Application of Support Vector Machine to the Forecasting of Dst Index During Geomagnetic Storm
-
摘要: 利用支持向量机(SVM)模型对大磁暴期间Dst指数进行预报研究.以1995-2014年期间的80次大磁暴(Dst≤-100nT)事件共2662组观测数据为研究对象,以对应时间的太阳风参数为模型输入参数,同时建立了神经网络模型和线性机模型进行对比,并利用交叉验证提高预测结果的可靠性.为比较不同模型的预测效果,选用相关系数(CC)、均方根误差(RMS)、磁暴期间Dst指数最小值预测结果的平均绝对误差以及Dst指数最小值出现时间预测结果的平均绝对误差等统计量作为对比参数.结果显示SVM模型的预测效果最好,其中相关系数为0.89,均方根误差为24.27nT,所有磁暴事件的最小Dst值预测平均绝对误差为17.35nT,最小Dst值出现时间的预测平均绝对误差为3.2h.为进一步检验模型对不同活动水平磁暴预报效果的可能差异,将所有磁暴事件分为大磁暴(-200 <Dst ≤-100nT)和特大磁暴(Dst≤-200nT)两组进行预测,发现两组事件的预测效果依然是SVM模型最好.Abstract: In this study the support vector machine is applied to the forecasting of Dst index during intense geomagnetic storms (Dst≤ -100nT) that occurred from 1995 to 2014. We collect 2662 Dst indices and use the corresponding solar wind data as model input. We also build Neural Network and Linear machine as comparison, and improve the reliability of the predicted results by using K-fold cross validation. For comparison, we calculate the Correlation Coefficient (CC), the RMS errors, the Mean Absolute Error of the minimum Dst (Em) and the Mean Absolute Error of the time when the minimum Dst occurred (Et) between the observed Dst data and the predicted one. As a result, we find that SVM shows the best prediction performance for all events: CC is 0.89, RMS is 24.27nT, Em is 17.35nT and Et is 3.2 hours respectively. For further comparison, the 80 storm events are divided into two groups depending on the minimum value of Dst index. It is shown that the forecasting performance of SVM is better than other models both in the intense (-200 <Dstmin ≤-100nT) and the super intense geomagnetic storm (Dstmin ≤-200nT) groups.
-
Key words:
- Support Vector Machine (SVM) /
- Forecasting /
- Geomagnetic storm /
- Dst index
-
[1] BURTON R K, MCPHERRON R L, RUSSELL C T. An empirical relationship between interplanetary conditions and Dst[J]. J. Geophys. Res., 1975, 80:4204-4214 [2] GONZALEZ W D, JOSELYN J A, KAMIDE Y, et al. What is a geomagnetic storm[J]. J. Geophys. Res., 1994, 99:5771 [3] MURAYAMA T. Coupling function between the solar wind and the Dst index//Solar Wind-Magnetosphere Coupling[M]. Tokyo: Springer, 1986:119-126 [4] THOMSEN M F, BOROVSKY J E, MCCOMAS D J, COLLIER M R. Variability of the ring current source population[J]. Geophys. Res. Lett., 1998, 25:3481 [5] KLIMAS A J, VASSILIADIS D, BAKER D N. Dst index prediction using data-derived analogues of the magnetospheric dynamics[J]. J. Geophys. Res., 1998, 103:20435 [6] BAN P P, SUN S J, CHEN C, ZHAO Z W. Forecasting of low-latitude storm-time ionospheric f0F2 using support vector machine[J]. Radio Sci., 2011, 46, RS6008. DOI: 10.1029/2010RS004633 [7] WU J G, LUNDSTEDT H. Prediction of geomagnetic storms from solar wind data using Elman recurrent neural networks[J]. Geophys. Res. Lett., 1996, 23:319 [8] LUNDSTEDT H, GLEISNER H, WINTOFT P. Operational forecasts of the geomagnetic Dst index[J]. Geophys. Res. Lett., 2002, 29(24):2181 [9] JI E Y, MOON Y J, PARK J, et al. Comparison of neural network and support vector machine methods for Kp forecasting[J]. J. Geophys. Res., 2013, 118:5109-5117 [10] LE G M, CAI Z, WANG H, ZHU Y. Solar cycle distribution of great geomagnetic storms[J]. Astrophys. Space Sci., 2012, 339:151-156 [11] LE G M, CAI Z, WANG H, YIN Z, LI P. Solar cycle distribution of major geomagnetic storms[J]. Res. Astron. Astrophys., 2013, 13(6):739-748 [12] VAPNIK V. Statistical Learning Theory[M]. New York: John Wiley, 1998 [13] VAPNIK V. The Nature of Statistical Learning Theory[M]. New York: Springer-Verlag, 1995 [14] GAVRISHCHAKA V V, GANGULI S B. Support vector machine as an efficient tool for high-dimensional data processing: application to substorm forecasting[J]. J. Geophys. Res., 2001, 106:29911-29914 [15] LI R, CUI Y, HE H, WANG H. Application of support vector machine combined with K-nearest neighbors in solar flare and solar proton events forecasting[J]. Adv. Space Sci., 2008, 42:1469-1474 [16] HUANG C, LIU D D, WANG J S. Forecast daily indices of solar activity, F10.7, using support vector regression method[J]. Res. Astron. Astrophys., 2009, 9:694-702 [17] CHEN C, WU Z S, XU Z W, et al. Forecasting the local ionospheric f0F2 parameter 1 hour ahead during disturbed geomagnetic conditions[J]. J. Geophys. Res., 2010, 115, A11315. DOI: 10.1029/2010JA015529 [18] LIU D D, HUANG C, LU J Y, WANG J S. The hourly average solar wind velocity prediction based on support vector regression method[J]. Mon. Not. Roy. Astron. Soc., 2011, 413:2877-2882 [19] CHOI S, MOON Y J, VIEN N A, PARK Y D. Application of support vector machine to the prediction of geo-effective halo CMEs[J]. J. Korean Astron. Soc., 2012, 45:31-38 [20] RODRÍGUEZ J D, PÉREZ A, LOZANO J A. Sensitivity analysis of k-fold cross validation in prediction error estimation[J]. IEEE Trans. Pattern Anal., 2010, 32(3):569-575 [21] YANG F H, WHITE M A, MICHAELIS A R, et al. Prediction of continental-scale evapotranspiration by combining MODIS and AmeriFlux data through support vector machine[J]. IEEE Trans. Geosci. Remote Sens., 2006, 44(11):3452-3461 [22] CRISTIANINI N, SHAWE-TAYLOR J. An Introduction to Support Vector Machines and Other Kernel-Based Learning Methods[M]. Cambridge: Cambridge University Press, 2000
点击查看大图
计量
- 文章访问数: 976
- HTML全文浏览量: 25
- PDF下载量: 767
- 被引次数: 0