Nantong Gympro Sports Co.,Ltd , https://www.masportsgympro.com
Research on Solar Wind Classification and Application of Space Weather Warning Based on Machine Learning
[ Instrument R & D of Instrument Network ] In 1959, the Soviet Union ’s Luna-1 satellite went to heaven, and humans were able to observe the solar wind in situ for the first time. Observations and studies in the following decades have shown that the near-Earth solar wind has different characteristics and originates in different source regions. The solar wind can be roughly divided into four categories: coronal cave wind, coronal flow wind, sector reversal zone wind and coronal ejection wind.
The classification of the solar wind according to the source area is of great significance to the solar and heliospheric physics research. First, in order to have a more comprehensive and accurate understanding of the nature of the solar wind, it is necessary to distinguish the type of solar wind in statistical research; second, the solar wind still "records" some characteristics of its source area during interplanetary propagation. A good understanding of the physical processes taking place in different regions of the sun; again, the effects of different types of solar wind on the ground are significantly different, and the solar wind classification information is also expected to enhance the effect of space weather warning.
Solar wind, astronomical, refers to the flow of charged particles of supersonic plasma emitted from the upper atmosphere of the sun. In the absence of the sun, this charged particle stream is often referred to as "stellar wind." The solar wind is a continuous flow of high-speed charged particles that comes from the sun and moves at a speed of 200-800 km / s. Although this substance is different from the air on the earth, it is not composed of gas molecules, but is composed of simpler elementary particles-protons and electrons, which are one level lower than atoms-but the effects generated when they flow The air flow is very similar, so it is called the solar wind. In March 2012, a strong solar storm erupted on the morning of the 7th in five years, and wireless communication was affected.
Traditionally, solar wind classification is accomplished by a number of experienced researchers combining many observational features of different types. For example, based on different observation characteristics, many researchers have separately published a list of interplanetary coronal mass ejection events, a list of magnetic clouds, a list of co-rotation interaction regions, a list of shock events, a list of heliospheric current sheets, and a heliospheric sector Boundary list etc. In addition, scholars have also tried to develop some empirical models for solar wind classification. For example, one-dimensional parameter space empirical model (solar wind speed Vp, Ptype), two-dimensional parameter space empirical model (O7 + / O6 +-Vp), three-dimensional parameter space empirical model (Sp-VA-Texp / Tp), etc.
In the past ten years, great progress has been made in the classification and identification of solar wind. Nevertheless, whether it is manual recognition or empirical model recognition, there are certain limitations. In the face of voluminous observation data, the standards for manual recognition are different and difficult to update in real time. For example, the authoritative Lepping magnetic cloud list has stopped updating since 2008, and some other lists also have different degrees of update lag. For the empirical model, it is difficult to be fully automated because of the frequent manual intervention. In addition, in order to facilitate the clear identification of the interface, the empirical model is carried out in the three-dimensional or below parameter space, there is room for improvement in recognition accuracy.
Recently, artificial intelligence technology has made great progress, and machine learning algorithms have begun to replace some tasks that require human intervention. As a typical task of machine learning, the multi-parameter space classification algorithm faces the era of spatial physics big data, and its advantages in the field of pattern recognition gradually become prominent, so it gradually becomes popular. In this context, Li Hui and Wang Chi, researchers in the weather room of the National Space Science Center of the Chinese Academy of Sciences, and Xu Fei, an associate professor at Nanjing University of Information Science and Technology, began a collaborative research on artificial intelligence to identify solar wind classification. The team used 10 internationally popular machine learning classification algorithms (KNN, LSVM, RBFSVM, DT, RF, AdaBoost, NN, GNB, QDA, XGBoost) to develop an automatic recognition algorithm for solar wind classification in the optimized 8-dimensional parameter space. The solar wind observation data can be automatically and quickly divided into four categories: coronal cave wind, coronal flow wind, sector reversal zone wind, and coronal ejection wind.
The core idea of ​​the kNN algorithm is that if most of the k nearest neighbor samples in a feature space belong to a certain category, the sample also belongs to this category and has the characteristics of the samples in this category. This method only determines the category of the samples to be classified according to the category of the nearest sample or samples in determining the classification decision. The kNN method is only relevant to a very small number of adjacent samples when making category decisions. Since the kNN method mainly depends on the limited neighboring samples around, rather than the method of discriminating the class domain to determine the category, the kNN method is more preferable than other methods for the sample set to be divided or overlapped in the class domain. As suitable.
Most of the common machine learning classification algorithms can get good classification results, and the classification accuracy of the KNN algorithm is 99.2%, 91.1%, 83.8%, and 92.9%, respectively. Compared with the previous good empirical model, the accuracy is improved by 2.3%, 21.2%, 11.8% and 5.4% respectively; the solar wind classification algorithm can identify the interplanetary small-scale flux rope (which lasts only a few hours), which is the subsequent small scale of the solar wind Research on the structural characteristics and the interaction with the background solar wind provides convenience; on this basis, through comparative studies, the team also confirmed that real-time solar wind classification information can be applied to space weather warning. When the satellite detects a strong solar wind farm (Ey> 5.0 mV / m) in real time, if the solar wind classification algorithm determines that it is carried by the coronal wind, then the probability of causing a medium magnetic storm is greater and accompanied by a higher satellite charging risk; If it is determined that it is carried by the corona projectile wind, then the probability of causing a strong magnetic storm is greater and is accompanied by a lower risk of satellite charging.
Machine learning is a multidisciplinary cross-discipline, involving multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and so on. Specially study how the computer simulates or realizes human learning behavior to acquire new knowledge or skills, and reorganize the existing knowledge structure to continuously improve its performance.
It is the core of artificial intelligence and the fundamental way to make computers intelligent.
Machine learning has the following definitions:
(1) Machine learning is a science of artificial intelligence. The main research object in this field is artificial intelligence, especially how to improve the performance of specific algorithms in empirical learning.
(2) Machine learning is the study of computer algorithms that can be improved automatically through experience.
(3) Machine learning uses data or past experience to optimize the performance standards of computer programs.
This research confirmed that the classification algorithm based on machine learning has the ability to efficiently and accurately identify four typical types of solar wind, and can obtain better classification results than previous empirical models. In addition, the classification algorithm only requires some basic solar wind parameter observations, such as plasma number density, temperature, velocity, and magnetic field strength. The fact that no solar wind component observation is needed also makes the algorithm have higher applicability and wider application prospects.
Source: Encyclopedia, National Space Science Center