Kp Index Prediction Based on Similarity Algorithm of Machine Learning
-
摘要: 基于机器学习中的相似度算法,建立了在历史太阳风数据中寻找与当前太阳风特征相近事例的推荐模型,用来预报地磁Kp指数。使用1998-2019年间随机选择的120个太阳风事例作为测试数据集,该模型能够推荐得到历史上与输入太阳风造成相似地磁影响的太阳风事例,最优事例的Kp指数与实际值的均方根误差为0.79,相关系数为0.93。本文的推荐模型不仅能获得推荐的太阳风事例的地磁Kp指数用作预报,还可以给出太阳风特征参数按时间序列变化情况对比,让预报员可以更好地结合自身经验进行预报。Abstract: The solar wind is the direct cause of the geomagnetic disturbance. In this paper, based on the feature selection and similarity algorithm of machine learning, a recommended model is established to search for cases whose characteristics are similar to the current solar wind in historical solar wind data, and to obtain the prediction of the geomagnetic Kp index. Tested on 120 solar wind cases randomly selected from 1998 to 2019, the results show that the solar wind cases which have similar geomagnetic effects to the input solar wind can be worked out successfully by proposed model . And the root mean square error between the Kp index of the optimal case recommended by the model and the actual value is 0.79, and the correlation coefficient is 0.93. Different from traditional forecast models, the proposed recommended model in this paper can not only provide a geomagnetic Kp index as a forecast, but also give a clearer and more intuitive comparison of the changes between the solar wind characteristic parameters according to the time series. Even because the historical events have already happened, we can artificially find more dimensional information of the similar historical cases, which makes forecasters better combine their own experience in Kp index forecasting.
-
Key words:
- Solar wind /
- Machine learning /
- Similarity algorithm /
- Kp index /
- Prediction
-
表 1 XGBoost对太阳风特征参数评分结果
Table 1. Scores of feature parameters using XGBoost
参数 v Bz N B T By 评分 1647 850 794 724 474 400 表 2 本文模型与现有典型模型比较
Table 2. Comparison between the model proposed in this paper and the existing typical models
模型 输入量 提前时间/h 相关系数 均方根误差 Costello[4] v, B, Bz 1 0.75 - Boberg[5] v, N, Bz 3 0.77 0.99 APL[6] v, N, B, Bz, Kp 1 0.92 - Bala Model 1[7] Boyle index, Kp 1 0.863 0.71 Model 2 Boyle index, Kp 2 0.854 0.82 Model 3 Boyle index 1 0.852 1.12 Model 4 Boyle index 3 0.845 1.12 Liuyang Model[3] v, N, B, By, Bz, d$\phi $/dt, n1/2v2 1~3.5 0.88 0.65 本文推荐模型 v, Bz 1~3 0.93 0.79 表 3 不同Kp指数下三个最相似事例的距离与平均Kp误差
Table 3. Distances of 3 recommanded cases and Kp error in different Kp indexes
Kp 测试事例
个数Top1
(平均距离4.45)
Kp平均绝对误差Top2
(平均距离5.08)
Kp平均绝对误差Top3
(平均距离5.47)
Kp平均绝对误差0 3 0.70 1.43 1.23 1 11 0.46 0.55 0.71 2 5 0.38 0.24 0.24 3 9 0.81 0.43 0.48 4 12 0.78 1.81 1.33 5 20 0.66 1.12 1.36 6 20 0.90 0.73 0.85 7 13 0.51 0.52 0.57 8 24 0.57 1.13 0.98 9 3 0.23 0.88 1.68 -
[1] KNIPP D J, MCQUADE M, KIRKPATRICK D. Understanding Space Weather and The Physics Behind it[M]. McGraw-Hill, 2011: 744 [2] 吕刚, 张伟. 基于深度学习的推荐系统应用综述[J]. 软件工程, 2020, 23(2): 5-8 doi: 10.19644/j.cnki.issn2096-1472.2020.02.002LV Gang, ZHANG Wei. Survey of deep learning applied in recommendation system[J]. Software Engineering, 2020, 23(2): 5-8 doi: 10.19644/j.cnki.issn2096-1472.2020.02.002 [3] 刘杨, 罗冰显, 刘四清, 等. 基于神经网络方法的Kp预报模型[J]. 载人航天, 2013, 19(2): 70-80 doi: 10.3969/j.issn.1674-5825.2013.02.011LIU Yang, LUO Bingxian, LIU Siqing, et al. Kp forecast models based on neural networks[J]. Manned Spaceflight, 2013, 19(2): 70-80 doi: 10.3969/j.issn.1674-5825.2013.02.011 [4] COSTELLO K A. Moving the Rice MSFM Into a Real-time Forecast Mode Using Solar Wind Driven Forecast Modules[D]. Houston: Rice University, 1998 [5] BOBERG F, WINTOFT P, LUNDSTEDT H. Real time Kp predictions from solar wind data using neural networks[J]. Physics and Chemistry of the Earth, Part C:Solar, Terrestrial & Planetary Science, 2000, 25(4): 275-280 [6] WING S, JOHNSON J R, JEN J, et al. Kp forecast models[J]. Journal of Geophysical Research, 2005, 110(A4): A04203 doi: 10.1029/2004JA010500 [7] BALA R, REIFF P H, LANDIVAR J E. Real-time prediction of magnetospheric activity using the Boyle index[J]. Space Weather, 2009, 7(4): S04003 doi: 10.1029/2008SW000407 [8] 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 [9] 李占山, 刘兆赓. 基于XGBoost的特征选择算法[J]. 通信学报, 2019, 40(10): 101-108 doi: 10.11959/j.issn.1000-436x.2019154LI Zhanshan, LIU Zhaogeng. Feature selection algorithm based on XGBoost[J]. Journal on Communications, 2019, 40(10): 101-108 doi: 10.11959/j.issn.1000-436x.2019154 [10] 涂传诒, 宗秋刚, 周煦之. 日地空间物理学[M]. 2版. 北京: 科学出版社, 2020TU Chuanyi, ZONG Qiugang, ZHOU Xuzhi. Solar-Terrestrial Space Physics[M]. 2 nd ed. Beijing: Science Press, 2020 [11] KINGMA D P, BA J. ADAM: a method for stochastic optimization[C]//Proceedings of the 3 rd International Conference on Learning Representations. San Diego: ICLR, 2015 [12] VAN DER HEIJDEN F, DUIN R P, DE RIDDER D. Classification, Parameter Estimation and State Estimation: An Engineering Approach Using MATLAB[M]. Hoboken: Wiley, 2004 [13] 倪庆千, 乔冀瑜, 连宗凯. 改进K-means聚类剪枝的DTW动态手势识别方法[J]. 现代计算机, 2020(27): 20-25 doi: 10.3969/j.issn.1007-1423.2020.27.004NI Qingqian, QIAO Jiyu, LIAN Zongkai. Improved K-means clustered pruning and DTW dynamic gesture recognition method[J]. Modern Computer, 2020(27): 20-25 doi: 10.3969/j.issn.1007-1423.2020.27.004