Three-dimensional thermal layer and ionized layer parameter inversion method and system based on electron density

Abstract

The invention discloses a three-dimensional thermal layer and ionized layer parameter inversion method and system based on electron density, and the method comprises the steps: obtaining original data, carrying out the preprocessing of the original data, and generating a preprocessed data set; constructing a multi-channel parameter inversion model based on the preprocessed data set; and performing deduction based on the multi-channel parameter inversion model to generate a deduction result. Therefore, according to the three-dimensional thermal layer and ionosphere parameter inversion method based on the electron density, global three-dimensional ionosphere electron density data easy to measure are fully utilized, high-quality deduction of thermal layer and ionosphere parameters is achieved, sequential output is replaced with synchronous output channels, the reasoning speed is obviously increased, the calculation cost is obviously reduced, and the method is suitable for large-scale popularization and application. And the model can be quickly updated and iterated to maintain timeliness.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of space weather modeling, and in particular to a method and system for inverting three-dimensional thermosphere and ionosphere parameters based on electron density. Background Art

[0002] Thermosphere and ionosphere are important components of the Earth's atmosphere. Understanding the dynamic changes of the thermosphere and ionosphere is crucial for optimizing satellite design, improving communication quality, enhancing navigation system accuracy, and predicting space weather. At present, the main modeling approaches for thermosphere and ionosphere parameters are theoretical modeling and empirical modeling. Theoretical modeling is generally based on physical laws and equations (such as fluid mechanics, thermodynamics, and electrodynamics), and is driven by physical parameters to solve the parameters of the thermosphere and ionosphere. However, the internal structure of the numerical simulation model in the traditional theoretical model is complex, and complex differential equations need to be solved, so it has problems such as high computational complexity and difficulty in updating and maintenance. Empirical modeling is generally based on observational data, which is limited by equipment and has incomplete data coverage. For example, data in oceans, polar regions, and other regions are partially or completely missing in time and space. Therefore, complex data interpolation and assimilation are required in advance, resulting in low efficiency in parameter acquisition. Summary of the invention

[0003] The embodiment of the present invention provides a three-dimensional thermosphere and ionosphere parameter inversion method and system based on electron density, which makes full use of the easily measurable global three-dimensional ionospheric electron density data to achieve high-quality deduction of thermosphere and ionosphere parameters, and uses synchronous output of each channel instead of sequential output, which significantly accelerates the reasoning speed, significantly reduces the computing cost, and can quickly update and iterate the model to maintain timeliness.

[0004] To achieve the above object, in a first aspect, the present invention provides a three-dimensional thermosphere and ionosphere parameter inversion method based on electron density, comprising:

[0005] Obtain the original data and preprocess it to generate a preprocessed data set;

[0006] Based on the preprocessed data set, construct a multi-channel parameter inversion model;

[0007] Deduction is performed based on the multi-channel parameter inversion model to generate deduction results.

[0008] In one embodiment of the present invention, the three-dimensional thermosphere and ionosphere parameter inversion method based on electron density further includes:

[0009] Step S400: adjusting and optimizing the multi-channel parameter inversion model by a first preset method.

[0010] In one embodiment of the present invention, obtaining raw data and preprocessing to generate a preprocessed data set includes:

[0011] Based on the TIEGCM model, various key parameters of the global thermosphere and ionosphere are obtained as raw data;

[0012] Preprocessing the raw data by a second preset method to generate the preprocessed data set;

[0013] The preprocessed data set includes electron density data, spatiotemporal information and omni data.

[0014] In one embodiment of the present invention, constructing a multi-channel parameter inversion model based on the preprocessed data set includes:

[0015] Design the architecture of the multi-channel parameter inversion model and determine the input layer, hidden layer, and output layer;

[0016] Build a preliminary model based on the preset deep learning framework and algorithm;

[0017] The preliminary model is trained based on the preprocessed data set to construct the multi-channel parameter inversion model.

[0018] In one embodiment of the present invention, the adjusting and optimizing the multi-channel parameter inversion model by a first preset method includes:

[0019] Analyzing the multi-channel parameter inversion model by the first preset method to generate analysis results;

[0020] Based on the analysis results, determine the model parameters that need to be adjusted and optimized;

[0021] The selected model parameters are gradually adjusted and optimized.

[0022] In a second aspect, the present invention provides a three-dimensional thermosphere and ionosphere parameter inversion system based on electron density, comprising: a first generation module, a construction module and a second generation module. The first generation module is used to obtain raw data and perform preprocessing to generate a preprocessed data set. The construction module is used to construct a multi-channel parameter inversion model based on the preprocessed data set. And the second generation module is used to perform deduction based on the multi-channel parameter inversion model to generate a deduction result.

[0023] In one embodiment of the present invention, the three-dimensional thermosphere and ionosphere parameter inversion system based on electron density further includes an adjustment and optimization module for adjusting and optimizing the multi-channel parameter inversion model by a first preset method.

[0024] In one embodiment of the present invention, the first generation module includes: an acquisition unit and a first generation unit. The acquisition unit is used to acquire multiple key parameters of the global thermosphere and ionosphere as raw data based on the TIEGCM model; and the first generation unit is used to preprocess the raw data by a second preset method to generate the preprocessed data set. The preprocessed data set includes electron density data, spatiotemporal information and omni data.

[0025] In a third aspect, the present invention provides an electronic device, comprising:

[0026] at least one processor; and

[0027] a memory communicatively coupled to the at least one processor;

[0028] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the three-dimensional thermosphere and ionosphere parameter inversion method based on electron density as described above.

[0029] In a fourth aspect, the present invention provides a computer-readable storage medium, comprising a computer program and instructions, which, when executed on a computer, enables the computer to execute the three-dimensional thermosphere and ionosphere parameter inversion method based on electron density as described above.

[0030] Compared with the prior art, the three-dimensional thermosphere and ionosphere parameter inversion method and system based on electron density according to the present invention fully utilizes the easily measurable global three-dimensional ionosphere electron density data to achieve high-quality deduction of thermosphere and ionosphere parameters, and uses synchronous output of each channel instead of sequential output, which significantly accelerates the reasoning speed, significantly reduces the computing cost, and can quickly update and iterate the model to maintain timeliness. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a flow chart of a method for inverting three-dimensional thermosphere and ionosphere parameters based on electron density in Embodiment 1 of the present invention;

[0032] Figure 2 It is a structural schematic diagram of a three-dimensional thermosphere and ionosphere parameter inversion system based on electron density in Embodiment 2 of the present invention;

[0033] Figure 3 It is a structural schematic diagram of an electronic device in Embodiment 3 of the present invention.

[0034] Figure 4 It is a schematic diagram of the structure inside the neural network of a multi-channel parameter inversion model in a specific embodiment of the present invention;

[0035] Figure 5 It is a logical flow diagram of neural network training and reasoning of a multi-channel parameter inversion model in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0036] The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention, rather than to limit the embodiments of the present invention. It is also necessary to explain that, for ease of description, only parts related to the embodiments of the present invention are shown in the accompanying drawings, rather than all structures.

[0037] To facilitate understanding, the main implementation concepts of the embodiments of the present invention are first briefly described.

[0038] The existing thermosphere-ionosphere parameter observation and reconstruction models have the following problems:

[0039] 1. Observational data are sparse: 18 important parameters of the current thermosphere and ionosphere (O, O 2 、N 2 D.N 4 S, N 2 , NO, HE, O + , O 2 + , T e , T i , T n , U n 、V n , W n , U i E×B 、V i E×B , W i E×B ) is difficult to observe, the data has a small global coverage, and is unevenly distributed around the world, especially in remote areas and high-latitude regions, leading to data sparsity problems;

[0040] 2. High computational complexity: Traditional numerical simulation methods need to solve complex differential equations, which have high computational complexity and may face the problem of insufficient computing resources, which will limit the computing speed and affect real-time performance;

[0041] 3. Model update lag: Existing thermosphere-ionosphere parameter models such as the IRI model have complex internal structures, rely on past data updates, have long update cycles, and cannot fully utilize timely data.

[0042] The inventors have discovered the above-mentioned defects in the prior art and have provided a method and system for inverting three-dimensional thermosphere and ionosphere parameters based on electron density. By using the accurate and readily available electron density (Ne) and deep learning of the thermosphere-ionosphere multi-channel inversion model, the following are achieved:

[0043] 1. It can obtain reliable and accurate values ​​of the 18 important parameters worldwide, providing high-quality driving conditions for models that require initial data driving;

[0044] 2. Reduced model calculation overhead. Different from the traditional numerical simulation, where each output channel has a strict sequential operation relationship, each channel of the present invention outputs synchronously, and the reasoning speed is fast;

[0045] 3. Maintenance and updating are simple, and newly acquired data can be used efficiently and quickly, making it easier to upgrade and iterate, and more timely.

[0046] Embodiment 1

[0047] Figure 1 is a flow chart of a method for inverting three-dimensional thermosphere and ionosphere parameters based on electron density in Embodiment 1 of the present invention, such as Figure 1 As shown, embodiment 1 provides a three-dimensional thermosphere and ionosphere parameter inversion method based on electron density, including:

[0048] Step S100, obtaining raw data and preprocessing it to generate a preprocessed data set;

[0049] Specifically, the TIEGCM model is first used to obtain a variety of key parameters of the thermosphere and ionosphere on a global scale, which constitute the raw data for subsequent analysis. Since the raw data may have problems such as different formats, missing data or outliers, preprocessing is required. The preprocessing process may include, for example, data cleaning, format conversion, and necessary standardization or normalization to ensure the consistency and analyzability of the data. Specifically, the main input electron density, omni data and output channels are standardized, and the altitude is normalized to eliminate the impact of dimensional differences between different parameters. At the same time, the missing values ​​that appear at the highest altitude are deleted to ensure the integrity of the data. After preprocessing, a preprocessed data set containing electron density data, spatiotemporal information and omni data is generated, providing a high-quality data basis for subsequent model construction and deduction. This step is a key link to ensure the accuracy and reliability of the entire inversion method.

[0050] Step S200, constructing a multi-channel parameter inversion model based on the preprocessed data set;

[0051] Specifically, first, the model architecture is designed. This includes determining the structure of the input layer, hidden layer, and output layer. The input layer is responsible for receiving preprocessed data, which includes electron density, spatiotemporal information (such as longitude, latitude, time), and possible other related parameters (such as omni data). The hidden layer is the middle layer in the neural network, responsible for extracting data features and performing complex nonlinear transformations. The output layer is responsible for outputting the deduction results, namely 18 important parameters of the thermosphere and ionosphere. In the present invention, the input layer is designed to receive 17 parameters, and the values ​​of 18 channels are finally output by reducing the number of neurons layer by layer. Next, a preliminary model is constructed based on a preset deep learning framework and algorithm. Deep learning frameworks such as TensorFlow and PyTorch provide tools for building and training neural networks. In terms of algorithms, deep learning algorithms suitable for processing time series data and spatial data can be selected, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), etc. The specific selection depends on the characteristics of the data and the deduction requirements. Finally, the preliminary model is trained using the preprocessed data set. During the training process, the model minimizes the loss function by adjusting internal parameters (weights and biases) to improve the inference accuracy. The training strategy may include gradually reducing the learning rate, adjusting the batch size, and using early stopping techniques to prevent overfitting. In addition, in view of the inconsistent convergence speed that may exist in multi-channel output, the present invention also proposes a weight adjustment strategy to ensure that the convergence speed of each channel is similar, so as to maintain the correlation between parameters.

[0052] Based on the above analysis, the present invention constructs a multi-channel parameter inversion model that can efficiently and accurately deduce multiple key parameters of the thermosphere and ionosphere. This model not only improves the deduction accuracy, but also significantly speeds up the reasoning speed and reduces the computing cost through the design of synchronous output of each channel. At the same time, because the model can be quickly updated and iterated, it can better adapt to time-sensitive data and provide strong support for space weather modeling.

[0053] Step S300, performing deduction based on the multi-channel parameter inversion model to generate a deduction result;

[0054] Specifically, step S300 is one of the key steps of the three-dimensional thermosphere and ionosphere parameter inversion method based on electron density in an embodiment of the present invention, which involves using the already constructed multi-channel parameter inversion model to perform actual deduction work to generate the required deduction results. The following is a detailed description of this step: Before performing the deduction, it is necessary to ensure that all necessary input data are ready. These input data include, for example, electron density data, spatiotemporal information (such as longitude, latitude, time, etc.) and possible other related parameters (such as omni data, etc.). These data should be preprocessed to ensure that they meet the input requirements of the model, such as data format, data range, data quality, etc. Load the trained multi-channel parameter inversion model, that is, read the model file from the storage medium and load it into the memory for subsequent calculations. Initialize the model, including setting the necessary parameters and states to ensure that the model can be deduced correctly. Input the prepared input data into the model, that is, by calling the interface function of the model, which accepts the input data and returns the deduction results. The model will process the input data according to its internal structure and algorithm, and generate corresponding deduction results. This process involves, for example, the calculation of multiple hidden layers and the final output of the output layer. The deduction results usually include a variety of key parameters of the thermosphere and ionosphere, such as the density of atoms, molecules, and ions, as well as the velocity of the fluid in the longitudinal, latitudinal, and vertical directions. The deduction results are extracted from the model and necessary post-processing is performed. This includes, for example, converting the results into an easy-to-understand format, or further analyzing and verifying the results. The processed results are output to the user or stored in a specified location for subsequent use and analysis. During or after the deduction, the performance of the model can be evaluated, that is, the difference between the deduction results and the actual observation data is compared to evaluate the accuracy and reliability of the model. Based on the evaluation results, the model can be further optimized and adjusted to improve its performance. This may include adjusting the model architecture, updating training data, improving preprocessing methods, etc.

[0055] Based on the above analysis, step S300 is the core link of the deduction based on the multi-channel parameter inversion model, which involves data preparation, model loading and initialization, deduction process, result output and post-processing, performance evaluation and optimization, etc. By carefully designing and implementing this step, high-quality thermosphere and ionosphere parameter deduction results can be generated, providing strong support for space weather modeling and prediction.

[0056] In one embodiment of the present invention, the three-dimensional thermosphere and ionosphere parameter inversion method based on electron density further includes:

[0057] Step S400, adjusting and optimizing the multi-channel parameter inversion model by a first preset method;

[0058] Specifically, adjusting and optimizing the multi-channel parameter inversion model by the first preset method is a key step to ensure the accuracy and performance of the model deduction. Specifically, this process first uses the first preset method, such as but not limited to model evaluation index analysis, cross-validation, gradient descent algorithm, etc., to comprehensively analyze the multi-channel parameter inversion model that has been constructed. By analyzing the difference between the output results of the model and the real data, a detailed analysis result is generated to identify the areas of insufficiency or overfitting in the model. Based on these analysis results, the model parameters that need to be adjusted and optimized are further determined. These parameters, for example, involve the weights, biases, learning rates, batch sizes, etc. of the neural network, as well as architecture adjustments specific to the model, such as increasing or decreasing hidden layers, adjusting the number of neurons, etc. The process of adjustment and optimization is carried out step by step, and the selected model parameters are gradually adjusted through repeated trials and verifications in order to achieve the best deduction effect. The goal of this step is to enable the multi-channel parameter inversion model to better adapt to the characteristics of the input data and improve the accuracy and reliability of the deduction results by finely adjusting the model parameters. By continuously optimizing the model, we can ensure that it can maintain a high degree of stability and accuracy when dealing with complex and changing thermosphere and ionosphere parameters, thereby providing more accurate and reliable initial driving data for space weather modeling. The implementation of this step not only improves the performance of the model, but also enhances its applicability and value in practical applications.

[0059] Among them, the first preset method is a comprehensive model optimization process, which aims to improve the prediction accuracy and generalization ability of the multi-channel parameter inversion model through comprehensive analysis and fine adjustment. The specific steps are as follows: Use a part of the preprocessed data set as a validation set to evaluate the performance of the model. Evaluation indicators include but are not limited to mean square error (MSE), root mean square error (RMSE), correlation coefficient (R), etc. to quantify the prediction accuracy of the model. Analyze the performance of the model in different geographical regions (such as the equator, polar regions, mid-high latitudes, etc.) and different time periods (such as day and night, seasonal changes, etc.), and identify areas and time periods with weaker model performance. Since the data distribution and convergence speed of each channel in the multi-channel output model may be different, it is necessary to set appropriate weights for the evaluation indicators of each output channel. For channels that converge faster, appropriately reduce their weights (such as 0.6 to 0.8) to balance the convergence speed of each channel and ensure that the model can optimize the parameters of all channels simultaneously. On the contrary, for channels that converge more slowly, their weights are appropriately increased (e.g., 1.1 to 1.3) so that the model focuses more on the optimization of these channels during training. Use grid search, random search, or Bayesian optimization to tune the model's hyperparameters, such as learning rate, batch size, and regularization coefficient. According to the performance of the model on the validation set, gradually adjust the hyperparameters until the optimal parameter combination is found. To prevent the model from overfitting, apply L1 or L2 regularization techniques to constrain the model's weights. During or after model training, prune the model weights according to their importance, remove unimportant neurons or connections, to reduce the model's complexity and improve generalization. During training, monitor the performance changes of the model on the validation set. If the performance of the model on the validation set does not improve significantly or even decreases over multiple consecutive epochs, terminate the training using an early stopping strategy. The early stopping strategy helps prevent the model from overfitting on the training set while retaining the model's best performance on the validation set. The above steps are used as an iterative process, and the model evaluation, weight adjustment, hyperparameter tuning, regularization and pruning steps are repeated continuously. After each iteration, the performance of the model on an independent test set is evaluated to ensure that the generalization ability of the model is improved.

[0060] In this embodiment, step S100 includes:

[0061] Step S101, obtaining various key parameters of the global thermosphere and ionosphere as raw data based on the TIEGCM model;

[0062] Specifically, step S101 is a key step in obtaining raw data in an embodiment of the present invention. This step mainly relies on TIEGCM (Thermosphere-Ionosphere-Electrodynamics General Circulation Model) to obtain various key parameters of the thermosphere and ionosphere on a global scale. TIEGCM is a numerical model for simulating the dynamic changes of the earth's thermosphere and ionosphere. It can comprehensively consider a variety of physical processes, such as solar radiation, the earth's magnetic field, atmospheric circulation, etc., so as to accurately simulate various parameters of the thermosphere and ionosphere. In this step, various key parameters of the thermosphere and ionosphere can be obtained through the TIEGCM model. These parameters include, for example, but are not limited to: neutral atmospheric components (such as O, O 2 、N 2 etc.), the density of ions (such as O + , O 2 + etc.) density, electron temperature (T e ), ion temperature (T i ) and neutral gas temperature (T n ), and the meridional wind (U n ), zonal wind (V n ), Vertical Wind (W n ) and other velocity parameters, as well as the ion drift velocity (U i E×B 、V i E×B , W i E×B ) and so on. By running the TIEGCM model, the global thermosphere and ionosphere states within a specified time range can be simulated. Then, the required parameter data is extracted from the simulation results. These parameter data usually have global coverage and can provide complete information from low latitudes to high latitudes and from low altitudes to high altitudes. The data is usually stored in a grid format, with each grid point corresponding to a geographic location and a set of parameter values. The obtained parameter data will be used as the raw data for constructing a multi-channel parameter inversion model in subsequent steps. By training the model, high-quality deductions of these parameters can be achieved, thereby more accurately understanding the dynamic changes of the thermosphere and ionosphere. For example, it is necessary to obtain parameter data for the global thermosphere and ionosphere on March 1, 2023. The TIEGCM model can be run and the corresponding simulation parameters (such as time, geographic location, physical process, etc.) can be set. After the simulation is completed, the required parameter data, such as O, O 2 、N 2 The density of O + , O 2 +The density, T e , T i , T n The temperature, and U n 、V n , W n These data will be used as input data for the subsequent construction of a multi-channel parameter inversion model.

[0063] Based on the above analysis, it can be seen that step S101 is a key step in obtaining raw data in the embodiment of the present invention, which relies on the TIEGCM model to simulate and obtain various key parameters of the thermosphere and ionosphere on a global scale. These parameter data will serve as the basis for the subsequent construction and training of the multi-channel parameter inversion model.

[0064] Step S102, preprocessing the original data by a second preset method to generate the preprocessed data set;

[0065] Wherein, the pre-processed data set includes electron density data, spatiotemporal information and omni data;

[0066] Specifically, step S102 is a key link in data preprocessing, and its goal is to process the raw data through a specific preset method (second preset method) to generate a data set suitable for subsequent model training and deduction. The preprocessed data set contains electron density data, spatiotemporal information and omni data, which are the basis for building a multi-channel parameter inversion model. Electron density data is the main input parameter, which is obtained by global three-dimensional ionospheric electron density measurement. Electron density is an important parameter in the ionosphere, and its change is closely related to other parameters of the thermosphere and ionosphere. Spatiotemporal information includes time (such as date, timestamp) and space (such as geographic location, altitude) information. This information is crucial to understanding the dynamic changes of the thermosphere and ionosphere because they vary with time and space. Omni data refers to data related to solar activity, such as solar wind speed, solar wind density, solar high-energy particle flow, etc. These data have a significant impact on the state of the ionosphere and are therefore included in the preprocessed data set. The purpose of preprocessing includes, for example, data cleaning (removing outliers, filling missing values), data conversion (such as standardization, normalization) and data integration (integrating data from different sources into a unified format and framework). These processing steps help improve the consistency and analyzability of data, thereby providing a reliable basis for subsequent model construction and deduction. For example, there is a set of raw data obtained based on the TIEGCM model, including a variety of key parameters of the thermosphere and ionosphere on a global scale, such as oxygen atomic density, oxygen molecular density, electron temperature, etc. These data may have the following problems when they are obtained: 1. Missing data: In some areas or time periods, some data may be missing due to limitations of observation equipment or data transmission problems; 2. Data anomalies: Due to instrument failure or environmental factors (such as abnormal solar activity), some data may have abnormal values; 3. Inconsistent data formats: Data from different sources may use different formats and units and need to be processed uniformly. To address these problems, the following preprocessing steps can be taken: Data cleaning: For missing values, interpolation can be performed based on data from adjacent areas or time points; for abnormal values, they can be eliminated or corrected based on physical laws or statistical methods. Data conversion: In order to eliminate the impact of dimensions between different parameters, the data can be standardized or normalized. For example, the electron density data can be converted into a deviation or percentage relative to a certain benchmark value. Data integration: The processed data is integrated into a unified database or data file to facilitate subsequent model training and deduction. At the same time, spatiotemporal information and omni data can be added as additional input features to enrich the content of the data set and improve the accuracy of the model. Through these preprocessing steps, a preprocessed data set containing electron density data, spatiotemporal information, and omni data can be generated, which will serve as the input of the subsequent multi-channel parameter inversion model. In this way, this high-quality data set can be used to train the model and achieve high-precision deduction of thermosphere and ionosphere parameters.

[0067] Among them, the second preset method is a systematic data preprocessing process, which aims to improve the quality and consistency of the original data and provide a reliable data foundation for subsequent model construction and deduction. The specific steps are as follows: Check whether there are missing values ​​(such as NaN or null values) in the original data, and use appropriate filling strategies (such as mean filling, median filling, mode filling or interpolation based on the trend of the previous and next data) to handle these missing values ​​according to the actual situation. Identify outliers through statistical methods (such as standard deviation method, IQR interquartile range method) or based on business logic, and decide whether to retain, correct or delete these outliers. Check and delete or merge duplicate records in the data to ensure the uniqueness of each record. Convert the data types of all fields to a form suitable for analysis, such as converting date strings to date formats, converting text data to categorical variables, etc. Ensure that the dimensions of all input data are consistent, such as converting values ​​of different magnitudes to the same scale through scaling, normalization, etc. For numerical data that conforms to the normal distribution, the Z-score standardization method is used to adjust the mean of each feature to 0 and the variance to 1. For data that does not conform to the normal distribution or has large dimensional differences, the minimum-maximum normalization method is used to scale the data to the interval [0,1]. In the present invention, for example, normalization can be used for altitude data. Based on domain knowledge and preliminary data analysis results, features useful for model prediction are selected. In the present invention, electron density data, spatiotemporal information (longitude, latitude, time) and OMNI data are mainly selected as input features. The preprocessed data set is divided into a training set, a validation set and a test set. Usually, the training set is used for model training, the validation set is used for model tuning (such as hyperparameter selection, early stopping, etc.), and the test set is used to finally evaluate the model performance. For categorical variables, one-hot encoding (One-Hot Encoding), label encoding (Label Encoding) or other appropriate encoding methods are used to convert them into numerical data. For time series data, time window division, sliding window sampling and other processing are required to meet the input requirements of the model.

[0068] In this embodiment, step S200 includes:

[0069] Step S201, designing the architecture of the multi-channel parameter inversion model, and determining the input layer, hidden layer and output layer;

[0070] Specifically, the design of the model architecture is based on the understanding of the physical processes of the thermosphere and ionosphere, and the application of deep learning technology in parameter inversion. Considering the complexity and interrelationship of the thermosphere and ionosphere parameters, the model needs to be able to process multiple input parameters at the same time and output multiple related physical quantities. In the present invention, the model architecture is designed as a deep learning neural network comprising an input layer, multiple hidden layers and an output layer. This architecture allows the model to extract complex features from the input data, and perform nonlinear transformations through the hidden layers to finally generate outputs. The input layer is the starting point of the model and is responsible for receiving the preprocessed data set. In the present invention, the data received by the input layer includes electron density data, spatiotemporal information and omni data. Electron density data is the main input parameter because it is an ionospheric parameter that is easy to measure and has strong global coverage. Spatiotemporal information includes timestamps and geographic location information, which is crucial for simulating the dynamic changes of the earth's atmosphere. Omni data may, for example, include other related physical parameters, such as solar activity index, geomagnetic field changes, etc., which may affect the state of the thermosphere and ionosphere. The number of neurons in the input layer is determined according to the dimension of the input data to ensure that the complexity of the input information can be fully represented. The hidden layer is the middle layer in the model, which is responsible for nonlinear transformation and feature extraction. In the present invention, the design of the hidden layer is intended to capture the complex relationship between the input parameters and the output parameters. The number of hidden layers and the number of neurons in each layer are key parameters for model design. Too many hidden layers may lead to overfitting of the model, while too few hidden layers may not be able to fully capture the complexity of the data. Through experiments and comparisons, the present invention selects the appropriate number of hidden layers and neurons to achieve the best performance. The selection of activation function is also an important aspect of hidden layer design. In the present invention, the ReLU function is selected as the activation function because it can accelerate training and alleviate the gradient disappearance problem. The output layer is the end point of the model and is responsible for generating the final deduction results. In the present invention, the output layer outputs 18 important parameters of the thermosphere and ionosphere. The number of neurons in the output layer corresponds to the number of parameters that need to be deduced. Each neuron represents an output parameter, and its value is calculated by the model based on the transformation of the input data and the hidden layer. Taking into account the correlation between different output parameters and the difference in convergence speed, the present invention sets appropriate weights for the evaluation indicators of each output channel during the training process to ensure that each parameter can converge synchronously and maintain physical consistency.

[0071] Based on the above analysis, step S201 provides a solid foundation for subsequent model training and deduction by designing the architecture of the multi-channel parameter inversion model and determining the specific configuration of the input layer, hidden layer, and output layer. This design fully utilizes the advantages of deep learning technology and achieves high-quality deduction of thermosphere and ionosphere parameters.

[0072] Step S202, constructing a preliminary model based on a preset deep learning framework and algorithm;

[0073] Specifically, the main purpose of step S202 is to build a model that can preliminarily deduce thermosphere and ionosphere parameters based on the designed multi-channel parameter inversion model architecture, using a deep learning framework and algorithm, and this preliminary model is the basis for subsequent training and optimization. Selecting a suitable deep learning framework is the basis for building a preliminary model. Commonly used deep learning frameworks include, for example, TensorFlow, PyTorch, etc. These frameworks provide a rich library of algorithms, data processing tools, and model training functions, which can support the construction and training of complex neural networks. When building a preliminary model, it is necessary to select a suitable algorithm. This includes the type of neural network (such as feedforward neural network, convolutional neural network, etc.), activation function (such as ReLU, Sigmoid, etc.), optimizer (such as Adam, SGD, etc.), etc. The choice of algorithm directly affects the performance and training effect of the model. In step S201, the architecture of the multi-channel parameter inversion model has been designed, and the input layer, hidden layer, and output layer have been determined. In this step, it is necessary to build a model according to the designed architecture, including setting the number of neurons in each layer, the connection mode, etc. According to actual needs and hardware conditions, select a suitable deep learning framework and algorithm. Considering the complexity of the model and the training efficiency, you can choose a framework that supports distributed training and has a rich algorithm library. According to the designed model architecture, build the model in the deep learning framework. This includes defining the structure of the input layer, hidden layer, and output layer, setting the number of neurons in each layer, the activation function, etc. At the same time, the model parameters need to be initialized to prepare for subsequent training. After the model is built, it is necessary to compile the model and set the training parameters such as loss function and optimizer. The selection of these parameters directly affects the training effect and convergence speed of the model. After the model is compiled, a small amount of preprocessed data sets can be used for preliminary testing to verify the feasibility and performance of the model. This helps to discover potential problems and deficiencies and provide guidance for subsequent optimization. Among them, before building the preliminary model, the original data needs to be preprocessed to generate a preprocessed data set. This includes operations such as data cleaning, standardization, and normalization to ensure the consistency and analyzability of the data. The preprocessed data set is the basis for building and training the model. Among them, when building the preliminary model, some hyperparameters such as learning rate and batch size need to be set. The choice of these hyperparameters has a significant impact on the performance and training results of the model. Therefore, it is necessary to determine the best hyperparameter combination through experiments and tuning. After the initial test, the model needs to be evaluated to determine whether it meets the expected performance requirements. If the model performance is poor, it is necessary to analyze the reasons and optimize it. This may include adjusting the model architecture, changing the algorithm, increasing the amount of training data, etc.

[0074] Based on the above analysis, by selecting a suitable framework and algorithm, building a model according to the designed model architecture, and conducting preliminary testing and evaluation, we can build a model that can preliminarily deduce the thermosphere and ionosphere parameters, laying the foundation for subsequent training and optimization.

[0075] Step S203, training the preliminary model based on the preprocessed data set to construct the multi-channel parameter inversion model;

[0076] Specifically, before this step, the raw data has been acquired and preprocessed by step S101 and step S102. The preprocessed data set contains key parameters such as electron density data, spatiotemporal information and omni data, which are the basis of model training. The preliminary model has been designed and constructed in step S201 and step S202. The model architecture includes an input layer, a hidden layer and an output layer, and a suitable activation function (such as ReLU function) and an optimizer (such as Adam optimizer) have been selected. In step S203, the preprocessed data set is input into the preliminary model for training. The training process is to optimize the model parameters by iteration so that the model can gradually learn the rules and features in the data set. During the training process, the model will calculate the loss function according to the input data and the expected output (i.e., the other 18 important parameters in the preprocessed data set), and update the model parameters by the back propagation algorithm to minimize the loss function. Due to the certain correlation and imbalance between the output parameters, it is necessary to set a suitable weight for the evaluation index of each output channel during the training process to ensure that each channel can converge synchronously and maintain the correlation between the parameters. During the training process, the performance of the model needs to be continuously evaluated, including accuracy, convergence speed, etc. If the model performance is found to be poor, the model architecture, hyperparameters or learning strategies may need to be adjusted. After the training is completed, the model needs to be comprehensively evaluated, including comparison with the output of the TIEGCM numerical simulation model to ensure that the physical characteristics and evolution laws of the model are consistent with the actual situation. The multi-channel parameter inversion model after training can receive new electron density data as input and simultaneously output 18 other important parameters of the thermosphere and ionosphere.

[0077] Based on the above analysis, step S203 is the core step in the entire parameter inversion method. It builds a multi-channel parameter inversion model by training a preliminary model, and realizes the deduction from electron density data to other key parameters. The success or failure of this step directly determines the performance and accuracy of the entire method.

[0078] In this embodiment, step S400 includes:

[0079] Step S401, analyzing the multi-channel parameter inversion model by the first preset method to generate analysis results;

[0080] Specifically, this step comprehensively analyzes the multi-channel parameter inversion model that has been constructed to evaluate its performance, identify potential problems, and provide a basis for subsequent adjustment and optimization. The first preset method here may include, for example, a series of analysis tools and techniques, such as cross-validation, model evaluation index calculation (such as accuracy, recall rate, F1 score, etc.), residual analysis, feature importance evaluation, etc. For the thermosphere and ionosphere parameter inversion model, special attention is paid to, for example, the prediction ability of the model for different regions and different time periods in the world, and the prediction accuracy and consistency of the model for each output channel (i.e., 18 kinds of important parameters). A part of the pre-processed data set is used as a validation set, and by comparing the model prediction results with the true value, global evaluation indicators such as mean square error (MSE), root mean square error (RMSE), etc. are calculated to evaluate the overall performance of the model. The performance differences of the analysis model in different geographical regions (such as the equator, the polar regions, mid-high latitudes, etc.) and different time periods (such as day and night, seasonal changes, etc.) are identified, and the model performs poorly in which regions or time periods. Evaluate the prediction accuracy of the model for 18 output parameters separately, identify which parameters are more difficult to predict and which parameters have more accurate prediction results. Analyze the residual distribution between the model prediction value and the true value, identify whether there are systematic deviations or outliers, and the possible causes of these deviations or outliers. Evaluate the influence of input features (such as electron density, spatiotemporal information, omni data, etc.) on the model prediction results, and identify which features contribute most to the model performance. Based on the above content, generate a detailed analysis report, including the model's global performance evaluation results, regional and time period performance evaluation results, channel performance evaluation results, residual analysis, and feature importance evaluation, etc. This analysis report will serve as an important reference for subsequent adjustment and optimization of the model.

[0081] Step S402, based on the analysis result, determining the model parameters that need to be adjusted and optimized;

[0082] Specifically, after building and training the multi-channel parameter inversion model, the model needs to be further adjusted and optimized to improve its performance and accuracy. Step S402 is carried out in this context, and its purpose is to determine which model parameters need to be adjusted and optimized by analyzing the output results and performance of the model, so as to further improve the accuracy and efficiency of the model. Use a part of the pre-processed data set as a validation set to evaluate the performance of the model. Evaluation indicators may include mean square error (MSE), root mean square error (RMSE), correlation coefficient (R), etc., which are used to quantify the prediction accuracy of the model. Compare the prediction results of the model with the real data (i.e., the target value in the pre-processed data set), and analyze the deviation and error distribution of the prediction results. Identify the part with larger error in the prediction result, such as a specific time, spatial region or specific parameter. By changing some parameters in the model, observe the change in model performance to determine which parameters have a greater impact on model performance. This can be done by methods such as grid search, random search or Bayesian optimization. If the overall performance of the model is not good, it may be necessary to adjust the architecture of the model, such as increasing or decreasing hidden layers, changing the number of neurons, etc. If the prediction accuracy of certain specific parameters is low, it may be necessary to adjust the input features or output weights of the model for these parameters. For the part with larger errors in the prediction results, analyze the possible reasons, such as data sparseness, model overfitting or underfitting, etc. According to the analysis results, determine the parameters that need to be adjusted, such as adding regularization terms, adjusting the learning rate, changing the activation function, etc. Select the parameters that have a greater impact on the performance of the model for adjustment and optimization. You can gradually adjust the parameter values ​​and observe the changes in model performance to determine the optimal parameter values. After determining the model parameters that need to be adjusted and optimized, you can enter step S403, gradually adjust the selected model parameters, and adjust and optimize. In this process, it is necessary to continuously evaluate the performance of the model to ensure that the direction of adjustment and optimization is correct.

[0083] Step S403, gradually adjusting the selected model parameters and performing adjustment optimization;

[0084] Specifically, in step S402, the model parameters that need to be adjusted and optimized have been determined based on the analysis results. These parameters may include, for example, the weights, biases, learning rates, batch sizes, etc. of the neural network. "Gradual adjustment" means that this process is not a one-time process, but requires multiple iterations and experiments. After each adjustment, the performance of the model needs to be re-evaluated to determine whether the optimization goal has been achieved. The adjusted parameter values ​​are usually fine-tuned within a predefined range to avoid excessive adjustments that cause the model performance to drop sharply. "Adjustment and optimization" is the core goal of step S403. By gradually adjusting the parameters, it is expected that a set of optimal parameter combinations can be found so that the performance of the model on a given task is optimal. During the optimization process, various performance indicators, such as loss function value, accuracy, recall rate, etc., are used to evaluate the pros and cons of the model. For the present invention, special attention may be paid to the accuracy and stability of the model in inverting the thermosphere and ionosphere parameters. In actual operation, the adjustment and optimization process may require the use of automated tools or algorithms, such as grid search, random search, or Bayesian optimization, to efficiently explore the parameter space. After each parameter adjustment, the model needs to be retrained and its performance evaluated. This may require a lot of computing resources and time, so reasonable planning is required in actual operation. The adjustment and optimization process is an iterative process until the preset optimization goal is reached or the performance is no longer significantly improved. During the iteration process, it is necessary to pay close attention to the performance changes of the model and the impact of adjusting parameters on performance so as to adjust the optimization strategy in time. After completing the parameter adjustment and optimization, the model needs to be fully verified and tested to ensure its stability and generalization ability in different scenarios. The verification and testing process may include using an independent test data set, performing cross-validation or leave-one-out validation, etc.

[0085] Based on the above analysis, we can further improve the performance and accuracy of the model by carefully adjusting and optimizing the model parameters, thereby achieving high-quality deduction of thermosphere and ionosphere parameters. This process requires the use of automated tools, algorithms, and a large amount of computing resources and time to ensure that the final model can meet the needs of practical applications.

[0086] In a specific embodiment of the present invention, the global three-dimensional ionospheric electron density (Ne) is an ionospheric parameter with a relatively mature measurement technology. Its measurement methods include the incoherent scatter radar (ISR) emitting radar waves, satellite observation of occultation technology, etc., with various measurement forms, high data accuracy, and strong global coverage. The present invention aims to use the high-precision ionospheric three-dimensional electron density data that is easy to measure and has high accuracy to deduce 18 other important parameters of the thermosphere-ionosphere with high credibility by training a neural network model, so as to provide more reliable initial driving data for the space weather model. Specifically, if Figure 5 As shown, the three-dimensional thermosphere and ionosphere parameter inversion method based on electron density in this embodiment is divided into three stages:

[0087] Stage 1, training data preparation:

[0088] In this stage, TIEGCM is used to obtain various key parameters of the thermosphere and ionosphere at a specified time and in the global range as raw data. The data is then preprocessed to solve the problem of outliers in the boundary area caused by the range of physical parameter values, and the simulation data of irregular grids are converted into regular grids to improve the consistency and analyzability of the data.

[0089] Data preprocessing and data cleaning include standardization of the main input electron density, omni data and output channels, and normalization of altitude (because they are all integers and have a small range of values); data cleaning mainly removes missing values ​​(represented by nan in the original data). Missing values ​​only appear at the highest altitude (700km) and can be directly deleted. Ensure the consistency and integrity of the parameter data at the same time. Prepare for the next stage of model construction, effectively reduce the impact of outliers on simulation results, and improve the reliability of the overall model.

[0090] Phase 2, thermosphere-ionosphere multi-channel parameter inversion model architecture and training:

[0091] The internal structure of the neural network of the thermosphere-ionosphere multi-channel parameter inversion model is as follows: Figure 4 As shown. The model architecture consists of: input layer (17-256), hidden layer 1 (256-128), hidden layer 2 (128-64), output layer (64-18). The input layer inputs 17 parameters, including spatiotemporal information, electron density and omni data. 256 is a suitable number of neurons selected after comparative testing; the hidden layer gradually reduces the number of neurons until the output layer outputs the value of 18 channels. The activation function selects the ReLU function, which can accelerate training and alleviate the gradient disappearance problem. It is applied after each layer except the output layer; the optimizer uses the Adam optimizer, which combines the advantages of momentum and adaptive learning rate and is relatively versatile.

[0092] There is a certain correlation between the parameters of the thermosphere-ionosphere. Compared with single-channel output, multi-channel output can avoid losing the correlation between parameters. However, multi-channel output increases the number of parameters of the neural network, which is more likely to lead to overfitting; the data distribution of different channels is uneven, which may cause the corresponding parameters of channels with more comprehensive data and stronger regularity to reach the local optimum too early, resulting in a decrease in model performance. Targeted optimization is needed for this. In the initial training stage, due to the different strengths of the regularity of each output channel, its convergence speed is also fast or slow, which is not conducive to the overall performance of the model. When the convergence speed of some channels is faster, the model will pay more attention to the update of the relevant parameters of these channels, while ignoring the characteristics of the channels with slower convergence. Therefore, in the further training process, appropriate weights are set for the evaluation indicators of each output channel so that their convergence speeds are close to synchronization, and the correlation between the channels of the thermosphere-ionosphere is retained to a greater extent.

[0093] Specifically, the 18 output channels include two categories, one is scalar, including the density of atoms, molecules, and ions; the other is vector, including the velocity of the fluid in the longitudinal, latitudinal, and vertical directions (the three directions are a group); the closeness of the connection between these channels is different, and the regularity of global distribution is also different. Close connection: For example, the velocity of the three directions of neutral wind is highly correlated; oxygen molecules, oxygen atoms, and oxygen ions are closely related, but the correlation with other channels is relatively weak. Global distribution regularity: For example, the electron density is larger in the sun-irradiated area and smaller in other areas. It is a parameter with strong regularity, which is easy for the neural network to learn the law. Due to these two differences, the convergence speed of each channel is different, which is reflected in the size of the loss value during the neural network training process. During the stochastic gradient descent process, the Adam optimizer will try to choose the parameter update direction with a low loss value. The strategy for adjusting weights is to set smaller weights (0.6 to 0.8) for channels that converge faster, and larger weights (1.1 to 1.3) for channels that converge faster. In this way, during training, channels that are difficult to converge have a greater impact on the loss, and parameter updates will focus on these channels, eventually adjusting to a similar convergence speed for each channel.

[0094] Stage 3, fine-tuning of thermosphere-ionosphere multi-channel parameter inversion model:

[0095] This stage is a further performance optimization after the model architecture is completed. The model accuracy is further improved by adjusting the network depth, model hyperparameters, and learning strategies. After multiple tests of deeper and shallower neural networks, it was found that a combination of 4 layers (17-256-128-64-18) of network depth is more suitable. For hyperparameters, the mainstream adjustment strategy is adopted. The learning rate is gradually reduced during the training process (1e-3 to 1e-5), and the batch size is adjusted according to the video memory. The learning strategy is to gradually reduce the learning rate so that the optimization parameters are reduced when the model is close to convergence, thereby improving the final accuracy.

[0096] At the same time, the output of the numerical simulation model of TIEGCM is compared and evaluated to ensure that the overall physical characteristics are accurate and conform to the laws of evolution.

[0097] Based on the above content, it can be seen that the present invention uses deep learning technology, based on the global three-dimensional ionospheric electron density (Ne) data, to achieve the simultaneous deduction of 18 other important parameters of the thermosphere-ionosphere, while maintaining the correlation between the parameters; through multi-channel output design, the model can output all parameters synchronously, avoiding the sequential operation relationship between the output channels in traditional numerical simulations, thereby improving the reasoning speed and reducing the computational overhead; and the newly acquired data can be used efficiently and quickly, which is easier to upgrade and iterate, and has stronger timeliness, overcoming the problems of high computational complexity and delayed update of traditional theoretical models.

[0098] The more mature existing technology is the International Reference Ionospheric Model (IRI), which relies on complex numerical simulations of physical processes and must strictly calculate the output of each channel in sequence, resulting in high computational complexity. IRI contains multiple sub-models, all of which are data-sensitive numerical simulation models. To update with the help of new data or new theories, a large amount of data processing and model docking are required, which results in a low update frequency of IRI and inability to fully utilize timely data.

[0099] Compared with the prior art, the present invention makes full use of the easily measurable global three-dimensional ionospheric electron density data to achieve high-quality deduction of thermosphere and ionospheric parameters, uses deep learning neural networks to explore the relationship between channels that may be ignored by numerical calculations, and replaces sequential output with synchronous output of each channel. The reasoning speed is significantly accelerated, the computing cost is significantly reduced, and the model can be quickly updated and iterated to maintain timeliness. This multi-channel synchronous output design overcomes the limitations of sequential operations in traditional numerical simulations, improves the reasoning speed and overall performance. Compared with traditional models that rely on past data updates and have a long update cycle (such as the IRI model), this method can adapt to time-sensitive data more quickly and achieve efficient and accurate inversion of local phenomena such as short-term disturbances.

[0100] Embodiment 2

[0101] Figure 2is a schematic diagram of the structure of a three-dimensional thermosphere and ionosphere parameter inversion system based on electron density in the second embodiment of the present invention, such as Figure 2 As shown, embodiment 2 provides a three-dimensional thermosphere and ionosphere parameter inversion system based on electron density, including: a first generation module 201, a construction module 202 and a second generation module 203. The first generation module 201 is used to obtain raw data and perform preprocessing to generate a preprocessed data set. The construction module 202 is used to construct a multi-channel parameter inversion model based on the preprocessed data set. And the second generation module 203 is used to perform deduction based on the multi-channel parameter inversion model to generate a deduction result.

[0102] In this embodiment, the electron density-based three-dimensional thermosphere and ionosphere parameter inversion system further includes an adjustment and optimization module 204 for adjusting and optimizing the multi-channel parameter inversion model by a first preset method.

[0103] In this embodiment, the first generation module 201 includes: an acquisition unit and a first generation unit. The acquisition unit is used to acquire multiple key parameters of the global thermosphere and ionosphere as raw data based on the TIEGCM model; and the first generation unit is used to preprocess the raw data by a second preset method to generate the preprocessed data set. The preprocessed data set includes electron density data, spatiotemporal information and omni data.

[0104] In this embodiment, the construction module 202 includes: a first determination unit, a first construction unit, and a second construction unit. The first determination unit is used to design the architecture of the multi-channel parameter inversion model and determine the input layer, the hidden layer, and the output layer. The first construction unit is used to construct a preliminary model based on a preset deep learning framework and algorithm. The second construction unit is used to train the preliminary model based on the preprocessed data set to construct the multi-channel parameter inversion model.

[0105] In this embodiment, the adjustment and optimization module 204 includes:

[0106] A second generating unit, configured to analyze the multi-channel parameter inversion model by using the first preset method to generate an analysis result;

[0107] A second determining unit, used to determine the model parameters that need to be adjusted and optimized based on the analysis result;

[0108] The adjustment and optimization unit is used to gradually adjust the selected model parameters and perform adjustment and optimization.

[0109] The various variations and specific examples of the three-dimensional thermosphere and ionosphere parameter inversion method based on electron density provided in Example 1 are also applicable to the three-dimensional thermosphere and ionosphere parameter inversion system based on electron density provided in this embodiment. Through the above detailed description of a three-dimensional thermosphere and ionosphere parameter inversion method based on electron density, those skilled in the art can clearly know the implementation method of a three-dimensional thermosphere and ionosphere parameter inversion system based on electron density in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here.

[0110] Embodiment 3

[0111] Figure 3 is a schematic diagram of the structure of an electronic device in Embodiment 3 of the present invention, such as Figure 3 As shown, the third embodiment further provides an electronic device 300 , which may include: a processor 301 and a memory 302 .

[0112] The memory 302 is used to store programs; the memory 302 may include volatile memory (English: volatile memory), such as random-access memory (English: random-access memory, abbreviated: RAM), such as static random-access memory (English: static random-access memory, abbreviated: SRAM), double data rate synchronous dynamic random access memory (English: Double Data Rate Synchronous Dynamic Random Access Memory, abbreviated: DDR SDRAM), etc.; the memory may also include non-volatile memory (English: non-volatile memory), such as flash memory (English: flash memory). The memory 302 is used to store computer programs (such as applications, functional modules, etc. that implement the above method), computer instructions, etc., and the above computer programs, computer instructions, etc. can be partitioned and stored in one or more memories 302. And the above computer programs, computer instructions, data, etc. can be called by the processor 301.

[0113] The above-mentioned computer programs, computer instructions, etc. may be stored in partitions in one or more memories 302 . And the above-mentioned computer programs, computer instructions, etc. may be called by the processor 301 .

[0114] The processor 301 is used to execute the computer program stored in the memory 302 to implement each step of the method involved in the above embodiment.

[0115] For details, please refer to the relevant description in the previous method embodiment.

[0116] The processor 301 and the memory 302 may be independent structures or integrated structures. When the processor 301 and the memory 302 are independent structures, the memory 302 and the processor 301 may be coupled and connected via a bus 303 .

[0117] The electronic device of this embodiment can execute the technical solution in the above method, and its specific implementation process and technical principle are the same, which will not be repeated here.

[0118] Embodiment 4

[0119] Embodiment 4 also provides a computer-readable storage medium, including a computer program and instructions. When the computer program or instructions are executed on a computer, the computer executes the three-dimensional thermosphere and ionosphere parameter inversion method based on electron density of any embodiment of the present invention.

[0120] Computer-readable storage media include: USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical disks, and other media that can store program codes.

[0121] This embodiment also provides a computer program product, which includes: a computer program, which is stored in a readable storage medium. At least one processor of an electronic device can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the electronic device executes the solution provided by any of the above embodiments.

[0122] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the disclosure of the present invention can be performed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and this document does not limit this.

[0123] In summary, the three-dimensional thermosphere and ionosphere parameter inversion method and system based on electron density of the present invention make full use of the easily measurable global three-dimensional ionospheric electron density data to achieve high-quality deduction of thermosphere and ionosphere parameters, and use synchronous output of each channel instead of sequential output, which significantly speeds up the reasoning speed, significantly reduces the computing cost, and can quickly update and iterate the model to maintain timeliness.

[0124] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A three-dimensional thermosphere and ionosphere parameter inversion method based on electron density, characterized in that: include: Obtain the original data and preprocess it to generate a preprocessed data set; Based on the preprocessed data set, construct a multi-channel parameter inversion model; Deduction is performed based on the multi-channel parameter inversion model to generate deduction results.

2. The method for inverting three-dimensional thermosphere and ionosphere parameters based on electron density according to claim 1, characterized in that: Also includes: The multi-channel parameter inversion model is adjusted and optimized by a first preset method.

3. The method for inverting three-dimensional thermosphere and ionosphere parameters based on electron density according to claim 1, characterized in that: The obtaining of raw data and preprocessing to generate a preprocessed data set includes: Based on the TIEGCM model, various key parameters of the global thermosphere and ionosphere are obtained as raw data; Preprocessing the raw data by a second preset method to generate the preprocessed data set; The preprocessed data set includes electron density data, spatiotemporal information and omni data.

4. The three-dimensional thermosphere and ionosphere parameter inversion method based on electron density according to claim 1, characterized in that: The method of constructing a multi-channel parameter inversion model based on the preprocessed data set includes: Design the architecture of the multi-channel parameter inversion model and determine the input layer, hidden layer, and output layer; Build a preliminary model based on the preset deep learning framework and algorithm; The preliminary model is trained based on the preprocessed data set to construct the multi-channel parameter inversion model.

5. The three-dimensional thermosphere and ionosphere parameter inversion method based on electron density as claimed in claim 2, characterized in that: The adjusting and optimizing the multi-channel parameter inversion model by the first preset method includes: Analyzing the multi-channel parameter inversion model by the first preset method to generate analysis results; Based on the analysis results, determine the model parameters that need to be adjusted and optimized; The selected model parameters are gradually adjusted and optimized.

6. A three-dimensional thermosphere and ionosphere parameter inversion system based on electron density, characterized in that: include: The first generation module is used to obtain the original data and perform preprocessing to generate a preprocessed data set; A construction module, used for constructing a multi-channel parameter inversion model based on the preprocessed data set; as well as The second generating module is used to perform deduction based on the multi-channel parameter inversion model to generate deduction results.

7. The three-dimensional thermosphere and ionosphere parameter inversion system based on electron density according to claim 6, characterized in that: Also includes: The adjustment and optimization module is used to adjust and optimize the multi-channel parameter inversion model through a first preset method.

8. The three-dimensional thermosphere and ionosphere parameter inversion system based on electron density according to claim 6, characterized in that: The first generation module comprises: An acquisition unit, used for acquiring various key parameters of the global thermosphere and ionosphere as raw data based on the TIEGCM model; and A first generating unit, configured to preprocess the original data by a second preset method to generate the preprocessed data set; The preprocessed data set includes electron density data, spatiotemporal information and omni data.

9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively coupled to the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the three-dimensional thermosphere and ionosphere parameter inversion method based on electron density as described in any one of claims 1-5.

10. A computer-readable storage medium, characterized in that: The invention comprises a computer program and instructions. When the computer program or the instructions are executed on a computer, the computer is enabled to execute the three-dimensional thermosphere and ionosphere parameter inversion method based on electron density as described in any one of claims 1 to 5.

Images