A reconstruction method for ionospheric peak electron density
By constructing the SRON2NN model and incorporating effective solar radiation and thermospheric parameters, the problem of insufficient accuracy of existing models under extreme conditions is solved. This enables high-precision reconstruction of ionospheric peak electron density and model interpretability, thereby improving prediction accuracy during geomagnetic storms.
Patent Information
- Application Number
- CN202610441466.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-05
- Publication Date
- 2026-06-26
AI Technical Summary
Existing ionospheric peak electron density prediction models lack accuracy under extreme space weather events, lack explicit modeling of photochemical source and loss terms, resulting in limited generalization ability of the models under strong disturbances such as geomagnetic storms, and existing models are difficult to reconstruct with high accuracy and physical interpretation.
A fully connected feedforward neural network SRON2NN model was constructed, and the effective solar radiation EUV×cosχ and thermosphere ΣO/N2 were introduced as key physical features. The model was trained with COSMIC-1 data and the Adam optimizer and early stopping strategy were used to achieve high-precision reconstruction of the peak electron density NmF2 of the ionospheric F2 layer.
It achieves high-precision reconstruction of the peak electron density NmF2 of the F2 layer of the ionosphere globally, reduces model error, improves prediction accuracy during geomagnetic storms, and provides a profound understanding of the ionosphere-thermosphere coupling mechanism.
Smart Images
Figure CN122287364A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of interdisciplinary technology of space weather detection and artificial intelligence, and more specifically, to a method for reconstructing the peak electron density of the ionosphere using a deep learning neural network in combination with physical parameters such as effective solar radiation and thermosphere ΣO / N2. Background Technology
[0002] The peak electron density (NmF2) of the F2 layer of the ionosphere is a key parameter affecting high-frequency radio communication, satellite navigation and positioning, and space weather early warning. Traditional prediction methods mainly rely on empirical models (such as IRI) or physical models. The former has limited accuracy under extreme space weather events, while the latter is computationally complex and difficult to accurately characterize the nonlinear evolution of the ionosphere. In recent years, deep learning methods have shown strong fitting capabilities in ionospheric modeling. However, existing models mostly use space weather indices as the main input and lack explicit modeling of key physical processes in the ionospheric continuity equation (such as photochemical generation and recombination losses), resulting in insufficient model interpretability and limited generalization ability under strong disturbances such as geomagnetic storms. Therefore, how to effectively embed the physical parameters characterizing photochemical source and loss terms into neural networks to construct an NmF2 reconstruction model with both high accuracy and physical interpretability has become an important research direction in current ionospheric modeling. Summary of the Invention
[0003] The purpose of this invention is to provide a method for reconstructing the peak electron density of the ionosphere, so as to overcome the defects of the prior art.
[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for reconstructing the peak electron density of the ionosphere includes the following steps: S1. Construct the peak electron density reconstruction neural network SRON2NN model. This model adopts a fully connected feedforward neural network, which includes an input layer, several hidden layers and an output layer. S2. Construct and introduce key physical features, including effective solar radiation EUV×cosχ and thermosphere ΣO / N2; S3. The model was trained using peak electron density NmF2 data, time information, geomagnetic index Dst, solar activity index F10.7, effective solar radiation EUV × cosχ, and ΣO / N2 obtained from COSMIC-1 radio occultation observations. S4. Based on the trained SRON2NN model, the peak electron density NmF2 of the ionosphere F2 layer was reconstructed and verified.
[0005] Furthermore, it also includes the following steps: S5. Apply the reconstruction results to ionospheric reporting, data missing filling, communication and navigation system performance evaluation, and space weather early warning.
[0006] Furthermore, in step S1: the input features of the input layer include time information, geographical location, geomagnetic index Dst, solar activity index F10.7, effective solar radiation, and thermosphere ΣO / N2; the hidden layer uses a nonlinear activation function; the output layer directly outputs the predicted value of the peak electron density NmF2 of the ionosphere F2 layer.
[0007] Furthermore, the time information includes year, annual day, hour, and minute; the geographical location is geographical latitude and longitude; and the effective solar radiation is EUV×cosχ, where EUV is solar extreme ultraviolet radiation and χ is the solar zenith angle.
[0008] Furthermore, in step S2: the effective solar radiation is composed of the product of solar EUV flux and the cosine of the solar zenith angle, used to simulate the electron generation rate of the ionosphere; the thermosphere ΣO / N2 is observed by the TIMED / GUVI satellite, and the electron loss rate and generation rate are jointly affected by adjusting the electron recombination loss coefficient and carrying atomic oxygen density information.
[0009] Furthermore, in step S3: the data used for model training includes: NmF2 data retrieved from COSMIC-1 radio occultation observations, solar EUV radiation data observed by the SDO / EVE instrument, thermospheric ΣO / N2 data observed by the TIMED / GUVI satellite, and geomagnetic index Dst and solar activity index F10.7 from the OMNI database; all data are matched and aligned according to time, location, and physical characteristics after quality control.
[0010] Furthermore, the model training data is divided into training, validation, and test sets in an 8:1:1 ratio; the Adam optimizer is used with mean squared error (MSE) as the loss function, along with an early stopping strategy to prevent model overfitting.
[0011] Furthermore, the Adam optimizer has an initial learning rate of 0.0002, a batch size of 1024, a maximum number of training epochs of 50, and an early stopping strategy that stops training if the verification loss does not decrease for 5 consecutive epochs.
[0012] Furthermore, the SRON2NN model has 4 hidden layers with 128, 256, 256 and 128 neurons respectively. The SRON2NN model is implemented based on the PyTorch framework.
[0013] Furthermore, in step S4: the model reconstruction range is the global region between 60° north and south latitude; the root mean square error (RMSE) of the SRON2NN model on the test set is 9.94 × 10⁻⁶.4 –2.09×10 5 el / cm³.
[0014] Compared with the prior art, the advantages of the present invention are as follows: The present invention provides an ionospheric peak electron density reconstruction method, which achieves high-precision and interpretable reconstruction of the peak electron density NmF2 of the F2 layer of the ionosphere globally by explicitly introducing effective solar radiation (representing the photochemical source term) and thermosphere ΣO / N2 (representing the electron loss term). Attached Figure Description
[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a schematic diagram comparing the structure of ANN and the SRON2NN model of this invention; Figure 2 This is a performance comparison chart of the SRON2NN model of this invention and the baseline model on the test set. Detailed Implementation
[0017] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.
[0018] See Figure 1 As shown, this embodiment discloses a method for reconstructing the peak electron density of the ionosphere, including the following steps: Step S1: Construct the peak electron density reconstruction neural network SRON2NN model. This model adopts a fully connected feedforward neural network, which includes an input layer, several hidden layers and an output layer.
[0019] In this embodiment, the input features of the input layer include time information (including year, annual day, hour, and minute), geographical location (geographical latitude and longitude), geomagnetic index Dst, solar activity index F10.7, effective solar radiation EUV×cosχ, where χ is the solar zenith angle, and thermosphere ΣO / N2; the hidden layer uses a nonlinear activation function to capture complex mapping relationships; the output layer directly outputs the predicted value of the peak electron density NmF2 of the ionospheric F2 layer.
[0020] Step S2: Construct and introduce key physical features, including effective solar radiation EUV×cosχ and thermosphere ΣO / N2; wherein, effective solar radiation is composed of the product of solar EUV flux and the cosine of solar zenith angle, and is used to simulate the electron generation rate of the ionosphere; thermosphere ΣO / N2 is observed by the TIMED / GUVI satellite, and thermosphere ΣO / N2 affects the electron loss rate and generation rate by adjusting the electron recombination loss coefficient and carrying atomic oxygen density information.
[0021] Step S3: Train the model using the peak electron density NmF2 data of the ionospheric F2 layer, time information, geomagnetic index Dst, solar activity index F10.7, effective solar radiation EUV × cosχ, and ΣO / N2 obtained from COSMIC-1 radio occultation observations.
[0022] In this embodiment, the data used for model training includes: NmF2 data (2008–2019) retrieved from COSMIC-1 radio occultation observations, solar EUV radiation data observed by the SDO / EVE instrument, thermospheric ΣO / N2 data observed by the TIMED / GUVI satellite, and the geomagnetic index Dst and solar activity index F10.7 from the OMNI database; all data were matched and aligned according to time, location, and physical characteristics after quality control.
[0023] The training data was divided into training, validation and test sets in an 8:1:1 ratio; the Adam optimizer was used with mean squared error (MSE) as the loss function, and an early stopping strategy was employed to prevent model overfitting.
[0024] Step S4: Reconstruct and verify the peak electron density NmF2 of the F2 layer of the ionosphere based on the trained SRON2NN model.
[0025] In this embodiment, the model reconstruction range is the global region between 60° north and south latitude; the root mean square error (RMSE) of the SRON2NN model on the test set is 9.94 × 10⁻⁶. 4 –2.09×10 5 The error was reduced by 2.37% (stable period) to 4.27% (disturbance period) compared to the baseline model (ANN) which only uses traditional space weather indices, with the most significant improvement in mid-latitude regions during geomagnetic storms (reduction of up to 6.70%).
[0026] Step S5: Apply the reconstruction results to ionospheric current reporting, data missing completion, communication and navigation system performance evaluation, and space weather early warning system input. The importance analysis of the physical features extracted by the model helps deepen the understanding of the ionosphere-thermosphere coupling mechanism.
[0027] The invention will be further described below in conjunction with the construction of the peak electron density reconstruction neural network SRON2NN model.
[0028] 1. Data preparation: NmF2 data retrieved using COSMIC-1 radio occultation observations (2008–2019, 60°S–60°N). Solar EUV radiation data were obtained from the SDO / EVE instrument (0.1–50 nm band). Thermospheric ΣO / N2 data are from the TIMED / GUVI satellite; The geomagnetic index (Dst) and solar activity index (F10.7) are from the OMNI database; After quality control, all data is matched and aligned according to time, location, and physical characteristics.
[0029] 2. Model Building: The input layer includes: time information (year, annual day, hour, minute), geographic latitude and longitude, Dst, F10.7, effective solar radiation (EUV×cosχ), ΣO / N2; The hidden layers are set to 4 layers, with 128, 256, 256 and 128 neurons respectively; The output layer contains NmF2 predictions; The model is implemented using the PyTorch framework.
[0030] 3. Model training: Using the Adam optimizer, with an initial learning rate of 0.0002 and a batch size of 1024; The maximum number of training rounds is 50, and an early stopping strategy is adopted (if the loss does not decrease for 5 consecutive rounds, the training is stopped). The loss function is the mean squared error (MSE).
[0031] 4. Result Verification: like Figure 2 As shown, SRON2NN's average RMSE on the test set for the entire year is 1.453 × 10⁻⁶. 5 el / cm³ (static stability period) and 1.998×10 5 The el / cm³ (perturbation period) is significantly better than the baseline ANN. ON2NN and SRNN represent the addition of ΣO / N2 and effective solar radiation respectively to the ANN, and can be used to compare the contribution of ΣO / N2 and effective solar radiation features to the improvement of model accuracy. Figure 2 It can be seen that, regardless of whether the geomagnetic disturbance period or the calm period, ON2NN is more effective than SRNN in reducing RMSE.
[0032] This invention can be used for real-time reconstruction and forecasting of global ionospheric NmF2; the importance of the physical features extracted by this invention helps to deepen the understanding of the ionospheric-thermosphere coupling mechanism.
[0033] Although embodiments of the present invention have been described in conjunction with the accompanying drawings, the patent owner may make various modifications or alterations within the scope of the appended claims, as long as they do not exceed the protection scope described in the claims of the present invention, they shall be within the protection scope of the present invention.
Claims
1. A method for reconstructing the peak electron density of the ionosphere, characterized in that, Includes the following steps: S1. Construct the peak electron density reconstruction neural network SRON2NN model. This model adopts a fully connected feedforward neural network, which includes an input layer, several hidden layers and an output layer. S2. Construct and introduce key physical features, including effective solar radiation EUV×cosχ and thermosphere ΣO / N2; S3. The model was trained using peak electron density NmF2 data, time information, geomagnetic index Dst, solar activity index F10.7, effective solar radiation EUV × cosχ, and ΣO / N2 obtained from COSMIC-1 radio occultation observations. S4. Based on the trained SRON2NN model, the peak electron density NmF2 of the ionosphere F2 layer was reconstructed and verified.
2. The method for reconstructing the peak electron density of the ionosphere according to claim 1, characterized in that, It also includes the following steps: S5. Apply the reconstruction results to ionospheric reporting, data missing filling, communication and navigation system performance evaluation, and space weather early warning.
3. The method for reconstructing the peak electron density of the ionosphere according to claim 1, characterized in that, In step S1: the input features of the input layer include time information, geographical location, geomagnetic index Dst, solar activity index F10.7, effective solar radiation, and thermosphere ΣO / N2; the hidden layer uses a nonlinear activation function; the output layer directly outputs the predicted value of the peak electron density NmF2 of the ionosphere F2 layer.
4. The method for reconstructing the peak electron density of the ionosphere according to claim 2, characterized in that, The time information includes year, annual day, hour, and minute; the geographical location is geographical latitude and longitude; the effective solar radiation is EUV×cosχ, where EUV is solar extreme ultraviolet radiation and χ is the solar zenith angle.
5. The method for reconstructing the peak electron density of the ionosphere according to claim 1, characterized in that, In step S2: the effective solar radiation is composed of the product of solar EUV flux and the cosine of the solar zenith angle, which is used to simulate the electron generation rate of the ionosphere. The thermosphere ΣO / N2 is observed by the TIMED / GUVI satellite. By adjusting the electron recombination loss coefficient and carrying atomic oxygen density information, the electron loss rate and generation rate are jointly affected.
6. The method for reconstructing the peak electron density of the ionosphere according to claim 1, characterized in that, In step S3: the data used for model training includes: NmF2 data retrieved from COSMIC-1 radio occultation observations, solar EUV radiation data observed by the SDO / EVE instrument, thermospheric ΣO / N2 data observed by the TIMED / GUVI satellite, and geomagnetic index Dst and solar activity index F10.7 from the OMNI database; all data are matched and aligned according to time, location and physical characteristics after quality control.
7. The method for reconstructing the peak electron density of the ionosphere according to claim 6, characterized in that, The model training data is divided into training, validation and test sets in an 8:1:1 ratio; the Adam optimizer is used with mean squared error (MSE) as the loss function, and an early stopping strategy is used to prevent model overfitting.
8. The method for reconstructing the peak electron density of the ionosphere according to claim 7, characterized in that, The Adam optimizer has an initial learning rate of 0.0002, a batch size of 1024, a maximum number of training epochs of 50, and an early stopping strategy that stops training if the verification loss does not decrease for 5 consecutive epochs.
9. The method for reconstructing the peak electron density of the ionosphere according to claim 1, characterized in that, The SRON2NN model has 4 hidden layers with 128, 256, 256, and 128 neurons respectively. The SRON2NN model is implemented based on the PyTorch framework.
10. The method for reconstructing the peak electron density of the ionosphere according to claim 1, characterized in that, In step S4: the model reconstruction range is the global region between 60° north and south latitude; the root mean square error (RMSE) of the SRON2NN model on the test set is 9.94 × 10⁻⁶. 4 –2.09×10 5 el / cm³.