A method, device, medium and product for predicting total electron content of ionosphere

Abstract

The application discloses a method, device, medium and product for predicting the total electron content of the ionosphere, which comprises the following steps: acquiring geomagnetic disturbance data of a region where a single-station global navigation satellite system is located, wherein the geomagnetic disturbance data comprises a magnetic north component and a magnetic east component; calculating the horizontal geomagnetic disturbance intensity based on the magnetic north component and the magnetic east component; inputting the horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the acquired solar activity index and the historical observation sequence of the total electron content into a preset prediction model, so as to obtain a plurality of subspace attention features through a multi-head attention mechanism, and to obtain an encoding feature matrix through linear change processing; decoding the encoding feature matrix through a decoder of the prediction model to obtain a context vector, and performing autoregressive processing on the context vector until a target total electron content prediction sequence is obtained. The application can improve the prediction accuracy of the total electron content in the ionosphere above the single-station global navigation satellite system.

Description

Technical Field

[0001] This invention relates to the field of ionospheric prediction technology, and in particular to a method, apparatus, medium, and product for predicting the total electron content of the ionosphere. Background Technology

[0002] The ionosphere is the region of Earth's atmosphere that is partially ionized by solar radiation. Its total electron content (TEC) is a key parameter affecting the performance of satellite navigation, communication, and remote sensing systems. In GNSS (Global Navigation Satellite System) applications, ionospheric delay is one of the main sources of positioning errors; its drastic fluctuations can lead to decreased positioning accuracy, cycle slips, and even signal loss. Therefore, achieving high-precision and timely ionospheric TEC prediction is of significant scientific and engineering value for improving GNSS positioning accuracy and ensuring aviation communications and space weather early warning.

[0003] In existing technologies, common ionospheric TEC prediction methods mainly rely on modeling and prediction by combining historical TEC data with global or regionally averaged space environment indices (such as geomagnetic indices Kp and Dst). However, this method has significant drawbacks: since the geomagnetic indices (such as Kp and Dst) it relies on are global or large-scale averages, they cannot reflect the fine geomagnetic disturbance structure occurring at specific geographical locations. Especially during extreme space weather events such as geomagnetic storms and substorms, local ionospheric TEC will change drastically and rapidly. Such global indices are difficult to characterize these local disturbances in a timely and accurate manner, resulting in slow model response to single-station TEC abrupt changes and significantly increased prediction errors, making it difficult to meet the application requirements of high-precision single-station forecasts. Summary of the Invention

[0004] This invention provides a method, apparatus, medium, and product for predicting the total electron content of the ionosphere, which can improve the prediction accuracy of the total electron content in the ionosphere above a single-station global navigation satellite system.

[0005] In a first aspect, an embodiment of the present invention provides a method for predicting the total electron content of the ionosphere, comprising: Acquire geomagnetic disturbance data of the area where a single-station global navigation satellite system is located, wherein the geomagnetic disturbance data includes a magnetic northward component and a magnetic eastward component; The intensity of the horizontal geomagnetic disturbance is calculated based on the magnetic north component and the magnetic east component. The horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the pre-acquired solar activity index, and the historical observation sequence of total electron content are input into a preset prediction model. Multi-head parallel attention calculation is performed through the multi-head attention mechanism of the encoder of the prediction model to obtain several subspace attention features. The subspace attention features are then linearly transformed to obtain an encoded feature matrix. The encoded feature matrix is ​​decoded by the decoder of the prediction model to obtain a context vector. The context vector is then subjected to autoregressive processing until a preset termination condition is met to obtain the target total electron content prediction sequence of the ionosphere above the single-station global navigation satellite system.

[0006] This approach, by acquiring high spatiotemporal resolution geomagnetic disturbance data of the region where a single-station global navigation satellite system is located, replaces the traditional method of relying on globally averaged geomagnetic indices (such as Kp and Dst). It directly captures the fine local magnetic field disturbances affecting that single station, providing a data foundation for solving the problem of being unable to capture the details of intense geomagnetic disturbances at specific locations. This fundamentally improves the ability to perceive local ionospheric disturbances and enhances prediction accuracy. Furthermore, by calculating the horizontal geomagnetic disturbance intensity based on the magnetic north and magnetic east components, the directional change of the geomagnetic vector is transformed into a comprehensive scalar intensity index, providing a more direct and stable basis for subsequent prediction models. Determining the intensity characteristics of geomagnetic activity helps the prediction model more accurately quantify the impact of geomagnetic disturbances on TEC, thus improving prediction accuracy. By inputting geomagnetic disturbance data, horizontal geomagnetic disturbance intensity, and pre-acquired solar activity index and historical observation sequences of total electron content into the pre-defined prediction model, a multi-level feature system covering local disturbance vector details, overall disturbance intensity, large-scale spatial environment, and its own evolution history is constructed. This enables the prediction model to comprehensively capture multiple factors affecting TEC changes, providing information support for high-precision single-station total electron content prediction. Multi-head parallel attention calculation is performed through the multi-head attention mechanism of the prediction model encoder. This process yields several subspace attention features, enabling the prediction model to extract information from different feature subspaces in parallel. This allows for distributed modeling of complex relationships between multi-source features, thus more comprehensively capturing various factors influencing TEC changes and improving prediction accuracy. Linear transformation of the attention features in each subspace yields an encoded feature matrix, achieving effective fusion and dimensionality reduction of attention information from different subspaces. This integrates scattered feature representations into a unified deep feature representation, enhancing the robustness and information density of the feature representation and contributing to improved prediction accuracy. Finally, the decoder of the prediction model decodes the encoded feature matrix to obtain the context. The context vector can dynamically extract the most relevant historical information and feature associations to the current prediction time from the encoded feature matrix, forming a context vector rich in temporal dependencies and causal relationships, thus improving prediction accuracy. Autoregressive processing is performed on the context vector until a preset termination condition is met, obtaining the target total electron content prediction sequence. This allows the prediction model to progressively and recursively generate TEC values ​​for future times based on this context vector, achieving continuous dynamic tracking of TEC change trends. Especially during extreme events such as geomagnetic storms, it can significantly improve the predictive ability and stability of TEC abrupt changes, thereby improving the overall prediction accuracy of single-station ionospheric TEC. This application can improve the prediction accuracy of the total electron content in the ionosphere above a single-station global navigation satellite system.

[0007] Furthermore, the horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the pre-acquired solar activity index, and the historical observation sequence of total electron content are input into a preset prediction model. Multi-head parallel attention calculation is then performed through the multi-head attention mechanism of the prediction model's encoder to obtain several subspace attention features, specifically including: The horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the solar activity index, and the historical observation sequence of the total electron content are merged into channels to obtain a multi-channel time series feature sequence; The multi-channel temporal feature sequence is input into the embedding layer of the encoder for mapping to obtain a first vector sequence; The first vector sequence is subjected to positional encoding injection to obtain the second vector sequence; The second vector sequence is input into the multi-head attention layer of the encoder for multi-head parallel attention calculation to obtain several subspace attention features.

[0008] By inputting geomagnetic disturbance data, horizontal geomagnetic disturbance intensity, and pre-acquired solar activity index and historical observation sequences of total electron content into a pre-defined prediction model, a multi-level feature system covering local disturbance vector details, overall disturbance intensity, large-scale spatial environment, and its own evolution history is constructed. This enables the prediction model to comprehensively capture multiple factors affecting TEC changes, providing information support for high-precision single-station total electron content prediction. Through the multi-head attention mechanism of the prediction model encoder, multi-head parallel attention calculation is performed to obtain several subspace attention features. This allows the prediction model to extract information from different feature subspaces in parallel, realizing distributed modeling of complex relationships between multi-source features, thereby more comprehensively capturing various factors affecting TEC changes and improving prediction accuracy.

[0009] Furthermore, the multi-head attention layer comprises several attention heads. The step of inputting the second vector sequence into the multi-head attention layer of the encoder for multi-head parallel attention computation to obtain several subspace attention features specifically includes: For each attention head, a query linear mapping is performed on the second vector sequence to obtain a query vector sequence, a key linear mapping is performed on the second vector sequence to obtain a key vector sequence, and a value linear mapping is performed on the second vector sequence to obtain a value vector sequence; The query vector sequence and the key vector sequence are respectively input into the scaling dot product attention formula to obtain the attention weight matrix; The attention weight matrix and the value vector sequence corresponding to each attention head are weighted and calculated to obtain the subspace attention features corresponding to each attention head.

[0010] By employing parallel linear mapping and attention computation across multiple heads, the prediction model can simultaneously capture complex dependencies and temporal evolution patterns among multi-source data from different feature representation subspaces. This approach is particularly suitable for handling nonlinear, multi-timescale coupling effects between geomagnetic disturbances, solar activity index, and total electron content. Consequently, it significantly enhances the prediction model's sensitivity to sudden space weather events (such as geomagnetic storms) and improves the accuracy of single-station total electron content prediction.

[0011] Furthermore, the step of decoding the encoded feature matrix using the decoder of the prediction model to obtain the context vector specifically includes: The encoded feature matrix is ​​input into the cross-attention layer of the decoder to perform a linear mapping on the encoded feature matrix through the cross-attention layer, thereby obtaining a key vector and a value vector; Based on the obtained state information of the decoder, a query vector is generated; The context vector is obtained by performing cross-attention calculation on the key vector, the value vector, and the query vector.

[0012] In this way, the decoder of the prediction model decodes the encoded feature matrix to obtain the context vector. The historical information and feature associations most relevant to the current prediction time can be dynamically extracted from the encoded feature matrix to form a context vector rich in temporal dependence and causal relationship, thereby improving the prediction accuracy.

[0013] Furthermore, the autoregressive processing of the context vector until a preset termination condition is met, to obtain the target total electron content prediction sequence of the ionosphere above the single-station global navigation satellite system, specifically includes: Based on the context vector, the predicted value of the total first electron content at the first prediction time is obtained. Add the predicted value of the first total electron content to the first prediction sequence; Determine whether the first prediction sequence meets the preset termination condition. If not, then based on the context vector and the first prediction sequence, predict the second total electron content value corresponding to the second prediction time. The second predicted total electron content value is integrated with the first predicted sequence to obtain the second predicted sequence; Determine whether the second prediction sequence meets the preset termination condition. If it does, then based on the second prediction sequence, determine the target total electron content prediction sequence of the ionosphere above the single-station global navigation satellite system.

[0014] By performing autoregressive processing on the context vector until a preset termination condition is met, the target total electron content prediction sequence is obtained. This allows the prediction model to generate TEC values ​​for future times step by step and recursively based on the context vector, achieving continuous dynamic tracking of TEC change trends. Especially during extreme events such as geomagnetic storms, it can significantly improve the prediction ability and stability of TEC abrupt changes, thereby improving the overall prediction accuracy of single-station ionospheric TEC.

[0015] Furthermore, the calculation of the horizontal geomagnetic disturbance intensity based on the magnetic north component and the magnetic east component specifically includes: Based on the magnetic north component, the squared value of the first disturbance amplitude is calculated; Based on the magnetic eastward component, the squared value of the second disturbance amplitude is calculated; Based on the squared value of the first disturbance amplitude and the squared value of the second disturbance amplitude, the disturbance fusion value is calculated; The intensity of the horizontal geomagnetic disturbance is obtained based on the disturbance fusion value.

[0016] The horizontal geomagnetic disturbance intensity is calculated based on the magnetic north and magnetic east components, transforming the directional change of the geomagnetic vector into a comprehensive scalar intensity index. This provides a more direct and stable geomagnetic activity intensity characteristic for subsequent prediction models, helping to more accurately quantify the impact of geomagnetic disturbances on TEC and improve prediction accuracy.

[0017] Furthermore, the linear transformation processing of the attention features of each subspace to obtain the encoded feature matrix specifically includes: The attention features of each subspace are concatenated according to their feature dimensions to obtain a concatenated feature vector. The concatenated feature vector is linearly transformed to obtain the encoded feature matrix.

[0018] By performing linear transformation on the attention features of each subspace to obtain the encoded feature matrix, the effective fusion and dimensionality reduction of attention information from different subspaces are achieved. This integrates the scattered feature representations into a unified deep feature representation, enhancing the robustness and information density of the feature representation and contributing to the improvement of prediction accuracy.

[0019] In a second aspect, the present invention provides a device for predicting the total electron content of the ionosphere, comprising a first module, a second module, and a third module; The first module is used to acquire geomagnetic disturbance data of the area where a single-station global navigation satellite system is located, wherein the geomagnetic disturbance data includes a magnetic north component and a magnetic east component; The second module is used to calculate the intensity of horizontal geomagnetic disturbance based on the magnetic north component and the magnetic east component; The third module is used to input the horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the pre-acquired solar activity index, and the historical observation sequence of total electron content into a preset prediction model. The prediction model's encoder performs multi-head parallel attention calculation to obtain several subspace attention features. Each subspace attention feature is then linearly transformed to obtain an encoded feature matrix. The prediction model's decoder decodes the encoded feature matrix to obtain a context vector. The context vector is then autoregressed until a preset termination condition is met to obtain the target total electron content prediction sequence of the ionosphere above the single-station global navigation satellite system.

[0020] This approach, employing the first module to acquire high spatiotemporal resolution geomagnetic disturbance data of the region where a single-station global navigation satellite system is located, replaces the traditional method of relying on globally averaged geomagnetic indices (such as Kp and Dst). It directly captures the fine local magnetic field disturbances affecting the single station, providing a data foundation for addressing the inability to capture the details of intense geomagnetic disturbances at specific locations. This fundamentally enhances the ability to perceive local ionospheric disturbances and improves prediction accuracy. The second module calculates the horizontal geomagnetic disturbance intensity based on the magnetic north and magnetic east components, transforming the directional change of the geomagnetic vector into a comprehensive scalar intensity index, providing a more comprehensive basis for subsequent prediction models. Direct and more stable geomagnetic activity intensity characteristics help the prediction model more accurately quantify the impact of geomagnetic disturbances on TEC, thus improving prediction accuracy. The third module inputs geomagnetic disturbance data, horizontal geomagnetic disturbance intensity, and pre-acquired solar activity index and historical observation sequences of total electron content into the preset prediction model, constructing a multi-level feature system covering local disturbance vector details, overall disturbance intensity, large-scale spatial environment, and its own evolution history. This enables the prediction model to comprehensively capture multiple factors affecting TEC changes, providing information support for high-precision single-station total electron content prediction. Multi-head attention mechanisms in the prediction model encoder are used for multi-head merging. Attention calculations are performed to obtain several subspace attention features, enabling the prediction model to extract information from different feature subspaces in parallel. This achieves distributed modeling of complex relationships between multi-source features, thus more comprehensively capturing various factors affecting TEC changes and improving prediction accuracy. Linear transformation processing is applied to the attention features of each subspace to obtain the encoded feature matrix, achieving effective fusion and dimensionality reduction of attention information from different subspaces. This integrates scattered feature representations into a unified deep feature representation, enhancing the robustness and information density of the feature representation and contributing to improved prediction accuracy. The decoder of the prediction model decodes the encoded feature matrix to obtain... The context vector can dynamically extract the most relevant historical information and feature associations to the current prediction time from the encoded feature matrix, forming a context vector rich in temporal dependence and causal relationship, thus improving prediction accuracy. The context vector is subjected to autoregressive processing until a preset termination condition is met to obtain the target total electron content prediction sequence. The prediction model can then generate TEC values ​​for future times step by step and recursively based on this context vector, realizing continuous dynamic tracking of TEC change trends. Especially during extreme events such as geomagnetic storms, it can significantly improve the prediction ability and stability of TEC mutations, thereby improving the overall prediction accuracy of single-station ionospheric TEC.

[0021] Thirdly, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device or apparatus where the computer-readable storage medium is located to perform a method for predicting the total electron content of the ionosphere.

[0022] Fourthly, another embodiment of the present invention provides a computer program product, including a computer program or instructions, which, when executed by a communication device, realizes a method for predicting the total electron content of the ionosphere. Attached Figure Description

[0023] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating an embodiment of a method for predicting the total electron content of the ionosphere provided in this application; Figure 2 This is a flowchart illustrating steps S201 to S204 provided in this application; Figure 3 This application provides a schematic diagram of the structure of a device for predicting the total electron content of the ionosphere. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0027] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0028] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0029] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0030] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0031] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0032] In the field of ionospheric prediction technology, drastic fluctuations in the total electron content (TEC) of the ionosphere can severely affect the performance of satellite navigation and communication systems. Therefore, achieving high-precision single-station TEC prediction is of great significance. Existing prediction methods mostly rely on historical TEC data and global geomagnetic indices (such as Kp and Dst), but they have the following limitations: First, global indices are difficult to characterize local fine geomagnetic disturbances, especially during extreme events such as geomagnetic storms; second, they are slow to respond to single-station TEC abrupt changes, leading to increased prediction errors and failing to meet the requirements for high-precision, real-time single-station forecasts.

[0033] See Figure 1 In order to improve the prediction accuracy of the total electron content in the ionosphere above a single-station global navigation satellite system, an embodiment of the present invention provides a method for predicting the total electron content in the ionosphere, including steps S101 to S103. Step S101: Obtain geomagnetic disturbance data of the area where the single-station global navigation satellite system is located, wherein the geomagnetic disturbance data includes a magnetic northward component and a magnetic eastward component; In some embodiments, geomagnetic disturbance data of the area where the single-station GNSS station is located is acquired. The geomagnetic disturbance data includes a magnetic north component and a magnetic east component. Specifically, it includes: acquiring geomagnetic station observation data from the SuperMAG geomagnetic observation network that is geographically close to the target single-station GNSS station. This includes three magnetometer three-component disturbance data, namely the magnetic north component, the magnetic east component, and the vertical component, which can reflect the disturbance of the magnetosphere-ionospheric current system after removing the geomagnetic main field and solar diurnal variation (Sq).

[0034] In some embodiments, after acquiring geomagnetic disturbance data for the area where a single-station global navigation satellite system is located, the method further includes: using thresholding and outlier statistics to detect outliers in the geomagnetic disturbance data; supplementing detected outliers or missing values ​​through linear interpolation during magnetically quiet days; and replacing them with data from neighboring geomagnetic stations with similar latitude and longitude to geomagnetic latitude during periods of magnetic disturbance; unifying the timestamps of the anomaly-processed geomagnetic disturbance data by substituting the hours corresponding to each data timestamp into preset time sine and time cosine functions to generate sine (HrS) and cosine (HrC) terms to construct independent time periodic characteristics for each geomagnetic disturbance data; and unifying the time resolution of the timestamp-processed geomagnetic disturbance data to a preset time value (e.g., 1 hour) to complete the preprocessing of the geomagnetic disturbance data.

[0035] It should be noted that, due to the differences in sampling frequencies for various types of data (e.g., geomagnetic data is 1 second or 1 minute, GNSSTEC is usually 30 seconds to 1 minute, and solar activity indices F10.7, SSN, and Lyα are daily values), it is necessary to reduce the resolution of high-frequency data by downsampling or window averaging, and to keep the daily values ​​of low-frequency solar activity indices expanded, so as to unify the resolution of all data to a preset time value (e.g., 1 hour).

[0036] In some embodiments, the relevant formulas following the acquisition of geomagnetic disturbance data for the area where a single-station global navigation satellite system is located include: Time sine function: ; Time cosine function: ; In the formula, This refers to the actual number of hours worked. The sine term represents the periodic characteristic. The cosine term represents the periodic characteristic.

[0037] It should be noted that the time period feature is used to characterize daily periodic information. By decomposing time (hours) into sine and cosine, the continuity and periodicity of the time feature can be maintained, avoiding the discontinuity problem that occurs at the time boundary (such as 23:00 to 0:00) in traditional time coding. During the acquisition and preprocessing process, it is necessary to ensure that the timestamps of all data are consistent to facilitate subsequent time alignment.

[0038] Step S102: Calculate the horizontal geomagnetic disturbance intensity based on the magnetic north component and the magnetic east component; In some embodiments, calculating the horizontal geomagnetic disturbance intensity based on the magnetic north component and the magnetic east component specifically includes: calculating a first squared disturbance amplitude value based on the magnetic north component; calculating a second squared disturbance amplitude value based on the magnetic east component; calculating a disturbance fusion value based on the first squared disturbance amplitude value and the second squared disturbance amplitude value; and obtaining the horizontal geomagnetic disturbance intensity based on the disturbance fusion value. Specifically, the squared values ​​of the magnetic north component and the magnetic east component are calculated separately to obtain the first squared disturbance amplitude value and the second squared disturbance amplitude value; the first squared disturbance amplitude value and the second squared disturbance amplitude value are added together to obtain the disturbance fusion value; and the square root operation is performed on the disturbance fusion value to obtain a scalar index that comprehensively characterizes the horizontal disturbance intensity of the local magnetic field, i.e., the horizontal geomagnetic disturbance intensity.

[0039] In some embodiments, the formula for calculating the horizontal geomagnetic disturbance intensity based on the magnetic north component and the magnetic east component specifically includes: Formula for calculating the intensity of horizontal geomagnetic disturbance: ; In the formula, This is the magnetic north component; This is the magnetic eastward component; This represents the intensity of horizontal geomagnetic disturbance.

[0040] It should be noted that the calculation process of horizontal geomagnetic disturbance intensity transforms the directional change of the geomagnetic vector into a single intensity feature, which can more intuitively reflect the disturbance level caused by the enhancement of current systems such as auroral photocurrents and jet streams, and provide a more stable and easier-to-process geomagnetic activity input for subsequent prediction models.

[0041] The horizontal geomagnetic disturbance intensity is calculated based on the magnetic north and magnetic east components, transforming the directional change of the geomagnetic vector into a comprehensive scalar intensity index. This provides a more direct and stable geomagnetic activity intensity characteristic for subsequent prediction models, helping to more accurately quantify the impact of geomagnetic disturbances on TEC and improve prediction accuracy.

[0042] Step S103: Input the horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the pre-acquired solar activity index, and the historical observation sequence of total electron content into a preset prediction model. Perform multi-head parallel attention calculation through the multi-head attention mechanism of the encoder of the prediction model to obtain several subspace attention features. Perform linear transformation processing on each subspace attention feature to obtain an encoded feature matrix. Decode the encoded feature matrix through the decoder of the prediction model to obtain a context vector. Perform autoregressive processing on the context vector until a preset termination condition is met to obtain the target total electron content prediction sequence of the ionosphere above the single-station global navigation satellite system. See Figure 2 In some embodiments, the step of inputting the horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the pre-acquired solar activity index and the historical observation sequence of total electron content into a preset prediction model, and performing multi-head parallel attention calculation through the multi-head attention mechanism of the encoder of the prediction model to obtain several subspace attention features, including steps S201 to S204. Step S201: Merge the horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the solar activity index, and the historical observation sequence of the total electron content to obtain a multi-channel time series feature sequence; In some embodiments, before merging the horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the solar activity index, and the historical observation sequence of total electron content to obtain a multi-channel time-series feature sequence, the method further includes: pre-processing (i.e., simultaneously acquiring the geomagnetic disturbance data) from the OMNIWEB database to obtain solar activity indices reflecting large-scale space environment changes, including sunspot number (SSN), solar 10.7 cm radio flux (F10.7), and Lyman alpha radiation (Lyα); acquiring historical observation sequences of total electron content within a historical time period from the target GNSS station; performing continuity checks and sliding window smoothing on the historical observation sequences of total electron content to detect outliers, and filtering outliers based on physical constraints (such as TEC being impossible to be negative and the daily variation range being limited), and replacing or interpolating by acquiring TEC data from neighboring GNSS stations. Missing data are filled using a method; outliers in the solar activity index are removed using a threshold method, and missing data are repaired using a linear interpolation method; all data after anomaly processing are processed in the same way as in step S101, i.e., while unifying the timestamps of the geomagnetic disturbance data in step S101, the hours corresponding to the timestamps of the historical observation sequences of the solar activity index and total electron content are substituted into preset time sine and time cosine functions respectively to generate sine terms (HrS) and cosine terms (HrC) to construct independent time periodic features; the historical observation sequences of total electron content and the solar activity index after timestamp processing are also processed in terms of time resolution to unify the resolution of all data to a preset time value (e.g., 1 hour); all data are mapped to the 0-1 interval by maximum and minimum value normalization to complete the data preprocessing.

[0043] In some embodiments, the relevant formulas preceding the channel merging of the horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the solar activity index, and the historical observation sequence of the total electron content to obtain a multi-channel time-series feature sequence include: Formula for normalizing maximum and minimum values: ; In the formula, The original value of a certain feature; This is the minimum value of the feature across the entire sample sequence; This is the maximum value of the feature across the entire sample sequence; This is the normalized value.

[0044] It should be noted that the time period feature is used to characterize daily periodic information. By decomposing time (hours) into sine and cosine, the continuity and periodicity of the time feature can be maintained, avoiding the discontinuity problem that occurs at the time boundary (such as 23:00 to 0:00) in traditional time coding. During the acquisition and preprocessing process, it is necessary to ensure that the timestamps of all data are consistent to facilitate subsequent time alignment.

[0045] For example, the training process of the prediction model specifically includes: acquiring geomagnetic disturbance data, solar activity index, and historical observation sequences of total electron content in the region where a single-station global navigation satellite system is located, wherein the geomagnetic disturbance data includes magnetic north and magnetic east components; performing data quality checks on the geomagnetic disturbance data, solar activity index, and historical observation sequences of total electron content to remove outliers or supplement missing values ​​through interpolation; calculating the horizontal geomagnetic disturbance intensity based on the magnetic north and magnetic east components; unifying the geomagnetic disturbance data, horizontal geomagnetic disturbance intensity, solar activity index, and historical observation sequences of total electron content to the same time resolution, and mapping all data to the 0-1 interval through maximum-minimum normalization to form standardized training samples; dividing the standardized training samples into training set, validation set, and test set, wherein the training set is used for model parameter learning, the validation set is used to monitor the training process and implement an early stopping strategy, and the test set is used to finally evaluate the model's prediction performance; and constructing a Transformer prediction model with a multi-head attention mechanism, whose encoder is used to extract the input sequence. The deep feature representation is used by the decoder to generate predictive sequences via autoregression. During the training phase, the input samples from the training set are input into the encoder. After embedding layer mapping and positional encoding injection, multi-head attention layers are used to perform multi-head parallel attention calculations to obtain several subspace attention features. The subspace attention features are then linearly transformed to obtain the encoded feature matrix. Based on the encoded feature matrix and decoder state information (or the predicted sequence generated in the previous iteration if it is a subsequent iteration), the decoder generates a context vector through a cross-attention layer and gradually generates the predicted sequence of total electron content for future time steps through autoregression. By minimizing the mean squared error loss function between the predicted sequence and the true sequence, the parameters of the Transformer prediction model are iteratively optimized using the backpropagation algorithm. At the same time, a validation set is used to monitor the loss change. Training is terminated early when the validation loss no longer decreases for several consecutive rounds. Finally, the prediction performance of the trained model is evaluated using a test set. The mean absolute error, root mean square error, and correlation coefficient are calculated to quantify the prediction accuracy, resulting in the trained prediction model.

[0046] For example, the relevant formulas for the training process of the prediction model specifically include: Mean Absolute Error Formula: Root mean square error formula: Correlation coefficient formula: In the formula, x i Indicates the first i Single-station TEC; Indicates the first i The TEC predicted by the model; and Don't mean x i and The average value; n and N Both represent the total number of TECs; It represents the standard deviation.

[0047] In some embodiments, the horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the solar activity index, and the historical observation sequence of the total electron content are merged to obtain a multi-channel time-series feature sequence. Specifically, this includes: precisely aligning the historical observation sequence of the total electron content (TEC), the solar activity index (including SSN, F10.7, and Lyα), the geomagnetic disturbance data (including the magnetic north component X, magnetic east component Y, and vertical component Z), and the horizontal geomagnetic disturbance intensity according to a unified timestamp to ensure that the multi-source feature data at the same time are not misaligned; and horizontally splicing and integrating all data in the feature channel dimension to merge the single-dimensional feature sequences into a multi-dimensional time-series feature sequence, forming a multi-channel time-series feature sequence with 8 feature channels (TEC, SSN, F10.7, Lyα, X, Y, Z, and horizontal geomagnetic disturbance intensity) at each time step.

[0048] Step S202: Input the multi-channel temporal feature sequence into the embedding layer of the encoder for mapping to obtain the first vector sequence; In some embodiments, the multi-channel temporal feature sequence is input into the embedding layer of the encoder for mapping to obtain a first vector sequence. Specifically, this includes: inputting the multi-channel temporal feature sequence into the embedding layer of the encoder; performing feature mapping through the learnable weight matrix of the embedding layer; transforming the discrete features in the multi-channel temporal features into a continuous vector representation that the prediction model can efficiently process; and finally obtaining a first vector sequence that satisfies... X∈R z×d The first vector sequence, where, z This represents the sequence length (e.g., the number of time steps corresponding to 24 hours). d For model dimensions.

[0049] Step S203: Position encoding injection is performed on the first vector sequence to obtain the second vector sequence; In some embodiments, positional encoding injection is performed on the first vector sequence to obtain a second vector sequence, specifically including: adding positional encoding to the first vector sequence to obtain the second vector sequence.

[0050] In some embodiments, positional encoding injection is performed on the first vector sequence to obtain the relevant formula for the second vector sequence, specifically including: Position-coded injection formula: ; In the formula, It is a positional encoding used to represent the relative positional information of each time step; This is the first vector sequence; This is the second vector sequence.

[0051] It should be noted that superimposing a preset position code on the first vector sequence can clarify the relative order of each time step in the first vector sequence, enabling the prediction model to capture the sequential dependencies of time series data.

[0052] Step S204: Input the second vector sequence into the multi-head attention layer of the encoder to perform multi-head parallel attention calculation to obtain several subspace attention features; In some embodiments, the multi-head attention layer includes several attention heads. The step of inputting the second vector sequence into the multi-head attention layer of the encoder for multi-head parallel attention computation to obtain several subspace attention features specifically includes: for each attention head, performing a query linear mapping on the second vector sequence to obtain a query vector sequence; performing a key linear mapping on the second vector sequence to obtain a key vector sequence; and performing a value linear mapping on the second vector sequence to obtain a value vector sequence; inputting the query vector sequence and the key vector sequence into a scaled dot product attention formula to obtain an attention weight matrix; and performing a weighted calculation on the attention weight matrix and the value vector sequence corresponding to each attention head to obtain the subspace attention feature corresponding to each attention head. Specifically, for each attention head in the multi-head attention layer, the second vector sequence is linearly mapped using three preset independent learnable linear transformation matrices to obtain the query vector sequence, key vector sequence, and value vector sequence. The query vector sequence and key vector sequence are then multiplied by a dot product and scaled by the square root of a scaling factor. The scaled result is then normalized using a softmax function to obtain the attention weight matrix representing the feature association strength at each time step. The attention weight matrix and value vector sequence are then weighted to obtain the subspace attention features corresponding to the current attention head. This process is repeated until all attention heads have completed their calculations, resulting in several subspace attention features corresponding to each attention head.

[0053] It should be noted that dividing the dot product result by the square root of the scaling factor can avoid the gradient vanishing problem caused by excessively large input values.

[0054] In some embodiments, the multi-head attention layer comprises a plurality of attention heads, and the step of inputting the second vector sequence into the multi-head attention layer of the encoder for multi-head parallel attention calculation to obtain relevant formulas for a plurality of subspace attention features specifically includes: The formula for the linear mapping of a query vector sequence is: ; The formula for the linear mapping of a key vector sequence is: ; The formula for the linear mapping of a value vector sequence is: ; In the formula, For the first The trainable query transformation weight matrix corresponding to each attention head. d For model dimensions, d k The query and key vector dimensions corresponding to each attention head ( dk =d / h , h (Note the total number of heads). For the first Each attention head corresponds to a trainable key transformation weight matrix; For the first The trainable transformation weight matrix corresponding to each attention head d v The dimension of the value vector corresponding to each attention head ( d v =d / h ); This is the second vector sequence; Scaling dot product attention formula: ; In the formula, d k The vector dimension corresponding to each attention head; For querying vector sequences; It is a sequence of key vectors; It is a sequence of value vectors.

[0055] By employing parallel linear mapping and attention computation across multiple heads, the prediction model can simultaneously capture complex dependencies and temporal evolution patterns among multi-source data from different feature representation subspaces. This approach is particularly suitable for handling nonlinear, multi-timescale coupling effects between geomagnetic disturbances, solar activity index, and total electron content. Consequently, it significantly enhances the prediction model's sensitivity to sudden space weather events (such as geomagnetic storms) and improves the accuracy of single-station total electron content prediction.

[0056] By inputting geomagnetic disturbance data, horizontal geomagnetic disturbance intensity, and pre-acquired solar activity index and historical observation sequences of total electron content into a pre-defined prediction model, a multi-level feature system covering local disturbance vector details, overall disturbance intensity, large-scale spatial environment, and its own evolution history is constructed. This enables the prediction model to comprehensively capture multiple factors affecting TEC changes, providing information support for high-precision single-station total electron content prediction. Through the multi-head attention mechanism of the prediction model encoder, multi-head parallel attention calculation is performed to obtain several subspace attention features. This allows the prediction model to extract information from different feature subspaces in parallel, realizing distributed modeling of complex relationships between multi-source features, thereby more comprehensively capturing various factors affecting TEC changes and improving prediction accuracy.

[0057] In some embodiments, the linear transformation processing of the attention features of each subspace to obtain the encoded feature matrix specifically includes: concatenating the attention features of each subspace along their feature dimensions to obtain a concatenated feature vector; and performing a linear transformation on the concatenated feature vector to obtain the encoded feature matrix. Specifically, the attention features of each subspace are concatenated along their feature dimensions to form a concatenated feature matrix, and a trainable linear transformation matrix is ​​used to perform a linear transformation on the concatenated feature matrix to map the feature dimensions back to the original hidden layer dimensions of the model, outputting the encoded feature matrix.

[0058] In some embodiments, the formula for performing linear transformation processing on the attention features of each subspace to obtain the encoding feature matrix specifically includes: Formula for calculating the coding feature matrix: ; In the formula, It is a trainable linear transformation matrix; For the first The attention features of the subspace corresponding to each attention head; This is the feature splicing function.

[0059] By performing linear transformation on the attention features of each subspace to obtain the encoded feature matrix, the effective fusion and dimensionality reduction of attention information from different subspaces are achieved. This integrates the scattered feature representations into a unified deep feature representation, enhancing the robustness and information density of the feature representation and contributing to the improvement of prediction accuracy.

[0060] In some embodiments, decoding the encoded feature matrix using the decoder of the prediction model to obtain a context vector specifically includes: inputting the encoded feature matrix into the cross-attention layer of the decoder to perform a linear mapping on the encoded feature matrix through the cross-attention layer, obtaining a key vector and a value vector; generating a query vector based on the acquired state information of the decoder; and performing cross-attention calculation on the key vector, the value vector, and the query vector to obtain the context vector. Specifically, the encoded feature matrix is ​​input into the cross-attention layer of the decoder, and the encoded feature matrix is ​​linearly mapped through two independent learnable linear transformation matrices to obtain the key vector and value vector of the cross-attention layer; a query vector is generated through learnable linear transformation based on the current state information of the decoder and the encoded feature matrix; and the query vector and the key vector are substituted into the cross-attention calculation formula to obtain a context vector that integrates historical feature association information.

[0061] In some embodiments, the step of decoding the encoded feature matrix through the decoder of the prediction model to obtain the relevant formula for the context vector specifically includes: The formula for calculating the query vector: ; Formula for calculating key vectors: ; The formula for calculating value vectors: ; In the formula, H enc This is the encoding feature matrix; H dec This is a matrix determined by the encoded feature matrix and the current state information of the decoder; , , All are learnable linear transformation matrices; Cross-attention calculation formula: ; In the formula, For query vector; The key vector; It is a value vector; For attention head dimension.

[0062] In this way, the decoder of the prediction model decodes the encoded feature matrix to obtain the context vector. The historical information and feature associations most relevant to the current prediction time can be dynamically extracted from the encoded feature matrix to form a context vector rich in temporal dependence and causal relationship, thereby improving the prediction accuracy.

[0063] In some embodiments, performing autoregressive processing on the context vector until a preset termination condition is met to obtain a target total electron content prediction sequence for the ionosphere above the single-station global navigation satellite system specifically includes: predicting a first total electron content prediction value corresponding to a first prediction time based on the context vector; adding the first total electron content prediction value to a first prediction sequence; determining whether the first prediction sequence meets the preset termination condition; if not, predicting a second total electron content prediction value corresponding to a second prediction time based on the context vector and the first prediction sequence; integrating the second total electron content prediction value with the first prediction sequence to obtain a second prediction sequence; determining whether the second prediction sequence meets the preset termination condition; if so, determining the target total electron content prediction sequence for the ionosphere above the single-station global navigation satellite system based on the second prediction sequence. Specifically, the context vector is input into the linear mapping layer of the decoder, and the first total electron content prediction value corresponding to the first prediction time (e.g., the first hour in the future) is generated through linear mapping. This first prediction value is used as the initial element to construct the first prediction sequence (e.g., length 1). It is determined whether the length of the first prediction sequence meets the preset termination condition (e.g., whether the prediction duration meets 24 hours). If not, the first prediction sequence is fed back to the decoder input, and features are fused with the original context vector to update the hidden state of the decoder. Based on the updated hidden state, a new query vector is regenerated, and the cross-attention calculation process is repeated to obtain a new context vector, thereby predicting the second total electron content prediction value corresponding to the second prediction time (e.g., the second hour in the future). The second prediction value is added to the first prediction sequence to obtain the second prediction sequence (e.g., length 2), and it is determined again whether the termination condition is met. This logic is used to perform autoregressive iteration until the preset termination condition is met (e.g., whether the prediction duration meets 24 hours), at which point the prediction stops and the sequence is output. This sequence is the target total electron content prediction sequence of the ionosphere above a single-station global navigation satellite system.

[0064] It should be noted that after the prediction is completed, the TEC values ​​in the predicted sequence are denormalized to restore them to the range of the actual physical quantities.

[0065] By performing autoregressive processing on the context vector until a preset termination condition is met, the target total electron content prediction sequence is obtained. This allows the prediction model to generate TEC values ​​for future times step by step and recursively based on the context vector, achieving continuous dynamic tracking of TEC change trends. Especially during extreme events such as geomagnetic storms, it can significantly improve the prediction ability and stability of TEC abrupt changes, thereby improving the overall prediction accuracy of single-station ionospheric TEC.

[0066] This approach, by acquiring high spatiotemporal resolution geomagnetic disturbance data of the region where a single-station global navigation satellite system is located, replaces the traditional method of relying on globally averaged geomagnetic indices (such as Kp and Dst). It directly captures the fine local magnetic field disturbances affecting that single station, providing a data foundation for solving the problem of being unable to capture the details of intense geomagnetic disturbances at specific locations. This fundamentally improves the ability to perceive local ionospheric disturbances and enhances prediction accuracy. Furthermore, by calculating the horizontal geomagnetic disturbance intensity based on the magnetic north and magnetic east components, the directional change of the geomagnetic vector is transformed into a comprehensive scalar intensity index, providing a more direct and stable basis for subsequent prediction models. Determining the intensity characteristics of geomagnetic activity helps the prediction model more accurately quantify the impact of geomagnetic disturbances on TEC, thus improving prediction accuracy. By inputting geomagnetic disturbance data, horizontal geomagnetic disturbance intensity, and pre-acquired solar activity index and historical observation sequences of total electron content into the pre-defined prediction model, a multi-level feature system covering local disturbance vector details, overall disturbance intensity, large-scale spatial environment, and its own evolution history is constructed. This enables the prediction model to comprehensively capture multiple factors affecting TEC changes, providing information support for high-precision single-station total electron content prediction. Multi-head parallel attention calculation is performed through the multi-head attention mechanism of the prediction model encoder. This process yields several subspace attention features, enabling the prediction model to extract information from different feature subspaces in parallel. This allows for distributed modeling of complex relationships between multi-source features, thus more comprehensively capturing various factors influencing TEC changes and improving prediction accuracy. Linear transformation of the attention features in each subspace yields an encoded feature matrix, achieving effective fusion and dimensionality reduction of attention information from different subspaces. This integrates scattered feature representations into a unified deep feature representation, enhancing the robustness and information density of the feature representation and contributing to improved prediction accuracy. Finally, the decoder of the prediction model decodes the encoded feature matrix to obtain the context. The context vector can dynamically extract the most relevant historical information and feature associations to the current prediction time from the encoded feature matrix, forming a context vector rich in temporal dependencies and causal relationships, thus improving prediction accuracy. Autoregressive processing is performed on the context vector until a preset termination condition is met, obtaining the target total electron content prediction sequence. This allows the prediction model to progressively and recursively generate TEC values ​​for future times based on this context vector, achieving continuous dynamic tracking of TEC change trends. Especially during extreme events such as geomagnetic storms, it can significantly improve the predictive ability and stability of TEC abrupt changes, thereby improving the overall prediction accuracy of single-station ionospheric TEC. This application can improve the prediction accuracy of the total electron content in the ionosphere above a single-station global navigation satellite system.

[0067] See Figure 3 Based on the above method embodiments, corresponding device embodiments are provided; An embodiment of the present invention provides a device for predicting the total electron content of the ionosphere, comprising a first module 100, a second module 200 and a third module 300; The first module 100 is used to acquire geomagnetic disturbance data of the area where a single-station global navigation satellite system is located, wherein the geomagnetic disturbance data includes a magnetic north component and a magnetic east component; The second module 200 is used to calculate the horizontal geomagnetic disturbance intensity based on the magnetic north component and the magnetic east component; The third module 300 is used to input the horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the pre-acquired solar activity index, and the historical observation sequence of total electron content into a preset prediction model. The prediction model uses a multi-head attention mechanism of the encoder to perform multi-head parallel attention calculation to obtain several subspace attention features. Each subspace attention feature is then linearly transformed to obtain an encoded feature matrix. The prediction model decodes the encoded feature matrix to obtain a context vector. The context vector is then subjected to autoregressive processing until a preset termination condition is met to obtain the target total electron content prediction sequence of the ionosphere above the single-station global navigation satellite system.

[0068] This approach, employing the first module to acquire high spatiotemporal resolution geomagnetic disturbance data of the region where a single-station global navigation satellite system is located, replaces the traditional method of relying on globally averaged geomagnetic indices (such as Kp and Dst). It directly captures the fine local magnetic field disturbances affecting the single station, providing a data foundation for addressing the inability to capture the details of intense geomagnetic disturbances at specific locations. This fundamentally enhances the ability to perceive local ionospheric disturbances and improves prediction accuracy. The second module calculates the horizontal geomagnetic disturbance intensity based on the magnetic north and magnetic east components, transforming the directional change of the geomagnetic vector into a comprehensive scalar intensity index, providing a more comprehensive basis for subsequent prediction models. Direct and more stable geomagnetic activity intensity characteristics help the prediction model more accurately quantify the impact of geomagnetic disturbances on TEC, thus improving prediction accuracy. The third module inputs geomagnetic disturbance data, horizontal geomagnetic disturbance intensity, and pre-acquired solar activity index and historical observation sequences of total electron content into the preset prediction model, constructing a multi-level feature system covering local disturbance vector details, overall disturbance intensity, large-scale spatial environment, and its own evolution history. This enables the prediction model to comprehensively capture multiple factors affecting TEC changes, providing information support for high-precision single-station total electron content prediction. Multi-head attention mechanisms in the prediction model encoder are used for multi-head merging. Attention calculations are performed to obtain several subspace attention features, enabling the prediction model to extract information from different feature subspaces in parallel. This achieves distributed modeling of complex relationships between multi-source features, thus more comprehensively capturing various factors affecting TEC changes and improving prediction accuracy. Linear transformation processing is applied to the attention features of each subspace to obtain the encoded feature matrix, achieving effective fusion and dimensionality reduction of attention information from different subspaces. This integrates scattered feature representations into a unified deep feature representation, enhancing the robustness and information density of the feature representation and contributing to improved prediction accuracy. The decoder of the prediction model decodes the encoded feature matrix to obtain... The context vector can dynamically extract the most relevant historical information and feature associations to the current prediction time from the encoded feature matrix, forming a context vector rich in temporal dependence and causal relationship, thus improving prediction accuracy. The context vector is subjected to autoregressive processing until a preset termination condition is met to obtain the target total electron content prediction sequence. The prediction model can then generate TEC values ​​for future times step by step and recursively based on this context vector, realizing continuous dynamic tracking of TEC change trends. Especially during extreme events such as geomagnetic storms, it can significantly improve the prediction ability and stability of TEC mutations, thereby improving the overall prediction accuracy of single-station ionospheric TEC.

[0069] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can realize the method for predicting the total electron content of the ionosphere provided by any of the above-described method embodiments of the present invention.

[0070] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0071] Based on the above-described embodiment of the method for predicting the total electron content of the ionosphere, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for predicting the total electron content of the ionosphere according to any embodiment of the present invention.

[0072] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0073] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0074] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0075] Based on the above-described method embodiments, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute a method for predicting the total ionospheric electron content as described in any of the above-described method embodiments of the present invention.

[0076] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0077] Based on the above-described method embodiments, another embodiment of the present invention provides a computer program product, including a computer program or instructions, which, when executed by a communication device, realizes a method for predicting the total electron content of the ionosphere.

[0078] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for predicting the total electron content of the ionosphere, characterized in that, include: Acquire geomagnetic disturbance data of the area where a single-station global navigation satellite system is located, wherein the geomagnetic disturbance data includes a magnetic northward component and a magnetic eastward component; The intensity of the horizontal geomagnetic disturbance is calculated based on the magnetic north component and the magnetic east component. The horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the pre-acquired solar activity index, and the historical observation sequence of total electron content are input into a preset prediction model. Multi-head parallel attention calculation is performed through the multi-head attention mechanism of the encoder of the prediction model to obtain several subspace attention features. The subspace attention features are then linearly transformed to obtain an encoded feature matrix. The encoded feature matrix is ​​decoded by the decoder of the prediction model to obtain a context vector. The context vector is then subjected to autoregressive processing until a preset termination condition is met to obtain the target total electron content prediction sequence of the ionosphere above the single-station global navigation satellite system.

2. The method for predicting the total electron content of the ionosphere as described in claim 1, characterized in that, The horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the pre-acquired solar activity index, and the historical observation sequence of total electron content are input into a preset prediction model. Multi-head parallel attention calculation is then performed through the multi-head attention mechanism of the prediction model's encoder to obtain several subspace attention features, specifically including: The horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the solar activity index, and the historical observation sequence of the total electron content are merged into channels to obtain a multi-channel time series feature sequence; The multi-channel temporal feature sequence is input into the embedding layer of the encoder for mapping to obtain a first vector sequence; The first vector sequence is subjected to positional encoding injection to obtain the second vector sequence; The second vector sequence is input into the multi-head attention layer of the encoder for multi-head parallel attention calculation to obtain several subspace attention features.

3. The method for predicting the total electron content of the ionosphere as described in claim 2, characterized in that, The multi-head attention layer comprises several attention heads. The step of inputting the second vector sequence into the multi-head attention layer of the encoder for multi-head parallel attention computation, resulting in several subspace attention features, specifically includes: For each attention head, a query linear mapping is performed on the second vector sequence to obtain a query vector sequence, a key linear mapping is performed on the second vector sequence to obtain a key vector sequence, and a value linear mapping is performed on the second vector sequence to obtain a value vector sequence; The query vector sequence and the key vector sequence are respectively input into the scaling dot product attention formula to obtain the attention weight matrix; The attention weight matrix and the value vector sequence corresponding to each attention head are weighted and calculated to obtain the subspace attention features corresponding to each attention head.

4. The method for predicting the total electron content of the ionosphere as described in claim 1, characterized in that, The step of decoding the encoded feature matrix using the decoder of the prediction model to obtain the context vector specifically includes: The encoded feature matrix is ​​input into the cross-attention layer of the decoder to perform a linear mapping on the encoded feature matrix through the cross-attention layer, thereby obtaining a key vector and a value vector; Based on the obtained state information of the decoder, a query vector is generated; The context vector is obtained by performing cross-attention calculation on the key vector, the value vector, and the query vector.

5. The method for predicting the total electron content of the ionosphere as described in claim 1, characterized in that, The process of performing autoregressive processing on the context vector until a preset termination condition is met, to obtain the target total electron content prediction sequence of the ionosphere above the single-station global navigation satellite system, specifically includes: Based on the context vector, the predicted value of the total first electron content at the first prediction time is obtained. Add the predicted value of the first total electron content to the first prediction sequence; Determine whether the first prediction sequence meets the preset termination condition. If not, then based on the context vector and the first prediction sequence, predict the second total electron content value corresponding to the second prediction time. The second predicted total electron content value is integrated with the first predicted sequence to obtain the second predicted sequence; Determine whether the second prediction sequence meets the preset termination condition. If it does, then based on the second prediction sequence, determine the target total electron content prediction sequence of the ionosphere above the single-station global navigation satellite system.

6. The method for predicting the total electron content of the ionosphere as described in claim 1, characterized in that, The calculation of the horizontal geomagnetic disturbance intensity based on the magnetic north component and the magnetic east component specifically includes: Based on the magnetic north component, the squared value of the first disturbance amplitude is calculated; Based on the magnetic eastward component, the squared value of the second disturbance amplitude is calculated; Based on the squared value of the first disturbance amplitude and the squared value of the second disturbance amplitude, the disturbance fusion value is calculated; The intensity of the horizontal geomagnetic disturbance is obtained based on the disturbance fusion value.

7. The method for predicting the total electron content of the ionosphere as described in claim 1, characterized in that, The linear transformation of the attention features of each subspace to obtain the encoded feature matrix specifically includes: The attention features of each subspace are concatenated according to their feature dimensions to obtain a concatenated feature vector. The concatenated feature vector is linearly transformed to obtain the encoded feature matrix.

8. A device for predicting the total electron content of the ionosphere, characterized in that, It includes Module 1, Module 2, and Module 3; The first module is used to acquire geomagnetic disturbance data of the area where a single-station global navigation satellite system is located, wherein the geomagnetic disturbance data includes a magnetic north component and a magnetic east component; The second module is used to calculate the intensity of horizontal geomagnetic disturbance based on the magnetic north component and the magnetic east component; The third module is used to input the horizontal geomagnetic disturbance intensity, the geomagnetic disturbance data, the pre-acquired solar activity index, and the historical observation sequence of total electron content into a preset prediction model. The prediction model's encoder performs multi-head parallel attention calculation to obtain several subspace attention features. Each subspace attention feature is then linearly transformed to obtain an encoded feature matrix. The prediction model's decoder decodes the encoded feature matrix to obtain a context vector. The context vector is then autoregressed until a preset termination condition is met to obtain the target total electron content prediction sequence of the ionosphere above the single-station global navigation satellite system.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device or apparatus containing the computer-readable storage medium to perform the method for predicting the total ionospheric electron content as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by the communication device, they implement the method for predicting the total electron content of the ionosphere as described in any one of claims 1 to 7.

Images