Testing and Validating Two Morphological Flare Predictors by Logistic Regression Machine Learning
Testing and Validating Two Morphological Flare Predictors by Logistic Regression Machine Learning
复制标题
DOI：
10.3389/fspas.2020.571186
复制
发表时间：
2020-12
期刊：
影响因子：
--
通讯作者：
M. Korsós;R. Erdélyi;Jiajia Liu;H. Morgan
中科院分区：
文献类型：
--
作者：
M. Korsós;R. Erdélyi;Jiajia Liu;H. Morgan
虽然已知最活跃的太阳活动区（ARs）经常发生耀斑，但预测单个耀斑的发生及其大小，是一个非常发展中的领域，具有很强的机器学习应用潜力。目前的工作是基于一种方法，该方法是用来定义具有相反极性的ar混合态的数值测量。该方法通过使用两个形态学参数：1)分离参数S l−f和2)水平磁梯度G S的总和，为给定AR的混合状态水平与该AR的太阳爆发概率水平之间的假设联系提供了令人信服的证据。在这项工作中，我们基于SOHO/MDI-Debrecen数据（SDD）和SDO/HMI -Debrecen数据（HMIDD）太阳黑子目录，研究了S l−f和G S作为耀斑预测器的效率。特别地，我们研究了大约1000个ar，以便通过应用逻辑回归机器学习方法测试和验证两个形态参数的联合预测能力。在这里，我们确认这两个参数及其阈值在一起应用时是很好的互补预测因子。此外，这些预测参数的预测概率在前一天至少为70%。
Whilst the most dynamic solar active regions (ARs) are known to flare frequently, predicting the occurrence of individual flares and their magnitude, is very much a developing field with strong potentials for machine learning applications. The present work is based on a method which is developed to define numerical measures of the mixed states of ARs with opposite polarities. The method yields compelling evidence for the assumed connection between the level of mixed states of a given AR and the level of the solar eruptive probability of this AR by employing two morphological parameters: 1) the separation parameter S l − f and 2) the sum of the horizontal magnetic gradient G S . In this work, we study the efficiency of S l − f and G S as flare predictors on a representative sample of ARs, based on the SOHO/MDI-Debrecen Data (SDD) and the SDO/HMI - Debrecen Data (HMIDD) sunspot catalogues. In particular, we investigate about 1,000 ARs in order to test and validate the joint prediction capabilities of the two morphological parameters by applying the logistic regression machine learning method. Here, we confirm that the two parameters with their threshold values are, when applied together, good complementary predictors. Furthermore, the prediction probability of these predictor parameters is given at least 70% a day before.