A fast simulation method and system for key parameters of space weather

Abstract

The present invention relates to the field of data processing technology, and provides a method and system for rapid simulation of key parameters of space weather, including: using a space weather model to simulate and obtain a multi-source key data set of the ionosphere and thermosphere; standardizing the multi-source key data set based on a preprocessing strategy, and testing the SWIN Transformer with multi-source key small sample data; iteratively optimizing the 3D convolution operator of the SWIN Transformer based on data test effect information; performing model adjustment and optimization of the Transformer channel based on a predetermined learning strategy, constructing a thermosphere mass density simulation prediction model, and simulating key parameters of space weather through the thermosphere mass density simulation prediction model. The method solves the technical problem that the numerical simulation model of space physics in the prior art is complex in operation, resulting in the simulation of thermosphere mechanism parameters that are not timely and accurate enough.

Description

Technical Field

[0001] The present invention relates to the technical field related to data processing, and in particular to a method and system for quickly simulating key parameters of space weather. Background Art

[0002] The thermosphere is closely related to human activities. A large number of low-orbit satellites operate in this area. Under extreme space weather conditions (such as magnetic storms), the mass density of the thermosphere will experience complex, instantaneous and violent disturbances. The sudden increase in density will lead to increased atmospheric drag, affecting satellite operations and even causing devastating blows. Therefore, how to accurately and timely simulate and predict the mass density of the thermosphere, especially under extreme space weather conditions such as magnetic storms, is extremely important for satellite precise orbit determination, orbit braking, collision prediction and life prediction, and is of great significance to the further independent development of my country's aerospace, military and other industries.

[0003] Traditional space physics numerical simulation models are complex and require a lot of HPC computing resources. They are currently mainly used for theoretical research and cannot meet the "timeliness" and "accuracy" required for space weather forecasting. Therefore, rapid simulation of thermospheric mechanism parameters is still a huge challenge.

[0004] In summary, the present application is used to solve the technical problem in the prior art that the calculation of the space physics numerical simulation model is complex, resulting in the simulation of the thermosphere mechanism parameters not being timely and accurate enough. Summary of the invention

[0005] This application provides a method and system for rapid simulation of key parameters of space weather, aiming to solve the technical problem in the prior art that the calculation of space physics numerical simulation models is complex, resulting in the simulation of thermosphere mechanism parameters not being timely and accurate enough.

[0006] In view of the above problems, an embodiment of the present application provides a method and system for rapid simulation of key parameters of space weather.

[0007] The first aspect disclosed in the present application provides a method for rapid simulation of key parameters of space weather, the method comprising: obtaining a multi-source key data set of the ionosphere-thermosphere by simulating a space weather model; standardizing the multi-source key data set based on a preprocessing strategy to obtain a multi-source standard key data set, wherein the preprocessing strategy includes structural conversion, data cleaning and normalization; dividing the multi-source standard key data set into sample proportions to obtain multi-source key small sample data, and using the multi-source key small sample data to test a SWIN Transformer to obtain data test effect information; iteratively optimizing a 3D convolution operator of the SWIN Transformer based on the data test effect information to obtain a Transformer channel that meets the expected accuracy; performing model adjustment and optimization of the Transformer channel based on a predetermined learning strategy, constructing a thermosphere mass density simulation prediction model, and simulating key parameters of space weather through the thermosphere mass density simulation prediction model.

[0008] Another aspect disclosed in the present application provides a rapid simulation system for key parameters of space weather, the system comprising: a data set acquisition module: used to use a space weather model to simulate and acquire a multi-source key data set of the ionosphere-thermosphere; a data preprocessing module: used to standardize the multi-source key data set based on a preprocessing strategy to acquire a multi-source standard key data set, wherein the preprocessing strategy includes structural conversion, data cleaning and normalization; a sample ratio division module: used to divide the multi-source standard key data set into sample ratios, acquire multi-source key small sample data, and use the multi-source key small sample data to test the SWIN Transformer to obtain data test effect information; an iterative optimization module: used to iteratively optimize the 3D convolution operator of the SWIN Transformer based on the data test effect information to obtain a Transformer channel that meets the expected accuracy; a key parameter simulation module: used to perform model adjustment and optimization of the Transformer channel based on a predetermined learning strategy, construct a thermosphere mass density simulation prediction model, and simulate key parameters of space weather through the thermosphere mass density simulation prediction model.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] By using a multi-source key data set of space weather model (such as TIEGCM to obtain the physics of the ionosphere-thermosphere) parameters; standardizing the multi-source key data set based on a preprocessing strategy, where the preprocessing strategy includes structure conversion, data cleaning and normalization; testing the Transformer channel of the multi-source standard data set, iteratively optimizing the 3D convolution operator based on the test results, and obtaining a Transformer channel that meets the expected accuracy; and performing model adjustment and optimization of the Transformer channel based on a predetermined learning strategy. The technical effect of improving the timely and accurate simulation of the thermosphere mechanism parameters has been achieved.

[0011] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A schematic flow chart of a method for rapid simulation of key space weather parameters is provided for an embodiment of the present application;

[0013] Figure 2 A schematic flow chart of obtaining a Transformer channel that meets the expected accuracy in a method for rapid simulation of key space weather parameters is provided for an embodiment of the present application;

[0014] Figure 3 A structural schematic diagram of a system for rapid simulation of key space weather parameters is provided for an embodiment of the present application.

[0015] Explanation of the reference numerals: data set acquisition module 11, data preprocessing module 12, sample ratio division module 13, iterative optimization module 14, key parameter simulation module 15. DETAILED DESCRIPTION

[0016] The overall idea of ​​the technical solution provided by this application is as follows:

[0017] The embodiment of the present application provides a method and system for rapid simulation of key parameters of space weather. First, multi-source data of the ionosphere-thermosphere are simulated by a space weather model and standardized by a preprocessing strategy. Subsequently, the SWIN Transformer model is tested using small sample data, and its 3D convolution operator is iteratively optimized based on the test results. Finally, based on the optimized Transformer channel and combined with a predetermined learning strategy, a thermosphere mass density simulation prediction model is constructed for simulation and prediction of key parameters of space weather. This method effectively integrates multi-source data and improves the accuracy and efficiency of space weather simulation.

[0018] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification.

[0019] Embodiment 1

[0020] like Figure 1 As shown, the embodiment of the present application provides a method for rapid simulation of key parameters of space weather, including:

[0021] Step S100: using a space weather model to simulate and obtain a multi-source key data set of the ionosphere and thermosphere;

[0022] Specifically: The space weather model refers to a magnetohydrodynamic model, particle dynamics model, statistical model, etc. that can simulate the physical processes of the ionosphere-thermosphere (such as TIEGCM obtaining the physical quantities of the ionosphere-thermosphere) based on actual research objectives and needs, thereby obtaining a multi-source key data set of the ionosphere-thermosphere, and the multi-source key data set includes temperature, pressure, electric field, magnetic field, etc.

[0023] Among them, TIEGCM (Thermosphere Ionosphere Electrodynamics General Circulation Model) is a theoretical model developed by NCAR / HAO to explore the interaction between the ionosphere and the thermosphere. The model is a three-dimensional dynamic theoretical model with time-varying characteristics. It can be divided into 29 pressure layers in the vertical direction, covering an altitude range of about 97 to 600 kilometers. In the horizontal direction, its resolution is 5°×5° in latitude and longitude.

[0024] TIEGCM can simulate and study the dynamic changes of the ionosphere-thermosphere system, such as the characteristics of atmospheric density changes. In addition, TIEGCM can also be used in future thermosphere atmospheric density forecasting work to provide security services for the on-orbit operation safety of low-orbit spacecraft.

[0025] Step S200: performing standardization processing on the multi-source key data set based on a preprocessing strategy to obtain a multi-source standard key data set, wherein the preprocessing strategy includes structure conversion, data cleaning and normalization processing;

[0026] Specifically: Based on the preprocessing strategy, the multi-source key data set is standardized, including structural conversion, data cleaning and normalization processing, wherein the structural conversion refers to converting the different data structures of different data sources into a unified format and structure, which is used to convert the simulation data of irregular grids into regular grids for subsequent processing. The data cleaning refers to removing noise, outliers, missing values, etc. in the data to ensure the accuracy and reliability of the data, and is used to check the consistency and integrity between parameter data at the same time, including the inspection of physical logical relationships. The normalization processing refers to converting the data to a unified scale, usually scaling the data to a range of 0 to 1, for subsequent model training and prediction. Thus, a multi-source standard key data set is obtained, which is used to study the dynamic changes, physical processes, and interactions with other space weather phenomena of the ionosphere-thermosphere, which has important scientific and application value for understanding the physical processes of the Earth's atmosphere and ensuring the normal operation of communication and navigation systems in modern society.

[0027] Furthermore, the step S200 of the present application of obtaining a multi-source standard key data set also includes:

[0028] Step S210: performing structural transformation on the multi-source key data set to obtain a rule key data set;

[0029] Step S220: obtaining a parameter boundary abnormality threshold, and performing abnormality identification on the rule key data set based on the parameter boundary abnormality threshold to obtain an abnormal key data set;

[0030] Step S230: performing data cleaning on the abnormal key data set to obtain a usable key data set;

[0031] Step S240: performing normalization processing based on the available key data set to obtain the multi-source standard key data set.

[0032] Specifically: the multi-source key data set is structurally converted, the purpose of which is to convert the data into a unified format and structure for subsequent processing and analysis. The structure may include renaming of data fields, conversion of data types, standardization of timestamps, etc. Based on this step, a regular key data set is obtained, which is a data set with a unified format, reliable quality, and comparable values. This data set contains the key parameters required for ionosphere-thermosphere research, namely temperature, pressure, electric field, magnetic field and other data in a unified format.

[0033] Furthermore, in the key data set, there may be outliers or erroneous data. Therefore, one or more parameter boundary anomaly thresholds are set for each key parameter. The parameter boundary anomaly threshold is determined based on physical principles, empirical rules or statistical methods. Further, based on the parameter boundary anomaly threshold, each data point in the rule key data set is identified as anomaly, all data points that exceed the threshold range are marked as outliers, and they are placed in the abnormal key data set, which refers to the data point set that is identified in the data cleaning and anomaly identification stage as not meeting expectations or not within the normal range. It may be caused by various reasons, such as equipment failure, data transmission error, measurement error, extreme weather conditions, etc.

[0034] Then, the abnormal key data set is cleaned to delete or correct the erroneous, redundant or inconsistent data in the data set, including: deleting abnormal values, replacing them with the mean or median, filling missing values ​​with interpolation methods, etc. The cleaned data set is the usable key data set, which contains valid and error-free data.

[0035] Finally, normalization is performed based on the available key data set to convert data of different ranges into the same range (such as between 0 and 1) for comparison and analysis. Normalization of the available key data set ensures that all parameters have the same numerical weight and reduces the impact of scale differences between different parameters on the analysis results. The normalized data set is the multi-source standard key data set, which can be directly used for subsequent data analysis and modeling.

[0036] Furthermore, the step S230 of obtaining the available key data set in the present application further includes:

[0037] Step 231: identifying missing data for the abnormal key data set, obtaining the missing key data set, and supplementing the missing key data set to obtain a complete key data set;

[0038] Step 232: Perform consistency calibration based on the complete key data set to obtain a standardized key data set;

[0039] Step 233: Performing a logical limit analysis on the normalized key data set to obtain an over-threshold key data set;

[0040] Step 234: Perform logic correction based on the super-threshold key data set to obtain the available key data set.

[0041] Specifically: missing data is identified for the abnormal key data set, and the missing data identification is the process of finding missing values. Once the missing data is identified, data supplementation is performed. There are many ways to supplement data, such as: interpolating the value of the previous or next data point; filling with the mean, median or mode of the entire data set; using machine learning algorithms (such as K nearest neighbors, decision trees, etc.) to predict missing values. After data supplementation, a complete key data set is obtained, which contains a complete key data set of all expected data points.

[0042] Furthermore, based on the above complete key data set, due to different units used by different data sources and inconsistent timestamp formats, the complete key data set is calibrated for consistency, wherein the consistency calibration is to convert these data into a unified format and standard to ensure that they can be correctly processed in subsequent analysis. Including: unit conversion (such as conversion from Celsius to Kelvin); unification of timestamp format; standardization of data types. Thus, a standardized key data set is obtained.

[0043] Furthermore, the normalized key data set is subjected to a logical limit analysis, which is a process of checking whether the data meets certain logical or physical limits. For example, the electron density in the ionosphere cannot be negative, the atmospheric temperature cannot exceed a reasonable upper limit, etc. Through the logical limit analysis, data points that exceed the logical limit can be identified and placed in the super-threshold key data set.

[0044] Then, logic correction is performed based on the over-threshold key data set, where the method of logic correction depends on the specific situation of the data. For example: if the over-threshold is caused by an error, these data points can be deleted or replaced; if the over-threshold is caused by some special conditions (such as extreme weather), special treatment of these data points can be considered during data analysis; if the over-threshold data points are few and have little impact on the overall analysis, they can also be ignored. After logic correction, a more accurate and reliable set of available key data is obtained for subsequent data analysis and modeling.

[0045] Through the above steps, an accurate, complete, consistent and reliable set of available key data can be obtained. The available key data set not only conforms to physical and logical laws, but also has a unified format and standard, which is convenient for subsequent data analysis and modeling work. The accuracy and reliability of data analysis are improved.

[0046] Step S300: dividing the multi-source standard key data set into sample proportions, obtaining multi-source key small sample data, and using the multi-source key small sample data to test the SWIN Transformer to obtain data test effect information;

[0047] Specifically: the sample ratio division refers to randomly extracting samples from a multi-source standard key data set in a certain ratio (such as 70% for training, 15% for verification, and 15% for testing) to form three independent data sets. In the division process, it is necessary to ensure that the distribution and characteristics of the data remain unchanged to avoid introducing additional deviations. Furthermore, the ratio division also includes sampling again on the basis of the divided training set, verification set or test set to obtain a smaller sample data set, that is, multi-source key small sample data. The multi-source small sample data set may contain only some categories of data (including: temperature, pressure, electric field, magnetic field and other data), or only extract data from some data sources to simulate small sample learning problems in real scenarios.

[0048] Further, the multi-source key small sample data is input into the SWIN Transformer model for training and / or testing. In the training phase, the training set is used to train the model, and the validation set is used to adjust the hyperparameters of the model. In the testing phase, the test set is used to evaluate the performance of the model. In the testing phase, the performance indicators of the model are collected and recorded. The performance indicators include accuracy, recall, F1 score, AUC-ROC curve, etc.

[0049] Step S400: iteratively optimizing the 3D convolution operator of the SWIN Transformer based on the data test effect information to obtain a Transformer channel that meets the expected accuracy;

[0050] Specifically: Based on the test effect information, the 3D convolution operator of SWIN Transformer is iteratively optimized. The 3D convolution operator is responsible for performing convolution operations on three-dimensional spatial data to extract spatial features. By adjusting the parameters of these operators, the performance of the model on specific tasks can be improved. The iterative optimization process includes multiple training, verification and testing cycles until the model reaches the expected accuracy or meets other stopping conditions.

[0051] Among them, in order to obtain a Transformer channel that meets the expected accuracy, it includes performing operator problem analysis on the data test effect information to obtain operator transformation learning parameters, and further improving and testing the 3D convolution operator based on the operator transformation learning parameters and multi-source key small sample data to obtain model effect feedback parameters; finally, the 3D convolution operator is iteratively optimized based on the model effect feedback parameters through the model optimization algorithm to obtain the Transformer channel that achieves the expected accuracy.

[0052] Furthermore, obtaining a Transformer channel that meets the expected accuracy in step S400 of the present application also includes:

[0053] Step S410: performing operator problem analysis on the data test effect information to obtain operator transformation learning parameters, where the operator transformation learning parameters include a convolutional layer, an encoder, and a decoder;

[0054] Step S420: performing an improvement test on the 3D convolution operator based on the operator transformation learning parameters and the multi-source key small sample data to obtain a model effect feedback parameter;

[0055] Step S430: Iteratively optimize the 3D convolution operator based on the model effect feedback parameter through a model optimization algorithm to obtain the Transformer channel that achieves the expected accuracy.

[0056] Specifically: In order to obtain a Transformer channel that meets the expected accuracy, the first step is to perform operator problem analysis on the data test effect information, that is, to deeply analyze the data test effect information and find out the key factors affecting model performance. This includes analyzing the accuracy, the changing trend of the loss function, the performance differences of the model on different categories or data sources, etc. Among them, the problem is located for different operators in the Transformer model (such as convolutional layers, encoders, and decoders). This is achieved by analyzing the output characteristics, gradient information, and interactions of the operator with other layers. According to the results of the problem analysis, determine the operators that need to be modified and their learning parameters, including the size of the convolution kernel, step size, padding method, and the structural configuration of the encoder and decoder.

[0057] In the second step, the learning parameters of the operator obtained in the first step are transformed to improve the design of the 3D convolution operator, including modifying the shape of the convolution kernel to adapt to the processing of 3D data, and introducing a new attention mechanism to enhance the feature extraction capability. The attention mechanism can help the model focus on the most relevant or important parts when processing a large amount of information, thereby improving the performance and efficiency of the model. The improved 3D convolution operator is tested using multi-source key small sample data, that is, by replacing the original operator at the appropriate position in the model, training it on the training set, and evaluating the performance on the validation set, and recording the model effect feedback parameters during the improvement test process, such as accuracy, loss function value, and visualization results of feature maps.

[0058] The third step is to evaluate whether the improved 3D convolution operator achieves the expected performance improvement based on the model effect feedback parameters. If the performance has not improved significantly or there are still problems, it is necessary to further analyze the cause of the problem. Use model optimization algorithms (such as gradient descent, Adam optimizer, etc.) to iteratively optimize the 3D convolution operator, that is, by adjusting hyperparameters such as learning rate, batch size, and regularization terms. In each iteration, use multi-source key small sample data to train and verify the model, and record the model effect feedback parameters. Adjust the optimization strategy according to the feedback parameters until the model performance reaches the expected accuracy. After multiple iterative optimizations, you will get a Transformer model with stable performance and expected accuracy. It can better handle multi-source key small sample data.

[0059] After the above steps, we deeply analyzed the data test effect information, adjusted the transformation learning parameters of key operators, used multi-source key small sample data for improved testing, and used the model optimization algorithm for iterative optimization, and obtained the Transformer channel that met the expected accuracy. This not only improved the accuracy and generalization ability of the model, but also improved the efficiency of iterative optimization and the operation efficiency of the model.

[0060] Furthermore, step S400 further includes:

[0061] Step 401: Collect and obtain key data information updated from multiple sources;

[0062] Step 402: input the multi-source updated key data information into the Transformer channel to obtain the prediction key parameter result;

[0063] Step 403: performing data loss analysis on the prediction key parameter results to obtain loss training data, and performing incremental learning updates on the Transformer channel based on the loss training data.

[0064] Specifically, it collects and obtains multi-source updated key data information, which refers to key data from different sources that need to be updated. By collecting multi-source updated key data information, it ensures that the model can process the latest and diverse data.

[0065] The multi-source updated key data information is input into the Transformer channel, which quickly processes the real-time updated data to provide timely prediction results of key parameters. Using the Transformer channel for prediction can maintain a high prediction accuracy when processing multi-source data. Through multiple iterations and optimizations, the Transformer channel can output the expected prediction key parameter results for a specific task.

[0066] Furthermore, by performing data loss analysis on the predicted key parameter results, the predicted key parameter results are compared with the real, known key parameter values. The difference between the two is calculated, i.e., the loss. The loss function includes mean square error (MSE), cross entropy loss, etc., and the loss training data is obtained. The loss training data describes which data points the model predicts inaccurately.

[0067] Based on the loss training data, the Transformer channel is updated with incremental learning. Incremental learning is a technique that allows the model to retain old knowledge while learning new data. During the update process, the model focuses on data points that cause large losses to improve its prediction ability. The parameters in the Transformer channel can be adjusted to minimize the loss function through optimization methods such as backpropagation algorithms and gradient descent. This process is iterated until the performance of the model reaches a preset standard or a certain termination condition is reached (such as the maximum number of iterations, the loss value no longer decreases significantly, etc.).

[0068] During the incremental learning update process, the performance of the model needs to be continuously monitored to ensure that it is not overfitting or underfitting. The performance of the model is evaluated through the validation set or test set, and hyperparameters such as learning rate and batch size are adjusted based on the evaluation results.

[0069] Based on the above, by collecting key data information from multiple sources and using the Transformer channel for prediction and incremental learning updates, the model achieves high accuracy, real-time, adaptability and continuous improvement.

[0070] Step S500: Execute model adjustment and optimization of the Transformer channel based on a predetermined learning strategy, construct a thermospheric mass density simulation prediction model, and simulate key parameters of space weather through the thermospheric mass density simulation prediction model.

[0071] Specifically, the predetermined learning strategy includes adjusting the hyperparameters of the training model, adjusting the model size, and changing the learning strategy, etc. The hyperparameters of the training model are adjusted, including grid search and random search. Adjusting the model size includes adjusting parameter accuracy, weight sharing and compression. Changing the learning strategy includes selecting a suitable loss function, setting a suitable learning rate, using batch training and regularization techniques.

[0072] Furthermore, in step S500 of the present application, a thermal layer mass density simulation prediction model is constructed, including:

[0073] Step S510: performing full sample data training on the Transformer channel based on the multi-source standard key data set to obtain a Transformer improvement channel;

[0074] Step S520: Execute the predetermined learning strategy to adjust and optimize the model parameters of the Transformer improvement channel to construct the thermal layer mass density simulation prediction model.

[0075] Specifically: Use the processed multi-source standard key data set to train the Transformer channel on full sample data. During the training process, the Transformer channel will learn the ability to extract features from the input data and predict the mass density of the thermal layer. Monitor indicators such as loss function changes and validation set performance during training to ensure effective training of the model. After training with full sample data, the Transformer improvement channel is obtained.

[0076] Furthermore, the predetermined learning strategy is executed to adjust and optimize the model parameters of the Transformer improvement channel, including adjusting the training model hyperparameters, i.e., adjusting the learning rate, batch size, regularization parameters, etc. Different data sets and model architectures may require different hyperparameter settings. It includes: Grid search: that is, by specifying a list of candidate values ​​for hyperparameters, exhaustively combining different hyperparameter values, and evaluating the performance of each set of hyperparameters. Finally, the hyperparameter combination with the best performance is selected. Random search: Randomly select some combinations from the specified range of hyperparameter values ​​for evaluation. Better hyperparameter combinations can be found in a relatively small number of evaluations, allowing for more flexible search space settings.

[0077] Adjusting the model size includes adjusting the parameter precision: the higher the parameter precision, the larger the model size. Select the appropriate parameter precision according to specific needs. Weight sharing and compression: In order to reduce the size of deep learning models, weight sharing and compression techniques can be used. Weight sharing refers to sharing some parameters between multiple layers in the model, thereby reducing storage space. Model compression methods include pruning, quantization, etc.

[0078] Changing the learning strategy includes choosing a suitable loss function: the loss function is an indicator used to measure the difference between the model output and the true value. Reasonable selection of the loss function can enable the model to better capture the characteristics and distribution of the data and make the training process more efficient. Set a suitable learning rate: the learning rate is an important parameter to control the speed and accuracy of model training. If the learning rate is too large, the model may not converge; if the learning rate is too small, the training time will be too long or fall into a local optimal solution. The appropriate learning rate can be set by learning rate decay, dynamic adjustment, or using adaptive adjustment methods. Use batch training and regularization techniques: batch training can reduce training time, but may cause the model to overfit to a specific batch of data. Regularization techniques (such as L1 regularization, L2 regularization, Dropout, etc.) can help prevent overfitting.

[0079] Based on the above adjustments and optimizations and achieving satisfactory performance, a thermospheric mass density simulation prediction model is obtained. It is used to simulate and predict changes in thermospheric mass density. Through the thermospheric mass density simulation prediction model, key parameters of space weather can be simulated, which may include solar wind speed, changes in the earth's magnetic field, ionosphere state, etc. The simulation process uses the prediction results of the model to generate time series or spatial distribution maps of these parameters in order to monitor and predict space weather in real time. The simulation results can be used to support applications in the fields of space weather forecasting, satellite orbit prediction, and communication system design.

[0080] Furthermore, the step S500 of the present application constructs a thermal layer mass density simulation prediction model, which also includes

[0081] Step S501: define a model loss function to verify the output accuracy of the Transformer improvement channel and determine a model loss verification result;

[0082] Step S502: performing strategy matching analysis on the predetermined learning strategy based on the model loss verification result to determine model optimization strategy parameters;

[0083] Step S503: using the back propagation algorithm and the model optimization strategy parameters, the Transformer improvement channel is subjected to model iterative optimization to obtain the thermal layer mass density simulation prediction model.

[0084] Specifically: define a model loss function to verify the output accuracy of the Transformer improvement channel, and the loss function is used to measure the difference between the model prediction value and the actual value. For the thermosphere mass density simulation prediction model, select Mean Squared Error (MSE), Mean Absolute Error (MAE) or Cross-Entropy Loss (if the output is in classification form) as the loss function. After defining the loss function, use the label in the training data set (that is, the actual thermosphere mass density value) and the model's prediction value to calculate the loss value. Further, use the validation data set (a data set independent of the training data set) to verify the output accuracy of the Transformer improvement channel. Pass the input data in the validation data set to the model to obtain the predicted output of the model. Compare the predicted output with the actual label in the validation data set, calculate the accuracy or other relevant performance indicators (such as precision, recall, F1 score, etc.), and obtain the model loss verification result, that is, evaluate the performance of the model on the validation data set, and calculate the validation loss (using the previously defined loss function) and accuracy.

[0085] Furthermore, according to the model loss verification results, the predetermined learning strategy is analyzed for strategy matching. That is, the trend of the loss value change, whether overfitting or underfitting occurs, etc., are analyzed to determine whether the hyperparameters in the learning strategy (such as learning rate, batch size, regularization strength, etc.) need to be adjusted. If the model performance is found to be poor, it may be necessary to try different optimization algorithms, adjust the learning rate strategy (such as using learning rate decay), add regularization terms or use other regularization techniques, etc., to guide the further optimization of the model, thereby determining the model optimization strategy parameters.

[0086] Finally, the backpropagation algorithm is used to calculate the gradient of the loss function to the model parameters. The model parameters are updated based on the calculated gradient and the model optimization strategy parameters (such as the learning rate). This process is repeated to gradually optimize the performance of the model through multiple iterations.

[0087] In summary, the method and system for rapid simulation of key space weather parameters provided by the embodiments of the present application have the following technical effects:

[0088] 1. The multi-source data of the ionosphere and thermosphere are simulated through the space weather model and standardized through the preprocessing strategy. The SWIN Transformer model is tested with small sample data, and its 3D convolution operator is iteratively optimized based on the test results. Based on the optimized Transformer channel and the predetermined learning strategy, a thermosphere mass density simulation prediction model is constructed for the simulation and prediction of key space weather parameters. The multi-source data is effectively integrated to improve the accuracy and efficiency of space weather simulation.

[0089] 2. Obtain a complete set of key data by identifying and supplementing missing data. Then, calibrate the complete data set for consistency to ensure data normalization. Perform logical boundary analysis on the normalized data to identify key data that exceeds the threshold. Perform logical correction on these over-threshold data to form a usable set of key data. This effectively improves the integrity and accuracy of the data and provides a reliable data foundation for subsequent analysis.

[0090] 3. Through in-depth analysis of data test effect information, adjustment of transformation learning parameters of key operators, improvement testing using multi-source key small sample data, and iterative optimization using model optimization algorithms, a Transformer channel with expected accuracy was obtained. This not only improves the accuracy and generalization ability of the model, but also improves the efficiency of iterative optimization and the operation efficiency of the model.

[0091] Embodiment 2

[0092] Based on the same inventive concept as the method for rapid simulation of key space weather parameters in the aforementioned embodiment, Figure 3 As shown, an embodiment of the present application provides a rapid simulation system for key parameters of space weather, the system comprising:

[0093] Acquisition data set module 11: used for simulating and acquiring multi-source key data sets of ionosphere and thermosphere by using space weather model;

[0094] Data preprocessing module 12: used to perform standardization processing on the multi-source key data set based on a preprocessing strategy to obtain a multi-source standard key data set, wherein the preprocessing strategy includes structure conversion, data cleaning and normalization processing;

[0095] The sample ratio division module 13 is used to divide the multi-source standard key data set into sample ratios, obtain multi-source key small sample data, and use the multi-source key small sample data to test the SWIN Transformer to obtain data test effect information;

[0096] Iterative optimization module 14: used for iteratively optimizing the 3D convolution operator of the SWIN Transformer based on the data test effect information to obtain a Transformer channel that meets the expected accuracy;

[0097] Key parameter simulation module 15: used to perform model adjustment and optimization of the Transformer channel based on a predetermined learning strategy, construct a thermospheric mass density simulation prediction model, and simulate key parameters of space weather through the thermospheric mass density simulation prediction model.

[0098] Furthermore, the system is also used for:

[0099] Performing structural transformation on the multi-source key data set to obtain a rule key data set;

[0100] Acquire a parameter boundary abnormality threshold, perform abnormality identification on the rule key data set based on the parameter boundary abnormality threshold, and obtain an abnormal key data set;

[0101] Performing data cleaning on the abnormal key data set to obtain a usable key data set;

[0102] Normalization processing is performed based on the available key data set to obtain the multi-source standard key data set.

[0103] Furthermore, the system is also used for:

[0104] Identifying missing data for the abnormal key data set, obtaining the missing key data set, and supplementing the missing key data set to obtain a complete key data set;

[0105] Perform consistency calibration based on the complete key data set to obtain a standardized key data set;

[0106] Performing a logical limit analysis on the normalized key data set to obtain an over-threshold key data set;

[0107] A logic correction is performed based on the super-threshold key data set to obtain the available key data set.

[0108] Furthermore, the system is also used for:

[0109] Performing operator problem analysis on the data test effect information to obtain operator transformation learning parameters, wherein the operator transformation learning parameters include a convolutional layer, an encoder, and a decoder;

[0110] Based on the operator transformation learning parameters and the multi-source key small sample data, the 3D convolution operator is improved and tested to obtain model effect feedback parameters;

[0111] The 3D convolution operator is iteratively optimized based on the model effect feedback parameters through a model optimization algorithm to obtain the Transformer channel that achieves the expected accuracy.

[0112] Furthermore, the system is also used for:

[0113] Collect and update key data information from multiple sources;

[0114] Input the multi-source updated key data information into the Transformer channel to obtain the predicted key parameter results;

[0115] By performing data loss analysis on the prediction key parameter results, loss training data is obtained, and incremental learning and updating of the Transformer channel is performed based on the loss training data.

[0116] Furthermore, the system is also used for:

[0117] Performing full sample data training on the Transformer channel based on the multi-source standard key data set to obtain a Transformer improvement channel;

[0118] The predetermined learning strategy is executed to adjust and optimize the model parameters of the Transformer improvement channel, and the thermal layer mass density simulation prediction model is constructed.

[0119] Furthermore, the system is also used for:

[0120] Define a model loss function to verify the output accuracy of the Transformer improvement channel and determine the model loss verification result;

[0121] Performing strategy matching analysis on the predetermined learning strategy based on the model loss verification result to determine model optimization strategy parameters;

[0122] The back propagation algorithm and the model optimization strategy parameters are used to perform model iterative optimization on the Transformer improvement channel to obtain the thermal layer mass density simulation prediction model.

[0123] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0124] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to multiple elements that can be selected individually or in full. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for rapid simulation of key space weather parameters, characterized in that: The method comprises: Use space weather model simulation to obtain multi-source key data sets of the ionosphere and thermosphere; Standardizing the multi-source key data set based on a preprocessing strategy to obtain a multi-source standard key data set, wherein the preprocessing strategy includes structure conversion, data cleaning and normalization processing; Dividing the multi-source standard key data set into sample proportions to obtain multi-source key small sample data, and using the multi-source key small sample data to test the SWIN Transformer to obtain data test effect information; Iteratively optimizing the 3D convolution operator of the SWIN Transformer based on the data test effect information to obtain a Transformer channel that meets the expected accuracy; The model adjustment and optimization of the Transformer channel is performed based on a predetermined learning strategy. After training with all sample data, the Transformer improvement channel is obtained, and then a thermospheric mass density simulation prediction model is constructed, and key space weather parameters are simulated through the thermospheric mass density simulation prediction model; The method of constructing a thermosphere mass density simulation prediction model comprises: Performing full sample data training on the Transformer channel based on the multi-source standard key data set to obtain a Transformer improvement channel; The predetermined learning strategy is executed to adjust and optimize the model parameters of the Transformer improvement channel, and the thermal layer mass density simulation prediction model is constructed, wherein the predetermined learning strategy includes adjusting the training model hyperparameters, adjusting the model size, and changing the learning strategy.

2. The method according to claim 1, characterized in that The step of obtaining a multi-source standard key data set includes: Performing structural transformation on the multi-source key data set to obtain a rule key data set; Acquire a parameter boundary abnormality threshold, perform abnormality identification on the rule key data set based on the parameter boundary abnormality threshold, and obtain an abnormal key data set; Performing data cleaning on the abnormal key data set to obtain a usable key data set; Normalization processing is performed based on the available key data set to obtain the multi-source standard key data set.

3. The method according to claim 2, characterized in that The obtaining of available key data sets includes: Identifying missing data for the abnormal key data set, obtaining the missing key data set, and supplementing the missing key data set to obtain a complete key data set; Perform consistency calibration based on the complete key data set to obtain a standardized key data set; Performing a logical limit analysis on the normalized key data set to obtain an over-threshold key data set; A logic correction is performed based on the super-threshold key data set to obtain the available key data set.

4. The method according to claim 1, characterized in that The Transformer channel that meets the expected accuracy is obtained, including: Performing operator problem analysis on the data test effect information to obtain operator transformation learning parameters, wherein the operator transformation learning parameters include a convolutional layer, an encoder, and a decoder; Based on the operator transformation learning parameters and the multi-source key small sample data, the 3D convolution operator is improved and tested to obtain model effect feedback parameters; The 3D convolution operator is iteratively optimized based on the model effect feedback parameters through a model optimization algorithm to obtain the Transformer channel that achieves the expected accuracy.

5. The method according to claim 4, characterized in that The method comprises: Collect and update key data information from multiple sources; Input the multi-source updated key data information into the Transformer channel to obtain the predicted key parameter results; By performing data loss analysis on the prediction key parameter results, loss training data is obtained, and incremental learning and updating of the Transformer channel is performed based on the loss training data.

6. The method of claim 1, wherein: The step of constructing the thermal layer mass density simulation prediction model comprises: Define a model loss function to verify the output accuracy of the Transformer improvement channel and determine the model loss verification result; Performing strategy matching analysis on the predetermined learning strategy based on the model loss verification result to determine model optimization strategy parameters; The back propagation algorithm and the model optimization strategy parameters are used to perform model iterative optimization on the Transformer improvement channel to obtain the thermal layer mass density simulation prediction model.

7. A rapid simulation system for key space weather parameters, characterized in that: For executing the method according to any one of claims 1 to 6, the system comprises: Data collection acquisition module: used to obtain multi-source key data collection of ionosphere and thermosphere by using space weather model simulation; Data preprocessing module: used to perform standardization processing on the multi-source key data set based on a preprocessing strategy to obtain a multi-source standard key data set, wherein the preprocessing strategy includes structure conversion, data cleaning and normalization processing; Sample ratio division module: used to divide the multi-source standard key data set into sample ratios, obtain multi-source key small sample data, and use the multi-source key small sample data to test SWIN Transformer to obtain data test effect information; Iterative optimization module: used for iteratively optimizing the 3D convolution operator of the SWIN Transformer based on the data test effect information to obtain a Transformer channel that meets the expected accuracy; Key parameter simulation module: used to perform model adjustment and optimization of the Transformer channel based on a predetermined learning strategy. After training with full sample data, the Transformer improvement channel is obtained, and then a thermospheric mass density simulation prediction model is constructed, and key parameters of space weather are simulated through the thermospheric mass density simulation prediction model.

Images