Two-Stage Solar Flare Forecasting Based on Convolutional Neural Networks

作     者：Jun Chen Weifu Li Shuxin Li Hong Chen Xuebin Zhao Jiangtao Peng Yanhong Chen Hao Deng 

作者机构：College of ScienceHuazhong Agricultural UniversityWuhan 430070China Hubei Key Laboratory of Applied MathematicsHubei UniversityWuhan 430062China Nation Space Science CenterChinese Academy of ScienceBeijing 100190China Key Laboratory of Science and Technology on Environment Space Situation AwarenessBeijing 100190China University of Chinese Academy of ScienceBeijing 100049China 

出 版 物：《Space(Science & Technology)》 (空间科学与技术（英文）)

年 卷 期：2022年第2022卷第1期

页      面：36-45页

核心收录：

学科分类：07[理学] 070401[理学-天体物理] 0704[理学-天文学] 

基　　金：This work is support in part by the National Natural Science Foundation of China under Grant Nos.12071166 and 42171351 Hubei Key Laboratory of Applied Mathematics under Grant Nos.HBAM 202004 and 201612 Hubei Provincial Natural Science Foundation of China under Grant No.2021CFA087 

主　　题：Solar forecasting storm 

摘      要：Solar flares are solar storm events driven by the magnetic field in the solar activity *** flare,often associated with solar proton event or CME,has a negative impact on ratio communication,aviation,and ***,its forecasting has attracted much attention from the academic *** to the limitation of the unbalanced distribution of the observation data,most techniques failed to effectively learn complex magnetic field characteristics,leading to poor forecasting *** the statistical analysis of solar flare magnetic map data observed by SDO/HMI from 2010 to 2019,we find that unsupervised clustering algorithms have high accuracy in identifying the sunspot group in which the positive samples account for the ***,for these identified sunspot groups,the ensemble model that integrates the capability of boosting and convolutional neural network(CNN)achieves high-precision prediction of whether the solar flares will occur in the next 48 *** on the above findings,a two-stage solar flare early warning system is established in this *** F1 score of our method is 0.5639,which shows that it is superior to the traditional methods such as logistic regression and support vector machine(SVM).