A method for constructing an ionospheric model during a storm in the Yunnan-Sichuan region of China

Abstract

The present invention discloses a method for constructing an ionosphere model during storms in the Yunnan-Sichuan region of China. The model is based on a sparrow search algorithm (SSA), a convolutional long short-term memory network (ConvLSTM) and a bidirectional long short-term memory network (BiLSTM), and can learn the complex nonlinear relationship between multiple factors of the ionosphere from a large number of sample events, thereby having a strong prediction capability and being able to simulate the complex spatiotemporal variation characteristics of the ionosphere in a small area of ​​low latitudes under extreme space weather, thereby enhancing the satellite navigation and positioning services in the region.

Description

Technical Field

[0001] The present invention relates to the technical field of model construction, and in particular to a method for constructing an ionospheric model during a storm in the Yunnan-Sichuan region of China. Background Art

[0002] Changes in the electron content of the ionosphere seriously affect the propagation of radio signals and pose a serious threat to the services of the Global Navigation Satellite System (GNSS). An accurate total electron content model (TEC) can effectively correct the delay effect of the ionosphere on radio signals and improve the navigation and positioning accuracy of the GNSS system. It also provides support for optimizing communication frequencies and propagation paths, and enhances the reliability and stability of the communication system. The magnitude of ionosphere changes is closely related to space weather. Solar activity and geomagnetic activity are the main factors affecting the drastic changes in the ionosphere. In addition, many geological disasters can also cause significant ionospheric fluctuations. Total electron content (TEC) is an important parameter to characterize ionospheric activity. Therefore, building a high-precision TEC model can not only reduce the impact of extreme weather and improve the accuracy of satellite navigation and positioning, but also warn of geological disasters and improve the social disaster prevention and mitigation capabilities. For example, from August 25 to 26, 2018, a large geomagnetic storm occurred on the earth, which not only caused a serious ionospheric delay effect, significantly reduced the accuracy of navigation and positioning, but also affected the stability of the low-orbit satellite orbit.

[0003] The Yunnan-Sichuan region is located in the southwest of China, spanning the middle and low latitudes, with rich landforms, including plateaus, mountains, basins and plains. The region has large altitude changes and is seriously affected by multipath effects. Therefore, studying the characteristics of the ionosphere in this region is important for optimizing radio communication systems and improving communication quality and coverage. In addition, the Yunnan-Sichuan region is located in the north of the Equatorial Ionization Anomaly (EIA). Under the combined effects of magnetic fields, electric fields, neutral winds, etc., the ionosphere has significant periodic changes. During the day, solar radiation is strong, ionization is enhanced, and electron density increases. At night, light ionization weakens and electron density decreases. In addition, seasonal changes will affect the angle and intensity of solar radiation, and thus affect the electron density distribution in the ionosphere. At the same time, the ionosphere in low-latitude regions is very sensitive to solar activity. Solar winds and solar flares can cause disturbances in the ionosphere, affecting disturbances in the ionosphere. During a magnetic storm, the TEC value will fluctuate violently, causing signal attenuation and scattering. This patent mainly uses the Dst index as a reference indicator. When Dst≤-30nT, it is defined as a weak magnetic storm. When Dst≤-50nT, it is defined as a moderate magnetic storm. When Dst≤-100, it is defined as a strong magnetic storm.

[0004] Regarding the research on the TEC model of the ionosphere in a small low-latitude area during a storm, the neural network models constructed by domestic scholars such as Convolutional Long Short-Term Memory Network (ConvLSTM) and Long Short-Term Memory Network based on Attention Mechanism (LSTM-Attention) have certain limitations. These traditional neural network models are usually difficult to fully capture and characterize the complex spatiotemporal evolution characteristics of the ionosphere in a small low-latitude area. In order to improve the prediction accuracy, it is necessary to build a composite model to enhance the ability to predict the time-varying characteristics of the ionosphere.

[0005] Therefore, this paper proposes a method to construct an ionospheric model during storm events in the Yunnan-Sichuan region of China. Summary of the invention

[0006] The purpose of the present invention is to design a method for constructing a storm ionosphere model in the Yunnan-Sichuan region of China, which can well combine the advantages of a convolutional long short-term memory network (ConvLSTM) and a bidirectional long short-term memory network (BiLSTM) and has a powerful ability to learn and predict the spatiotemporal evolution of the storm ionosphere.

[0007] In order to achieve the above technical effects, the present invention is implemented by the following technical solutions: A method for constructing a storm ionosphere model in the Yunnan-Sichuan region of China, characterized in that it includes the following steps:

[0008] S1. Select appropriate GNSS stations in the Yunnan-Sichuan region, which cover the target research area and can stably receive Beidou satellite navigation system signals; download observation data files in RINEX format from the data center;

[0009] S2. Combine median filtering and bicubic interpolation techniques to reduce data noise, pre-process the original ionospheric electron content data, and improve data smoothness;

[0010] S3, using XGBoost algorithm to extract the features that best match the ionospheric total electron content (TEC) data;

[0011] S4. The sparrow search algorithm (SSA), convolutional long short-term memory network (ConvLSTM) and bidirectional long short-term memory network (BiLSTM) are combined to form an ionospheric model of the Yunnan-Sichuan region of China during storms based on SSA-ConvLSTM-BiLSTM, which can better simulate the changes in the total electron content (TEC) of the ionosphere in the Yunnan-Sichuan region during storms.

[0012] Furthermore, in S1, the observation data includes pseudorange, carrier phase and signal strength of the satellite signal.

[0013] Furthermore, in S1, the observation data comes from the observation data of the dual-frequency GNSS receiver. The dual-frequency observation can eliminate the influence of ionospheric delay. For signals of two different frequencies, the ionospheric delay can be calculated by the following formula:

[0014]

[0015] Among them, P1 and P2 are pseudoranges measured at two different frequencies; f1 and f2 are signals of two different frequencies;

[0016] Ionospheric delay is also affected by other error sources, such as multipath effect, satellite clock error, etc.; these errors can be corrected by the following formula:

[0017] P corrected =P smooth +d iono +d multi +d clock

[0018] Among them, P corrected is the corrected pseudorange observation, d iono is the ionospheric delay correction term, d multi is the multipath correction term, d clock is the satellite clock error correction term.

[0019] Furthermore, in S2, the median filter reduces data noise specifically by: using the median filter for each data point in the signal, taking the median of the data points of a certain window size around it as a new data point; this is suitable for signals carrying impulse noise and can effectively remove outliers; it is expressed as follows:

[0020] y i =median(x i-w / / 2 ,…,x i+w / / 2 )

[0021] Among them, y i is the value at the middle position of the data point, w represents the window size, which determines the number of neighboring data points involved in the calculation; i is the middle value after sorting, and for odd window sizes, it is exactly the (w / / 2+1)th data point.

[0022] Further, in S2, the bicubic interpolation technology reduces data noise specifically by: using a cubic polynomial to interpolate four nearest neighbor data points; bicubic interpolation can produce a smooth interpolation result, effectively eliminate the jagged effect of bilinear interpolation, and is suitable for scenarios requiring high-quality interpolation; the bicubic interpolation formula can be expressed as:

[0023]

[0024] Among them, f(x,y) is the interpolation value at the point (x,y); a ij There are 16 coefficients that need to be solved; B i (x) and B j (y) is a cubic basis function.

[0025] Furthermore, in S3, the use of the XGBoost algorithm to extract the features that best match the ionospheric total electron content (TEC) data is specifically as follows: 23 solar-geomagnetic activity characteristic parameters are imported into the XGBoost algorithm; in the XGBoost model, the number of trees, the depth of the trees, and the learning rate are defined; the XGBoost model is trained using the selected optimal parameter combination and evaluated on the test set, and the evaluation indicator can be selected as the root mean square error (RMSE) or the R square indicator.

[0026] Furthermore, in S4, the better simulation of the changes in the total electron content (TEC) of the ionosphere in the Yunnan-Sichuan region during the storm is as follows: the selected solar-geomagnetic parameters and TEC data are organized into sequence data in chronological order, and each time contains all channel data; the data is then normalized to ensure that the data of different channels are in the same numerical range, and the data is imported into the SSA-ConvLSTM-BiLSTM model; the sparrow search algorithm (SSA) is used to determine the parameter matching of the model in the first stage, and the ConvLSTM-BiLSTM network is used in the second stage to learn the extracted feature sequence and model the spatiotemporal relationship between the sequences; in the process of determining the parameters, K-fold cross validation is used, and the intervention of the test set is not required.

[0027] The beneficial effects of the present invention are:

[0028] (1) Compared with the traditional subjective selection of solar-geomagnetic parameters as multi-channel features, the XGBoost algorithm used in this patent can better select the most suitable solar-geomagnetic parameter combination according to the characteristics of TEC data and improve the prediction accuracy;

[0029] (2) The ConvLSTM-BiLSTM model constructed by the present invention can better learn the TEC distribution characteristics of low-latitude Chinese regions during storms, combined with the most suitable solar-geomagnetic parameters selected by XGBoost, such as: SunspotR, LymanAlpha, Dst, F10.7, etc.; it is helpful to reduce the impact of ionospheric TEC anomalies caused by storms, and can help reduce the negative impact of magnetic storms on the ionosphere in small low-latitude areas in China;

[0030] (3) Compared with traditional ConvLSTM, CNN-GRU and other models, the SSA-ConvLSTM-BiLSTM model shows higher accuracy in ionospheric TEC prediction in small low-latitude areas of China. Traditional neural network models are usually unable to effectively capture the spatiotemporal dependencies in time series data, while the SSA-ConvLSTM-BiLSTM model can better learn and model complex spatiotemporal relationships by combining spatiotemporal feature extraction, convolutional long short-term memory network (ConvLSTM) and bidirectional long short-term memory network (BiLSTM), thereby improving prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0032] Figure 1 It is a flow chart of data processing of the raw ionospheric data of the Beidou Navigation Satellite System (BDS) received by the GNSS station according to the present invention;

[0033] Figure 2 It is the ionospheric TEC fitting diagram of the Yunnan-Sichuan region at 0:00 and 12:00 on the 248th day to the 251st day during the storm period in 2017 described in Example 2 of the present invention;

[0034] Figure 3 This is a heat map of the correlation coefficients between the three models and the true values ​​during the 2017 and 2018 storms;

[0035] Figure 4 It is the TEC fitting diagram of the ionosphere in the Yunnan-Sichuan region of each model from 0:00 on the 248th day to the 251st day during the storm in 2017 described in Example 2 of the present invention;

[0036] Figure 5 It is a line comparison chart of the predicted values ​​of each model described in the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] Example 1

[0039] Based on the extraction and processing of ionospheric total electron content (TEC) data in low-latitude areas of China by the Beidou Navigation Satellite System (BDS), this paper constructs an optimized neural network model. First, the original ionospheric TEC observation data in the Yunnan-Sichuan region are collected, and the median filter and bicubic interpolation techniques are combined to reduce data noise and improve data smoothness; then, the XGBoost algorithm is used to extract the features that best match the TEC data; finally, the convolutional long short-term memory network (ConvLSTM) and the bidirectional long short-term memory network (BiLSTM) are combined to better simulate the changes in ionospheric TEC in the Yunnan-Sichuan region during the storm.

[0040] The specific steps are:

[0041] Step 1: Dealing with signal propagation errors

[0042] This patent uses dual-frequency observation to eliminate the influence of ionospheric delay. For two signals with different frequencies (such as f1 and f2), the ionospheric delay can be calculated by the following formula. Among them, P1 and P2 are pseudoranges measured at two different frequencies;

[0043]

[0044] Ionospheric delay is also affected by other error sources, such as multipath effect, satellite clock error, etc. These errors can be corrected by the following formula;

[0045] P corrected =P smooth +d iono +d multi +d clock

[0046] Among them, P corrected is the corrected pseudorange observation, d iono is the ionospheric delay correction term, d multi is the multipath correction term, d clock is the satellite clock error correction term.

[0047] Step 2: Median filtering to remove noise

[0048] Median filtering is a nonlinear filtering method mainly used in signal processing. Median filtering has a good suppression effect on impulse noise and can effectively remove isolated spike noise in images or signals. The expression is as follows:

[0049] y i =median(x i-w / / 2 ,…,x i+w / / 2 )

[0050] Among them, y iis the value at the middle position of the data point, w represents the window size, which determines the number of neighboring data points involved in the calculation; i is the middle value after sorting. For odd window sizes, it is exactly the (w / / 2+1)th data point, and the symbol “ / / ” means rounding down.

[0051] Step 3: Bicubic interpolation to grid the TEC data

[0052] Bicubic interpolation is an interpolation method commonly used in image processing and digital signal processing. This interpolation uses a cubic polynomial to estimate the value of the interpolation point in each interpolation interval, and uses an adjacent 4*4 network to calculate the value of each interpolation point. The bicubic interpolation formula can be expressed as:

[0053]

[0054] Where f(x,y) is the interpolated value at the point (x,y). ij There are 16 coefficients that need to be solved. i (x) and B j (y) is a cubic basis function.

[0055] Step 4: XGBoost parameter selection

[0056] XGboost parameters can be divided into three categories: general parameters, boost parameters, and learning target parameters. In the general parameters, select "gbtree", a tree-based model. In the boost parameters, set "max_depth", i.e. the maximum depth of the tree, to 8. The evaluation indicator selects the root mean square error (RMSE) as the final test method.

[0057] Step 5: SSA-ConvLSTM-BiLSTM

[0058] The sparrow search algorithm (SSA) and K-fold cross validation help determine the parameters and structure of the neural network model. When the TEC data passes through the ConvLSTM network model, its spatiotemporal features are fully captured. When it passes through the BiLSTM layer, the neural network model can better understand the changes in the context of the data.

[0059] Example 2

[0060] This embodiment is a specific implementation case, which is as follows:

[0061] This patent is further explained in detail using the BDS monitoring station data in Yunnan and Sichuan provinces.

[0062] The data from the 248th to the 251st day during the 2017 storm are selected as examples to illustrate the use of this patent;

[0063] Step 1: Select the monitoring station data from the 248th to the 251st day of 2017 and process them using the TEC data processing script code to reduce the interference caused by ionospheric delay, multipath effect and other problems;

[0064] Step 2: Use median filtering to remove outliers and ensure that the data is within the normal fluctuation range;

[0065] Step 3: Use the bicubic interpolation method to grid the TEC data to facilitate subsequent ionospheric modeling and simulation work;

[0066] Step 4: Use the XGBoost algorithm to select the solar-geomagnetic parameters that are most suitable for the prediction of this period;

[0067] Step 5: Combine the sparrow search algorithm (SSA) with K-fold cross validation, set the range of parameters such as filters and kernel_size, so as to facilitate the subsequent matching of different parameters; then adjust all channel data to the target format and import it into the model to obtain the prediction results.

[0068] Example 3

[0069] In this embodiment, two traditional models (Convolutional Long Short-Term Memory Network Model (ConvLSTM) and Bidirectional Long Short-Term Memory Network Model (BiLSTM)) are used as comparison models, and the correlation coefficients between them and the SSA-ConvLSTM-BiLSTM model described in this application are studied under the storm environment in 2017 and 2018 respectively.

[0070] Specific steps: Calculate the correlation coefficient between the true value, SSA-ConvLSTM-BiLSTM and the comparison models CNN-GRU and LSTM-CNN; the closer the correlation coefficient is to 1, the higher the correlation is. At the same time, the correlation coefficient can help determine the number of outliers. If there are more outliers, the correlation coefficient will also decrease.

[0071] Get as Figure 3 As shown in the figure, the correlation coefficients of the predicted values ​​of the three models and the true values ​​are compared. The results show that the correlation coefficient of this model is significantly higher than that of the two traditional models in the magnetic storm environment in 2017 and 2018, that is, the prediction effect of this model is better than that of the traditional model.

Claims

1. A method for constructing a storm ionospheric model in the Yunnan-Sichuan region of China, characterized in that: The following steps are involved: S1. Select GNSS stations in the Yunnan-Sichuan region, which cover the target research area and can stably receive Beidou satellite navigation system signals; download observation data files in RINEX format from the data center; The observation data comes from the observation data of the dual-frequency GNSS receiver. The dual-frequency observation eliminates the influence of ionospheric delay. For signals of two different frequencies, the ionospheric delay is calculated by the following formula: Among them, P1 and P2 are pseudoranges measured at two different frequencies; f1 and f2 are signals of two different frequencies; Ionospheric delay is also affected by other error sources, including multipath effects and satellite clock errors; these errors are corrected by the following formula: P corrected =P smooth +d iono +d multi +d clock Among them, P corrected is the corrected pseudorange observation, d iono is the ionospheric delay correction term, d multi is the multipath correction term, d clock is the satellite clock error correction term; S2. Combine median filtering and bicubic interpolation techniques to reduce data noise, pre-process the original ionospheric electron content data, and improve data smoothness; S3. Use the XGBoost algorithm to extract the features that best match the TEC data of the ionosphere, by importing 23 solar-geomagnetic activity characteristic parameters into the XGBoost algorithm; in the XGBoost model, define the number of trees, the depth of the tree, and the learning rate; use the selected optimal parameter combination to train the XGBoost model and evaluate it on the test set, using the root mean square error RMSE or R square indicator as the evaluation indicator; S4. Combine the sparrow search algorithm SSA, the convolutional long short-term memory network ConvLSTM and the bidirectional long short-term memory network BiLSTM to form a storm-time ionosphere model for the Yunnan-Sichuan region of China based on SSA-ConvLSTM-BiLSTM; then combine the selected solar-geomagnetic parameters and TEC data into sequence data in chronological order, and each time contains all channel data; then normalize the data to ensure that the data of different channels are in the same numerical range, and import the data into the SSA-ConvLSTM-BiLSTM model; in the first stage, the sparrow search algorithm SSA is used to determine the parameter matching of the model, and in the second stage, the ConvLSTM-BiLSTM network is used to learn the extracted feature sequences and model the spatiotemporal relationship between the sequences; in the process of determining the parameters, K-fold cross validation is used, and the intervention of the test set is not required; better simulate the changes in the total electron content TEC of the ionosphere in the Yunnan-Sichuan region during storms.

2. The method for constructing a storm ionospheric model in the Yunnan-Sichuan region of China according to claim 1, characterized in that: In S1, the observation data includes pseudorange, carrier phase and signal strength of satellite signals.

3. The method for constructing a storm ionospheric model in the Yunnan-Sichuan region of China according to claim 1, characterized in that: In S2, the median filter reduces data noise specifically by: using the median filter for each data point in the signal, taking the median of the data points of a certain window size around it as a new data point; this is suitable for signals carrying impulse noise and can effectively remove outliers; it is expressed as follows: y i =median(x i-w / / 2 ,…,x i+w / / 2 ) Among them, y i is the value at the middle position of the data point, w represents the window size, which determines the number of adjacent data points involved in the calculation; i is the middle value after sorting. For odd window sizes, it is exactly the w / / 2+1th data point, and the symbol " / / " means rounding down.

4. The method for constructing a storm ionospheric model in the Yunnan-Sichuan region of China according to claim 1, characterized in that: In S2, the bicubic interpolation technology reduces data noise by using a cubic polynomial to interpolate four nearest neighbor data points. Bicubic interpolation can produce smooth interpolation results and effectively eliminate the jagged effect of bilinear interpolation, which is suitable for scenarios requiring high-quality interpolation. The bicubic interpolation formula is expressed as: Among them, f(x,y) is the interpolation value at the point (x,y); a ij There are 16 coefficients that need to be solved; B i (x) and B j (y) is a cubic basis function.

Images