Ionosphere parameter and short wave frequency band prediction method and electronic equipment

Abstract

The invention provides an ionosphere parameter and short wave frequency band prediction method and electronic equipment, and the method comprises the steps: obtaining the ionosphere parameter data in a corresponding IRI model above the midpoint of a great circle through obtaining the geographic coordinates of two places of a communication link and the midpoint of the great circle, obtaining the geomagnetic solar activity data of an open-source data website, and obtaining the geomagnetic solar activity data of the open-source data website; the method comprises the following steps: carrying out ionosphere oblique detection by using ionosphere detectors at two communication places, obtaining an ionosphere oblique detection ionogram, obtaining frequency band data available for short-wave communication from the oblique detection ionogram, carrying out data preprocessing, selecting different tasks, setting corresponding Transform-LSTM model hyper-parameters, carrying out model training, and storing optimal model parameters; and performing an ionosphere parameter prediction task or a short-wave communication available frequency band prediction task by using the corresponding model. The ionosphere detection data, the environment monitoring data and the like are fused in the deep learning field and the short-wave communication field, and the method has important significance and application prospects.

Description

Technical Field

[0001] This invention relates to the field of ionospheric detection technology, and in particular to a method and electronic device for predicting ionospheric parameters and shortwave band frequencies. Background Technology

[0002] Shortwave communication is an important form of radio communication. After being transmitted, the radio waves are reflected by the ionosphere to reach the receiving equipment. It has a long communication distance and is the only long-range communication method not limited by active repeaters. Through multiple reflections from the ionosphere, shortwave communication can cover the entire globe, including special regions such as mountainous areas, deserts, and oceans that are inaccessible to ultra-shortwave radio. Shortwave communication does not require satellites, has low construction and maintenance costs, and possesses extremely high flexibility, resilience, and stability.

[0003] The ionosphere is a crucial channel for shortwave communication, and its time-varying dispersive properties significantly impact communication quality. Because the ionosphere's state is greatly influenced by solar activity and seasonal variations, and anomalous states are frequent, real-time shortwave communication practices necessitate on-site switching of communication frequencies based on the ionospheric state to ensure communication quality and stability. Therefore, frequency selection based on ionospheric parameters is critical for shortwave communication.

[0004] Time series forecasting is a method that predicts future values ​​by analyzing historical patterns in time series data, and it has wide applications in many fields. Ionospheric parameter forecasting can be broadly divided into long-term and short-term forecasts. The International Reference Ionospheric (IRI) model is internationally recognized as a long-term forecasting model, capable of predicting multiple ionospheric parameters globally. During periods of geomagnetic quiet or stable ionospheric changes, the estimates provided by the IRI model are relatively accurate. However, due to the ionospheric variation characteristics, when ionospheric disturbances occur caused by factors such as solar activity or geomagnetic storms, relying solely on the IRI model's estimates will result in significant errors.

[0005] The autonomous frequency selection technology currently used in shortwave communication integrates spectrum sensing technology and historical communication data, demonstrating superior performance. However, this technology fails to consider the ionospheric factor, which has a significant impact on shortwave transmission, and instead relies solely on noise and historical experience data for real-time frequency selection. Because it ignores the influence of ionospheric parameters, frequency selection based on a single historical communication data source suffers from poor stability during ionospheric anomalies. Summary of the Invention

[0006] This invention provides a method and electronic device for predicting ionospheric parameters and shortwave frequency bands, addressing the deficiencies in existing technologies. It acquires ionospheric parameter data corresponding to the IRI model near the relay points of two shortwave stations, obtains parameter data related to global geomagnetic and solar activity, and acquires historical oblique detection data from the two shortwave stations using ionospheric detectors. Data preprocessing is performed, and depending on the task type (e.g., ionospheric parameter prediction or shortwave communication available frequency band prediction), different hyperparameters and processed data are used to train the deep learning model Transformer-LSTM. By inputting historical ionospheric parameter data, historical geomagnetic and solar activity index data, and historical detection data for a specified duration, and combining the trained model parameters, the predicted results of ionospheric parameters or shortwave communication available frequency bands for a specified future duration can be obtained.

[0007] In a first aspect, the present invention provides a method for predicting ionospheric parameters and shortwave band frequencies, comprising: Obtain the geographical coordinates of the transmitting and receiving stations, as well as the ionospheric parameters corresponding to the international reference ionospheric model near the relay point, and obtain global geomagnetic activity parameters and solar activity parameters; Using the ionospheric detectors at the transmitting and receiving stations, oblique frequency sweep detection is performed to obtain oblique ionosphere ionization maps; Post-processing of the ionosphere oblique probe ionization map yields oblique probe frequency band data; Preprocessing of global geomagnetic activity parameters, solar activity parameters, and oblique detection frequency band data is performed to select data with the same time resolution within consecutive time periods; Construct a Transformer-LSTM combined model, train the model using specified preprocessed data, and obtain an ionospheric parameter prediction model or a shortwave band prediction model. Input historical data of a specified duration and type into the ionospheric parameter prediction model or the shortwave band prediction model, and output the ionospheric parameter prediction results or the shortwave band prediction results.

[0008] According to the method for predicting ionospheric parameters and shortwave frequencies provided by the present invention, the geographical coordinates of transmitting and receiving stations are obtained, including: The transmitting station and the receiving station are represented by geographic coordinates, and the latitude and longitude coordinates of the midpoint of the great circle between the two stations are calculated using geographic coordinates. Input the time range, resolution, latitude and longitude coordinates of the midpoint of the great circle, altitude range above the ground, and resolution into the International Reference Ionospheric Model.

[0009] According to the present invention, an ionospheric parameter and shortwave band prediction method is provided to obtain ionospheric parameters corresponding to the international reference ionospheric model near the relay point of the transmitting and receiving stations, including: Historical data of ionospheric parameters were obtained using the international reference ionospheric model. By calling the international reference ionospheric model and inputting the time range, resolution, latitude and longitude coordinates of the midpoint of the great circle, altitude range above the ground, and resolution, the critical frequency, peak height, half thickness, and distribution of electron concentration in space of the ionosphere at the midpoint of the great circle between the two stations can be obtained.

[0010] According to the present invention, an ionospheric parameter and shortwave band prediction method is provided, which utilizes ionospheric detectors at transmitting and receiving stations to perform oblique frequency sweep detection to obtain oblique ionosphere ionization maps, including: The transmitting station uses an inverted V antenna to transmit an m-sequence swept frequency modulation signal. The start frequency, end frequency, and step frequency of the sweep are set according to different ionospheric states. The oblique sounding signal is received at a remote location using a two-wire dipole antenna. By processing the received oblique sounding signal, an oblique sounding ionosphere map is drawn.

[0011] According to the ionospheric parameter and shortwave band prediction method provided by the present invention, the ionosphere oblique probe ionogram is post-processed to obtain oblique probe frequency band data, including: The projection method is used to focus on the maximum signal amplitude value corresponding to each frequency point within the swept frequency band. Using the threshold method, the start and end frequencies corresponding to the signal frequency bands are extracted using a preset threshold and used as oblique detection frequency band data.

[0012] According to the ionospheric parameter and shortwave band prediction method provided by the present invention, global geomagnetic activity parameters, solar activity parameters and oblique sounding frequency band data are preprocessed, and data with the same time resolution in continuous time periods are selected, including: By using a unified time range and time resolution, missing data interpolation and outlier data removal are performed to obtain preprocessed data.

[0013] According to the ionospheric parameter and shortwave band prediction method provided by the present invention, a Transformer-LSTM combined model is constructed, including: The Transformer model is determined to include only the encoder part of the original Transformer. The LSTM layer in the LSTM model is combined with the feedforward neural network, residual connection and layer normalization of the decoder of the original Transformer to form the Transformer-LSTM combined model. The key model parameters, including sequence length and prediction length, are determined for model training. After model training is completed, the optimal model parameters are saved.

[0014] According to the present invention, an ionospheric parameter and shortwave band prediction method is provided, which uses specified preprocessed data for model training to obtain an ionospheric parameter prediction model or a shortwave band prediction model, including: A model for predicting ionospheric parameters was obtained by training the model using global geomagnetic activity parameters and solar activity parameters. A shortwave band prediction model was obtained by training the model using global geomagnetic activity parameters, solar activity parameters, and ionosphere oblique probe ionization maps.

[0015] According to the present invention, an ionospheric parameter and shortwave band prediction method is provided, which inputs historical data of specified duration and type into an ionospheric parameter prediction model or a shortwave band prediction model, and outputs ionospheric parameter prediction results or shortwave band prediction results, including: Load the optimal parameters of the trained model and determine the historical data corresponding to the ionospheric parameter prediction and / or shortwave band prediction; After inputting historical data of sequence length, the model outputs predicted data of the predicted length following the end time of the sequence length. The start time of the time series corresponding to the sequence length and the end time of the time series corresponding to the predicted length are within the time range of the test set corresponding to the model training. The model outputs the prediction results of ionospheric parameters or shortwave bands.

[0016] In a second aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the ionospheric parameter and shortwave band prediction method as described above.

[0017] The ionospheric parameter and shortwave frequency band prediction method and electronic device provided by this invention introduces ionospheric factors, which are highly dependent on shortwave communication, into the task of predicting available shortwave frequency bands. Based on historical ionospheric state change information, it provides available frequency bands at different time points in the future, which can further improve the accuracy and stability of frequency selection. By selecting multiple deep learning models for training and prediction, the Transformer-LSTM model is used to perform ionospheric parameter prediction and shortwave frequency band prediction tasks under various conditions such as different environments, times or locations, and different amounts of data accumulation. It can also enrich the research content in the field of data fusion. It is an attempt and exploration in the fields of deep learning and shortwave communication to integrate multiple influencing factors such as ionospheric detection data and environmental monitoring data for parameter prediction and frequency band selection. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram illustrating the principle of the ionospheric parameter prediction and shortwave frequency selection method provided by the present invention; Figure 2 This is a schematic diagram of the ionospheric parameter prediction and shortwave frequency selection process provided by the present invention; Figure 3 This is a framework diagram of the Transformer-LSTM deep learning hybrid model for ionospheric parameter prediction and shortwave frequency selection provided by the present invention. Figure 4 The ionosphere oblique detection ionization map results obtained by detection provided by this invention; Figure 5 This invention provides information on available frequency bands and optimal available frequencies in oblique detection ionization maps obtained after processing using the projection threshold method. Figure 6 This invention provides the result of using a trained model to predict the ionospheric parameter foF2 in the ionospheric parameter prediction task. Figure 7 This invention provides the result of using a trained model to predict the maximum available frequency (MUF) in the task of predicting available frequency bands for shortwave communication. Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] To address the shortcomings of existing technologies, this invention provides a method for ionospheric parameter prediction and shortwave frequency selection. This method utilizes ionospheric oblique sounding, combined with historical ionospheric model data, geomagnetic solar activity data, and ionospheric oblique sounding data. It incorporates ionospheric factors, which are highly dependent on shortwave communication, into frequency selection, providing predicted communication frequency bands for a future period based on historical ionospheric information, historical geomagnetic solar activity information, and historical communication frequency information. This method integrates multi-source data and uses a multi-model fusion deep learning model, thereby further improving the accuracy and stability of frequency band prediction. Furthermore, this method enriches research in data fusion, deep learning, and shortwave communication, representing an attempt and exploration of ionospheric parameter prediction and shortwave communication frequency band prediction using multiple factors such as ionospheric model data, geomagnetic solar activity data, and shortwave communication data. Specifically, it includes: Step 100: Obtain the geographical coordinates of the transmitting and receiving stations, as well as the ionospheric parameters corresponding to the international reference ionospheric model near the relay point, and obtain global geomagnetic activity parameters and solar activity parameters; Step 200: Use the ionospheric detectors at the transmitting and receiving stations to perform oblique frequency sweep detection to obtain the oblique ionosphere ionization map; Step 300: Post-process the ionosphere oblique probe ionization map to obtain oblique probe frequency band data; Step 400: Preprocess global geomagnetic activity parameters, solar activity parameters, and oblique detection frequency band data, and select data with the same time resolution within consecutive time periods; Step 500: Construct a Transformer-LSTM combined model, train the model using the specified preprocessed data, and obtain an ionospheric parameter prediction model or a shortwave band prediction model. Step 600: Input historical data of specified duration and type into the ionospheric parameter prediction model or the shortwave band prediction model, and output the ionospheric parameter prediction results or the shortwave band prediction results.

[0022] The basic principles of the embodiments of the present invention are as follows: Figure 1 As shown, historical ionospheric parameter data from the ionospheric IRI model are obtained, geomagnetic solar activity data from open-source data websites are acquired, oblique probing is performed using an ionospheric detection system to obtain oblique probing ionographs, communication frequency band data is obtained through the projection threshold method, data preprocessing is performed, including data integration, data interpolation, and outlier removal, appropriate parameters are selected to train the deep learning model Transformer-LSTM, and finally, ionospheric parameters or shortwave communication frequency bands are predicted.

[0023] like Figure 2 As shown, the embodiments of the present invention include the following steps: Step S1: Obtain the geographical coordinates of the two locations of the communication link transmitting station and receiving station, as well as the midpoint of the great circle distance between the two locations, which is the relay point. Combine geographical coordinates, time, space and other information to obtain ionospheric parameter data above the relay point. At the same time, obtain parameter data related to global geomagnetic activity and solar activity from open source websites. It should be noted that the locations of the two stations are represented by geographical latitude and longitude coordinates. The midpoint of the great circle between the two points is calculated using geographical coordinates. One station serves as the transmitting station for the detection signal in the subsequent steps, and the other station serves as the receiving station for the detection signal in the subsequent steps, receiving the ionospheric oblique detection signal.

[0024] It should be noted that when obtaining the ionospheric parameters from the International Ionospheric Reference Model (IRI model), the specific ionospheric parameters selected are hmF2, NmE, TEC, M3000F2, and foF2. Here, hmF2 is the distance from the ground to the point of highest electron concentration in the F2 layer of the ionosphere; NmE is the peak electron concentration in the E layer of the ionosphere; TEC is the total electron content of the ionosphere; M3000F2 is the frequency at which the electron density reaches its maximum at an altitude of approximately 3000 km in the F2 layer; and foF2 is the critical frequency of the F2 layer. The selected time parameters specifically cover the period from January 1, 2011 to December 31, 2025, with a time resolution of 3 hours.

[0025] It should be noted that the data on geomagnetic and solar activity parameters are sourced from the Helmholtz Centre for Geosciences in Germany (https: / / kp.gfz.de / en / data#c222) and the Kyoto World Geomagnetic Data Center (https: / / wdc.kugi.kyoto-u.ac.jp / wdc / Sec3.html). By specifying the time range and time resolution longitude, geomagnetic and solar activity data for the corresponding time period can be obtained, including Kp, Ap, SN, F10.7obs, F10.7adj, and Dst. Wherein: Kp is the global geomagnetic activity index, which is the degree of disturbance of the geomagnetic field measured at different locations on the Earth's surface; Ap is the geomagnetic activity index over a 24-hour period, calculated based on the daily average value of the Kp index; SN is the number of sunspots on the solar surface (photosphere); F10.7obs is the observed solar radiation intensity at a wavelength of 10.7 cm; F10.7adj is the corrected F10.7 radiation intensity; Dst represents the axially symmetrical perturbation magnetic field at the Earth's North and South Poles and the polar equator, used to measure the intensity and variation of geomagnetic storms.

[0026] Step S2: Use the ionospheric detectors at the transmitting and receiving stations to perform oblique frequency sweep detection to obtain the oblique ionosphere ionization map; It should be noted that the transmitting station uses an inverted V antenna to transmit m-sequence swept frequency modulated signals. The inverted V antenna is a type of transmitting antenna, shaped like an inverted letter V. The m-sequence is one of the code patterns selected for transmission. The start frequency, end frequency, and step frequency of the sweep are set reasonably according to different ionospheric states. The oblique detection signal is received at a remote location using a two-wire dipole antenna. By processing the received signal, an oblique detection ionization map is plotted.

[0027] Step S3: Post-process the drawn oblique probe ionization map to extract the available frequency band information from the oblique probe ionization map; It should be noted that the main methods for extracting available frequency bands are as follows: First, the projection method is used to focus only on the maximum signal amplitude value corresponding to each frequency point within the frequency sweep range under all different distance thresholds. Then, the threshold method is used to select an appropriate threshold to extract the start and end frequencies corresponding to the signal frequency bands, which are used as historical detection available frequency band data.

[0028] It should be noted that at certain times, due to strong geomagnetic storms or weather conditions, the ionospheric detectors at receiving stations may acquire weak or missing signals, causing the projection threshold method to fail to correctly obtain usable frequency band data. The missing data will be filled in using data interpolation in subsequent steps.

[0029] Step S4, data preprocessing, includes data integration from three different sources, missing data interpolation, outlier data removal, etc., selecting data with the same time resolution within a continuous time period; It should be noted that data integration involves combining data from three different sources: historical ionospheric parameter data from the IRI model, historical geomagnetic solar activity index data from the Helmholtz Centre for Earth Sciences in Germany and the Kyoto World Geomagnetic Data Center, and historical ionospheric detection data obtained from oblique probes of the ionospheric detector. This integration aims to unify the time range and temporal resolution of the data. Outliers were then removed, and missing data was filled in by interpolation from the other data sources to maintain temporal continuity.

[0030] Step S5: Train the deep learning model. You can choose to train using only historical ionospheric parameter data and historical geomagnetic solar activity index data; the trained model will then be used solely for ionospheric parameter prediction. Alternatively, you can choose to train using all three types of historical data; the trained model can then be used for both ionospheric parameter prediction and shortwave communication frequency band prediction. The trained model is a Transformer-LSTM. Select some hyperparameters relevant to the model, and save the optimal model parameters after training. It should be noted that this method can be used for different tasks depending on the type of data used in model training. That is, if only historical data on ionospheric parameters and historical data on geomagnetic solar activity index are used, the trained model can only be used for the ionospheric parameter prediction task; if historical data on ionospheric parameters, historical data on geomagnetic solar activity index, and historical data on ionospheric detection are used, the trained model can be used for two tasks: the ionospheric parameter prediction task and the shortwave frequency selection available frequency band prediction task.

[0031] It should be noted that the deep learning model selected and trained is the Transformer-LSTM model. Several key hyperparameters were selected during training, including the sequence length (seq_len) and prediction length (pred_len), which are related to prediction in subsequent steps. These two lengths are mainly determined based on the size of the dataset. The default sequence length and prediction length are both 96. When the dataset is small (e.g., less than 5000 data points), the sequence length and prediction length can be set to 48, or even 24. The optimal recommendation is to determine these lengths based on the final model training results, using RMSE as a specific metric to measure the model's performance after training. After training the selected model, the optimal model parameters are automatically saved.

[0032] It should be noted that the part of the model that uses Transformer mainly employs an attention mechanism, the principle of which is roughly as follows. Where, Q For query, n × d , n For the number of queries, d (length) K For key (key, m × d , m The number of key-value pairs. d (length) V value m × d , m The number of key-value pairs. d (where the length is used). To prevent values ​​from being too large or too small during calculation, a scaling factor is used. Introduced into the denominator, where It is the dimension of the key.

[0033] (1) It should be noted that the general principle of LSTM in the model is as follows. Among them, h Indicates the number of hidden units. For batch size, If the input dimension is used, then at the time step t Input data It is A 3D matrix, representing t The input at any given moment; and the hidden state. Input gate Output gate Forgotten Gate All for A dimensional matrix representing the current gating value and the corresponding hidden state at the previous time step.

[0034] (2) in, For sigmoid gate functions, Candidate memory units, , These are the current and previous memory units, respectively. and These represent the current and previous hidden states, respectively. , , , and , , , For weight parameters, , , , This is the bias parameter.

[0035] It should be noted that the Transformer-LSTM model mentioned above is a fusion of the Transformer model and the LSTM model. The Transformer part only uses the encoder part of the original Transformer, while the Transformer-LSTM model combines the LSTM layer of the LSTM model with the feedforward neural network, residual connections and layer normalization of the decoder of the original Transformer, while removing the masked multi-head attention layer, thus realizing the establishment of the Transformer-LSTM model.

[0036] Step S6: Input historical data of a specified duration into the model. The type of historical data to be input must be consistent with that used in the training in step S5. The model will then provide the corresponding prediction results for ionospheric parameters or shortwave communication frequency bands.

[0037] It should be noted that the main steps to obtain the prediction results are as follows: First, load the optimal parameters of the trained model. Then, select the task type based on the previous steps: either a single ionospheric parameter prediction task or a combined task of ionospheric parameter prediction and shortwave frequency selection prediction. Given the task type, select the key parameters seq_len and pred_len according to the previous steps. After inputting a known time series data segment of length seq_len into the model, the model can automatically output a predicted time series of length pred_len immediately following the end of seq_len. Simultaneously, it is required that the start time of the time series corresponding to seq_len and the end time of the time series corresponding to pred_len must be within the time range of the test set corresponding to the model training.

[0038] Based on the above embodiments, this embodiment of the invention combines historical data of ionospheric parameters and historical data of geomagnetic solar activity index from the IRI model. After using an ionospheric detection system to perform oblique ionospheric detection, the ionograph data is post-processed to obtain historical ionospheric detection data. A series of processing is performed on the three different types of data, and the processed data is used to train the Transformer-LSTM model. Finally, the trained model is used for prediction.

[0039] In an embodiment of the invention, a practical observation example is to first establish a general Transformer-LSTM deep learning hybrid model framework, such as... Figure 3 As shown, it includes: (1) Input layer: receives raw data input; (2) Data Embedding: Maps discrete inputs to continuous dense vectors; (3) Positional Encoding: Add positional information to each position in the sequence. Since the Transformer itself does not have the ability to sense order, positional encoding injects positional information into the embedding vector through sine / cosine functions or learnable parameters. (4) Each Long Short-Term Memory (LSTM) block is determined based on the model's performance after training. The optimal values ​​are typically between 2 and 4: LSTM captures the temporal dependencies and sequential structure of sequences. It processes sequence data through gating mechanisms (input gate, forget gate, output gate), making it suitable for modeling long-term dependencies while also being order-sensitive. Each Long Short-Term Memory (LSTM) network block includes multiple LSTM cells and a residual connection + layer normalization module (Add + LayerNorm). The optimal number of LSTM cells here is around 1 to 3. (5) One Transformer encoder (determined based on the performance results of the model after training) The optimal values ​​are selected based on the parameters that best enable the model's performance (typically, the optimal range is around 1 to 3). Each Transformer encoder includes a multi-head attention mechanism + residual connection + layer normalization (Add + LayerNorm) and a feedforward neural network + residual connection + layer normalization (Add + LayerNorm). (6) Convolution and Avg Pooling Block: Further extract local features and reduce sequence length, including two one-dimensional convolution Conv + GELU activation function + Avg average pooling, and one dropout unit + Avg average pooling. (7) Output layer: Connect fully connected layers or other output layers according to the task (classification, regression, generation, etc.) to output the final processing result.

[0040] Understandably, this invention proposes a solution that combines the advantages of both LSTM and Transformer models. First, an LSTM layer is used to extract the short-term temporal dependencies of the sequence. Then, the output of the LSTM is fed back into the multi-head attention mechanism of the Transformer model to extract the long-term and global temporal dependencies of the sequence. Next, the output of the Transformer model is passed through one-dimensional convolutional and pooling layers and then merged with the output of the LSTM layer. Finally, a fully connected layer is used to output the prediction result. Figure 3 As shown, this paper proposes a framework for predicting ionospheric parameters and maximum usable frequency (MUF) based on a hybrid LSTM and Transformer model.

[0041] Compared to the original Transformer model, the LSTM-Transformer hybrid model makes the following improvements, which aim to improve the model's prediction accuracy and adaptability to dynamic time-varying ionospheric environments: Adjustment 1: The input part is coupled to the LSTM layer, then connected to the input part of the Transformer encoder, and finally the outputs of the LSTM and Transformer are merged in a residual connection manner; Adjustment 2: The Transformer decoder part is removed, and the model finally generates the output through a fully connected layer; Adjustment 3: After the output of the Transformer encoder, a convolutional layer and a global average pooling layer are added to simplify the information output of the Transformer and match the dimension of the LSTM output.

[0042] The LSTM-Transformer ionospheric parameter and MUF prediction framework mainly includes the following processes: Step 1: In this step, we first collect all the necessary ionospheric model parameter data, geomagnetic solar activity index data, and HF communication data.

[0043] Step 2: Perform data filtering and preprocessing to obtain the raw time series data. Then, perform location encoding and data encoding on this data.

[0044] Step 3: The processed time series is processed through an LSTM layer. The LSTM layer is responsible for processing the sequence data, extracting time features, and effectively capturing short-term dependencies, while preserving the sequence order of the data.

[0045] Step 4: The output sequence of the LSTM passes through a series of Transformer encoder layers. These encoder layers are mainly responsible for further feature extraction and capturing long-range and global dependencies within the sequence. The multi-head self-attention mechanism of these layers allows for simultaneous attention to information from all locations in the sequence.

[0046] Step 5: The output of the Transformer encoder layer, after passing through convolutional and global pooling layers, is merged with the output of the LSTM layer, and finally mapped to the predicted value through a linear layer. The linear layer acts as a bridge connecting the predicted output and the model feature output, transforming the complex features processed by the LSTM layer and the Transformer encoder layer into the final output.

[0047] Ionospheric IRI model data and geomagnetic solar activity index data were acquired at 3-hour intervals from January 1, 2011 to December 31, 2025. Using a conventional ionospheric sounding system, an m-sequence swept-frequency modulated signal was transmitted via an inverted-V antenna to perform oblique ionospheric sounding, obtaining oblique ionographs. A particular oblique ionospheric sounding operation is shown below. Figure 4 As shown, the projection thresholding method is used to obtain the start and end frequencies corresponding to the signal frequency band, such as... Figure 5 The red triangles and red pentagrams in the image represent the available frequency band data. Data preprocessing then involves integrating data from three different sources, interpolating missing data, and removing outliers, selecting data with the same temporal resolution within consecutive time periods. The processed data includes datasets of three types: ionospheric IRI model data, geomagnetic solar activity index data, and available frequency band data for ionospheric detection; or datasets containing only two types of data: ionospheric IRI model data and geomagnetic solar activity index data. Finally, the dataset for the ionospheric parameter prediction task contains 43,832 data entries, and the dataset for the mixed task of ionospheric parameter prediction and shortwave communication available frequency band prediction contains 3,882 data entries.

[0048] In this embodiment of the invention, under the task of predicting ionospheric parameters, after selecting appropriate hyperparameters, only two types of data—ionospheric IRI model data and geomagnetic solar activity index data—were used to output partial results of training the Transformer-LSTM model. The selected model hyperparameters are: sequence length seq_len = 96, label length label_len = 48, prediction length pred_len = 336, the optimizer used is Adam, and the loss function used in training is... The final training results are detailed for each metric. , The corresponding calculation method is as follows: (3) in, This represents the difference between the actual value and the predicted value. This represents the difference between the true value and its average. This represents the difference between the predicted value and its mean, where n represents the number of samples. MSE and RMSE range from 0 (perfect fit) to +∞ (no fit); MAE is the average absolute error, describing the difference between the true and predicted values; MAPE is the average absolute percentage error, ranging from 0 (perfect fit) to 1 (no fit). The value range is from -1 to 1, where -1 to 0 indicates that the model is not fitting, and the closer it is to 1, the closer the predicted trend is to the actual trend. In other words, the closer the model is to 1, the better it fits.

[0049] In embodiments of the present invention, such as Figure 6As shown, in the ionospheric parameter prediction task, using a sequence length of seq_len of 96 to predict the ionospheric parameter foF2 corresponding to 336 future time points with a pred_len length, the vertical axis represents the value of foF2, with the corresponding unit being MHz. The blue line (dark) represents Ground Truth, indicating the true value, and the orange line (light) represents the predicted value. The final prediction result is composed of 96 original data inputs and 336 predicted outputs.

[0050] In embodiments of the present invention, such as Figure 7 As shown, this is the result of predicting the available shortwave communication frequency band (MUF) parameters (i.e., the maximum available frequency) for 24 future times using a sequence length of seq_len of 24, under a hybrid task of predicting out-of-situ parameters and shortwave communication available frequency bands. The vertical axis represents the MUF value, with the corresponding unit being MHz. The blue line (dark) represents the ground truth, and the orange line (light) represents the predicted value. The final prediction result is composed of 24 original data inputs and 24 predicted outputs.

[0051] The model was trained and predicted using only data from the IRI model and the global geomagnetic solar activity index. The results are shown in Table 1. In the test set, the LSTM-Transformer model showed better performance, with an RMSE of 0.2875MHz, which is better than the 0.5660MHz of the traditional LSTM model and the 0.5567MHz of the MLP (Multilayer Perceptron), and also better than the 0.3782MHz of the Transformer model. The same trend was observed in MAE and MAPE metrics. Therefore, the model performance ranking from best to worst is: LSTM-Transformer > Transformer > MLP > LSTM.

[0052] Table 1. Comparison results of multiple models

[0053] To further verify the effectiveness of ionospheric parameters and the MUF prediction framework, this invention also constructed multiple high-frequency (HF) communication test lines for experiments: the first line runs from City A to City B (roughly northward), the second line runs from City A to City C (roughly eastward), and the third line runs from City D to City E (roughly northwestward).

[0054] In evaluating model performance, similar to when using only the IRI dataset, higher values ​​for RMSE, MAE, and MAPE indicate worse model performance. Therefore, the performance ranking from best to worst is mostly LSTM-Transformer > Transformer > LSTM > MLP, with the DE link showing LSTM-Transformer > Transformer > MLP > LSTM. Overall, LSTM-Transformer remains superior to the other models. Furthermore, as shown in Table 2, all models achieved good prediction results in the DE shortwave communication link, followed by the AB link, while the AC link showed slightly worse prediction results.

[0055] Table 2 Comparison Results of HF Link Experiments

[0056] Finally, the present invention also completed ablation experiments for each part of the model to verify the effectiveness of each component of the hybrid model and its contribution to model performance. Model 1 corresponds to all parts related to LSTM, that is, Model 1 removes all LSTM-related parts from the original LSTM-Transformer model; Model 2 corresponds to the residual connection part between the LSTM and Transformer encoder modules, that is, Model 2 removes the residual connection at the end of the LSTM-Transformer model, and only inputs the data into the LSTM before directly outputting it through the Transformer encoder module; Model 3 corresponds to the Transformer decoder part, that is, adds a Transformer decoder part to the Transformer encoder module in the LSTM-Transformer model; Model 4 corresponds to the convolutional layer and global average pooling layer part after the output of the Transformer encoder, that is, removes the convolutional layer and global average pooling layer after the output of the Transformer encoder from the LSTM-Transformer model. Among them, Model 1 and Model 2 correspond to Adjustment 1 mentioned above, Model 3 corresponds to Adjustment 2, and Model 4 corresponds to Adjustment 3. The specific results are shown in Table 3.

[0057] Table 3 Comparison results of ablation experiments

[0058] As shown in Table 3, compared to the original model, removing any component degrades model performance, manifested as an increase in metrics such as RMSE, MAE, and MAPE. Furthermore, Model 3 experiences the largest performance drop, indicating that removing the Transformer decoder component has the greatest impact on improving model performance. Similarly, the model performance ranking from best to worst is: Model 2... Model 4 > Model 1 > Model 3, respectively. The removal of the Transformer decoder has the greatest impact on model performance, followed by the addition of LSTM layers, then the addition of convolutional layers and global average pooling layers, and finally the residual connection between LSTM and Transformer.

[0059] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logic instructions in the memory 830 to execute ionospheric parameter and shortwave band prediction methods. These methods include: acquiring the geographic coordinates of the transmitting and receiving stations, as well as the ionospheric parameters corresponding to the international reference ionospheric model near the relay point, and acquiring global geomagnetic activity parameters and solar activity parameters; using the ionospheric detectors at the transmitting and receiving stations to perform oblique frequency sweep detection to obtain oblique ionosphere maps; post-processing the oblique ionosphere maps to obtain oblique ionosphere frequency band data; preprocessing the global geomagnetic activity parameters, solar activity parameters, and oblique ionosphere frequency band data, selecting data with the same temporal resolution within a continuous time period; constructing a Transformer-LSTM combined model, training the model using specified pre-processed data, and obtaining an ionospheric parameter prediction model or a shortwave band prediction model; inputting historical data of specified duration and type into the ionospheric parameter prediction model or the shortwave band prediction model, and outputting the ionospheric parameter prediction results or the shortwave band prediction results.

[0060] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0061] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0062] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting ionospheric parameters and shortwave frequency bands, characterized in that, include: Obtain the geographical coordinates of the transmitting and receiving stations, as well as the ionospheric parameters corresponding to the international reference ionospheric model near the relay point, and obtain global geomagnetic activity parameters and solar activity parameters; Using the ionospheric detectors at the transmitting and receiving stations, oblique frequency sweep detection is performed to obtain oblique ionosphere ionization maps; Post-processing of the ionosphere oblique probe ionization map yields oblique probe frequency band data; Preprocessing of global geomagnetic activity parameters, solar activity parameters, and oblique detection frequency band data is performed to select data with the same time resolution within consecutive time periods; Construct a Transformer-LSTM combined model, train the model using specified preprocessed data, and obtain an ionospheric parameter prediction model or a shortwave band prediction model. Input historical data of a specified duration and type into the ionospheric parameter prediction model or the shortwave band prediction model, and output the ionospheric parameter prediction results or the shortwave band prediction results.

2. The method for predicting ionospheric parameters and shortwave band frequencies according to claim 1, characterized in that, Obtain the geographical coordinates of the transmitting and receiving stations, including: The transmitting station and the receiving station are represented by geographic coordinates, and the latitude and longitude coordinates of the midpoint of the great circle between the two stations are calculated using geographic coordinates. Input the time range, resolution, latitude and longitude coordinates of the midpoint of the great circle, altitude range above the ground, and resolution into the International Reference Ionospheric Model.

3. The method for predicting ionospheric parameters and shortwave band frequencies according to claim 2, characterized in that, Obtain the ionospheric parameters corresponding to the international reference ionospheric model near the relay point between the transmitting and receiving stations, including: Historical data of ionospheric parameters were obtained using the international reference ionospheric model. By calling the international reference ionospheric model and inputting the time range, resolution, latitude and longitude coordinates of the midpoint of the great circle, and the range and resolution of the altitude above the ground, the critical frequency, peak height, half thickness of the ionosphere at the midpoint of the great circle between the two stations, as well as the distribution of electron concentration in space, can be obtained.

4. The method for predicting ionospheric parameters and shortwave bands according to claim 1, characterized in that, Using ionospheric detectors at both the transmitting and receiving stations, oblique frequency sweep detection is performed to obtain oblique ionotropic maps of the ionosphere, including: The transmitting station uses an inverted V antenna to transmit an m-sequence swept frequency modulation signal. The start frequency, end frequency, and step frequency of the sweep are set according to different ionospheric states. The oblique sounding signal is received at a remote location using a two-wire dipole antenna. By processing the received oblique sounding signal, an oblique sounding ionosphere map is drawn.

5. The method for predicting ionospheric parameters and shortwave band frequencies according to claim 1, characterized in that, Post-processing of the ionosphere oblique probe ionogram yields oblique probe frequency band data, including: The projection method is used to focus on the maximum signal amplitude value corresponding to each frequency point within the swept frequency band. Using the threshold method, the start and end frequencies corresponding to the signal frequency bands are extracted using a preset threshold and used as oblique detection frequency band data.

6. The method for predicting ionospheric parameters and shortwave band frequencies according to claim 1, characterized in that, Preprocessing was performed on global geomagnetic activity parameters, solar activity parameters, and oblique detection frequency band data. Data with the same time resolution within consecutive time periods were selected, including: By using a unified time range and time resolution, missing data interpolation and outlier data removal are performed to obtain preprocessed data.

7. The method for predicting ionospheric parameters and shortwave bands according to claim 1, characterized in that, Constructing a Transformer-LSTM combined model includes: The Transformer model is determined to include only the encoder part of the original Transformer. The LSTM layer in the LSTM model is combined with the feedforward neural network, residual connection and layer normalization of the decoder of the original Transformer to form the Transformer-LSTM combined model. The key model parameters, including sequence length and prediction length, are determined for model training. After model training is completed, the optimal model parameters are saved.

8. The method for predicting ionospheric parameters and shortwave bands according to claim 7, characterized in that, Model training is performed using specified preprocessed data to obtain ionospheric parameter prediction models or shortwave band prediction models, including: A model for predicting ionospheric parameters was obtained by training the model using global geomagnetic activity parameters and solar activity parameters. A shortwave band prediction model was obtained by training the model using global geomagnetic activity parameters, solar activity parameters, and ionosphere oblique probe ionization maps.

9. The method for predicting ionospheric parameters and shortwave bands according to claim 1, characterized in that, Input historical data of specified duration and type into the ionospheric parameter prediction model or shortwave band prediction model, and output the ionospheric parameter prediction results or shortwave band prediction results, including: Load the optimal parameters of the trained model and determine the historical data corresponding to the ionospheric parameter prediction and / or shortwave band prediction; After inputting historical data of sequence length, the model outputs predicted data of the predicted length following the end time of the sequence length. The start time of the time series corresponding to the sequence length and the end time of the time series corresponding to the predicted length are within the time range of the test set corresponding to the model training. The model outputs the prediction results of ionospheric parameters or shortwave bands.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the ionospheric parameter and shortwave band prediction method as described in any one of claims 1 to 9.

Images