Application Of Machine Learning In Space Environment Feature Recognition And Analysis

In recent years,more and more researchers in different fields have applied machine learning methods to the research in this field.Space environment is one of the environments closely related to human life on earth.Many physical phenomena in space environment have rich observation data,but their physical generation mechanism is not completely clear at present.Therefore,using machine learning method to study space physics problems is a new trend in recent years.Firstly,the deep learning method is applied to the feature recognition of magnetic types in solar active areas to realize the automatic classification of magnetic types of sunspot groups.At the same time,the machine learning model based on statistical theory can also be used to analyze the characteristics of space environment and study the physical mechanism.In this paper,the interpretable machine learning model is used to study the generation mechanism of ionospheric plasma bubble for the first time.The correlation between plasma bubble and multiple parameters is analyzed.The main work and results are as follows:The main research contents and results are as follows:(1)The automatic recognition of magnetic types in active areas based on Mount Wilson classification is realized by using the deep learning method(Xception).Based on the SDO / HMI sharp sunspot map data and magnetic map data from 2010 to 2017,a method for automatically identifying the magnetic type of sunspot group is proposed,and a series of model training is carried out under the Xception structure.Generally speaking,the recognition effect of the model is good,among which α The F1 score of type sunspots is 96.50%,β Class is 93.20%,and other types of sunspots are 84.65%.By using the method of integrated network learning,we can further learn the magnetic map information,especially for improving the β Magnetic type and β-x The recognition accuracy of X magnetic type is very helpful.(2)According to the data of ROCSAT-1,Firstly,we introduce a sigmoid function and develop a method to identify plasma bubbles in combination with the standard deviation of plasma density,which can better realize the automatic identification of satellite plasma bubbles.The results show that the plasma bubbles in middle and low latitudes mainly occur in spring and autumn,and mainly appear after18:00 and before 08:00 in local time.(3)The interpretable lasso regression model and sparse additive model(spam)are used for the first time to explore the changes of various parameters during the occurrence of plasma bubbles and their correlation with plasma bubbles.The results show that the vertical drift of plasma caused by eastward polarized electric field has a good correlation with plasma bubbles,which confirms the existence of Rayleigh Taylor instability,and there is a good correlation between temperature and the generation of plasma bubbles,No + ions may also have a potential relationship with plasma bubbles.