Ionospheric scintillation online prediction method and system based on width learning

Abstract

The application discloses an ionospheric scintillation online prediction method and system based on width learning, and belongs to the technical field of navigation. The method first performs time alignment and fusion on multi-source heterogeneous data from a super double aurora radar, an altimeter, a geomagnetic activity index and ionospheric phase scintillation, and fills in missing data through a self-attention neural network. Then, a width learning system prediction model is constructed, the model is trained using historical time series data to predict future phase scintillation indexes. Finally, through the designed incremental learning mechanism, the model can quickly update the weight with the arrival of new data, realizing minute-level prediction and millisecond-level response. The application effectively solves the problems of poor real-time performance, weak adaptability and high calculation cost of the existing prediction method, significantly improves the stability and reliability of the navigation system in a complex ionospheric environment, and the model is light and efficient, suitable for deployment and application in resource-limited terminal devices.

Description

Technical Field

[0001] This invention belongs to the field of navigation technology, specifically relating to an online prediction method and system for ionospheric scintillation based on width learning. Background Technology

[0002] Ionospheric scintillation is a space weather phenomenon caused by irregularities in electron density within the ionosphere. It leads to amplitude fading and rapid phase fluctuations in radio signals (such as GNSS signals) traversing the ionosphere. This phenomenon is particularly pronounced in high-latitude regions and the magnetic equator, and intensifies during years of high solar activity. Ionospheric scintillation severely degrades the signal tracking performance of GNSS receivers, causing carrier phase lock-up and increased ranging errors, thus directly impacting the reliability and security of high-precision navigation and positioning applications in aviation, maritime, and surveying. Therefore, accurate and timely prediction of ionospheric scintillation is of significant practical importance for ensuring the continuity and accuracy of GNSS services.

[0003] Currently, methods for predicting ionospheric scintillation can be mainly divided into three categories: empirical models, physical models, and data-driven machine learning models. Empirical models, such as the International Reference Ionospheric (IRI) model, rely on long-term historical observational statistics and can provide relatively stable background predictions when the ionosphere is calm or normal. However, when sudden space weather events occur (such as solar storms and geomagnetic storms), the ionospheric state deviates sharply from historical statistical patterns, leading to a significant decrease in the prediction accuracy of these models. Furthermore, empirical models typically cannot achieve short-term, detailed predictions on timescales of minutes or even shorter, making them unsuitable for real-time navigation applications. Physical models, such as WBMOD and GISM, attempt to simulate the generation and evolution of scintillation by numerically solving the ionospheric plasma physics equations. These models have well-defined physical mechanisms, but their computational processes are extremely complex, requiring enormous computational resources. A single simulation typically takes several hours or even longer, making real-time or near-real-time online prediction impossible. Moreover, physical models are extremely sensitive to initial and boundary conditions, making it difficult to accurately capture sudden, small-scale scintillation events.

[0004] In recent years, with the development of artificial intelligence technology, machine learning models, such as Long Short-Term Memory (LSTM) networks and Random Forest (RF), have been introduced into ionospheric scintillation prediction. These methods can learn the nonlinear relationship between scintillation and various driving factors (such as geomagnetic indices, solar activity indices, and radar observation data) from historical data, improving prediction accuracy compared to traditional models and achieving hourly-level predictions. However, existing machine learning methods still have significant limitations. First, they typically rely on large amounts of high-quality, well-labeled training data, while ionospheric scintillation observation data itself suffers from missing data, noise, and significant differences in scintillation characteristics across different regions and times, limiting the model's generalization ability. Second, the parameters of traditional machine learning and deep learning methods are usually fixed after model training. To incorporate new observation data to adapt to the dynamic changes in ionospheric states, the entire model often needs to be retrained, a computationally expensive process that cannot meet the needs of online, real-time updates. Finally, deploying complex deep learning models on embedded devices such as GNSS receivers with limited computing resources faces significant challenges. Summary of the Invention

[0005] To address the shortcomings of existing technologies, such as poor real-time prediction performance, insufficient model robustness, and high computational resource requirements, this invention aims to provide an online ionospheric scintillation prediction method based on width learning. This method integrates multi-source heterogeneous observation data and innovatively introduces a width learning model with sparse regularization and incremental learning capabilities. It aims to achieve high-precision prediction within minutes and rapid model updates within milliseconds, thereby effectively improving the stability and reliability of navigation systems in complex and variable ionospheric environments.

[0006] To solve the above-mentioned technical problems, the technical solution proposed in this application is as follows:

[0007] This invention provides an online prediction method for ionospheric scintillation based on width learning, comprising the following steps: S1. Time alignment is performed on data from four types of observation sources: super dual auroral radar, altimeter, geomagnetic solar activity index, and ionospheric phase scintillation, to form a fusion dataset with unified time resolution; S2. Calculate the dynamic window midpoint for the vector field data of the super dual-polarization radar to suppress noise; S3. A SAITS neural network based on a self-attention mechanism is used to complete and optimize the quality of missing data in the fused dataset; S4. Divide the preprocessed data into training set, validation set, and test set; S5. Establish a width learning system model for online prediction of ionospheric scintillation, input historical time series data, and output the phase scintillation index at future time. S6. Evaluate the root mean square error (RMSE) of the width learning system model using the test set data; S7. Using the newly acquired observation data, the connection weights of the width learning system model are quickly updated using an incremental learning mechanism, and the corresponding predicted phase flicker index is output.

[0008] Further, in step S1, the unified time resolution is a 1-minute resolution; wherein, The super dual-polarization radar extracts the first valid record every minute; The data from the altimeter is expanded to a 1-minute resolution using a copying method; The geomagnetic solar activity index data and the ionospheric phase scintillation data retain their original time series.

[0009] Furthermore, the vector field data of the super dual aurora radar in step S2 includes power, coordination speed, fitted spectral width, and elevation angle.

[0010] Furthermore, in step S3, the SAITS neural network learns the temporal distribution characteristics of the altimeter data through a self-attention mechanism and interpolates the missing positions of the altimeter data.

[0011] Furthermore, step S5 specifically includes the following sub-steps: S51. The vector field data, altimeter data, and geomagnetic solar activity index data from the super dual auroral radar are used as input features, and the ionospheric phase scintillation data is used as the output target. S52. Based on the satellite observation station data, divide all input data into several batches, denoted as... , where a is the batch number; S53. For the first batch of input data Calculate the first batch of feature nodes and the first batch of enhanced feature nodes :

[0012]

[0013] in, and These are the weights for the first batch of feature nodes and the first batch of enhanced feature nodes, respectively. and These are the weight biases for the first batch of feature nodes and the first batch of enhanced feature nodes, respectively. and These are the activation functions for the feature nodes and the activation functions for the augmentation nodes, respectively.

[0014] S54. Merge the feature nodes and enhanced feature nodes This forms the first batch of intermediate feature layers. and connect to the output layer ; S55. Calculate the first batch of intermediate feature layers. to output layer Connection weights :

[0015] in, and It is the norm regularization coefficient. It is the sparsity coefficient; S56. Based on the obtained network parameters and connection layer weights, and according to the data predicted... The model predicts the phase scintillation index. The value is:

[0016] Furthermore, in step S7, the incremental learning mechanism specifically includes: S71. Define the memory module for the first batch of input data as follows:

[0017]

[0018] in This is the feature memory matrix for the first batch. This is the knowledge memory matrix for the first batch; S72. For the data input in batch a... Where a≥2, calculate the a-th batch of feature nodes. Batch a of enhanced feature nodes and the intermediate feature layer of the a-th batch formed ; S73. Generate the memory matrix for the incremental portion:

[0019]

[0020] in This is the incremental feature memory matrix for the a-th batch. This is the incremental knowledge memory matrix for the a-th batch; S74. Calculate the memory fast update module under the new data stream:

[0021]

[0022] in Let a be the feature memory matrix of the a-th batch. This is the knowledge memory matrix for the a-th batch; S75. Calculate the updated connection weights:

[0023] in, λ is the connection weight of the a-th batch, I is the identity matrix, and λ is the regularization coefficient.

[0024] Furthermore, the root mean square error (RMSE) is used to evaluate the model's predictive performance.

[0025] On the other hand, this application also claims protection for an online ionospheric scintillation prediction system based on width learning, comprising: The data fusion and preprocessing module is used to perform time alignment on data from four types of observation sources, namely super dual auroral radar, altimeter, geomagnetic solar activity index and ionospheric phase scintillation, to form a fused dataset. It calculates the dynamic window midpoint for the vector field data of the super dual auroral radar to suppress noise, and uses a SAITS neural network based on self-attention mechanism to complete and optimize the quality of missing data in the fused dataset. The model building and training module is used to divide the preprocessed data into training set, validation set and test set, establish a width learning system model for online prediction of ionospheric scintillation, and evaluate the root mean square error (RMSE) of the width learning system model using the test set data. The online prediction and update module is used to quickly update the connection weights of the width learning system model using newly acquired observation data and an incremental learning mechanism, and output the corresponding predicted phase flicker index.

[0026] On the other hand, this application also claims protection for an electronic device, comprising: Memory, used to store computer programs; A processor is configured to execute a computer program stored in the memory to implement the online prediction method for ionospheric scintillation based on width learning as described in any of the preceding descriptions.

[0027] On the other hand, this application also claims protection for a computer-readable storage medium storing a computer program that, when executed by a processor, implements the ionospheric scintillation online prediction method based on width learning as described in any of the preceding claims.

[0028] Compared with the prior art, the present invention achieves the following beneficial technical effects: This application presents an online ionospheric scintillation prediction method based on width learning. By fusing multi-source heterogeneous data and introducing a width learning model with sparse regularization and incremental learning capabilities, it effectively overcomes the bottlenecks of poor real-time prediction, weak model adaptability, and high computational cost in existing technologies. This method can achieve high-precision short-term predictions at the minute level, and with its efficient online update mechanism, it can complete model self-optimization with extremely low latency after acquiring new data, continuously tracking dynamic changes in the ionosphere. Ultimately, it significantly improves the stability and reliability of navigation systems in complex space weather environments, and the lightweight model design facilitates deployment and application in resource-constrained terminal devices. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart of the online prediction method for ionospheric scintillation based on width learning proposed in this application.

[0031] Figure 2 This is an analytical process diagram of the online prediction method for ionospheric scintillation based on width learning provided in the embodiments of this application.

[0032] Figure 3 This is a schematic diagram of the width learning system model structure for ionospheric scintillation prediction in an embodiment of this application.

[0033] Figure 4 This is a schematic diagram illustrating the principle of the width learning system model in this application, which uses an incremental learning mechanism for rapid updates.

[0034] Figure 5 This is a block diagram of the ionospheric scintillation online prediction system based on width learning proposed in this application. Detailed Implementation To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] like Figure 1As shown in the figure, this invention proposes an online prediction method for ionospheric scintillation based on width learning. This method aims to solve the problems of insufficient real-time prediction, poor adaptability to sudden disturbance events, and high computational complexity that makes it difficult to deploy on terminals in existing technologies. The following is a combination of... Figures 1 to 5 The technical solution of the present invention will be described in detail and coherently.

[0036] S1. Time alignment is performed on data from four types of observation sources: super dual auroral radar, altimeter, geomagnetic solar activity index, and ionospheric phase scintillation, to form a fusion dataset with a unified time resolution.

[0037] Ionospheric scintillation is influenced by multiple factors, and a single data source cannot fully characterize its complex dynamics. Therefore, this invention integrates four types of complementary observational data: super dual auroral radar data, ionospheric altimeter data, geomagnetic and solar activity indices (such as the Kp index and F10.7 index), and the ionospheric phase scintillation index (usually obtained by GNSS receiver monitoring), which directly reflects scintillation intensity. These data differ in spatiotemporal resolution, physical meaning, and missing modes, representing typical multi-source heterogeneous data.

[0038] First, a high-quality fused dataset is constructed. Due to the inconsistent time resolutions of the various data sources, direct fusion would lead to timeline misalignment. Therefore, strict time alignment is required to unify all data onto a common time reference. This application selects 1 minute as the unified time resolution, a compromise that balances data detail richness and computational feasibility. The specific alignment strategy is as follows: For super dual auroral radar data, whose time resolution is typically higher than 1 minute, the first valid record within each minute is extracted as the representative value for that minute. For altimeter data with lower time resolution (e.g., once every 15 minutes), a data duplication method is used, copying and expanding each 15-minute data point to fill the corresponding 15 1-minute time slots, thereby increasing its resolution to 1 minute. For geomagnetic solar activity index and ionospheric phase scintillation data, which are already 1-minute resolution, their original time series are directly retained without special processing. Through these operations, all four types of data are aligned to the same series of 1-minute timestamps in the time dimension, forming a preliminary fused time series dataset.

[0039] S2. Calculate the dynamic window midpoint for the vector field data of the super dual-polarization radar to suppress noise; The vector field data (including echo power, line-of-sight velocity, spectral width, and elevation angle) provided by the super dual-polarization radar are important indicators reflecting the dynamic state of the ionosphere. However, observations inevitably contain noise and outliers. To improve the robustness of the input features, this application applies dynamic window median filtering to each vector field feature (power, velocity, spectral width, and elevation angle) of the radar. Specifically, for a radar feature value at the current time t, all valid observations within a time window centered at t (e.g., a 5-minute window before and after t) are selected, the median of these values ​​is calculated, and this median is used as the final value of the feature at time t. Median filtering effectively resists impulse noise and outlier interference, resulting in a smoother and more reliable feature sequence.

[0040] S3. A SAITS neural network based on a self-attention mechanism is used to complete and optimize the quality of missing data in the fused dataset; In real-world observations, especially with altimeter data, a significant amount of data is missing due to equipment malfunctions, weather interference, and other factors. Directly training a model using data containing numerous missing values ​​leads to a severe performance degradation. To address this, this application introduces an advanced data completion neural network based on a self-attention mechanism—SAITS. The core of the SAITS neural network utilizes the self-attention mechanism in the Transformer encoder structure to learn the inherent distribution and dependencies of complete time-series data. We input the spatiotemporally aligned altimeter data sequence containing missing values ​​into the SAITS network. The network identifies the missing locations in the data through a masking mechanism and uses a self-attention layer to aggregate known information (including past and future information) throughout the sequence to infer and generate the most probable values ​​for the missing locations. After processing by SAITS, we obtain a complete, high-quality altimeter data sequence. This step is crucial, as it provides clean and continuous input for subsequent model training.

[0041] S4. Divide the preprocessed data into training set, validation set, and test set; After data preprocessing and completion, the resulting complete fused dataset is divided into three non-overlapping parts in chronological order: a training set, a validation set, and a test set. Typically, this division can be done in a certain ratio (e.g., 7:2:1). The training set is used to build the initial width-learning model and learn its parameters; the validation set is used to adjust some hyperparameters of the model, such as the number of feature nodes and augmentation nodes in subsequent width-learning models, and the magnitude of the regularization coefficient; the test set is used to objectively evaluate the model's generalization prediction performance, simulating the model's performance on unseen data.

[0042] S5. Establish a width learning system model for online prediction of ionospheric scintillation, input historical time series data, and output the phase scintillation index at future time. This step is the core of this application. The width learning system is an efficient flat network structure. It maps the input data to a feature space spanned by a large number of randomly generated feature nodes and augmentation nodes, and then directly solves the output weights through linear regression, thereby avoiding the cumbersome backpropagation iterative training process of deep learning models.

[0043] The specific implementation process is as follows: S51. The vector field data, altimeter data, and geomagnetic solar activity index data from the super dual auroral radar are used as input features, and the ionospheric phase scintillation data is used as the output target; The model's input features include the super dual auroral radar vector field data (power, velocity, spectral width, elevation angle) processed in the aforementioned steps, complete altimeter data, and geomagnetic solar activity index data. The model's output target is the ionospheric phase scintillation index.

[0044] S52. Based on the satellite observation station data, divide all input data into several batches, denoted as... , where a is the batch number; Due to the potentially large volume of data, and for the convenience of subsequent incremental updates, we divide all input data into several consecutive data batches based on their chronological order or satellite observation station grouping. We denote the data in batch a as... .

[0045] S53. For the first batch of input data Calculate feature nodes and enhanced feature nodes : like Figure 3 As shown, for the first batch of input data We perform the following calculations to construct the initial model:

[0046]

[0047] in, and These are the weights of the feature mapping node and the augmented feature node, respectively. and These are the weight biases for the feature mapping nodes and the augmented feature nodes, respectively; n is the number of feature nodes, and m is the number of augmented feature nodes.

[0048] S54. Merge the feature nodes and enhanced feature nodes Forming an intermediate feature layer and connect to the output layer ; S55. Calculate the intermediate feature layer to output layer Connection weights :

[0049] Among them, and These are norm regularization coefficients, used in width learning. ,Right now This represents the prediction error of the model under the L2 norm. It is a sparsity coefficient used to control the degree of sparsity of connection weights, filtering out redundant information in multi-source heterogeneous data while retaining useful information.

[0050] S56. Based on the obtained network parameters and connection layer weights, and according to the data predicted... The model predicts the phase flicker index value as follows:

[0051] S6. Evaluate the root mean square error (RMSE) of the width learning system model using the test set data; After model training, performance is evaluated using a reserved test set. The input features of the test set are fed into the trained wide-range learning model to obtain the predicted phase flicker index sequence. This predicted sequence is compared with the actual phase flicker index sequence in the test set, and the root mean square error (RMSE) is calculated as the primary evaluation metric. RMSE comprehensively reflects the magnitude of the prediction error; the smaller the value, the higher the model's prediction accuracy. This step objectively quantifies the predictive ability of this method on unknown data.

[0052] S7. Using the newly acquired observation data, the connection weights of the width learning system model are quickly updated using an incremental learning mechanism, and the corresponding predicted phase flicker index is output.

[0053] This is a key step in achieving the "online prediction" and "strong adaptability" of this invention, and its principle is as follows: Figure 4As shown, the ionospheric state is constantly changing, and the predictive performance of models with fixed parameters decays over time. Traditional methods require periodically retraining the model using all historical data, which is costly. This invention utilizes the characteristics of wide-range learning models to design an efficient incremental learning algorithm, enabling the model to quickly update itself with the arrival of new data. After each update, the model represents knowledge of all observational data up to the current time, allowing it to better adapt to the current ionospheric state. When a real-time prediction request is received, the phase scintillation index can be directly predicted using the updated network weights, thus obtaining a prediction result based on the latest knowledge.

[0054] like Figure 2 As shown, the following is a further description with reference to the embodiments: 1. Construct a fused dataset For the super dual-auroral radar data, the first valid record per minute is extracted to match a 1-minute baseline; for altimeter data with a 15-minute resolution, a replication method is used to extend it to a 1-minute resolution; geomagnetic-solar activity index and ionospheric phase scintillation data (both with a 1-minute resolution) are directly retained in their original time series. For the vector field characteristic data of the super dual-auroral radar (including power...) Coordination speed , Fitted spectral width With elevation angle The current time-instance data is calculated using a dynamic window midpoint to improve feature robustness and suppress outlier interference. Through this method, all data are unified to a 1-minute time base while preserving key physical features, providing high-quality input for subsequent modeling.

[0055] 2. Use the SAITS neural network to complete the missing data of the electronic altimeter. A self-attention mechanism-based SAITS neural network is used to impute missing values ​​in the raw data collected by an electronic altimeter. The SAITS neural network learns the temporal distribution characteristics of the altimeter data through self-attention, generating imputed values ​​consistent with the actual data distribution while preserving the original distribution characteristics. Specifically, the network captures the temporal dependencies of the data through a multi-layer Transformer encoder, combines this with a masking mechanism to learn the missing locations, and finally outputs a complete altimeter dataset.

[0056] 3. Divide the data into training, validation, and test sets. The preprocessed dataset is divided into training, validation, and test sets. The training set is used for learning model parameters, the validation set is used for tuning hyperparameters (such as learning rate and network depth), and the test set is used to evaluate the final performance of the model and ensure its generalization ability.

[0057] 4. Establish a width-learning system model for online prediction of ionospheric scintillation, inputting historical data and outputting the future scintillation phase. A prediction model is built based on width learning for each batch of data. Learn in sequence: For the first batch of input data Calculate feature nodes and enhanced feature nodes ;

[0058]

[0059] Merging feature nodes and enhanced feature nodes , forming a feature layer and connect to the output layer ; Calculate feature layer and output layer Connection weights :

[0060] in and It is the norm regularization coefficient. It is used in width learning. ,Right now This represents the prediction error of the model under the L2 norm. It is a sparsity coefficient used to control the degree of sparsity of connection weights, filtering out redundant information in multi-source heterogeneous data while retaining useful information; Based on the obtained network parameters and connection layer weights, and according to the data to be predicted... The model predicts the phase scintillation index value as follows:

[0061] Incremental learning and updating. Width-based learning networks can be updated quickly based on a memory matrix, defining the first batch of input data. memory module

[0062]

[0063] in This is the feature memory matrix for the first batch. This is the knowledge memory matrix for the first batch; For the data input in batch a (a>=2) Calculate feature nodes and enhanced feature nodes ; Forming new feature layers Generate the memory matrix for the incremental part.

[0064]

[0065] in This is the incremental feature memory matrix for the a-th batch. The incremental knowledge memory matrix for batch a Computational memory fast update module

[0066]

[0067] in Let a be the feature memory matrix of the a-th batch. The knowledge memory matrix for batch a Then it can be from

[0068] Calculate the updated connection weights.

[0069] 5. Evaluate the RMSE of the model The root mean square error (RMSE) is used to evaluate the model's predictive performance. 6. Update model connection weights For new data, process the data directly using the same procedure; use the same calculation procedure as batch a (a≥2) in step 4 to perform incremental learning and updating.

[0070] In summary, this invention constructs a complete, efficient, and adaptive online ionospheric scintillation prediction system through multi-source data fusion, SAITS data completion, sparse regularized width learning modeling, and efficient incremental learning updates. This method not only significantly improves prediction accuracy but, more importantly, achieves real-time prediction and continuous model evolution capabilities. Furthermore, it is resource-friendly and has promising practical application prospects, particularly suitable for various critical fields requiring highly reliable navigation.

[0071] like Figure 5 As shown, this application also discloses an online ionospheric scintillation prediction system based on width learning, used to implement the above method, including: The data fusion and preprocessing module is used to perform time alignment on data from four types of observation sources, namely super dual auroral radar, altimeter, geomagnetic solar activity index and ionospheric phase scintillation, to form a fused dataset. It calculates the dynamic window midpoint for the vector field data of the super dual auroral radar to suppress noise, and uses a SAITS neural network based on self-attention mechanism to complete and optimize the quality of missing data in the fused dataset. The model building and training module is used to divide the preprocessed data into training set, validation set and test set, establish a width learning system model for online prediction of ionospheric scintillation, and evaluate the root mean square error (RMSE) of the width learning system model using the test set data. The online prediction and update module is used to quickly update the connection weights of the width learning system model using newly acquired observation data and an incremental learning mechanism, and output the corresponding predicted phase flicker index.

[0072] Those skilled in the art will understand that the steps described in the above specific embodiments can be implemented in the form of a computer program, which can be stored on a computer-readable storage medium (such as an optical disc, hard disk, flash memory, etc.). When the program is executed by a computer or processor, the method described in this invention can be implemented. Similarly, the system implementing this method can also be integrated into a dedicated electronic device (such as a high-performance GNSS receiver, an ionospheric monitoring station server, etc.), which includes a memory and a processor. The memory is used to store a program for executing the method of this invention, and the processor is used to execute the program to complete the prediction task.

[0073] The electronic device may include one or more processors, memory, and transmission devices, and the terminal may also include input / output devices.

[0074] Those skilled in the art will understand that the terminal can be a smartphone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile internet device (MID), a PAD, and other terminal devices.

[0075] 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 online prediction of ionospheric scintillation based on width learning, characterized in that, Includes the following steps: S1. Time alignment is performed on data from four types of observation sources: super dual auroral radar, altimeter, geomagnetic solar activity index, and ionospheric phase scintillation, to form a fusion dataset with unified time resolution; S2. Calculate the dynamic window midpoint for the vector field data of the super dual-polarization radar to suppress noise; S3. A SAITS neural network based on a self-attention mechanism is used to complete and optimize the quality of missing data in the fused dataset; S4. Divide the preprocessed data into training set, validation set, and test set; S5. Establish a width learning system model for online prediction of ionospheric scintillation, input historical time series data, and output the phase scintillation index at future time. S6. Evaluate the root mean square error (RMSE) of the width learning system model using the test set data; S7. Using the newly acquired observation data, the connection weights of the width learning system model are quickly updated using an incremental learning mechanism, and the corresponding predicted phase flicker index is output.

2. The online prediction method for ionospheric scintillation based on width learning according to claim 1, characterized in that, In step S1, the unified time resolution is a 1-minute resolution; wherein, The super dual-polarization radar extracts the first valid record every minute; The data from the altimeter is expanded to a 1-minute resolution using a copying method; The geomagnetic solar activity index data and the ionospheric phase scintillation data retain their original time series.

3. The online prediction method for ionospheric scintillation based on width learning according to claim 1, characterized in that, The vector field data of the super dual aurora radar mentioned in step S2 includes power, coordination velocity, fitted spectral width and elevation angle.

4. The online prediction method for ionospheric scintillation based on width learning according to claim 1, characterized in that, In step S3, the SAITS neural network learns the temporal distribution characteristics of the altimeter data through a self-attention mechanism and interpolates the missing positions of the altimeter data.

5. The online prediction method for ionospheric scintillation based on width learning according to claim 1, characterized in that, Step S5 specifically includes the following sub-steps: S51. The vector field data, altimeter data, and geomagnetic solar activity index data from the super dual auroral radar are used as input features, and the ionospheric phase scintillation data is used as the output target. S52. Based on the satellite observation station data, divide all input data into several batches, denoted as... , where a is the batch number; S53. For the first batch of input data Calculate the first batch of feature nodes and the first batch of enhanced feature nodes : in, and These are the weights for the first batch of feature nodes and the first batch of enhanced feature nodes, respectively. and These are the weight biases for the first batch of feature nodes and the first batch of enhanced feature nodes, respectively. and These are the activation functions for the feature nodes and the activation functions for the augmentation nodes, respectively. S54. Merge the feature nodes and enhanced feature nodes This forms the first batch of intermediate feature layers. and connect to the output layer ; S55. Calculate the first batch of intermediate feature layers. to output layer Connection weights : in, and It is the norm regularization coefficient. It is the sparsity coefficient; S56. Based on the obtained network parameters and connection layer weights, and according to the data predicted... The model predicts the phase scintillation index. The value is:

6. The online prediction method for ionospheric scintillation based on width learning according to claim 1, characterized in that, In step S7, the incremental learning mechanism specifically includes: S71. Define the memory module for the first batch of input data as follows: in This is the feature memory matrix for the first batch. This is the knowledge memory matrix for the first batch; S72. For the data input in batch a... Where a≥2, calculate the a-th batch of feature nodes. Batch a of enhanced feature nodes and the intermediate feature layer of the a-th batch formed ; S73. Generate the memory matrix for the incremental portion: in This is the incremental feature memory matrix for the a-th batch. This is the incremental knowledge memory matrix for the a-th batch; S74. Calculate the memory fast update module under the new data stream: in Let a be the feature memory matrix of the a-th batch. This is the knowledge memory matrix for the a-th batch; S75. Calculate the updated connection weights: in, λ is the connection weight of the a-th batch, I is the identity matrix, and λ is the regularization coefficient.

7. The online prediction method for ionospheric scintillation based on width learning according to claim 6, characterized in that, The root mean square error (RMSE) is used to evaluate the model's predictive performance.

8. An online prediction system for ionospheric scintillation based on width learning, characterized in that, include: The data fusion and preprocessing module is used to perform time alignment on data from four types of observation sources, namely super dual auroral radar, altimeter, geomagnetic solar activity index and ionospheric phase scintillation, to form a fused dataset. It calculates the dynamic window midpoint for the vector field data of the super dual auroral radar to suppress noise, and uses a SAITS neural network based on self-attention mechanism to complete and optimize the quality of missing data in the fused dataset. The model building and training module is used to divide the preprocessed data into training set, validation set and test set, establish a width learning system model for online prediction of ionospheric scintillation, and evaluate the root mean square error (RMSE) of the width learning system model using the test set data. The online prediction and update module is used to quickly update the connection weights of the width learning system model using newly acquired observation data and an incremental learning mechanism, and output the corresponding predicted phase flicker index.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute a computer program stored in the memory to implement the online prediction method for ionospheric scintillation based on width learning as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the online prediction method for ionospheric scintillation based on width learning as described in any one of claims 1 to 7.

Images