A Smart Ionospheric Forecasting and Monitoring Method and System Based on an Integrated Model

Abstract

This invention provides a method and system for intelligent ionospheric forecasting and monitoring based on an ensemble model, comprising: constructing an ionospheric feature vector based on historical ionospheric data; preprocessing the ionospheric feature vector to obtain a preprocessed ionospheric feature vector, and normalizing the preprocessed ionospheric feature vector to obtain a training dataset; constructing an ensemble model by sequentially stacking a Transformer model, an LSTM model, and a tree model; training the ensemble model using the training dataset; determining the optimal parameters of the ensemble model using a grid search method to obtain an ionospheric prediction model; and inputting real-time ionospheric data into the ionospheric prediction model to obtain the ionospheric prediction result. This invention, by applying ensemble model technology to intelligent ionospheric prediction, can cover multiple application scenarios, achieve multi-dimensional parameter enhancement to improve prediction accuracy, and enable real-time monitoring of ionospheric changes, meeting the need for real-time correction of ionospheric delay in precision navigation and positioning.

Description

Technical Field

[0001] This invention relates to the field of space physics technology, and in particular to an intelligent ionospheric forecasting and monitoring method and system based on an integrated model. Background Technology

[0002] The ionosphere, a crucial component of the solar-terrestrial space environment, plays a vital role in modern space weather research, with its variations significantly impacting positioning and navigation, wireless communication, and meteorological monitoring. However, influenced by factors such as solar activity, atmospheric molecular concentration, upper-atmosphere wind speed, and geomagnetic activity, the ionosphere exhibits various regular and irregular variations in both time and space, posing a significant challenge to ionospheric forecasting. Currently, solar activity has entered a new active cycle, with a sharp increase in global ionospheric activity, making the ionosphere even more difficult to predict. Therefore, highly reliable ionospheric forecasting is essential for near-Earth space environment disaster monitoring and early warning, high-precision positioning and navigation, and wireless communication. Minimizing the ionosphere's impact on navigation and positioning is also key to enabling satellite systems to empower autonomous driving, aviation, and other fields.

[0003] Currently, there are many studies on ionospheric forecasting. Traditional research primarily relies on mathematical models, such as least squares collocation models, autoregressive moving average models, Holt-Winter models, and empirical orthogonal functions. In recent years, machine learning and deep learning models have been widely applied to ionospheric prediction, mainly using artificial neural networks (ANNs), recurrent neural networks (RNNs), gated recurrent units (GRUs), and long short-term memory networks (LSTMs). However, all of these models have some limitations. Because the ionosphere, as a nonlinear and non-stationary system, has irregular characteristics and complex mechanisms of local variation, using a single model cannot accurately describe its spatiotemporal variations and complex large-scale anomalies, such as equatorial ionization anomalies, mid-latitude summer night anomalies, and winter anomalies. Especially in complex space environments, such as during geomagnetic storms, improving the accuracy of ionospheric forecasts is particularly difficult.

[0004] Therefore, in view of the shortcomings of the above methods, it is urgent to propose new methods for ionospheric prediction. Summary of the Invention

[0005] This invention provides an intelligent ionospheric forecasting and monitoring method and system based on an integrated model, which solves the problems of existing technologies that use a single machine learning model for ionospheric forecasting, which cannot accurately reflect the characteristics of ionospheric changes and have shortcomings such as short forecast time and poor forecast accuracy.

[0006] In a first aspect, the present invention provides a method for intelligent ionospheric forecasting and monitoring based on an integrated model, comprising:

[0007] Collect historical data of the ionosphere and construct an ionosphere feature vector based on the historical data of the ionosphere;

[0008] The ionospheric feature vector is preprocessed to obtain a preprocessed ionospheric feature vector, and the preprocessed ionospheric feature vector is normalized to obtain a training dataset.

[0009] An ensemble model is constructed by stacking Transformer, LSTM and tree models in sequence. The ensemble model is trained using the training dataset. The optimal parameters of the ensemble model are determined by grid search method to obtain the ionospheric prediction model.

[0010] Real-time ionospheric data is input into the ionospheric prediction model to obtain ionospheric prediction results.

[0011] According to the present invention, an intelligent ionospheric forecasting and monitoring method based on an ensemble model is provided, which involves collecting historical ionospheric data and constructing an ionospheric feature vector based on the historical ionospheric data, including:

[0012] Ionospheric data, solar activity index data, geomagnetic activity index data, and interplanetary magnetic field index data are acquired to construct the ionospheric feature vector.

[0013] According to the intelligent ionospheric forecasting and monitoring method based on an integrated model provided by the present invention, the ionospheric feature vector is preprocessed to obtain a preprocessed ionospheric feature vector, including:

[0014] Outliers in the ionospheric feature vector are removed, and the feature data is fitted using a classical spline function. The fitted feature data is then used to supplement the missing values ​​in the ionospheric feature vector to obtain the initial preprocessed ionospheric feature vector.

[0015] The initial preprocessed ionospheric feature vector is resampled using polynomial interpolation to align the temporal information of different types of data, and the preprocessed ionospheric feature vector is output.

[0016] According to the present invention, an intelligent ionospheric forecasting and monitoring method based on an ensemble model is provided, which normalizes the preprocessed ionospheric feature vectors to obtain a training dataset, including:

[0017] The preprocessed ionospheric feature vectors are divided into a training set, a validation set, and a test set according to a preset ratio.

[0018] Based on the maximum and minimum eigenvalues ​​of the training set, a normalized training set is obtained;

[0019] Based on the training set, determine the maximum and minimum feature values ​​of the validation set, as well as the maximum and minimum feature values ​​of the test set; obtain a normalized validation set based on the maximum and minimum feature values ​​of the validation set; and obtain a normalized test set based on the maximum and minimum feature values ​​of the test set.

[0020] The training dataset is constructed from the normalized training set, the normalized validation set, and the normalized test set.

[0021] According to the present invention, an intelligent ionospheric forecasting and monitoring method based on an ensemble model is provided, which constructs an ensemble model by sequentially stacking a Transformer model, an LSTM model, and a tree model, including:

[0022] The primary learner of the tree model is constructed using the random forest algorithm, the secondary learner of the tree model is constructed using the gradient boosting decision tree algorithm, and the ensemble model is constructed using the long short-term memory network LSTM.

[0023] The Transformer model includes an encoder and a decoder. The decoder is connected to the encoder through an attention mechanism. The encoder receives historical time series, and the decoder extracts historical values ​​from the historical time series using a masked self-attention mechanism.

[0024] The LSTM model includes an input layer, a hidden layer, and an output layer. The input layer includes historical ionospheric feature vectors. The hidden layer includes two LSTM network layers, two Dropout layers, and one fully connected layer. The output layer includes predicted ionospheric data.

[0025] The tree model uses the RF algorithm to construct the initial learner and the XGBoost algorithm to construct the secondary learner.

[0026] According to the present invention, an intelligent ionospheric forecasting and monitoring method based on an ensemble model is provided, which trains the ensemble model using the training dataset, determines the optimal parameters of the ensemble model using a grid search method, and obtains an ionospheric prediction model, comprising:

[0027] Determining the parameters of the ensemble model includes the learning rate, maximum number of decision trees, random state, maximum tree depth, minimum sample weight in leaf nodes, training batch size, number of training epochs, number of samples per epoch, number of neurons in the LSTM network, training dropout rate, activation function, and loss function.

[0028] Determine the fixed parameter combinations of the Transformer model, the LSTM model, and the tree model in the ensemble model parameters, as well as the initial values ​​of the parameters to be adjusted;

[0029] Using the grid search method, the optimal value of the parameter to be adjusted is determined based on the combination of the fixed parameters;

[0030] Repeat the steps of using the grid search method to determine the optimal value of the parameter to be adjusted based on the fixed parameter combination until the Transformer model, the LSTM model and the tree model reach local optimal parameters respectively;

[0031] The Transformer model, the LSTM model, and the tree model that have reached the local optimal parameters are weighted and calculated with preset weights to obtain the ionospheric prediction model.

[0032] According to the present invention, an intelligent ionospheric forecasting and monitoring method based on an integrated model is provided, wherein real-time ionospheric data is input into the ionospheric prediction model to obtain ionospheric prediction results, including:

[0033] Obtain the preset historical time series ionospheric feature vector of the real-time ionospheric data;

[0034] The preset historical time series ionospheric feature vector is input into the ionospheric prediction model, and the predicted data of the total electron content of the ionosphere for a preset future time period is output.

[0035] Secondly, the present invention also provides an intelligent ionospheric forecasting and monitoring system based on an integrated model, comprising:

[0036] The acquisition module is used to acquire historical ionospheric data and construct an ionospheric feature vector based on the historical ionospheric data.

[0037] The preprocessing module is used to preprocess the ionospheric feature vector to obtain a preprocessed ionospheric feature vector, and to normalize the preprocessed ionospheric feature vector to obtain a training dataset.

[0038] The training module is used to construct an ensemble model by stacking Transformer model, LSTM model and tree model in sequence, train the ensemble model using the training dataset, and determine the optimal parameters of the ensemble model using the grid search method to obtain the ionospheric prediction model.

[0039] The prediction module is used to input real-time ionospheric data into the ionospheric prediction model to obtain ionospheric prediction results.

[0040] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the ionospheric intelligent forecasting and monitoring method based on the integrated model as described above.

[0041] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the ionospheric intelligent forecasting and monitoring method based on an integrated model as described above.

[0042] The intelligent ionospheric forecasting and monitoring method and system based on integrated models provided by this invention can cover multiple application scenarios by applying integrated model technology to intelligent ionospheric prediction. It realizes multi-dimensional parameter improvement of prediction accuracy and real-time monitoring of ionospheric changes, meeting the needs of real-time correction of ionospheric delay in precision navigation and positioning. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0044] Figure 1 This is one of the flowcharts of the intelligent ionospheric forecasting and monitoring method based on an integrated model provided by the present invention;

[0045] Figure 2 This is the second flowchart of the intelligent ionospheric forecasting and monitoring method based on an integrated model provided by the present invention;

[0046] Figure 3 This is a Transformer structure diagram of the Transformer ensemble model provided by the present invention;

[0047] Figure 4 This is an LSTM structure diagram of the Transformer ensemble model provided by the present invention;

[0048] Figure 5 This is a diagram of the integrated model structure provided by the present invention;

[0049] Figure 6 This is a schematic diagram of the structure of the intelligent ionospheric forecasting and monitoring system based on an integrated model provided by the present invention;

[0050] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0052] To address the limitations of existing machine learning models used in ionospheric prediction, this invention proposes an intelligent ionospheric forecasting and monitoring method based on an ensemble model. Using real-time data on the total electron content of the ionosphere and related indices such as solar activity and geomagnetic activity, the method employs an ensemble model to intelligently forecast the total electron content of the ionosphere over a future period. Furthermore, this method can forecast both one-dimensional and multi-dimensional data, offering advantages such as high forecasting speed and high accuracy.

[0053] Figure 1 This is one of the flowcharts illustrating the intelligent ionospheric forecasting and monitoring method based on an integrated model provided in this invention. Figure 1 As shown, it includes:

[0054] Step 100: Collect historical ionospheric data and construct an ionospheric feature vector based on the historical ionospheric data;

[0055] Step 200: Preprocess the ionospheric feature vector to obtain a preprocessed ionospheric feature vector, and normalize the preprocessed ionospheric feature vector to obtain a training dataset;

[0056] Step 300: Construct an ensemble model by stacking Transformer model, LSTM model and tree model in sequence, train the ensemble model using the training dataset, determine the optimal parameters of the ensemble model using grid search method, and obtain the ionospheric prediction model;

[0057] Step 400: Input the real-time ionospheric data into the ionospheric prediction model to obtain the ionospheric prediction results.

[0058] It should be noted that the ionospheric prediction method based on an ensemble model proposed in this invention is based on the following considerations: 1) the excellent performance of different models is due to independent training; 2) the excellent performance of different models is because each model views the data from a different perspective and makes predictions, that is, each model obtains some characteristics of the data rather than all characteristics; 3) the data features obtained by different models are different. Taking the above factors into account, this invention utilizes different algorithms to observe nonlinear and non-stationary ionospheric data from multiple dimensions, using the Transformer model, which performs particularly well in various fields, as well as the Long Short-Term Memory (LSTM) network model optimized for time series, and Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) in machine learning, to construct an ensemble model for intelligent ionospheric prediction with the Transformer model as the core.

[0059] Specifically, such as Figure 2 The flowchart shown first constructs feature vectors using ionospheric data, solar activity, and geomagnetic index data. Preprocessing of this data includes outlier removal, missing value filling, and data resampling. Then, the dataset is divided and normalized to form a training dataset. Next, an ensemble model is constructed by stacking Transformer, LSTM, and tree models sequentially. The ensemble model is trained, and parameters are adjusted for each part. This process includes an iterative process; if the model does not provide relevant feedback after several consecutive training iterations, the iteration stops and the model is output, resulting in a trained ionospheric prediction model. Finally, real-time ionospheric feature vectors are input into the ionospheric prediction model to obtain a high-precision ionospheric prediction result.

[0060] This invention applies ensemble modeling technology to intelligent ionospheric forecasting. The ensemble model combines the advantages of multiple models, providing high-precision and highly reliable forecast results. After a single training iteration, it can be directly deployed to various scenarios, requiring minimal memory for rapid subsequent computation. By considering multiple parameters influencing ionospheric activity, such as the solar activity index and geomagnetic activity index, the selection of characteristic variables is more representative, improving forecast accuracy while enabling intelligent ionospheric forecasting based on these parameters. Furthermore, the method proposed in this invention can be extended to other fields to meet the needs of different scenarios, such as temperature forecasting and tropospheric forecasting, enabling both one-dimensional and multi-dimensional forecasting. High-precision ionospheric forecasting allows for real-time monitoring of ionospheric changes, meeting the requirement for real-time correction of ionospheric delays in precise navigation and positioning.

[0061] Based on the above embodiments, step 100 includes:

[0062] Ionospheric data, solar activity index data, geomagnetic activity index data, and interplanetary magnetic field index data are acquired to construct the ionospheric feature vector.

[0063] Specifically, by acquiring data in real time, data such as ionospheric data, solar activity index, and geomagnetic activity index corresponding to time information are obtained. Ionospheric data sources support two methods: (1) downloading published ionospheric data from organizations such as the International GPS Service (IGS) via the Internet; (2) extracting ionospheric data by self-calculation from satellite observation data of the Global Navigation Satellite System (GNSS).

[0064] Ionospheric data includes Total Electron Content (TEC) data, solar activity indices including the F10.7 index (10.7 cm wavelength index), geomagnetic activity indices including Kp and Dst data, and also includes the Interplanetary Magnetic Field (IMF) index. All of the above data are combined to construct an ionospheric feature vector.

[0065] Based on the above embodiments, step 200, which involves preprocessing the ionospheric feature vector to obtain a preprocessed ionospheric feature vector, includes:

[0066] Outliers in the ionospheric feature vector are removed, and the feature data is fitted using a classical spline function. The fitted feature data is then used to supplement the missing values ​​in the ionospheric feature vector to obtain the initial preprocessed ionospheric feature vector.

[0067] The initial preprocessed ionospheric feature vector is resampled using polynomial interpolation to align the temporal information of different types of data, and the preprocessed ionospheric feature vector is output.

[0068] Specifically, the constructed ionospheric characteristic vector is preprocessed, including outlier removal, missing value imputation, and resampling of data at different resolutions to the same time dimension. For missing value imputation, a classic spline function is used to fit the feature data. The missing data time is substituted into the fitted function to calculate the missing value. After data imputation, the data for each feature should be consistent. If some data times are not aligned or the sampling rates are different, the feature is resampled. During resampling, polynomial interpolation is used to align the time information of different types of data.

[0069] Based on the above embodiments, step 200, which involves normalizing the preprocessed ionospheric feature vectors to obtain the training dataset, includes:

[0070] The preprocessed ionospheric feature vectors are divided into a training set, a validation set, and a test set according to a preset ratio.

[0071] Based on the maximum and minimum eigenvalues ​​of the training set, a normalized training set is obtained;

[0072] Based on the training set, determine the maximum and minimum feature values ​​of the validation set, as well as the maximum and minimum feature values ​​of the test set; obtain a normalized validation set based on the maximum and minimum feature values ​​of the validation set; and obtain a normalized test set based on the maximum and minimum feature values ​​of the test set.

[0073] The training dataset is constructed from the normalized training set, the normalized validation set, and the normalized test set.

[0074] Specifically, a mapping relationship is constructed for the preprocessed rule data, which is then divided into a training set, a validation set, and a test set according to a certain ratio, the ratio of which is determined based on the data size. After division, the dataset is normalized using the following normalization formula:

[0075]

[0076] Where Z represents the data to be normalized, Zi represents the normalized data, and Zmax and Zmin represent the maximum and minimum values ​​in the data to be normalized, respectively. The maximum and minimum features of the validation and test sets need to be normalized using the results from the training set.

[0077] Based on the above embodiments, step 300, which involves sequentially stacking the Transformer model, LSTM model, and tree model to construct an ensemble model, includes:

[0078] The primary learner of the tree model is constructed using the random forest algorithm, the secondary learner of the tree model is constructed using the gradient boosting decision tree algorithm, and the ensemble model is constructed using LSTM.

[0079] The Transformer model includes an encoder and a decoder. The decoder is connected to the encoder through an attention mechanism. The encoder receives historical time series, and the decoder extracts historical values ​​from the historical time series using a masked self-attention mechanism.

[0080] The LSTM model includes an input layer, a hidden layer, and an output layer. The input layer includes historical ionospheric feature vectors. The hidden layer includes two LSTM network layers, two Dropout layers, and one fully connected layer. The output layer includes predicted ionospheric data.

[0081] The tree model uses the RF algorithm to construct the initial learner and the XGBoost algorithm to construct the secondary learner.

[0082] Specifically, in this embodiment of the invention, the stacking method in model ensemble technology is used to construct an ensemble model for ionospheric prediction.

[0083] The ensemble model consists of three parts: a Transformer model, an encoder that takes historical time series data as input, and an decoder that uses ionospheric feature vectors constructed from historical time series data prior to the forecast time. The time span of the historical data can be adjusted according to the actual scenario to achieve the highest correlation between the forecast and the data. The decoder then connects to the encoder using an attention mechanism and performs predictions in an autoregressive manner. Figure 3 As shown. In the LSTM model, the input layer is also an ionospheric feature vector constructed from a historical time series prior to the forecast time; the hidden layer consists of two LSTM layers, two Dropout layers, and one fully connected layer. The number of cells in the two LSTM layers will be determined in the next step. The Dropout layer discards or deactivates the learned parameters with a certain probability to prevent overfitting; the output layer outputs the predicted value within a given time period. For example, if forecasting ionospheric data for the next two days with a time sampling rate of 2 hours, the output layer outputs data for the next 24 epochs, such as... Figure 4 As shown. In the tree model part, the Random Forest (RF) algorithm is used as the initial learner to obtain a predicted value, which is then input as one of the features into the secondary learner composed of XGBoost. Finally, the three models are combined to obtain the final ensemble model.

[0084] Based on the above embodiments, step 300, which involves training the ensemble model using the training dataset, determining the optimal parameters of the ensemble model using a grid search method, and obtaining the ionospheric prediction model, includes:

[0085] Determining the parameters of the ensemble model includes the learning rate, maximum number of decision trees, random state, maximum tree depth, minimum sample weight in leaf nodes, training batch size, number of training epochs, number of samples per epoch, number of neurons in the LSTM network, training dropout rate, activation function, and loss function.

[0086] Determine the fixed parameter combinations of the Transformer model, the LSTM model, and the tree model in the ensemble model parameters, as well as the initial values ​​of the parameters to be adjusted;

[0087] Using the grid search method, the optimal value of the parameter to be adjusted is determined based on the combination of the fixed parameters;

[0088] Repeat the steps of using the grid search method to determine the optimal value of the parameter to be adjusted based on the fixed parameter combination until the Transformer model, the LSTM model and the tree model reach local optimal parameters respectively;

[0089] The Transformer model, the LSTM model, and the tree model that have reached the local optimal parameters are weighted and calculated with preset weights to obtain the ionospheric prediction model.

[0090] Specifically, in this embodiment of the invention, a Transformer ensemble model is trained using a training set, and the optimal model parameters are determined using a grid search method. The Transformer ensemble model makes predictions based on ionospheric feature vectors constructed from a historical time series prior to the prediction time. The prediction result is then the future total electron content data of the ionosphere.

[0091] For example, taking 0:00 on February 27th of a certain year as an example, the total electron content of the ionosphere over a certain region can be predicted using the ionospheric feature vector data from February 22nd to February 26th of a certain year (a total of five days). During training, the actual ionospheric data obtained from February 27th to February 28th of a certain year is used as the objective function for calculation.

[0092] like Figure 5 As shown, the core idea of ​​the model algorithms is to minimize the residual between the training result and the target result. Therefore, it is necessary to determine the parameters in the training process. These parameters are responsible for fitting the samples as closely as possible and avoiding overfitting. Training the model means finding the optimal model parameters for the given training data. Based on the features, the range of hyperparameter selection for the model is first determined. The training process uses a grid search to obtain the optimal model parameters. Features 1 to 5, i.e., the ionospheric feature vectors in the previous embodiment, are used as inputs to the random forest model and the XGBoost model, as well as the Transformer model and the LSTM model. Feature 6 is the output of the primary learner of the random forest model and is used as input to the XGBoost model. Finally, the prediction results of the tree model, the Transformer model, and the LSTM model are weighted and calculated to obtain the final result of the ensemble model.

[0093] The Transformer ensemble model parameters include: learning_rate, n_estimators (maximum number of decision trees), random_state, max_depth (maximum tree depth), min_child_weight (minimum sum of weights of samples in leaf nodes), batch_size (training batch size), epochs (number of training epochs), steps_per_epoch (number of samples per epoch), LSTM_nodes (number of neurons in the LSTM network), dropout (training dropout rate), activation (activation function), and loss (loss function).

[0094] The three parts of the Transformer ensemble model are trained separately, and the specific tuning steps are the same, as follows:

[0095] Step 1: Determine the fixed combinations of some parameters and set the initial values ​​of the parameters to be adjusted;

[0096] Step 2: Following the grid search algorithm, determine the optimal value of each parameter sequentially while keeping other parameters fixed. The random_state parameter typically ranges from [50, 500]. Then, use the built-in cv function to find the optimal n_estimators parameter values ​​that match the learning_rate. For a given learning rate and maximum number of decision trees, select the parameters max_depth (typically [3, 10]) and min_child_weight (typically [0, 30]). The batch_size is typically chosen as a power of 2 (typically [8, 512]). The epochs are used to find the balance point between training set loss and test set loss based on the training effect. The steps_per_epoch typically ranges from [100, 1000]. The LSTM_nodes are determined based on the impact of different values ​​on the training results (typically [10, 128]). The dropout typically ranges from [0.2, 0.5]. The activation function is typically the Rule function, and the loss function is typically the rmse function.

[0097] Step 3: Repeat step 2 until the model reaches a local optimum. If there is no improvement in model performance after 5 or more consecutive training runs (this value can be adjusted as needed), the training is considered complete.

[0098] Step 4: After all parts of the model have been trained, they are combined into the final Transformer ensemble model.

[0099] Based on the above embodiments, step 400 includes:

[0100] Real-time ionospheric data is input into the ionospheric prediction model to obtain ionospheric prediction results, including:

[0101] Obtain the preset historical time series ionospheric feature vector of the real-time ionospheric data;

[0102] The preset historical time series ionospheric feature vector is input into the ionospheric prediction model, and the predicted data of the total electron content of the ionosphere for a preset future time period is output.

[0103] Specifically, in this embodiment of the invention, ionospheric feature vectors constructed from real-time acquired ionospheric data and historical time series prior to the forecast time are input into the trained ensemble model to obtain ionospheric forecast results for future times. For example, taking 00:00 on March 27th of a certain year as an example, inputting the ionospheric feature vector data for five days from March 22nd to March 26th of a certain year over a certain region into the trained ensemble model yields high-precision ionospheric total electron content forecast data for two days, from March 27th to March 28th of that year.

[0104] The intelligent ionospheric forecasting and monitoring system based on an integrated model provided by the present invention will be described below. The intelligent ionospheric forecasting and monitoring system based on an integrated model described below can be referred to in correspondence with the intelligent ionospheric forecasting and monitoring method based on an integrated model described above.

[0105] Figure 6 This is a schematic diagram of the structure of the intelligent ionospheric forecasting and monitoring system based on an integrated model provided in an embodiment of the present invention, as shown below. Figure 6 As shown, it includes: a data acquisition module 61, a preprocessing module 62, a training module 63, and a prediction module 64, wherein:

[0106] The acquisition module 61 is used to acquire historical ionospheric data and construct an ionospheric feature vector based on the historical ionospheric data; the preprocessing module 62 is used to preprocess the ionospheric feature vector to obtain a preprocessed ionospheric feature vector, and normalize the preprocessed ionospheric feature vector to obtain a training dataset; the training module 63 is used to construct an ensemble model by stacking a Transformer model, an LSTM model, and a tree model in sequence, train the ensemble model using the training dataset, and determine the optimal parameters of the ensemble model using a grid search method to obtain an ionospheric prediction model; the prediction module 64 is used to input real-time ionospheric data into the ionospheric prediction model to obtain ionospheric prediction results.

[0107] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7As shown, the electronic device may include a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute an ionospheric intelligent forecasting and monitoring method based on an integrated model. This method includes: collecting historical ionospheric data; constructing an ionospheric feature vector based on the historical ionospheric data; preprocessing the ionospheric feature vector to obtain a preprocessed ionospheric feature vector; normalizing the preprocessed ionospheric feature vector to obtain a training dataset; constructing an integrated model by sequentially stacking a Transformer model, an LSTM model, and a tree model; training the integrated model using the training dataset; determining the optimal parameters of the integrated model using a grid search method to obtain an ionospheric prediction model; and inputting real-time ionospheric data into the ionospheric prediction model to obtain ionospheric prediction results.

[0108] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0109] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the ionospheric intelligent forecasting and monitoring method based on the ensemble model provided by the above methods. The method includes: collecting historical ionospheric data and constructing an ionospheric feature vector based on the historical ionospheric data; preprocessing the ionospheric feature vector to obtain a preprocessed ionospheric feature vector and normalizing the preprocessed ionospheric feature vector to obtain a training dataset; constructing an ensemble model by sequentially stacking a Transformer model, an LSTM model, and a tree model; training the ensemble model using the training dataset; determining the optimal parameters of the ensemble model using a grid search method to obtain an ionospheric prediction model; and inputting real-time ionospheric data into the ionospheric prediction model to obtain an ionospheric prediction result.

[0110] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the intelligent ionospheric forecasting and monitoring method based on the ensemble model provided by the above-described methods. This method includes: collecting historical ionospheric data; constructing an ionospheric feature vector based on the historical ionospheric data; preprocessing the ionospheric feature vector to obtain a preprocessed ionospheric feature vector; normalizing the preprocessed ionospheric feature vector to obtain a training dataset; constructing an ensemble model by sequentially stacking a Transformer model, an LSTM model, and a tree model; training the ensemble model using the training dataset; determining the optimal parameters of the ensemble model using a grid search method to obtain an ionospheric prediction model; and inputting real-time ionospheric data into the ionospheric prediction model to obtain an ionospheric prediction result.

[0111] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0112] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart ionospheric forecasting and monitoring method based on an integrated model, characterized in that, include: Collect historical data of the ionosphere and construct an ionosphere feature vector based on the historical data of the ionosphere; The ionospheric feature vector is preprocessed to obtain a preprocessed ionospheric feature vector, and the preprocessed ionospheric feature vector is normalized to obtain a training dataset. An ensemble model is constructed by stacking Transformer model, Long Short-Term Memory (LSTM) model and tree model in sequence. The ensemble model is trained using the training dataset. The optimal parameters of the ensemble model are determined by grid search method to obtain the ionospheric prediction model. Real-time ionospheric data is input into the ionospheric prediction model to obtain ionospheric prediction results; An ensemble model is constructed by stacking Transformer, LSTM, and tree models sequentially, including: The primary learner of the tree model is constructed using the random forest algorithm, the secondary learner of the tree model is constructed using the gradient boosting decision tree algorithm, and the ensemble model is constructed using the long short-term memory network LSTM. The Transformer model includes an encoder and a decoder. The decoder is connected to the encoder through an attention mechanism. The encoder receives historical time series, and the decoder extracts historical values ​​from the historical time series using a masked self-attention mechanism. The LSTM model includes an input layer, a hidden layer, and an output layer. The input layer includes historical ionospheric feature vectors. The hidden layer includes two LSTM network layers, two Dropout layers, and one fully connected layer. The output layer includes predicted ionospheric data. The tree model uses the Random Forest (RF) algorithm to construct the initial learner and the Extreme Gradient Boosting Tree (XGBoost) algorithm to construct the secondary learner. The ensemble model is trained using the training dataset, and the optimal parameters of the ensemble model are determined using a grid search method to obtain the ionospheric prediction model, including: Determining the parameters of the ensemble model includes the learning rate, maximum number of decision trees, random state, maximum tree depth, minimum sample weight in leaf nodes, training batch size, number of training epochs, number of samples per epoch, number of neurons in the LSTM network, training dropout rate, activation function, and loss function. Determine the fixed parameter combinations of the Transformer model, the LSTM model, and the tree model in the ensemble model parameters, as well as the initial values ​​of the parameters to be adjusted; Using the grid search method, the optimal value of the parameter to be adjusted is determined based on the combination of the fixed parameters; Repeat the steps of using the grid search method to determine the optimal value of the parameter to be adjusted based on the fixed parameter combination until the Transformer model, the LSTM model and the tree model reach local optimal parameters respectively; The Transformer model, the LSTM model, and the tree model that have reached the local optimal parameters are weighted and calculated with preset weights to obtain the ionospheric prediction model.

2. The intelligent ionospheric forecasting and monitoring method based on an integrated model according to claim 1, characterized in that, Collect historical ionospheric data, and construct an ionospheric feature vector based on the historical ionospheric data, including: Ionospheric data, solar activity index data, geomagnetic activity index data, and interplanetary magnetic field index data are acquired to construct the ionospheric feature vector.

3. The intelligent ionospheric forecasting and monitoring method based on an integrated model according to claim 1, characterized in that, The ionospheric feature vector is preprocessed to obtain a preprocessed ionospheric feature vector, including: Outliers in the ionospheric feature vector are removed, and the feature data is fitted using a classical spline function. The fitted feature data is then used to supplement the missing values ​​in the ionospheric feature vector to obtain the initial preprocessed ionospheric feature vector. The initial preprocessed ionospheric feature vector is resampled using polynomial interpolation to align the temporal information of different types of data, and the preprocessed ionospheric feature vector is output.

4. The intelligent ionospheric forecasting and monitoring method based on an integrated model according to claim 3, characterized in that, The preprocessed ionospheric feature vectors are normalized to obtain the training dataset, which includes: The preprocessed ionospheric feature vectors are divided into a training set, a validation set, and a test set according to a preset ratio. Based on the maximum and minimum eigenvalues ​​of the training set, a normalized training set is obtained; Based on the training set, determine the maximum and minimum feature values ​​of the validation set, as well as the maximum and minimum feature values ​​of the test set; obtain a normalized validation set based on the maximum and minimum feature values ​​of the validation set; and obtain a normalized test set based on the maximum and minimum feature values ​​of the test set. The training dataset is constructed from the normalized training set, the normalized validation set, and the normalized test set.

5. The intelligent ionospheric forecasting and monitoring method based on an integrated model according to claim 1, characterized in that, Real-time ionospheric data is input into the ionospheric prediction model to obtain ionospheric prediction results, including: Obtain the preset historical time series ionospheric feature vector of the real-time ionospheric data; The preset historical time series ionospheric feature vector is input into the ionospheric prediction model, and the predicted data of the total electron content of the ionosphere for a preset future time period is output.

6. An intelligent ionospheric forecasting and monitoring system based on an integrated model, comprising the intelligent ionospheric forecasting and monitoring method based on an integrated model as described in any one of claims 1 to 5, characterized in that, include: The acquisition module is used to acquire historical ionospheric data and construct an ionospheric feature vector based on the historical ionospheric data. The preprocessing module is used to preprocess the ionospheric feature vector to obtain a preprocessed ionospheric feature vector, and to normalize the preprocessed ionospheric feature vector to obtain a training dataset. The training module is used to construct an ensemble model by stacking Transformer model, LSTM model and tree model in sequence, train the ensemble model using the training dataset, and determine the optimal parameters of the ensemble model using the grid search method to obtain the ionospheric prediction model. The prediction module is used to input real-time ionospheric data into the ionospheric prediction model to obtain ionospheric prediction results.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the intelligent ionospheric forecasting and monitoring method based on an integrated model as described in any one of claims 1 to 5.

8. A non-transitory 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 intelligent ionospheric forecasting and monitoring method based on the integrated model as described in any one of claims 1 to 5.

Images