CN-122021723-A - Method and device for predicting duration of sub-storm expansion period based on machine learning algorithm

CN122021723ACN 122021723 ACN122021723 ACN 122021723ACN-122021723-A

Abstract

The application discloses a machine learning algorithm-based method for predicting the duration of a sub-storm expansion period, which comprises the steps of receiving sub-storm event data to be detected, inputting a pre-established and trained sub-storm expansion period duration prediction model based on the machine learning algorithm to obtain a sub-storm expansion period duration prediction result, wherein the sub-storm expansion period duration prediction model based on the machine learning algorithm comprises an LSTM layer, an output result of the LSTM layer passes through a convolution layer, and the output result of the convolution layer is input into a full-connection layer, and the predicted sub-storm expansion period duration result is obtained through the full-connection layer. The application also discloses a device for predicting the duration of the sub-storm expansion period, which comprises a memory for receiving and storing data and a program, and a processor for executing the program to realize the method for predicting the duration of the sub-storm expansion period. After the model is trained, the prediction result of the corresponding event time period can be output only by inputting the corresponding time period data into the model without adjusting the parameters, and the modeling prediction is performed on the critical parameters of the whole sub-storm for the first time.

Inventors

LU YANG
SHI KE
ZOU ZIMING

Assignees

中国科学院国家空间科学中心

Dates

Publication Date: 20260512
Application Date: 20251223

Claims (10)

1. The method for predicting the duration of the sub-storm expansion period based on the machine learning algorithm is characterized by comprising the steps of receiving sub-storm event data to be detected, inputting a pre-established and trained sub-storm expansion period duration prediction model based on the machine learning algorithm, and obtaining a sub-storm expansion period duration prediction result; The machine learning algorithm-based sub-storm expansion period duration prediction model comprises an LSTM layer, wherein an output result of the LSTM layer passes through a convolution layer, the output result of the convolution layer is input into a full-connection layer, and a predicted sub-storm expansion period duration result is obtained through the full-connection layer.
2. The method for predicting the duration of the sub-storm expansion period based on the machine learning algorithm according to claim 1, wherein the receiving of the sub-storm event data to be detected comprises the steps of utilizing solar wind and data of an inter-planetary magnetic field as input data of a prediction model, establishing a corresponding relation between the input data and the duration of the sub-storm expansion period, and constructing tag data of the duration of the sub-storm expansion period.
3. The machine learning algorithm-based sub-storm expansion period duration prediction method according to claim 2, wherein constructing the tag data comprises screening out artificially marked sub-storm events and corresponding expansion period aurora image sequences by using extreme ultraviolet measurement data, finding out a time period for continuously increasing the total aurora intensity according to the total aurora intensity, and marking the expansion period duration as the sub-storm expansion period time period.
4. A method of predicting a period of sub-storm surge based on a machine learning algorithm as claimed in any one of claims 1-3 further comprising preprocessing received sub-storm event data to be detected including data outlier processing, missing value filling, and auris total intensity calculation.
5. The method for predicting the duration of the sub-storm surge period based on the machine learning algorithm as claimed in claim 4 wherein training the model for predicting the duration of the sub-storm surge period based on the machine learning algorithm comprises the steps of: Constructing a data set according to the starting time of a sub-storm event list to be observed, extracting solar wind and inter-planetary magnetic field observation data for T minutes from S minutes before the starting time, and constructing as T K matrix tensors, wherein the matrix tensors comprise time sequence data of T minutes and K physical parameters of solar wind and inter-planetary magnetic field observation data, the K physical parameters comprise magnetic field intensity average values, components of magnetic fields in x, y and z directions in an earth geocentric inertial system and an earth geocentric solar magnetic system, root mean square standard deviation of magnetic field intensity scalar quantities, root mean square standard deviation of magnetic field vectors, speed components of solar wind in x, y and z directions, number density of protons in the solar wind, temperature of the protons in the solar wind, dynamic pressure of the solar wind, electric field intensity in the solar wind, plasma parameters, alfen Mach number and magnetoacoustic Mach number, and the corresponding observation data are mapped into a [0,1] interval, so that original distribution characteristics of the data are reserved; Transmitting matrix tensors of the data set into an LSTM layer for calculation; extracting the output of the LSTM layer and performing convolution operation through the convolution layer; And (3) transmitting the output of the convolution layer into the full-connection layer for linear transformation, and mapping the output of the convolution layer to a classification result to obtain the predicted duration of the sub-storm expansion period.
6. The method for predicting the duration of a sub-storm surge period based on a machine learning algorithm as claimed in claim 4 wherein said abnormal value and said missing value of said data are processed by linear interpolation.
7. The method for predicting the duration of the sub-storm expansion period based on the machine learning algorithm according to claim 5, further comprising the step of evaluating a prediction model, wherein the sample data of the marked data set is used as the input of the prediction model of the duration of the sub-storm expansion period, the parameter training duration prediction model is adjusted to obtain a model extraction result and the model extraction result is evaluated, and the evaluation is based on the error and the average error of the predicted duration of the expansion period and the actual duration of the expansion period; The step of evaluating the prediction model further comprises the step of dividing the data set into N equal parts with the same data volume in average, training and evaluating the model each time, taking different parts as test sets and other N-1 equal parts as training sets, repeating training and testing and evaluating for N times, and ensuring that the data set of each equal part is subjected to testing, so that an average result of N times is obtained as a cross-validation result.
8. A machine learning algorithm-based sub-storm inflation period duration prediction device, comprising: The data set unit is used for receiving and storing the sub-storm event data to be detected; The sub-storm expansion period duration prediction model is used for receiving sub-storm event data to be detected after training and obtaining a sub-storm expansion period duration prediction result based on a machine learning algorithm; the machine learning algorithm-based sub-storm expansion period duration prediction model comprises an LSTM layer, wherein an output result of the LSTM layer passes through a convolution layer, the output result of the convolution layer is input into a full-connection layer, and a predicted sub-storm expansion period duration result is obtained through the full-connection layer; and the machine learning algorithm-based substorm expansion period duration prediction model is trained according to the method of claim 5.
9. A machine learning algorithm-based sub-storm inflation period duration prediction device, comprising: a memory for receiving and storing data and a computer program; A processor for executing the computer program to implement the machine learning algorithm-based method for predicting a sub-storm surge period duration according to claim 1.
10. A computer readable storage medium comprising a computer program executable by a processor to implement the method of predicting a sub-storm surge period duration as set forth in claim 1.

Description

Method and device for predicting duration of sub-storm expansion period based on machine learning algorithm Technical Field The application relates to the technical field of application and research of machine learning technology in the field of space physics, in particular to a method and a device for predicting duration of a sub-storm expansion period based on a machine learning algorithm. Background Geomagnetic sub-storm (substorm), hereinafter referred to as sub-storm, is a geomagnetic disturbance event of a sudden burst of energy in the geospatial environment, mainly occurs at the tail part (night side) of a magnetic layer near the earth, and is a transient response of the earth magnetic field to solar energy and wind energy input. Sub-storms occur about 3 to 4 times per day, which, in addition to causing severe changes in the regional environment of the earth, excite strong perturbations in the earth's ionosphere, affecting the reception of GPS navigation signals, and communication between satellites and the earth. Sub-storms are often accompanied by an increase in aurora phenomena, and the most intense region of the earth where aurora intensities are most varied is an elliptical band-shaped region of 65 degrees to 70 degrees north-south latitude, which is connected with the region where the magnetic layer sub-storms are most active. The duration of the sub-storm is typically 2-3 hours, and is typically divided into an initial phase, an expanded phase and a recovery phase. The initial phase of the sub-storm is in an initial state and is characterized in that the local aurora area is rapidly enhanced, and the initial phase generally lasts for about 5-10 minutes. During the sub-explosion expansion phase, the area of the main polar ovum area is obviously expanded, the equatorial boundary and the polar boundary are outwards expanded, and the duration of the sub-explosion expansion phase is usually 10-100 minutes along with the increase of the polar light intensity. In the subriot recovery period, that is, geomagnetic activity is gradually gentle, the aurora intensity and area gradually recover to the normal state. In general, the initiation and evolution of a sub-storm is believed to be highly related to the magnetic layer top-reconnection activity, often in the southerly of the interplanetary magnetic fieldWhen solar energy and wind energy are injected into the earth magnetic layer, the efficiency is high, and the magnetic layer storm is easy to trigger. Other physical parameters such as solar dynamic pressure, velocity and inter-planetary magnetic field strength all affect the evolution of the sub-storm. Sub-storms are an important process of the earth's magnetic layer activity, and observation and modeling of sub-storms can help people recognize the process of the entire magnetic layer activity. In the prior art, for example, patent application publication 1 of China, publication No. 2024.04.19, publication No. 1, publication No. CN117910656A, discloses a magnetic storm short-term prediction method based on a mixed deep learning model, which includes the steps of firstly obtaining Dst data information, processing abnormal data and missing data by using a linear interpolation technology, converting Dst and external input characteristic parameters into supervision sequence pairs by looking back history data, dividing a data set, performing zero mean value standardized preprocessing, constructing a mixed deep learning Dst prediction model formed by a pre-training network and a prediction network, training the model to obtain an optimal model structure, finally using a trained Dst prediction test set to output results, evaluating model performance, and finishing magnetic storm short-term prediction. The method utilizes the historical Dst data information to construct the mixed deep learning model consisting of the pre-training network and the prediction network, improves the prediction accuracy, adopts the Monte Carlo discarding technology to estimate the uncertainty of the model, provides a prediction interval, realizes relatively accurate short-term prediction of the magnetic storm, and is suitable for Dst prediction 1-4 hours in advance. Further, as disclosed in Chinese patent application publication No. 2:CN119919989A with publication date 2025.05.02, "an aurora sub-storm identification method and device based on visual eye movement pattern and deep learning is disclosed, which mainly solves the problems of low accuracy and no basis for model design of the existing aurora sub-storm identification method. The method comprises the steps of generating expert eye movement fixation pattern diagrams of various aurora sub-riots by collecting eye movement data of multiple-name space physicists, establishing an aurora sub-riot eye movement data set, further designing a visual eye movement pattern prediction module for learning a visual fixation pattern of an expert, and setting a sub-riot sequence recognition m