Deep learning-based methods, devices, and equipment for predicting total ionospheric electron content

Abstract

This application relates to a method, apparatus, and device for predicting total ionospheric electron content based on deep learning. The method includes: acquiring an M-day historical time-series IGS-TEC map; where M is a positive integer not less than 3; generating a corresponding label map for the historical time-series IGS-TEC map using semantic segmentation; the label map is used to indicate ionospheric peak structure regions and non-ionospheric peak structure regions; inputting the historical time-series IGS-TEC map and the corresponding label map into a trained TEC prediction model to obtain a future M-day TEC prediction map output by the TEC prediction model; the TEC prediction model is a pix2pixhd deep learning model trained on global TEC data. This significantly improves the accuracy of TEC prediction in low-latitude regions.

Description

Technical Field

[0001] This invention belongs to the field of ionospheric detection technology, and relates to a method, apparatus and equipment for predicting the total electron content of the ionosphere based on deep learning. Background Technology

[0002] Total Electron Content (TEC) represents the total number of electrons along the path from the satellite to the receiver and is one of the main parameters indicating ionospheric disturbances in space weather. Accurate TEC measurement is crucial for various fields, including aviation, satellite positioning, remote sensing systems, communications, and defense. Therefore, ionospheric TEC forecasting has significant practical implications. Currently, using neural networks for ionospheric TEC forecasting modeling has become a very popular method. Neural networks, by learning the complexity between variables and undergoing multiple training iterations, can obtain the functional relationship between the input vector and the output result. Some researchers have also used conditional adversarial neural networks to establish TEC forecasting models one day in advance. Furthermore, researchers have analyzed the performance of three deep learning networks (ARIMA, LSTM, and seq2seq models) for TEC forecasting, showing that all three network models perform poorly during geomagnetic storms. Others have used LSTM models to predict TEC for the next 5 hours. However, in developing this invention, the inventors discovered that currently used neural network modeling methods typically use TEC maps as model input for forecasting, which suffers from low accuracy in TEC forecasting for low-latitude regions. Summary of the Invention

[0003] To address the problems existing in the above-mentioned traditional methods, this invention proposes a deep learning-based method for predicting total ionospheric electron content, a deep learning-based device for predicting total ionospheric electron content, and a computer device that can significantly improve the accuracy of TEC forecasting in low-latitude regions.

[0004] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0005] On the one hand, a deep learning-based method for predicting the total electron content of the ionosphere is provided, including the following steps:

[0006] Obtain the historical time series IGS-TEC map for M days; M is a positive integer not less than 3;

[0007] Semantic segmentation is used to generate corresponding label maps for historical time-series IGS-TEC maps; the label maps are used to indicate ionospheric peak structure regions and non-ionospheric peak structure regions.

[0008] The historical time series IGS-TEC map and the corresponding label map are input into the trained TEC forecast model to obtain the TEC forecast map for the next M days output by the TEC forecast model; the TEC forecast model is a model based on the deep learning method pix2pixhd, trained on global TEC data.

[0009] In one embodiment, the step of generating a label map corresponding to a historical time-series IGS-TEC map using semantic segmentation includes:

[0010] The first mask is used to cover the ionospheric peak structure in the historical time series IGS-TEC map, and the second mask is used to cover the remaining areas in the historical time series IGS-TEC map other than the ionospheric peak structure, thus obtaining the corresponding label map.

[0011] In one embodiment, the method further includes:

[0012] Obtain global TEC data;

[0013] Batch processing of global TEC data to create a global ionospheric TEC map;

[0014] Using semantic segmentation, generate a label map corresponding to the global ionospheric TEC map;

[0015] A training dataset was constructed based on the global ionospheric TEC map and its corresponding label map;

[0016] The TEC prediction model was trained using the training dataset;

[0017] The training of the TEC forecast model is complete when the discriminator of the TEC forecast model cannot distinguish between the real and fake maps in the model output map.

[0018] In one embodiment, the step of constructing a training dataset based on a global ionospheric TEC map and a corresponding label map includes:

[0019] Multiple sets of input data were constructed using a global ionospheric TEC map and corresponding label maps; each set of input data included an M-day historical time series IGS-TEC map, an M-day future time series IGS-TEC map, and corresponding label maps.

[0020] In one embodiment, the first mask is a red mask and the second mask is a blue mask.

[0021] In one embodiment, M = 3.

[0022] In one embodiment, the step of obtaining an M-day historical time series IGS-TEC map includes:

[0023] Obtain the forecastable TEC data for M days of historical time;

[0024] Batch processing of the forecast TEC data is used to create a historical time series IGS-TEC map.

[0025] On the other hand, a deep learning-based device for predicting the total electron content of the ionosphere is also provided, comprising:

[0026] The input acquisition module is used to obtain the M-day historical time series IGS-TEC map; M is a positive integer not less than 3.

[0027] The label conversion module is used to generate corresponding label maps for historical time-series IGS-TEC maps using semantic segmentation; the label maps are used to indicate ionospheric peak structure regions and non-ionospheric peak structure regions.

[0028] The forecast output module is used to input the historical time series IGS-TEC map and the corresponding label map into the trained TEC forecast model to obtain the TEC forecast map for the next M days output by the TEC forecast model; the TEC forecast model is a model based on the deep learning method pix2pixhd, trained on global TEC data.

[0029] On the other hand, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-mentioned deep learning-based method for predicting the total electron content of the ionosphere.

[0030] Furthermore, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the aforementioned deep learning-based method for predicting the total electron content of the ionosphere.

[0031] One of the above technical solutions has the following advantages and beneficial effects:

[0032] The aforementioned deep learning-based method, apparatus, and equipment for forecasting total ionospheric electron content first processes the TEC data to be forecasted, specifically by acquiring M-day historical time-series IGS-TEC maps. Then, semantic segmentation is used to generate corresponding label maps for these historical time-series IGS-TEC maps, which serve as input to a trained TEC forecasting model. Subsequently, a TEC forecasting model based on the pix2pixhd deep learning method, trained using global TEC data, is invoked. The historical time-series IGS-TEC maps and their corresponding label maps are input into the TEC forecasting model, outputting a TEC forecast map for the next M days.

[0033] Compared to traditional methods, the above approach utilizes a pre-trained pix2pixhd deep learning-based TEC forecasting model trained on global TEC data. It uses historical time series of global IGS-TEC maps and their corresponding label maps as input data for training the pix2pixhd model, dividing the global region into TEC peak structure regions and non-TEC peak structure regions. This significantly improves the accuracy of TEC forecasts in low-latitude regions and enables high-accuracy TEC forecasts even during geomagnetic storms. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application or the conventional technology, the drawings used in the description of the embodiments or the conventional technology 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 based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart illustrating a deep learning-based method for predicting total ionospheric electron content in one embodiment.

[0036] Figure 2 This is a schematic diagram of the training process for a TEC forecasting model in one embodiment;

[0037] Figure 3 This is a schematic diagram illustrating the process of converting an original map into a labeled map in one embodiment.

[0038] Figure 4 This is a schematic diagram of the training process framework for the TEC forecasting model in one embodiment;

[0039] Figure 5 Here is an example of an input historical time series IGS-TEC map; where (a) is the IGS-TEC map for the previous two days, (b) is the IGS-TEC map for the previous day, and (c) is the IGS-TEC map for the current day.

[0040] Figure 6 Here is a target (actual) IGS-TEC map in one embodiment; where (a) is the IGS-TEC map for the next day, (b) is the IGS-TEC map for the next two days, and (c) is the IGS-TEC map for the next three days.

[0041] Figure 7 Here is an output map generated by the TEC forecasting model in one embodiment; where (a) is the IGS-TEC map for the next day, (b) is the IGS-TEC map for the next two days, and (c) is the IGS-TEC map for the next three days.

[0042] Figure 8 Here is an IRI-TEC map in one embodiment; where (a) is the IRI-TEC map for the following day, (b) is the IRI-TEC map for the following two days, and (c) is the IRI-TEC map for the following three days.

[0043] Figure 9 This is a schematic diagram of the module structure of a deep learning-based ionospheric total electron content prediction device in one embodiment. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0045] 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 belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0046] It should be noted that, in this document, the reference to "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The presentation of this phrase in various locations throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments.

[0047] Those skilled in the art will understand that the embodiments described herein can be combined with other embodiments. The term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items, and all possible combinations thereof.

[0048] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0049] Please see Figure 1 In one embodiment, this application provides a deep learning-based method for predicting the total electron content of the ionosphere, including the following processing steps S12 to S16:

[0050] S12, obtain the M-day historical time series IGS-TEC map; M is a positive integer not less than 3;

[0051] S14, use semantic segmentation to generate the corresponding label map of the historical time series IGS-TEC map; the label map is used to indicate the ionospheric peak structure region and the non-ionospheric peak structure region.

[0052] S16. Input the historical time series IGS-TEC map and the corresponding label map into the trained TEC forecast model to obtain the TEC forecast map for the next M days output by the TEC forecast model; the TEC forecast model is a model based on the deep learning method pix2pixhd, trained on global TEC data.

[0053] It can be understood that the M-day historical time series IGS-TEC map refers to the IGS-TEC map for each day of the past (M-1) days, starting from the current day (inclusive). This M-day historical time series IGS-TEC map, composed of each date (time) and its corresponding IGS-TEC map, forms the aforementioned historical time series IGS-TEC map used to predict the future M-day TEC forecast map. The IGS-TEC map is an ionospheric TEC map obtained based on the Global Ionospheric Map (GIM) provided by the IGS Working Group (an important application data of the ionosphere).

[0054] After obtaining the M-day historical time series IGS-TEC map that needs to be predicted, the existing semantic segmentation principle in this field can be used to generate the corresponding label map of the historical time series IGS-TEC map. That is, to generate the label map corresponding to each IGS-TEC map, so as to determine the ionospheric peak structure region and the remaining non-ionospheric peak structure region corresponding to each IGS-TEC map, thereby realizing the detailed division of the global ionospheric region and ensuring the model's TEC forecasting effect in low-latitude regions.

[0055] The label map can use different colors to mark the ionospheric peak structure region and the remaining non-ionospheric peak structure region, or it can use masks of different colors or textures to mark the ionospheric peak structure region and the remaining non-ionospheric peak structure region, or it can use numerical marking or other division methods to mark it, as long as it can effectively and accurately divide the ionospheric peak structure region and the remaining non-ionospheric peak structure region of the corresponding IGS-TEC map.

[0056] The TEC forecasting model based on the deep learning method pix2pixhd can be pre-trained on global TEC data and stored in the system for quick and direct retrieval during forecasting. Specifically, to forecast TEC data for a certain number of days in the future, historical time-series IGS-TEC maps up to the present can be obtained. Then, corresponding label maps for these historical time-series IGS-TEC maps are generated. Finally, these historical time-series IGS-TEC maps and their label maps are input into the trained TEC forecasting model, which will output a TEC forecast map for the corresponding number of days in the future. M can be 3 days, 4 days, 5 days, or other numbers of days, selected based on actual forecasting needs and forecast accuracy metrics.

[0057] The aforementioned deep learning-based method for predicting total ionospheric electron content first processes the TEC data to be predicted, specifically by acquiring M-day historical time-series IGS-TEC maps. Then, semantic segmentation is used to generate corresponding label maps for these historical time-series IGS-TEC maps, which serve as input to a trained TEC prediction model. Finally, a TEC prediction model based on the pix2pixhd deep learning method, trained using global TEC data, is invoked. By inputting the historical time-series IGS-TEC maps and their corresponding label maps into the TEC prediction model, the model outputs a TEC prediction map for the next M days.

[0058] Compared to traditional methods, the above approach utilizes a pre-trained pix2pixhd deep learning-based TEC forecasting model trained on global TEC data. It uses historical time series of global IGS-TEC maps and their corresponding label maps as input data for training the pix2pixhd model, dividing the global region into TEC peak structure regions and non-TEC peak structure regions. This significantly improves the accuracy of TEC forecasts in low-latitude regions and enables high-accuracy TEC forecasts even during geomagnetic storms.

[0059] In one embodiment, step S12 may specifically include the following processing steps:

[0060] Obtain the forecastable TEC data for M days of historical time;

[0061] Batch processing of the forecast TEC data is used to create a historical time series IGS-TEC map.

[0062] Specifically, the TEC data to be forecasted is preprocessed, that is, the TEC data to be forecasted over M days of historical time is acquired and batch-processed using existing batch processing scripts to generate the corresponding M-day historical time series IGS-TEC map. The M-day historical TEC data to be forecasted can be obtained directly from the IGS working group or other data providers.

[0063] Through the above processing steps, the acquired raw TEC data can be quickly converted into IGS-TEC map data that the model can directly process, thereby improving the forecast output efficiency.

[0064] In one embodiment, step S14 may specifically include the following processing steps:

[0065] The first mask is used to cover the ionospheric peak structure in the historical time series IGS-TEC map, and the second mask is used to cover the remaining areas in the historical time series IGS-TEC map other than the ionospheric peak structure, thus obtaining the corresponding label map.

[0066] It is understood that in this embodiment, a mask is used to divide the ionospheric peak structure region and the remaining non-ionospheric peak structure region. The first mask and the second mask are two different masks, such as, but not limited to, image patches with different colors, image patches with different textures, or other forms of different image patches. The two different masks are used to cover the ionospheric peak structure region and the remaining region in the historical time series IGS-TEC map, respectively, to distinguish them into two types of image features. The resulting map with masks is called the aforementioned label map.

[0067] Through the above processing steps, the ionospheric peak structure region and the remaining region in the IGS-TEC map can be efficiently and accurately delineated.

[0068] In one embodiment, such as Figure 2 As shown, the above method may also include the following processing steps:

[0069] S01, obtain global TEC data;

[0070] S02, batch process global TEC data to create a global ionospheric TEC map;

[0071] S03, using semantic segmentation to generate a label map corresponding to the global ionospheric TEC map;

[0072] S04. Construct a training dataset based on the global ionospheric TEC map and the corresponding label map;

[0073] S05, Train the TEC forecast model using the training dataset;

[0074] S06. When the discriminator of the TEC forecast model cannot distinguish between the real and fake maps in the model output map, the training of the TEC forecast model is completed.

[0075] Understandably, training the TEC forecasting model can begin by directly acquiring the required global TEC data from the Global Ionospheric Map (GIM) provided by the IGS working group. This global TEC data is then batch-processed and plotted to obtain the corresponding global ionospheric TEC maps. Next, semantic segmentation is used to convert these global ionospheric TEC maps into corresponding label maps. Finally, based on these global ionospheric TEC maps and their corresponding label maps, the training dataset required for model training is constructed.

[0076] In some implementations, step S04 may specifically include the following processing steps:

[0077] Multiple sets of input data were constructed using a global ionospheric TEC map and corresponding label maps; each set of input data included an M-day historical time series IGS-TEC map, an M-day future time series IGS-TEC map, and corresponding label maps.

[0078] Specifically, based on the obtained global ionospheric TEC map, corresponding label maps are first created. For example, two different masks are used to divide the ionospheric peak structure and remaining areas in the global ionospheric TEC map into two categories: the ionospheric peak structure area is masked with the first mask, and the remaining area is masked with the second mask. This generates corresponding label maps for each global ionospheric TEC map, such as... Figure 3 As shown.

[0079] Next, a training dataset is created. An M-day historical time-series IGS-TEC map, an M-day future time-series IGS-TEC map, and corresponding label maps constitute one set of input data, which is also one set of data in the training dataset. Similarly, multiple sets of input data can be constructed using the obtained global ionospheric TEC map and corresponding label maps; the dataset composed of all these input sets is the training dataset. The size of the training dataset can be determined according to the actual needs of model training, such as determining the required number of input data sets based on the number of empirical or statistical epochs completed during model training.

[0080] It should be noted that the future time in the IGS-TEC map of the M-day future time series in this embodiment is relative to the "now (or current day)" selected from the selected M-day historical time. It is a historical time that is later than the M-day historical time selected from the existing global ionospheric TEC map in order to complete the model training, and does not strictly refer to the future time in the real world relative to the actual current day.

[0081] After obtaining the required training dataset, the input data from each group of this training dataset is used as the input to train the TEC prediction model based on the deep learning method pix2pixhd. The TEC prediction model is based on the deep learning method pix2pixhd. pix2pixhd has two generators and three discriminators. The loss function of pix2pixhd consists of two loss functions: CGAN loss and Feature Matching (FM) loss. The CGAN loss function is derived from the following formula:

[0082] L CGAN (G,D)=E (s,x) [logD(s,x)]+E s[log(1-D(s,G(s)))] (1)

[0083] Where s, x, and G(s) are the input image (map), the actual image (map), and the model output image (map), respectively. G represents the generator, and D represents the discriminator. G minimizes L CGAN D then maximizes it. E (s,x) [·] and E s [·] represents the probability of real data and the probability of generated data, respectively.

[0084] The feature matching loss function is as follows:

[0085]

[0086] Where T and N i These represent the total number of layers and the number of elements in the i-th layer, respectively. This represents the feature of the i-th layer extracted by the k-th discriminator.

[0087] The total loss function is shown below:

[0088]

[0089] Where λ is the control L CGAN and L FM Importance-weighted hyperparameters. CGAN focuses on the details of the content, while FM ensures the consistency of the content.

[0090] like Figure 4 The diagram shows the framework of the model training process. At the start of training, the generator parameters are initialized with random values. As the training time increases, the output map of the TEC prediction model gradually becomes more consistent with the actual map. When the discriminator can no longer distinguish between the output map and the actual map, the TEC prediction model training is complete.

[0091] Through the above processing steps

[0092] In one embodiment, the first mask is a red mask and the second mask is a blue mask. Preferably, in this embodiment, the ionospheric peak structure region in the global ionospheric TEC map can be masked with a red mask, and the remaining areas can be masked with a blue mask, thus generating a corresponding labeled map for the global ionospheric TEC map. Label creation is simple and efficient.

[0093] In one embodiment, M = 3. In this embodiment, optionally, a three-day historical time-series IGS-TEC map and the corresponding label map are used as input to the trained TEC forecast model. The TEC forecast model then outputs the corresponding TEC forecast map for the next three days, as shown in the forecast results. Figures 5 to 8As shown, comparisons are also displayed between the IGS-TEC map in Universal Time (UT), the output map of the TEC forecast model of this application, and the International Reference Ionosphere (IRI)-TEC map. Here, d represents the time unit as days; Figure 5 The input is a historical time series IGS-TEC map ((a) is the IGS-TEC map of the previous two days, (b) is the IGS-TEC map of the previous day, and (c) is the IGS-TEC map of the current day); Figure 6 The target IGS-TEC map ((a) is the IGS-TEC map for the next day, (b) is the IGS-TEC map for the next two days, and (c) is the IGS-TEC map for the next three days); Figure 7 Output maps generated for the TEC forecast model of this application ((a) is the IGS-TEC map for the next day, (b) is the IGS-TEC map for the next two days, and (c) is the IGS-TEC map for the next three days); Figure 8 The IRI-TEC map is shown in (a) for the next day, (b) for the next two days, and (c) for the next three days.

[0094] Compared with existing technologies, the method described in this application improves the accuracy of model forecasts in low-latitude regions and can effectively generate ionospheric peak structures, demonstrating superior prediction performance in low-latitude regions compared to mid- and high-latitude regions. Satisfactory results were achieved during geomagnetic storm periods (with root mean square errors of TEC forecasts for the next three days of 0.9662, 0.9633, and 0.9561, respectively) and during years of high solar activity (with root mean square errors of TEC forecasts for the next three days of 0.9660, 0.9615, and 0.9552, respectively), which is unattainable by existing technologies.

[0095] It should be understood that, although Figure 1 , Figure 2 and Figure 4 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order requirement for the execution of these steps; they can be executed in other orders. Figure 1 , Figure 2 and Figure 4At least some of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0096] Please see Figure 9 In one embodiment, a deep learning-based ionospheric total electron content forecasting device 100 is provided, including an input acquisition module 11, a label conversion module 13, and a forecast output module 15. The input acquisition module 11 acquires an M-day historical time-series IGS-TEC map; M is a positive integer not less than 3. The label conversion module 13 generates corresponding label maps for the historical time-series IGS-TEC map using semantic segmentation; the label maps indicate ionospheric peak structure regions and non-ionospheric peak structure regions. The forecast output module 15 inputs the historical time-series IGS-TEC map and the corresponding label maps into a trained TEC forecasting model to obtain a future M-day TEC forecast map output by the TEC forecasting model; the TEC forecasting model is a pix2pixhd deep learning model trained on global TEC data.

[0097] The aforementioned deep learning-based ionospheric total electron content forecasting device 100, through the collaboration of its various modules, first processes the TEC data to be forecasted, specifically acquiring the M-day historical time-series IGS-TEC map to be forecasted. Then, semantic segmentation is used to generate corresponding label maps for these historical time-series IGS-TEC maps, which serve as input to the trained TEC forecasting model. Subsequently, the TEC forecasting model, trained based on the pix2pixhd deep learning method using global TEC data, is invoked. The historical time-series IGS-TEC map and its corresponding label map are input into the TEC forecasting model, outputting the TEC forecast map for the next M days.

[0098] Compared to traditional methods, the above approach utilizes a pre-trained pix2pixhd deep learning-based TEC forecasting model trained on global TEC data. It uses historical time series of global IGS-TEC maps and their corresponding label maps as input data for training the pix2pixhd model, dividing the global region into TEC peak structure regions and non-TEC peak structure regions. This significantly improves the accuracy of TEC forecasts in low-latitude regions and enables high-accuracy TEC forecasts even during geomagnetic storms.

[0099] In one embodiment, the label conversion module 13 is specifically used to cover the ionospheric peak structure in the historical time series IGS-TEC map with a first mask and to cover the remaining areas outside the ionospheric peak structure in the historical time series IGS-TEC map with a second mask, so as to obtain the corresponding label map.

[0100] In one embodiment, the deep learning-based ionospheric total electron content forecasting device 100 further includes a data module, a batch processing module, a label processing module, a dataset module, a training module, and a training completion module. The data module acquires global TEC data. The batch processing module processes the global TEC data in batches to create a global ionospheric TEC map. The label processing module generates corresponding label maps for the global ionospheric TEC map using semantic segmentation. The dataset module constructs a training dataset based on the global ionospheric TEC map and the corresponding label maps. The training module trains the TEC forecasting model using the training dataset. The training completion module completes the training of the TEC forecasting model when the model's discriminator cannot distinguish between the model's output map and the real map.

[0101] In one embodiment, the aforementioned dataset module is specifically used to construct multiple sets of input data using a global ionospheric TEC map and corresponding label maps; each set of input data includes an M-day historical time series IGS-TEC map, an M-day future time series IGS-TEC map, and corresponding label maps.

[0102] In one embodiment, the first mask is a red mask and the second mask is a blue mask.

[0103] In one embodiment, M = 3.

[0104] In one embodiment, the input acquisition module 11 is specifically used to acquire TEC data to be predicted over M historical time periods; to batch process and plot the TEC data to be predicted to obtain a historical time series IGS-TEC map.

[0105] For specific limitations regarding the deep learning-based ionospheric total electron content forecasting device 100, please refer to the corresponding limitations of the deep learning-based ionospheric total electron content forecasting method described above, which will not be repeated here. Each module in the aforementioned deep learning-based ionospheric total electron content forecasting device 100 can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of a device with specific data processing functions, or stored in software in the memory of the aforementioned device, so that the processor can call and execute the operations corresponding to each module. The aforementioned device can be, but is not limited to, various types of TEC forecasting devices already existing in the art.

[0106] In one embodiment, a computer device is also provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the following processing steps: acquiring an M-day historical time series IGS-TEC map; M is a positive integer not less than 3; generating a corresponding label map for the historical time series IGS-TEC map using semantic segmentation; the label map is used to indicate ionospheric peak structure regions and non-ionospheric peak structure regions; inputting the historical time series IGS-TEC map and the corresponding label map into a trained TEC forecast model to obtain a future M-day TEC forecast map output by the TEC forecast model; the TEC forecast model is a model based on the deep learning method pix2pixhd, trained based on global TEC data.

[0107] It is understood that, in addition to the memory and processor mentioned above, the computer equipment described above also includes other hardware and software components not listed in this specification. The specific components can be determined according to the model of the computer equipment in different application scenarios, and will not be listed and described in detail in this specification.

[0108] In one embodiment, when the processor executes the computer program, it can also implement the steps or sub-steps added in the various embodiments of the deep learning-based ionospheric total electron content prediction method.

[0109] In one embodiment, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, it performs the following processing steps: acquiring an M-day historical time series IGS-TEC map; M is a positive integer not less than 3; generating a label map corresponding to the historical time series IGS-TEC map using semantic segmentation; the label map is used to indicate ionospheric peak structure regions and non-ionospheric peak structure regions; inputting the historical time series IGS-TEC map and the corresponding label map into a trained TEC forecast model to obtain a future M-day TEC forecast map output by the TEC forecast model; the TEC forecast model is a model based on the deep learning method pix2pixhd, trained based on global TEC data.

[0110] In one embodiment, when the computer program is executed by a processor, it can also implement the steps or sub-steps added to the various embodiments of the deep learning-based ionospheric total electron content prediction method described above.

[0111] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), memory bus DRAM (RDRAM), and interface DRAM (DRDRAM), etc.

[0112] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0113] The above embodiments merely illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, all of which fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for predicting the total electron content of the ionosphere based on deep learning, characterized in that, Including the following steps: Obtain the historical time series IGS-TEC map for M days; M is a positive integer not less than 3; Semantic segmentation is used to generate a corresponding label map for the historical time series IGS-TEC map; the label map is used to indicate the ionospheric peak structure region and the non-ionospheric peak structure region. The historical time series IGS-TEC map and the corresponding label map are input into the trained TEC forecast model to obtain the TEC forecast map for the next M days output by the TEC forecast model; the TEC forecast model is a model based on the deep learning method pix2pixhd, trained on global TEC data; The step of generating the corresponding label map for the historical time series IGS-TEC map using semantic segmentation includes: The ionospheric peak structure in the historical time series IGS-TEC map is masked using a first mask, and the remaining areas outside the ionospheric peak structure in the historical time series IGS-TEC map are masked using a second mask to obtain the corresponding label map.

2. The method for predicting total ionospheric electron content based on deep learning according to claim 1, characterized in that, The method further includes: Obtain global TEC data; Batch processing and plotting of the global TEC data yields a global ionospheric TEC map; Using semantic segmentation, a corresponding labeled map is generated for the global ionospheric TEC map; A training dataset is constructed based on the global ionospheric TEC map and the corresponding label map; The TEC prediction model is trained using the training dataset. The training of the TEC forecast model is completed when the discriminator of the TEC forecast model cannot distinguish between the real and fake maps in the model output map.

3. The method for predicting the total electron content of the ionosphere based on deep learning according to claim 2, characterized in that, The steps for constructing a training dataset based on the global ionospheric TEC map and the corresponding label map include: Using the global ionospheric TEC map and the corresponding label map, multiple sets of input data are constructed; each set of input data includes an M-day historical time series IGS-TEC map, an M-day future time series IGS-TEC map, and the corresponding label map.

4. The method for predicting total ionospheric electron content based on deep learning according to claim 1, characterized in that, The first mask is a red mask, and the second mask is a blue mask.

5. The method for predicting total ionospheric electron content based on deep learning according to claim 1, characterized in that, M=3。 6. The method for predicting total ionospheric electron content based on deep learning according to claim 1, characterized in that, The steps to obtain an M-day historical time series IGS-TEC map include: Obtain the forecastable TEC data for M days of historical time; The TEC data to be predicted is batch-processed and plotted to obtain the historical time series IGS-TEC map.

7. A deep learning-based device for predicting the total electron content of the ionosphere, characterized in that, include: The input acquisition module is used to acquire an M-day historical time series IGS-TEC map; M is a positive integer not less than 3; The label conversion module is used to generate a corresponding label map for the historical time series IGS-TEC map using semantic segmentation; wherein, the ionospheric peak structure in the historical time series IGS-TEC map is masked with a first mask, and the remaining areas in the historical time series IGS-TEC map other than the ionospheric peak structure are masked with a second mask to obtain the corresponding label map, which is used to indicate the ionospheric peak structure area and the non-ionospheric peak structure area; The forecast output module is used to input the historical time series IGS-TEC map and the corresponding label map into the trained TEC forecast model to obtain the TEC forecast map for the next M days output by the TEC forecast model; the TEC forecast model is a model based on the deep learning method pix2pixhd, trained on global TEC data.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the deep learning-based method for predicting total ionospheric electron content as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the deep learning-based method for predicting the total electron content of the ionosphere as described in any one of claims 1 to 6.

Images