Earth ionosphere electron quantity prediction method

Abstract

The invention provides a method for predicting the electron quantity of an earth ionosphere. According to the technical scheme, global ionosphere electron quantity data obtained from an IGS data center is used for training a neural network model, a principal component analysis method is used for conducting dimension reduction on the global ionosphere electron quantity data, and a new data set is obtained; adding the geomagnetic index and the radiation flux F10.7 data into the new data set through a vector splicing method to form an input data set; inputting the input data set into the neural network model provided by the invention for training to obtain a prediction model; the neural network model provided by the invention is formed by adding the last full connection layer to the neural network of the Informer model. The method is high in prediction precision, prediction data in a real ionosphere space is obtained, and direct subsequent application is facilitated.

Description

Technical Field

[0001] The invention relates to the technical field of earth ionosphere detection, and in particular to a method for predicting the amount of electrons in the earth's ionosphere. Background Art

[0002] The ionosphere usually refers to the Earth's atmosphere above 60 kilometers and below 1,000 kilometers from the surface. Atoms, molecules and other particles in the ionosphere are partially or completely ionized due to the irradiation of radiation from outside the Earth (mainly solar radiation). The amount of electrons in the ionosphere will affect the electromagnetic signals in satellite communications, thereby affecting the quality of satellite communication services. Therefore, predicting the change in the amount of electrons has great application significance. The existing ionospheric electron quantity prediction methods are mainly based on the current electron quantity to predict the electron quantity at the future moment, such as the physical numerical model method (see reference: ADRichmond, ECRidley, and RGRobie. A THERMOSPHERE / IONOSPHERE GENERAL CIRCULATION MODELWITH COUPLED ELECTRODYNAMICS. GEOPHYSICAL RESEARCH LETTERS, VOL. 19, NO. 6, PAGES 601-604, MARCH 20, 1992.) and the empirical model method (see reference: Bilitza, D., Altadill, D., Zhang, Y., Mertens, C., Truhlik, V., Richards, P., McKinnell, L., & Reinisch, B. (2014). The International Reference Ionosphere 2012–a model of international collaboration, J SPACE WEATHER SPAC,4,A7.https: / / doi.org / 10.1051 / swsc / 2014004),The main drawback of these two methods is their low prediction accuracy.

[0003] With the rise of deep learning technology, deep learning models based on neural networks have been applied to the prediction of ionospheric electron content (see the literature: Zhou, H., Zhang, S., Peng, J., Zhang, S., Li, J., Siyu, T., Hui X. & Wancai Z. (2021). Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting. https: / / doi.org / 10.48550 / arXiv.2012.07436). However, the time series model proposed in this paper can only calculate the electron content of a small number of ionospheric grid points at the same time during one operation. Summary of the invention

[0004] The purpose of the present invention is to propose a prediction method for the electron content of the Earth's ionosphere based on a neural network that combines deep learning with principal component analysis, which can simultaneously calculate the electron content of a large number of ionospheric grid points in one operation with high calculation accuracy.

[0005] The technical solution of the present invention is a method for predicting the amount of electrons in the Earth's ionosphere, characterized in that it comprises the following steps:

[0006] The neural network model is trained using global ionospheric electron content data obtained from the IGS data center, which includes the following steps:

[0007] The principal component analysis method is used to reduce the dimension of the global ionospheric electron content data to obtain a new data set; the geomagnetic index and radiation flux F10.7 data are added to the new data set by the vector splicing method to form an input data set; the input data set is input into the neural network model proposed in the present invention for training to obtain a prediction model; the neural network model proposed in the present invention is formed by adding the last fully connected layer to the neural network of the Informer model;

[0008] When predicting the amount of electrons in the Earth's ionosphere, the latest global ionosphere electron data is obtained from the IGS data center, and the principal component analysis method is used to reduce the dimension of the latest global ionosphere electron data to obtain a new data set at the latest moment; the geomagnetic index and radiation flux F10.7 data are added to the new data set at the latest moment by vector splicing to form the input data set at the latest moment; the input data set at the latest moment is input into the prediction model to obtain the prediction result of the amount of electrons in the Earth's ionosphere.

[0009] The beneficial effects of the present invention are: the principal component analysis method is introduced into the prediction of the electron quantity of the earth's ionosphere, and a group of possibly related variables is converted into a group of linearly unrelated variables through orthogonal transformation, that is, the main characteristics of the data can be described with less data. After the ionosphere electron quantity data is reduced in dimension, the relevant parameters are added by using the splicing method, and the data is input into the prediction model proposed by the present invention, and the prediction result after dimensionality increase is directly obtained. This prediction result is not only highly accurate, but also is the data in the real ionosphere space, which is convenient for direct subsequent application. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic diagram of the principle flow of the present invention;

[0011] Figure 2 These are the prediction results of the existing method and the prediction results of the present invention. DETAILED DESCRIPTION

[0012] The present invention is further described below in conjunction with specific implementation methods.

[0013] Figure 1 It is a schematic diagram of the principle flow of the present invention.

[0014] The principal component analysis method is used to reduce the dimension of the global ionospheric electron content data, including steps ① to ⑤:

[0015] Step 1: Standardize data:

[0016] Since the principal component analysis method is affected by the data scale, the global ionospheric electron content data needs to be standardized before the principal component analysis method is performed, that is, each electron content data is subtracted from the mean of all electron content data, and then divided by the variance of all electron content data to obtain the electron content data set. This ensures that the mean of the electron content data set is 0 and the standard deviation is 1.

[0017] Step ②: Calculate the covariance matrix: Calculate the covariance matrix of the electron quantity data set.

[0018] Step 3: Calculate eigenvalues ​​and eigenvectors: Calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix.

[0019] Step ④: Select the principal component: According to the size of the eigenvalue, select the eigenvectors corresponding to the first few largest eigenvalues. These eigenvectors are the principal components, which represent the main change direction in the data. In the embodiment of the present invention, 63 eigenvectors are selected as the principal components.

[0020] Step ⑤: Convert to a new space: Project the global ionospheric electron content data onto the principal components to obtain a new data set. The dimension of this new data set is usually lower than the original data set, achieving dimensionality reduction.

[0021] Then, a data set is formed by vector concatenation, using the following steps: The new data set is represented by a matrix, and the number of rows in the matrix is ​​N, which is the number of eigenvectors of the principal component. The geomagnetic index data obtained from the Tokyo Geomagnetic Center is added to the N+1th row of the matrix, and the solar radiation flux F10.7 at 10.7 cm obtained by the Celestrak satellite is added to the N+2th row of the matrix to obtain the input data set. Each column of the matrix is ​​data at the same time point.

[0022] The obtained input data set is input into the neural network model proposed in the present invention (wherein the dimension of the added fully connected layer is the dimension of the global ionospheric electron content data, which is 5183 in the embodiment of the present invention) for training. In a training embodiment of the present invention, M days of data are used for training, the data of the first M-2 days are input, and the data of the next 2 days are predicted (the data from Monday to Thursday are used to predict the data of Friday and Saturday in the embodiment), and the relevant parameters of the neural network model are obtained. When the value of the loss function does not decrease for three consecutive rounds (the input corresponding to one round of training is all M-2 days of data), the training is stopped, and the obtained model is the optimal neural network model, and the neural network model parameters are stored. The matrix expression of the input neural network is as follows: X = [X 0, ,X1,X2,…X i , X I ], where X0 represents the first column of data at the start time of the first day, and the time interval between each column of data is T. In the embodiment, T1 = 2 hours. The matrix expression of the neural network output is as follows: Y = [Y 0, ,Y1,Y2,…Y j , Y J ], where Y0 represents the first column of data at the starting time of the M-2th day, and the time interval between each column of data is also T.

[0023] When using the neural network, the input to the neural network is 4 days of global ionospheric electron content data obtained from the IGS data center, and the ionospheric electron content data for the next 2 days is predicted.

[0024] Figure 2It is the result of a comparative experiment. Among them, the first row of sub-graphs (a) is the electronic quantity data to be predicted, the second row (b) is the electronic quantity data predicted by the present invention, and Figures (c), (d) and (e) are the prediction results of the first deep learning model (Informer model), the second deep learning model (LSTM model) and the empirical model in the past, respectively, for comparison with the present invention. The experimental results are that the variance of the prediction results of the present invention is 2.35, the variance of the prediction results of the first deep learning model is 2.83, the variance of the prediction results of the second deep learning model is 2.48, and the variance of the prediction results of the empirical model is 4.57. The experimental results prove that the prediction accuracy of the present invention is high.

Claims

1. A method for predicting the amount of electrons in the Earth's ionosphere, characterized in that: The steps include: Obtaining global ionospheric electron content data to train the neural network model includes the following steps: The principal component analysis method is used to reduce the dimension of the global ionospheric electron content data to obtain a new data set; the geomagnetic index and radiation flux F10.7 data are added to the new data set by the vector splicing method to form an input data set; the input data set is input into the neural network model for training to obtain a prediction model; the neural network model is formed by adding the last fully connected layer to the neural network of the Informer model; When predicting the electron content of the Earth's ionosphere, the latest global ionosphere electron content data is obtained, and the principal component analysis method is used to reduce the dimension of the latest global ionosphere electron content data to obtain a new data set at the latest time; the geomagnetic index and radiation flux F10.7 data are added to the new data set at the latest time by the vector splicing method to form the latest input data set; The latest input data set is input into the prediction model to obtain the prediction result of the electron content in the Earth's ionosphere.

2. The method for predicting the amount of electrons in the Earth's ionosphere according to claim 1, characterized in that: During training, M days of global ionospheric electron content data are used for training, the data of the first M-2 days are input, and the data of the next 2 days are predicted.

3. The method for predicting the amount of electrons in the Earth's ionosphere according to claim 2, characterized in that: M is greater than 4 or equal to 4.

Images