A Machine Learning-Based Method for Modeling the Global Ionospheric Top Elevation
By constructing a global ionospheric top elevation model using machine learning technology, the challenge of modeling the ionospheric top elevation was solved, enabling accurate description of the top elevation and analysis of its spatial and temporal variations, thus improving the accuracy and reliability of the ionospheric model.
Patent Information
- Application Number
- CN202311283779.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-29
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2043-09-29
AI Technical Summary
Existing technologies are insufficient to establish reasonable elevation models above the top of the global ionosphere, lack the ability to analyze the variation characteristics of the top elevation at different times and geographical locations, and have insufficient research on its impact on solar activity.
Using machine learning methods, this study constructs a global ionospheric top elevation model using COSMIC occultation electron density profile data and an international reference ionospheric model, employing a random forest algorithm. The model uses solar and geomagnetic observation data, geographical location, and solar activity index as inputs to predict the temporal and spatial variations of the top elevation.
It has achieved accurate modeling of the top elevation of the global ionosphere, improved the reliability and accuracy of ionospheric models, and provided effective support for the design of ground-to-air radio information system links traversing the ionospheric environment.
Smart Images

Figure CN117554995B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of space environment situational awareness, and specifically relates to a machine learning-based method for modeling the global ionospheric top elevation. Background Technology
[0002] The ionosphere is a crucial space environment factor influencing satellite and shortwave communication, satellite navigation, radar, SAR, and other radio systems. A thorough understanding of the ionosphere is fundamental for the rational utilization of our living environment. Based on altitude, the ionosphere can be divided into the D, E, F, and upper ionosphere layers. Due to limitations in observation technology, traditional ionospheric research has primarily focused on the area below the F layer (approximately 250-350 km). Research on the upper ionosphere relies heavily on top-mounted vertical sounders (DLS) on satellites or space stations, but DLS data is extremely scarce, making it challenging to establish a more accurate upper ionospheric model. Ionospheric elevation is an inherent parameter for establishing electron density profile models above the upper ionosphere. Establishing a global model of the elevation parameters above the upper ionosphere is of significant practical value for improving the performance of radio systems traversing the ionosphere.
[0003] Elevation, containing information on ion and electron temperature and dynamic processes, is a crucial parameter for constructing ionospheric profile models. Currently, most studies use data from observatories for fixed-point elevation analysis, lacking the ability to analyze elevation variations across different times and geographical locations. The impact of solar activity levels on elevation also requires further investigation. There is a lack of research on the correlation between occultation data elevation and the peak height hmF2 and peak density NmF2 of the F2 layer. Identifying the key control parameters governing ionospheric elevation, and using these parameters as inputs to construct an optimized mathematical model based on large-scale space-based observational data, is a major challenge in ionospheric modeling. Summary of the Invention
[0004] The technical problem to be solved by this invention is to provide a method for modeling the global ionospheric top elevation based on machine learning.
[0005] The present invention adopts the following technical solution:
[0006] An improved method for modeling the global ionospheric top elevation based on machine learning includes the following steps:
[0007] Step 1, Acquisition and Preprocessing of Solar and Geomagnetic Observation Data:
[0008] Step 11, download the solar F10.7 and geomagnetic Dst indices;
[0009] Step 12: Arrange the solar F10.7 and geomagnetic Dst indices in ascending order of time;
[0010] Step 13: Set the missing values of solar F10.7 and geomagnetic Dst index to nan;
[0011] Step 14: Use linear interpolation to fill in the missing value nan.
[0012] Step 15: Output the observation time, solar F10.7 and geomagnetic Dst index and store them in a text file;
[0013] Step 2: Acquisition and preprocessing of occultation observation data, calculating the top elevation Hsc using the occultation electron density profile:
[0014] Step 21: Download COSMIC occultation electron density profile IonPrf observation data;
[0015] Step 22: Extract the observation time, collision point latitude, longitude, altitude coordinates, and ionospheric density value from the electron density profile;
[0016] Step 23: Use the average deviation data screening method to screen the initial data to remove irregular profile data:
[0017]
[0018] In the above formula, Ne i Let be the COSMIC electron density at point i. The electron density is the electron density after the moving average of the original electron density. The moving window is set to 8, and N is the total electron density on a profile.
[0019] Step 24: Extract the peak electron density NmF2 and peak ionospheric height hmF2 from the electron density profile;
[0020] NmF2=max{Ne i ,i=1,2,...,n}
[0021] hmF2=h j (Ne j ==NmF2) j=1,2,...n
[0022] In the above formula: max means taking the maximum value, h is the height of the collision point, and n is the total number of sampling points included in one occultation event;
[0023] Step 25: Define the ionosphere hmF2 as valid electron density profile data within the range of 150km-500km, and calculate the top elevation Hsc of the valid profile data:
[0024] Hsc = hsc - hmF2
[0025] In the above formula, hsc represents the height of the ionosphere NmF2 after it drops by 1 / e, where e is a natural number;
[0026] Step 26: Output the longitude and latitude coordinates of the collision point, the observation time, the ionosphere NmF2, the ionosphere hmF2, and the top elevation Hsc and store them in a text file;
[0027] Step 3: Calculate the ionospheric NmF2 and ionospheric hmF2 using the international reference ionospheric model:
[0028] Step 31: Input the altimeter coordinates, occultation collision point coordinates, observation time, and sunspot number into the international reference ionospheric model to calculate the ionospheric NmF2 at the corresponding time and location;
[0029] Step 32: Input the altimeter coordinates, occultation collision point coordinates, observation time, and sunspot number into the international reference ionospheric model to calculate the ionospheric hmF2 at the corresponding time and location;
[0030] Step 33: Output the ionosphere NmF2 and ionosphere hmF2 calculated by the international reference ionosphere model and store them in a text file;
[0031] Step 4: Machine learning to build a model of the global top elevation of the ionosphere:
[0032] Step 41: Normalize the annual day of birth (DOY) and local time (LT), and calculate their sine and cosine components respectively.
[0033]
[0034]
[0035]
[0036]
[0037] In the above formula, DOYs represents the sine component of the annual day DOY, DOYc represents the cosine component of the annual day DOY, LTs represents the sine component of the local time LT, and LTc represents the cosine component of the local time LT.
[0038] Step 42: Use DOYs, DOYc, LTs, LTc, geographic latitude, geographic longitude, solar activity index F10.7, geomagnetic Dst index, and ionospheric NmF2 and ionospheric hmF2 calculated by the international reference ionospheric model as training inputs for machine learning, and use the top elevation Hsc of the actual occultation observation as the output.
[0039] Step 43: Use the Random Forest (RF) algorithm for learning, with 6 leaves and a forest size of 330; use the bootstrap aggregating algorithm to train the learning network; randomly select 70% of the top ionospheric elevation (Hsc) samples for model training, 15% of the top ionospheric elevation (Hsc) samples for model testing, and 15% of the top ionospheric elevation (Hsc) samples for model accuracy evaluation.
[0040] Step 44: Store the trained Random Forest (RF) learning network model for calculating the top elevation Hsc of the ionosphere.
[0041] Furthermore, step 5, the accuracy evaluation of the Random Forest (RF) learning network model, is also included.
[0042] Step 51: Select the location of the occultation collision point, reconstruct the spatial distribution of the top elevation Hsc of the ionosphere, and compare the observed values with the model predictions;
[0043] Step 52: Calculate the root mean square error (RMSE) between the observed value and the model prediction of the ionospheric top elevation Hsc. The calculation formula is as follows:
[0044]
[0045] In the above formula, y obsi Let y be the i-th observation of the top elevation Hsc of the ionosphere. forei Let be the i-th model prediction value of the top elevation Hsc of the ionosphere, and m be the total number of samples;
[0046] Step 53, calculate the correlation coefficient ρ between the observed value and the model prediction of the ionospheric top elevation Hsc:
[0047]
[0048] In the above formula, This represents the mean of the observed values of the top elevation (Hsc) of the ionosphere. This represents the mean value of the Hsc model prediction for the top elevation of the ionosphere;
[0049] Step 54: Output the model accuracy evaluation results. The model accuracy is qualified if the RMSE is less than 30km and the ρ is greater than 0.8.
[0050] The beneficial effects of this invention are:
[0051] The modeling method disclosed in this invention is based on long-term electron density profile data of COSMIC occultation and uses machine learning methods to model the temporal and spatial variations of the global ionospheric top elevation, which can provide model support for the design of ground-to-air radio information system links traversing the ionospheric environment. Attached Figure Description
[0052] Figure 1 This is a flowchart illustrating the modeling method disclosed in this invention;
[0053] Figure 2 This is a schematic diagram of the top elevation Hsc of the ionosphere;
[0054] Figure 3 This is a diagram illustrating the input and output settings of the machine learning algorithm during the modeling process;
[0055] Figure 4 This is a comparison chart of the observed values and model predictions of the top elevation at LT12:00-13:00 during the spring equinox, summer solstice, autumn equinox, and winter solstice in 2013.
[0056] Figure 5 This is a schematic diagram of the model accuracy evaluation results. Detailed Implementation
[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0058] Currently, the amount of observational data available for the global ionosphere is increasing. Since the establishment of the COSMIC constellation in 2006, the number of satellites using radio occultation for ionospheric exploration has also been increasing, and the occultation electron density profile data they provide has laid an important foundation for global ionospheric modeling.
[0059] Machine learning can guide computers to find patterns in big data, learn suitable models from massive amounts of data, and fit complex nonlinear functions. It is often used to predict or estimate other unknown data. With the improvement of computer performance and the continuous increase in data volume, machine learning has made great strides. Applying machine learning technology to ionospheric modeling can further improve the reliability and accuracy of ionospheric modeling.
[0060] Example 1 discloses a machine learning-based method for modeling the global ionospheric top elevation, such as... Figure 1 As shown, it includes the following steps:
[0061] Step 1, Acquisition and Preprocessing of Solar and Geomagnetic Observation Data:
[0062] Step 11, download the solar F10.7 and geomagnetic Dst indices;
[0063] Step 12: Arrange the solar F10.7 and geomagnetic Dst indices in ascending order of time;
[0064] Step 13: Set the missing values of solar F10.7 and geomagnetic Dst index to nan;
[0065] Step 14: Use linear interpolation to fill in the missing value nan.
[0066] Step 15: Output the observation time, solar F10.7 and geomagnetic Dst index and store them in a text file;
[0067] Step 2: Acquisition and preprocessing of occultation observation data, calculating the top elevation Hsc using the occultation electron density profile:
[0068] Step 21: Download COSMIC occultation electron density profile IonPrf observation data;
[0069] Step 22: Extract the observation time, collision point latitude, longitude, altitude coordinates, and ionospheric density value from the electron density profile;
[0070] Step 23: Use the mean deviation data screening method to screen the initial data to remove irregular profile data:
[0071]
[0072] In the above formula, Ne i Let Ne be the COSMIC electron density at the i-th point. i The electron density is the electron density after the moving average of the original electron density. The moving window is set to 8, and N is the total electron density on a profile.
[0073] Step 24: Extract the peak electron density NmF2 and peak ionospheric height hmF2 from the electron density profile;
[0074] NmF2=max{Ne i ,i=1,2,...,n}
[0075] hmF2=h j (Ne j ==NmF2) j=1,2,...n
[0076] In the above formula: max means taking the maximum value, h is the height of the collision point, and n is the total number of sampling points included in one occultation event;
[0077] Step 25: Define the ionosphere hmF2 as valid electron density profile data within the range of 150km-500km, and calculate the top elevation Hsc of the valid profile data:
[0078] Hsc = hsc - hmF2
[0079] In the above formula, hsc represents the height of the ionosphere NmF2 after a drop of 1 / e, where e is a natural number. Specific parameters are as follows: Figure 2 As shown;
[0080] Step 26: Output the required occultation parameters, including the longitude and latitude coordinates of the collision point, the observation time, the ionosphere NmF2, the ionosphere hmF2, and the top elevation Hsc, and store them in a text file.
[0081] Step 3: Calculate the ionospheric NmF2 and ionospheric hmF2 using the international reference ionospheric model:
[0082] Step 31: Input the altimeter coordinates, occultation collision point coordinates, observation time, sunspot number and other parameters into the international reference ionospheric model to calculate the ionospheric NmF2 at the corresponding time and location;
[0083] Step 32: Input the altimeter coordinates, occultation collision point coordinates, observation time, sunspot number and other parameters into the international reference ionospheric model to calculate the ionospheric hmF2 at the corresponding time and location;
[0084] Step 33: Output the ionosphere NmF2 and ionosphere hmF2 calculated by the international reference ionosphere model and store them in a text file;
[0085] Step 4: Machine learning to build a model of the global top elevation of the ionosphere:
[0086] Step 41: Normalize the annual day of birth (DOY) and local time (LT), and calculate their sine and cosine components respectively.
[0087]
[0088]
[0089]
[0090]
[0091] In the above formula, DOYs represents the sine component of the annual day DOY, DOYc represents the cosine component of the annual day DOY, LTs represents the sine component of the local time LT, and LTc represents the cosine component of the local time LT.
[0092] Step 42, as Figure 3As shown, DOYs, DOYc, LTs, LTc, geographic latitude, geographic longitude, solar activity index F10.7, geomagnetic Dst index, and ionospheric NmF2 and ionospheric hmF2 calculated by the International Reference Ionospheric Model (IRI) are used as training inputs for machine learning, and the top elevation Hsc of the actual occultation observation is used as the output.
[0093] Step 43: Use the Random Forest (RF) algorithm for learning, with 6 leaves and a forest size of 330; use the bootstrap aggregating algorithm to train the learning network; randomly select 70% of the top ionospheric elevation (Hsc) samples for model training, 15% of the top ionospheric elevation (Hsc) samples for model testing, and 15% of the top ionospheric elevation (Hsc) samples for model accuracy evaluation.
[0094] Step 44: Store the trained Random Forest (RF) learning network model in the computer for calculating the top elevation Hsc of the ionosphere.
[0095] Step 5, Accuracy evaluation of the Random Forest (RF) learning network model:
[0096] Step 51: Select the location of the occultation collision point, reconstruct the spatial distribution of the top ionospheric elevation Hsc, and compare the observed values with the model predictions. The comparison results are as follows: Figure 4 As shown;
[0097] Step 52: Calculate the root mean square error (RMSE) between the observed value and the model prediction of the ionospheric top elevation Hsc. The calculation formula is as follows:
[0098]
[0099] In the above formula, y obsi Let y be the i-th observation of the top elevation Hsc of the ionosphere. forei Let be the i-th model prediction value of the top elevation Hsc of the ionosphere, and m be the total number of samples; RMSE can reflect the estimation performance of the model;
[0100] Step 53, calculate the correlation coefficient ρ between the observed value and the model prediction of the ionospheric top elevation Hsc:
[0101]
[0102] In the above formula, This represents the mean of the observed values of the top elevation (Hsc) of the ionosphere. ρ represents the mean of the model-predicted values of the top elevation of the ionosphere, Hsc; ρ reflects the correlation between the observed values and the model predictions.
[0103] Step 54: Output the model accuracy evaluation results, which will serve as the basis for model prediction error analysis. The accuracy evaluation results are shown below. Figure 5 The model is considered to have acceptable accuracy if the RMSE is less than 30km and the ρ is greater than 0.8.
Claims
1. A method for modeling the global ionospheric top elevation based on machine learning, characterized in that, Includes the following steps: Step 1, Acquisition and Preprocessing of Solar and Geomagnetic Observation Data: Step 11, download the solar F10.7 and geomagnetic Dst indices; Step 12: Arrange the solar F10.7 and geomagnetic Dst indices in ascending order of time; Step 13: Set the missing values of solar F10.7 and geomagnetic Dst index to nan; Step 14: Use linear interpolation to fill in the missing value nan. Step 15: Output the observation time, solar F10.7 and geomagnetic Dst index and store them in a text file; Step 2: Acquisition and preprocessing of occultation observation data, calculating the top elevation Hsc using the occultation electron density profile: Step 21: Download COSMIC occultation electron density profile IonPrf observation data; Step 22: Extract the observation time, collision point latitude, longitude, altitude coordinates, and ionospheric density value from the electron density profile; Step 23: Use the average deviation data screening method to screen the initial data to remove irregular profile data: In the above formula, Let be the COSMIC electron density at point i. The electron density is the electron density after the moving average of the original electron density. The moving window is set to 8, and N is the total electron density on a profile. Step 24: Extract the peak electron density NmF2 and peak ionospheric height hmF2 from the electron density profile; In the above formula: max represents taking the maximum value. Where is the height of the collision point, and n is the total number of sampling points included in one occultation event; Step 25: Define the ionosphere hmF2 as valid electron density profile data within the range of 150km-500km, and calculate the top elevation Hsc of the valid profile data: In the above formula, hsc represents the height of the ionosphere NmF2 after a drop of 1 / e, and e is the natural constant; Step 26: Output the longitude and latitude coordinates of the collision point, the observation time, the ionosphere NmF2, the ionosphere hmF2, and the top elevation Hsc and store them in a text file; Step 3: Calculate the ionospheric NmF2 and ionospheric hmF2 using the international reference ionospheric model: Step 31: Input the altimeter coordinates, occultation collision point coordinates, observation time, and sunspot number into the international reference ionospheric model to calculate the ionospheric NmF2 at the corresponding time and location; Step 32: Input the altimeter coordinates, occultation collision point coordinates, observation time, and sunspot number into the international reference ionospheric model to calculate the ionospheric hmF2 at the corresponding time and location; Step 33: Output the ionosphere NmF2 and ionosphere hmF2 calculated by the international reference ionosphere model and store them in a text file; Step 4: Machine learning to build a model of the global top elevation of the ionosphere: Step 41: Normalize the annual day of birth (DOY) and local time (LT), and calculate their sine and cosine components respectively. In the above formula, This represents the sine component of the accumulated days DOY. This represents the cosine component of the accumulated day DOY. This represents the sinusoidal component of local time LT. Represents the cosine component of the local time LT; Step 42: Use DOYs, DOYc, LTs, LTc, geographic latitude, geographic longitude, solar activity index F10.7, geomagnetic Dst index, and ionospheric NmF2 and ionospheric hmF2 calculated by the international reference ionospheric model as training inputs for machine learning, and use the top elevation Hsc of the actual occultation observation as the output. Step 43: Use the Random Forest (RF) algorithm for learning, with 6 leaves and a forest size of 330; use the bootstrap aggregation algorithm to train the learning network; randomly select 70% of the top elevation (Hsc) samples of the ionosphere for model training, 15% of the top elevation (Hsc) samples of the ionosphere for model testing, and 15% of the top elevation (Hsc) samples of the ionosphere for model accuracy evaluation. Step 44: Store the trained Random Forest (RF) learning network model for calculating the top elevation Hsc of the ionosphere.
2. The method for modeling the global ionospheric top elevation based on machine learning according to claim 1, characterized in that: It also includes step 5, accuracy evaluation of the Random Forest (RF) learning network model: Step 51: Select the location of the occultation collision point, reconstruct the spatial distribution of the top elevation Hsc of the ionosphere, and compare the observed values with the model predictions; Step 52: Calculate the root mean square error (RMSE) between the observed value and the model prediction of the ionospheric top elevation Hsc. The calculation formula is as follows: In the above formula, Let Hsc be the i-th observation of the top elevation of the ionosphere. Let Hsc be the i-th model prediction value of the top elevation of the ionosphere, where m is the total number of samples. Step 53: Calculate the correlation coefficient between the observed value and the model prediction of the top elevation Hsc of the ionosphere. : In the above formula, This represents the mean of the observed values of the top elevation (Hsc) of the ionosphere. This represents the mean value of the Hsc model prediction for the top elevation of the ionosphere; Step 54: Output the model accuracy evaluation results, setting the RMSE to be less than 30 km. Models with an accuracy greater than 0.8 are considered to have acceptable accuracy.
Citation Information
Patent Citations
Ionized layer tomography method based on multi-source data driving
CN114384564A
Three-dimensional ionosphere electron concentration distribution model training method based on deep learning
CN115062526A
