Solar Proton Events Short-time Forecasting Based on Ensemble Learning
-
摘要: 太阳质子事件是一种由太阳活动爆发时喷射并传播到近地空间的高能粒子引起的空间天气现象。这些高能粒子会对航天器和宇航员产生严重危害,对太阳质子事件进行准确的短期预报是航天活动灾害预防的重要内容。针对当前主要预报模型中普遍存在的高虚报率问题,提出了一种基于集成学习的太阳质子事件短期预报方法,利用第23个太阳活动周数据,建立了一种集成8种机器学习模型的太阳质子事件短期预报系统。实验结果表明,本文方法在取得了80.95%的报准率的同时,将虚报率降低至19.05%,相比现有的预报系统具有较为明显的优势。Abstract: Solar proton event is a space weather phenomenon caused by energetic particles ejected and propagated into near-Earth space during bursts of solar activity. These high-energy particles can cause serious harm to spacecraft and astronauts, therefore, accurate short-term forecasting of solar proton events is very necessary as part of disaster prevention for space activities. The short-time forecasting of solar proton events still faces a lot of challenges, one of which is the high false alarm rate. To solve this problem, we adopted a whole new set of methods-machine learning. As a branch of computer science, many excellent algorithms have emerged in the field of machine learning in recent years, and have achieved successful applications in many fields. In this study, an ensemble model based on 8 widely used machine learning models is established to make precise forecasting of solar proton events. An experiment on the 23rd solar cycle shows that this model gets a probability of detection of 80.95% and a false alarm rate of 19.05%.
-
Key words:
- Solar proton events /
- Short-time prediction /
- Ensemble learning /
- False alarm rate
-
表 1 本文使用的8种机器学习模型
Table 1. Machine learning models applied in this study
表 2 实验数据的划分方式
Table 2. Dataset division strategy
子数据集 数量 作用 训练集 交叉训练集 400条
训练集的4/5交叉验证中训练
模型训练集成
模型交叉验证集 100条
训练集的1/5交叉验证中测试
模型能力测试集 164条 测试集成模型的能力 表 3 八种模型各自在交叉验证中的表现
Table 3. Performance of 8 individual models on cross-validation dataset
模型 报准率 虚报率 F1值 LR 0.5325 0.3602 0.5753 DT 0.619 0.3810 0.6190 RF 0.5714 0.1429 0.6857 Ada 0.5238 0.2143 0.6286 GBDT 0.6667 0.1765 0.7368 XGB 0.7619 0.3333 0.7111 GNB 1.0000 0.6719 0.4941 KNN 0.5714 0.3333 0.6154 表 4 集成模型对测试集中的太阳质子事件的预报结果
Table 4. SPEs prediction result of the ensemble model on the test set
质子事件
发生时间峰值
通量是否
漏报质子事件
发生时间峰值
通量是否
漏报1998-04-20 1700 否 2002-08-14 26 是 1998-11-07 11 否 2002-08-24 317 否 1999-06-04 64 否 2003-10-26 466 否 2000-11-24 94 否 2003-11-02 30 否 2001-04-02 1110 否 2003-12-02 89 是 2001-04-12 51 否 2004-07-25 2090 否 2001-11-17 34 是 2004-11-01 63 是 2001-11-22 25 否 2005-01-20 1860 否 2002-03-15 13 否 2005-08-22 337 否 2002-04-21 2520 否 2005-09-09 1880 否 2002-07-15 234 否 表 5 集成模型在测试集上的预报结果
Table 5. Result of the ensemble model on the test set
预报/实际 无SPE 有SPE 总计 无SPE 139 4 143 有SPE 4 17 21 总计 143 21 164 表 6 本文模型与现行其他预报方法的对比
Table 6. Comparison with other current prediction methods
模型 报准率 虚报率 F1值 PPS 85.71% 50.00% 0.6316 Laurenza模型 63% 42% 0.6040 UMASEP 80.72% 33.99% 0.7263 Qiuzhen模型 80.00% 25.85% 0.7696 本文方法 80.95% 19.05% 0.8085 -
[1] ARAN A, SANAHUJA B, LARIO D. SOLPENCO: a solar particle engineering code[J]. Advances in Space Research, 2006, 37(6): 1240-1246 doi: 10.1016/j.asr.2005.09.019 [2] LUHMANN J G, LEDVINA S A, ODSTRCIL D, et al. Cone model-based SEP event calculations for applications to multipoint observations[J]. Advances in Space Research, 2010, 46(1): 1-21 doi: 10.1016/j.asr.2010.03.011 [3] KAHLER S W, CLIVER E W, LING A G. Validating the Proton Prediction System (PPS)[J]. Journal of Atmospheric and Solar-Terrestrial Physics, 2007, 69(1/2): 43-49 [4] LAURENZA M, CLIVER E W, HEWITT J, et al. A technique for short-term warning of solar energetic particle events based on flare location, flare size, and evidence of particle escape[J]. Space Weather, 2009, 7(4): S04008 [5] NÚÑEZ M. Predicting solar energetic proton events (E > 10 MeV)[J]. Space Weather, 2011, 9(7): S07003 [6] HUANG X, WANG H N, LI L P. Ensemble prediction model of solar proton events associated with solar flares and coronal mass ejections[J]. Research in Astronomy and Astrophysics, 2012, 12(3): 313-321 doi: 10.1088/1674-4527/12/3/007 [7] ZHONG Q Z, WANG J J, MENG X J, et al. Prediction model for solar energetic proton events: analysis and verification[J]. Space Weather, 2019, 17(5): 709-726 doi: 10.1029/2018SW001915 [8] KAHLER S W. The role of the big flare syndrome in correlations of solar energetic proton fluxes and associated microwave burst parameters[J]. Journal of Geophysical Research: Space Physics, 1982, 87(A5): 3439-3448 doi: 10.1029/JA087iA05p03439 [9] 乐贵明, 王鸿雁, 白铁男. 太阳质子事件与太阳耀斑的关系[J]. 空间科学学报, 2018, 38(4): 437-443 doi: 10.11728/cjss2018.04.437LE Guiming, WANG Hongyan, BAI Tienan. Relationship between solar proton events and the associated solar flares[J]. Chinese Journal of Space Science, 2018, 38(4): 437-443 doi: 10.11728/cjss2018.04.437 [10] KAHLER S W, SHEELEY N R JR, HOWARD R A, et al. Associations between coronal mass ejections and solar energetic proton events[J]. Journal of Geophysical Research: Space Physics, 1984, 89(A11): 9683-9693 doi: 10.1029/JA089iA11p09683 [11] 王聪, 崔延美, 敖先志, 等. 全晕CME与太阳质子事件的关系[J]. 空间科学学报, 2018, 38(1): 9-18 doi: 10.11728/cjss2018.01.009WANG Cong, CUI Yanmei, AO Xianzhi, et al. Relationship of halo CME and solar proton events[J]. Chinese Journal of Space Science, 2018, 38(1): 9-18 doi: 10.11728/cjss2018.01.009 [12] FALCO M, COSTA P, ROMANO P. Solar flare forecasting using morphological properties of sunspot groups[J]. Journal of Space Weather and Space Climate, 2019, 9: A22 doi: 10.1051/swsc/2019019 [13] MCCLOSKEY A E, GALLAGHER P T, BLOOMFIELD D S. Flare forecasting using the evolution of McIntosh sunspot classifications[J]. Journal of Space Weather and Space Climate, 2018, 8: A34 doi: 10.1051/swsc/2018022 [14] HOWARD T. Coronal Mass Ejections[M]. New York: Springer, 2011 [15] 白铁男, 乐贵明, 赵浩峰. 第23至24周太阳质子事件的统计特征[J]. 空间科学学报, 2017, 37(6): 649-658 doi: 10.11728/cjss2017.06.649BAI Tienan, LE Guiming, ZHAO Haofeng. Statistical properties of solar proton events during solar cycle 23 and 24[J]. Chinese Journal of Space Science, 2017, 37(6): 649-658 doi: 10.11728/cjss2017.06.649 [16] MCINTOSH P S. The classification of sunspot groups[J]. Solar Physics, 1990, 125(2): 251-267 doi: 10.1007/BF00158405 [17] 徐振中, 王伟民, 张韧, 等. 第23太阳活动周武汉站电离层TEC特征分析[J]. 空间科学学报, 2013, 33(1): 28-33 doi: 10.11728/cjss2013.01.028XU Zhenzhong, WANG Weimin, ZHANG Ren, et al. Characteristic analysis of ionosphere TEC at Wuhan station during 23 rd solar cycle[J]. Chinese Journal of Space Science, 2013, 33(1): 28-33 doi: 10.11728/cjss2013.01.028 [18] HARRISON E, RIINU P. Logistic regression[M]// R for Health Data Science. New York: Chapman and Hall/CRC, 2020. [19] WANG P H, TU Y S, TSENG Y J. PgpRules: a decision tree based prediction server for P-glycoprotein substrates and inhibitors[J]. Bioinformatics, 2019, 35(20): 4193-4195 doi: 10.1093/bioinformatics/btz213 [20] FEZAI R, BOUZRARA K, MANSOURI M, et al. Random forest-based nonlinear improved feature extraction and selection for fault classification[C]//Proceedings of 2021 18th International Multi-Conference on Systems, Signals & Devices (SSD). Monastir: IEEE, 2021: 601-606 [21] FREUND Y, SCHAPIRE R E. A decision-theoretic generalization of on-line learning and an application to boosting[J]. Journal of Computer and System Sciences, 1997, 55(1): 119-139 doi: 10.1006/jcss.1997.1504 [22] BIAU G, CADRE B, ROUVIÈRE L. Accelerated gradient boosting[J]. Machine Learning, 2019, 108(6): 971-992 doi: 10.1007/s10994-019-05787-1 [23] CHEN T Q, GUESTRIN C. XGBoost: a scalable tree boosting system[C]//Proceedings of the 22 nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco: ACM, 2016: 785-794 [24] ZHU H J, ZHU L H. Encrypted network behaviors identification based on dynamic time warping and k-nearest neighbor[J]. Cluster Computing, 2019, 22(2): 2571-2580 [25] PÉREZ A, LARRAÑAGA P, INZA I. Supervised classification with conditional Gaussian networks: increasing the structure complexity from naive Bayes[J]. International Journal of Approximate Reasoning, 2006, 43(1): 1-25 doi: 10.1016/j.ijar.2006.01.002
计量
- 文章访问数: 207
- HTML全文浏览量: 103
- PDF下载量: 53
- 被引次数: 0