基于长短期记忆神经网络的太阳耀斑短期预报

何欣燃; 钟秋珍; 崔延美; 刘四清; 石育榕; 闫晓辉; 王子思禹

doi:10.11728/cjss2022.05.210315028

基于长短期记忆神经网络的太阳耀斑短期预报

doi: 10.11728/cjss2022.05.210315028

何欣燃^{1, 2, 3,},
钟秋珍^{1, 2, 3, ,},
崔延美^{1, 3},
刘四清^{1, 2, 3},
石育榕^{1, 2, 3},
闫晓辉^{1, 2, 3},
王子思禹^{1, 2, 3}

1.
中国科学院国家空间科学中心　北京　100190
2.
中国科学院大学　北京　100049
3.
中国科学院空间环境态势感知技术重点实验室　北京　100190

详细信息

作者简介:
何欣燃：E-mail：hexinran@baidu.com

通讯作者:
钟秋珍，E-mail：zhongqz@nssc.ac.cn

中图分类号: P354
计量
- 文章访问数: 110
- HTML全文浏览量: 66
- PDF下载量: 51
- 被引次数: 0
出版历程
- 收稿日期: 2021-03-12
- 录用日期: 2021-05-19
- 修回日期: 2021-05-17
- 网络出版日期: 2022-09-22

Solar Flare Short-term Forecast Model Based on Long and Short-term Memory Neural Network

HE Xinran^{1, 2, 3
,},
ZHONG Qiuzhen^{1, 2, 3
, ,},
CUI Yanmei^{1, 3},
LIU Siqing^{1, 2, 3},
SHI Yurong^{1, 2, 3},
YAN Xiaohui^{1, 2, 3},
WANG Zisiyu^{1, 2, 3}

1.
National Space Science Center, Chinese Academy of Sciences, Beijing 100190
2.
University of Chinese Academy of Sciences, Beijing 100049
3.
Key Laboratory of Science and Technology on Environmental Space Situation Awareness, Chinese Academy of Sciences, Beijing 100190

摘要

摘要: 提出了一个基于长短期记忆神经网络的耀斑预报模型，利用过去24 h太阳活动区的磁场变化时序构建样本，通过长短期记忆神经网络对磁场特征时序演化进行分析，预报未来48 h内是否发生≥M级别耀斑事件。使用的数据集为2010年5月到2017年5月所有活动区样本，选取了SDO/HMI SHARP的10个磁场特征参量。在建模过程中通过XGBoost方法选取权重、增益率和覆盖率均较高的6个特征参量作为输入参数。通过测试对比，模型的虚报率和准确率与传统机器学习模型相近，报准率和临界成功指数分别为0.7483和0.7402，优于传统机器学习模型。模型总体效果优于传统机器学习模型。
- 太阳耀斑预报 /
- 深度学习 /
- 长短期记忆神经网络
Abstract: Solar flares are a kind of violent solar eruptive activity phenomenon and an important warning device of space weather disturbance. In space weather forecasting, flare forecasting is an important forecast content. This paper proposes a flare prediction model based on long and short-term memory neural network, which uses the time sequence of magnetic field changes in the solar active area in the past 24 h to construct samples, and analyzes the time series evolution of magnetic field characteristics through the long and short-term memory neural network to predict whether ≥M-level flares will occur in the next 48 h. This paper uses a data set for all active area samples from May 2010 to May 2017, and selects 10 magnetic field characteristic parameters of SDO/HMI SHARP. In the modeling process, six feature parameters with high weight, gain rate and coverage rate were selected as input parameters through XGBoost method. Through test comparison, the false report rate and accuracy rate of the model are similar to the traditional machine learning model, and the accuracy rate and critical success index are better than the traditional machine learning model, which are 0.7483 and 0.7402 respectively. The overall effect of the model is better than that of the traditional machine learning model.
- Solar flare prediction /
- Deep learning /
- Long and short-term memory neural network
注释:

1) * http://jsoc.stanford.edu/doc/data/hmi/sharp.htm

2) *ftp://ftp.ngdc.noaa.gov/STP/space-weather/solar-data/solar-features/solar-flares/x-rays/goes/xrs/

HTML全文

图 1 滑动窗口原理

Figure 1. Sliding window principle

下载: 全尺寸图片幻灯片

图 2 LSTM中隐含层重复模块结构

Figure 2. Architecture of repeat module in the hidden layer of LSTM

下载: 全尺寸图片幻灯片

图 3 10个物理参量在所有弱学习器中的权重

Figure 3. Weights of ten parameters in all weak learners

下载: 全尺寸图片幻灯片

图 5 10个物理参量在所有弱学习中的覆盖率

Figure 5. Cover rate of ten parameters in all weak learners

下载: 全尺寸图片幻灯片

图 4 10个物理参量在所有弱学习器中的增益率

Figure 4. Gain rate of ten parameters in all weak learners

下载: 全尺寸图片幻灯片

图 6 LSTM耀斑预报模型结构

Figure 6. LSTM flare prediction model structure

下载: 全尺寸图片幻灯片

图 7 LSTM耀斑预报模型ROC曲线，图中红点为该模型的最佳阈值所对应的TPR和FPR

Figure 7. ROC curve of the LSTM flare prediction model. The red dots in the figure are the TPR and FPR corresponding to the optimal threshold of the model

下载: 全尺寸图片幻灯片

图 8 正样本（2012年3月8日09:36 UT至9日09:36 UT）各特征参数的变化（a）和GOES卫星测量（2012年3月9日00:00至12日00:00 UT）在该样本未来48 h的X射线通量变化（b）。橙色垂直虚线为模型预测范围，绿色虚线为耀斑事件（2012年3月10日17:44 UT）

Figure 8. Positive sample (from 09:36 UT on 8 March 2012 to 09:36 UT on 9 March 2012) characteristic parameter change (a) and the GOES satellite measurement (from 00:00 UT on 9 March 2012 to 00:00 UT on 12 March, 2012) of the X-ray flux change of the sample in the next 48 h (b). Orange dashed line is the model prediction range, and the green dashed line is the flare event (17:44 UT on 10 March 2012)

下载: 全尺寸图片幻灯片

图 9 负样本（2011年11月16日12:48 UT至17日12:48 UT）各特征参数变化（a）与GOES卫星测量（2011年11月17日00:00 UT至20日00:00 UT）该样本未来48 h的X射线通量变化（b），橙色虚线为模型预测范围，显示该样本未来48 h内无≥M级耀斑发生

Figure 9. Negative sample (from 12:48 UT on 16 November 2011 to 12:48 UT on 17 November 2011) characteristic parameter change (a) and the GOES satellite measurement (from 00:00 UT on 17 November 2011 to 00:00 on 20 November 2011) of the X-ray flux change of the sample in the next 48 h (b). Orange dashed line is the model prediction range, and shows that there is no ≥M class flares occurrance in the next 48 h for this sample

下载: 全尺寸图片幻灯片

表 1 图2中各符号含义的说明

Table 1. Implication of some symbols in Figure 2

数学符号	含义
$ {x}_{t} $	当前t 时刻的输入
$ {h}_{t} $	t 时刻的输出
$ {h}_{t-1} $	$t-1 $ 时刻的输出
$ {C}_{t} $	t 时刻的细胞状态
$ {C}_{t-1} $	$t-1 $ 时刻的单元状态
$ {\tilde {C}}_{t} $	t时刻的候选状态值
$ {f}_{t} $	遗忘门函数
$ {i}_{t} $	输入门函数
$ {o}_{t} $	输出门函数
$ \sigma ,\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h} $	激活函数

下载: 导出CSV

表 2 不同模型预报结果评估（1）

Table 2. Evaluation of forecast results of different models （1）

预报方法	TPR（报准率）	FPR（虚报率）	Accuracy（准确率）	TSS（临界成功指数）
LSTM	0.7483	0.0081	0.9894	0.7402
XGBoost	0.5033	0.0143	0.9896	0.4890
SVM	0.3311	0.0082	0.9899	0.3229
RandomForest	0.3245	0.0173	0.9891	0.3072
Logistic Regression	0.1987	0.0024	0.9906	0.1963
C4.5	0.4834	0.0387	0.9806	0.4447

下载: 导出CSV

表 3 不同模型预报结果评估（2）

Table 3. Evaluation of forecast results of different models （2）

预报方法	真实为正且预测为正（TP）	真实为正预测为负（FN）	真实为负预测为负（TN）	真实为负预测为正（FP）
LSTM	113	38	13840	113
XGBoost	76	74	13880	73
SVM	49	102	13912	41
RandomForest	48	103	13896	57
Logistic Regression	30	121	13941	12
C4.5	66	85	13743	210

下载: 导出CSV

表 4 CNN与LSTM模型预报结果评估（1）

Table 4. Evaluation of forecast results of CNN and LSTM models （1）

预报方法	TPR（报准率）	FPR（虚报率）	Accuracy（准确率）	TSS（临界成功指数）
LSTM	0.7483	0.0081	0.9894	0.7402
Ref. [27] CNN	0.8050	0.1860	0.7919	0.6190
Ref. [28] CNN	0.8100	0.1900	0.8111	0.6200

下载: 导出CSV

表 5 LSTM与CNN模型预报结果评估（2）

Table 5. Evaluation of forecast results of LSTM and CNN models （2）

预报方法	真实为正且预测为正（TP）	真实为正预测为负（FN）	真实为负预测为负（TN）	真实为负预测为正（FP）
LSTM	113	38	13840	113
Ref. [27] CNN	1008	291	978	231
Ref. [28] CNN	1614	378	37711	8783

下载: 导出CSV

参考文献(37)

[1]	MCINTOSH P S. The classification of sunspot groups[J]. Solar Physics, 1990, 125(2): 251-267 doi: 10.1007/BF00158405
[2]	CROWN M D. Validation of the NOAA space weather prediction center’s solar flare forecasting look-up table and forecaster-issued probabilities[J]. Space Weather, 2012, 10(6): S06006
[3]	MURRAY S A, BINGHAM S, SHARPE M, et al. Flare forecasting at the met office space weather operations Centre[J]. Space Weather, 2017, 15(4): 577-588 doi: 10.1002/2016SW001579
[4]	DEVOS A, VERBEECK C, ROBBRECHT E. Verification of space weather forecasting at the regional warning center in Belgium[J]. Journal of Space Weather and Space Climate, 2014, 4: A29 doi: 10.1051/swsc/2014025
[5]	BLOOMFIELD D S, HIGGINS P A, MCATEER R T J, et al. Toward reliable benchmarking of solar flare forecasting methods[J]. The Astrophysical Journal Letters, 2012, 747(2): L41 doi: 10.1088/2041-8205/747/2/L41
[6]	SHIN S, LEE J Y, MOON Y J, et al. Development of daily maximum flare-flux forecast models for strong solar flares[J]. Solar Physics, 2016, 291(3): 897-909 doi: 10.1007/s11207-016-0869-2
[7]	ANASTASIADIS A, PAPAIOANNOU A, SANDBERG I, et al. Predicting flares and solar energetic particle events: the FORSPEF tool[J]. Solar Physics, 2017, 292(9): 134 doi: 10.1007/s11207-017-1163-7
[8]	SONG H, TAN C Y, JING J, et al. Statistical assessment of photospheric magnetic features in imminent solar flare predictions[J]. Solar Physics, 2009, 254(1): 101-125 doi: 10.1007/s11207-008-9288-3
[9]	YANG X, LIN G H, ZHANG H Q, et al. Magnetic nonpotentiality in photospheric active regions as a predictor of solar flares[J]. The Astrophysical Journal Letters, 2013, 774(2): L27 doi: 10.1088/2041-8205/774/2/L27
[10]	MURANUSHI T, SHIBAYAMA T, MURANUSHI Y H, et al. UFCORIN: a fully automated predictor of solar flares in GOES X-ray flux[J]. Space Weather, 2015, 13(11): 778-796 doi: 10.1002/2015SW001257
[11]	COLAK T, QAHWAJI R. Automated solar activity prediction: a hybrid computer platform using machine learning and solar imaging for automated prediction of solar flares[J]. Space Weather, 2009, 7(6): S06001
[12]	AHMED O W, QAHWAJI R, COLAK T, et al. Solar flare prediction using advanced feature extraction, machine learning, and feature selection[J]. Solar Physics, 2013, 283(1): 157-175 doi: 10.1007/s11207-011-9896-1
[13]	HADA-MURANUSHI Y, MURANUSHI T, ASAI A, et al. A deep-learning approach for operation of an automated realtime flare forecast[OL]. arXiv: 1606.01587, 2016
[14]	NISHIZUKA N, SUGIURA K, KUBO Y, et al. Deep flare net (DeFN) model for solar flare prediction[J]. The Astrophysical Journal, 2018, 858(2): 113 doi: 10.3847/1538-4357/aab9a7
[15]	HUANG X, WANG H N, XU L, et al. Deep learning based solar flare forecasting model. I. results for line-of-sight magnetograms[J]. The Astrophysical Journal, 2018, 856(1): 7 doi: 10.3847/1538-4357/aaae00
[16]	QAHWAJI R, COLAK T. Automatic short-term solar flare prediction using machine learning and sunspot associations[J]. Solar Physics, 2007, 241(1): 195-211 doi: 10.1007/s11207-006-0272-5
[17]	LEKA K D, BARNES G, WAGNER E. The NWRA classification infrastructure: description and extension to the discriminant analysis flare forecasting system (DAFFS)[J]. Journal of Space Weather and Space Climate, 2018, 8: A25 doi: 10.1051/swsc/2018004
[18]	DOMIJAN K, BLOOMFIELD D S, PITIÉ F. Solar flare forecasting from magnetic feature properties generated by the solar monitor active region tracker[J]. Solar Physics, 2019, 294(1): 6 doi: 10.1007/s11207-018-1392-4
[19]	AL-GHRAIBAH A, BOUCHERON L E, MCATEER R T J. An automated classification approach to ranking photospheric proxies of magnetic energy build-up[J]. Astronomy & Astrophysics, 2015, 579: A64
[20]	WANG J X, ZHOU G P, JIN C L, et al. Solar intranetwork magnetic elements: bipolar flux appearance[J]. Solar Physics, 2012, 278(2): 299-322 doi: 10.1007/s11207-012-9950-7
[21]	RUST D M, SAKURAI T, GAIZAUSKAS V, et al. Preflare state[J]. Solar Physics, 1994, 153(1/2): 1-17
[22]	HUANG X, YU D R, HU Q H, et al. Short-term solar flare prediction using predictor teams[J]. Solar Physics, 2010, 263(1/2): 175-184
[23]	HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786): 504-507 doi: 10.1126/science.1127647
[24]	LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436-444 doi: 10.1038/nature14539
[25]	COLLOBERT R, WESTON J, BOTTOU L, et al. Natural language processing (almost) from scratch[J]. The Journal of Machine Learning Research, 2011, 12: 2493-2537
[26]	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90 doi: 10.1145/3065386
[27]	李蓉, 黄鑫. 一种黑子特征自动提取的太阳耀斑模型[J]. 中国科学: 物理学力学天文学, 2018, 48(11): 119601 LI Rong, HUANG Xin. Solar flare forecasting model based on automatic feature extraction of sunspots[J]. Scientia Sinica Physica, Mechanica & Astronomica, 2018, 48(11): 119601
[28]	LIU H, LIU C, WANG J T L, et al. Predicting solar flares using a long short-term memory network[J]. The Astrophysical Journal, 2019, 877(2): 121 doi: 10.3847/1538-4357/ab1b3c
[29]	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
[30]	LEE C H, LIN C R, CHEN M S. Sliding-window filtering: an efficient algorithm for incremental mining[C]//Proceedings of the Tenth International Conference on Information and Knowledge Management. Atlanta: ACM, 2001: 263-270
[31]	GOLAB L, DEHAAN D, DEMAINE E D, et al. Identifying frequent items in sliding windows over on-line packet streams[C]//Proceedings of the 3 rd ACM SIGCOMM Conference on Internet Measurement. Miami Beach: ACM, 2003: 173-178
[32]	CHANG J H, LEE W S. A sliding window method for finding recently frequent Itemsets over online data streams[J]. Journal of Information Science and Engineering, 2004, 20(4): 753-762
[33]	LIU Y J, FANG Y J, ZHU X M. Modeling of hydraulic turbine systems based on a bayesian-gaussian neural network driven by sliding window data[J]. Journal of Zhejiang University Science C, 2010, 11(1): 56 doi: 10.1631/jzus.C0910176
[34]	HINTON G E, SRIVASTAVA N, KRIZHEVSKY A, et al. Improving neural networks by preventing co-adaptation of feature detectors[OL]. arXiv: 1207.0580, 2012
[35]	SCHAPIRE R E. The strength of weak learnability[J]. Machine Learning, 1990, 5(2): 197-227
[36]	BREIMAN L, FRIEDMAN J H, OLSHEN R A, et al. Classification and regression trees[J]. Biometrics, 1984, 40(3): 874
[37]	FAWCETT T. An introduction to ROC analysis[J]. Pattern Recognition Letters, 2006, 27(8): 861-874 doi: 10.1016/j.patrec.2005.10.010