region is reconstructed in three dimensions, with longitude ranging from 110° to 155° and latitude ranging from -45° to -5°. Solar activity parameters (F10.7 index, solar flare flux) and geomagnetic activity parameters (Kp, Dst index) are simultaneously acquired. Specifically, FORMOSAT-7 / COSMIC-2 data can be obtained from the COSMIC Data Analysis and Management Center; SwarmA / B / C satellite data can be obtained from the ESA's Swarm data portal; and solar geomagnetic driving parameters are obtained from publicly available datasets provided by the Space Flight Center.

[0052] Furthermore, before aligning the electron density profile data and driving parameters according to the spatiotemporal grid, the process includes: performing profile quality checks using Chapman-α to remove electron density profile data that does not conform to the Chapman-α function, and removing driving data that satisfies extreme space background events.

[0053] Specifically, quality control requires the removal of data from extreme space background events (data with Kp≥8 or Dst≤-150nT), and the use of Chapman-α profile quality control to remove electron density profiles that do not conform to the Chapman-α function. The ionospheric Chapman-α function model is shown in Equation (1), and the occultation electron density profile quality control meets the following requirements:

[0054]

[0055] Goodness-of-fit test:

[0056]

[0057] Where RMSE ≤ threshold (threshold: 1.0-1.5×10¹¹el / m) 3 );

[0058] Correlation coefficient test:

[0059]

[0060] Where |R|≥0.85.

[0061] Physical parameter rationality verification:

[0062]

[0063] in, It is the electron density value simulated by the Chapman-α function. This is the electron density observed at height hi. At height h i The electron density calculated by the model, ion_N m Peak electron density (e / m 3 ),ion_h m σ is the peak height (km), ion_H is the elevation (km), and σ is the peak height (km). obs , σ model , It is the average value of the observed electron density. The average electron density of the Chapman model.

[0064] Furthermore, the electron density dataset includes: spatial coordinates, timestamps, and driving parameters.

[0065] Specifically, in S2, the dataset is constructed by aligning the electron density profile data with the driving parameters using a spatiotemporal grid (1°×1°×20km, 10-minute intervals). Dataset construction includes input features: spatial coordinates (x, y, z), timestamps (t), and solar / geomagnetic parameters (driving parameters such as Dst, F10.7, and Kp). The dataset is randomly divided into three parts: a training set (70%), a validation set (15%), and a test set (in this example, the test set is divided into three categories based on different storm intensities: weak storms, large storms, and strong storms).

[0066] Furthermore, before inputting the electron density dataset into the spatiotemporally encoded implicit neural network for training, the process includes: using the XGBoost model to analyze the nonlinear influence weights of the driving parameters on the electron density in the electron density profile data, quantifying the parameter contribution through SHAP values, and obtaining the target driving parameters.

[0067] Specifically, in S3, physical driving feature analysis and fusion are performed. The XGBoost model is used to analyze the nonlinear influence weights of solar / geomagnetic parameters on electron density. The parameter contribution is quantified by SHAP values. Finally, five solar / geomagnetic driving parameters are selected: Dst, F10.7, plasma temperature, AE, and Kp. (Note: This step is an offline analysis and only provides a basis for feature design. It does not participate in the final model training.)

[0068] Furthermore, the three-dimensional electron density model of the ionosphere includes: a spatiotemporal coding module and a multilayer perceptron module;

[0069] The spatiotemporal coding module is used to perform spatiotemporal coding on the electron density dataset;

[0070] The multilayer perceptron module is used to receive high-dimensional feature vectors after spatiotemporal encoding and learn their evolution patterns or features.

[0071] Furthermore, the spatiotemporal coding module includes: a temporal coding unit and a spatial coding unit;

[0072] The time encoding unit is used to time-encode the electron density dataset by incorporating different periodic variations.

[0073] Spatial coding unit, used to spatially code the electron density dataset using Gaussian random Fourier transform mapping.

[0074] Specifically, in S4, an implicit neural representation neural network model for ionospheric 3D reconstruction is constructed based on the principles of spatiotemporal coding and multilayer perceptron (MLP) algorithm. Its core lies in the application of spatiotemporal coding and MLP.

[0075] Time coding: Time coding takes into account different periodic changes, and the coding formula is shown in (5).

[0076]

[0077] Where, γ doy This is the annual day cycle code; DOY represents the annual day. H It is an hourly cycle code, where H represents the hour; γ T It is a minute-period encoding, T m It's minutes.

[0078] Spatial coding: The position encoder uses a Gaussian random Fourier mapping, as shown in formula (6):

[0079] γ(p)=[sin(2πBp),cos(2πBp)] T (6)

[0080] Where p = [x, y, z] are the position coordinates of any electron density. The elements of the N(0,σ) sequence follow the N(0,σ) sequence. 2 The distribution of features, the number of feature dimensions m (ranging from 128 to 512), and the spectral scaling factor σ (ranging from 8 to 10).

[0081] Furthermore, the multilayer perceptron module includes: an input layer, a hidden layer, and an output layer;

[0082] The input layer is used to input the encoded coordinate features and the encoded target driving parameters;

[0083] The hidden layer is used to receive ionospheric electron density information from the input layer and extract the spatiotemporal evolution characteristics of ionospheric electron density.

[0084] The output layer is used to receive the spatiotemporal evolution characteristics of the ionospheric electron density output from the last hidden layer, and finally predicts the three-dimensional electron density value of the target ionosphere.

[0085] Specifically, the design of a multilayer perceptron (MLP):

[0086] Input layer: high-dimensional spatiotemporal encoded features after spatiotemporal encoding: γ(x,y,z,t), and solar / geomagnetic driving parameters (F10.7, Dst and plasma temperature, AE and Kp, 5 solar / geomagnetic driving parameters);

[0087] Hidden layers: 5 fully connected layers (number of neurons: 512→256→128→64→32), using the periodic activation function (SIREN):

[0088] φ(x)=sin(ω0·(Wx+b))(7)

[0089] Wherein, ω0 is set decreasingly according to the network depth, with ω0=30 for the input layer, ω0=15 for the intermediate layers, and ω0=5 for the first layer of the output layer, representing the weight initialization range. Bias initialization b ~ U[-π,π];

[0090] Output layer: Linearly outputs the predicted value of ionospheric electron density at arbitrary spatiotemporal resolution.

[0091] Furthermore, in S5, model training and optimization involve training the model using the created dataset. The trained implicit neural representation model is then used to reconstruct the 3D electron density of the ionosphere. The input is the spatial region coordinates, time, and corresponding solar and geomagnetic driving factors; the output is the 3D electron density value of the ionosphere in the instance region. In S5, the model optimizer for this instance is AdamW, with a learning rate of 10. -4 The batch size is 512, the epoch is 15, the training process adopts an early stopping mechanism, and the patience value is 3.

[0092] In S6, the output three-dimensional electron density is verified and quality checked. The output results are compared with the actual observed values, and the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²) are calculated. 2 The control standards for indicators such as relative error of foF2 and absolute error of hmF2 are shown in Table 1.

[0093] Table 1

[0094]

[0095] in, It is the i-th value simulated by the model. It is the i-th actual observation, foF2 pred It is the peak electron density of the F layer output by the model, foF2 obs It is the actual peak electron density of the F layer, hmF2 pred It is the height of layer F output by the model, hmF2obs This represents the actual observed height of layer F, and N is the total number of layers.

[0096] Figure 2 This is a flowchart of the importance analysis of solar / geomagnetic driving parameters based on the XGBoost model.

[0097] The XGBoost model is used for selecting optimal driving parameters, and model performance is evaluated through cross-validation. Furthermore, it utilizes...

[0098] SHAP value analysis quantifies the contribution of each driving parameter. Based on the analysis results, it is determined whether to delete unimportant parameters. If deleted, the parameter combination is updated; otherwise, the parameters are reselected until the optimal driving parameter combination is obtained.

[0099] Figure 3 This is a schematic diagram of the spatiotemporal encoded implicit neural representation method for three-dimensional electron density modeling of the ionosphere, as described in this invention. Figure 3 (a)-(d) are schematic diagrams of the reconstruction process. Figure 3 (e) is a schematic diagram showing the details of the model structure.

[0100] Figure 4 Images (a)-(c) show the scatter plots illustrating the correlation distribution of the model on the test set under different extreme space environment conditions. Clearly, in this example, the model demonstrates excellent simulation performance under different levels of geomagnetic storms, with a correlation coefficient R0. 2 All are above 0.97, with the lowest root mean square error (RMSE) reaching 0.75 (10). 11 el / m 3 The mean absolute error can be as low as 0.45 (10). 11 el / m 3 These indicators all demonstrate that the model of this invention has excellent simulation effect of three-dimensional electron density of the ionosphere.

[0101] Figure 5 Images (a)-(f) show a comparison between the model's three-dimensional electron density profile and the actual electron density obtained from the ionospheric plumb line. In this embodiment, the electron density values ​​output by the model are compared with those obtained from the Perth ionospheric plumb line at six times in 2025: DOY 102, 104, 106, 108, 110, and 112, at 2:30 UTC. Figure 5 It can be seen that the electron density output by the model is highly consistent with the measured electron density profile of the vertical station, demonstrating the accuracy and reliability of the model.

[0102] Figure 6Images (a)-(f) show the reconstruction results of the ionospheric three-dimensional electron density based on the implicit neural representation method of this invention. In this embodiment, visualizations of the reconstruction results for the example region at six time points in 2025: DOY102, DOY104, DOY106, DOY108, DOY110, and DOY112, at 2:30 UTC. Figure 6 It can be seen that the ionospheric electron density value output by the model can well reflect the distribution of ionospheric electron density at different heights in the instance region.

[0103] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for three-dimensional electron density tomography of the ionosphere based on a spatiotemporal coding neural network, characterized in that, include: Acquire electron density profile data and driving parameters; The electron density profile data and driving parameters are aligned according to a spatiotemporal grid to obtain an electron density dataset. The electron density dataset is input into an implicit neural network for training to obtain a three-dimensional electron density model of the ionosphere. The implicit neural network is constructed based on spatiotemporal coding and multilayer perceptron algorithms. The three-dimensional electron density value of the ionosphere is simulated or predicted using the aforementioned three-dimensional electron density model of the ionosphere.

2. The method for three-dimensional electron density tomography of the ionosphere based on a spatiotemporal coding neural network according to claim 1, characterized in that, Before aligning the electron density profile data and driving parameters according to the spatiotemporal grid, the process further includes: performing profile quality checks using Chapman-α to remove electron density profile data that does not conform to the Chapman-α function, and removing driving data that satisfies extreme space background events.

3. The method for three-dimensional electron density tomography of the ionosphere based on a spatiotemporal coding neural network according to claim 2, characterized in that, The method for using Chapman-α to perform profile quality control and remove electron density profile data that does not conform to the Chapman-α function is as follows: in, The electron density observed at height h, ion_N m It is the peak electron density, ion_h m ion_H is the peak height, and ion_H is the elevation.

4. The method for three-dimensional electron density tomography of the ionosphere based on a spatiotemporal coding neural network according to claim 2, characterized in that, The electron density dataset includes: spatial coordinates, timestamps, and driving parameters.

5. The method for three-dimensional electron density tomography of the ionosphere based on a spatiotemporal encoded neural network according to claim 4, characterized in that, Before inputting the electron density dataset into the spatiotemporally encoded implicit neural network for training, the process further includes: using the XGBoost model to analyze the nonlinear influence weights of the driving parameters on the electron density in the electron density profile data, quantifying the parameter contribution through SHAP values, and obtaining the target driving parameters.

6. The method for three-dimensional electron density tomography of the ionosphere based on a spatiotemporal coding neural network according to claim 1, characterized in that, The three-dimensional electron density model of the ionosphere includes: a spatiotemporal coding module and a multilayer perceptron module; The spatiotemporal coding module is used to perform spatiotemporal coding on the electron density dataset; The multilayer perceptron module is used to receive high-dimensional feature vectors after spatiotemporal encoding and learn their evolution patterns.

7. The method for three-dimensional electron density tomography of the ionosphere based on a spatiotemporal encoded neural network according to claim 6, characterized in that, The spatiotemporal coding module includes: a time coding unit and a spatial coding unit; The time encoding unit is used to perform time encoding on the electron density dataset by combining different periodic changes; The spatial coding unit is used to spatially code the electron density dataset using a Gaussian random Fourier transform mapping.

8. The method for three-dimensional electron density tomography of the ionosphere based on a spatiotemporal encoded neural network according to claim 7, characterized in that, The multilayer perceptron module includes: an input layer, a hidden layer, and an output layer; The input layer is used to input the encoded coordinate features and the encoded target driving parameters; The hidden layer is used to receive the ionospheric electron density information of the input layer and extract the spatiotemporal evolution characteristics of the ionospheric electron density. The output layer is used to receive the spatiotemporal evolution characteristics of the ionospheric electron density output from the last hidden layer, and finally predict the three-dimensional electron density value of the target ionosphere.

Images