基于神经网络方法的Kp预报模型  

Kp Forecast Models Based on Neural Networks

作　　者：刘杨[1,2] 罗冰显[1] 刘四清[1] 龚建村[1] 

机构地区：[1]中国科学院空间科学与应用研究中心,北京100190 [2]中国科学院研究生院,北京100049

出　　处：《载人航天》2013年第2期70-80,共11页Manned Spaceflight

基　　金：国家重点基础研究计划项目(2012CB825606;2011CB811406)

摘　　要：为了更加准确地预报磁暴（Kp〉5）的发生，充分利用ACE卫星积累的上游行星际条件的数据，以开磁通生成速率函数dφMp／dt和太阳风磁层粘滞作用项n^1/2v^2为主要输入参数，应用神经网络方法，构建了三个模型，预报三小时时段的Kp值。根据实际需要，这三个模型采用了不同的训练集构造方法和提前时间量。模型1输入当前的开磁通生成率，粘滞作用项，太阳风速度、密度，和行星际磁场总强度、By分量、Bz分量，提前1～3．5h预报Kp；模型2在模型1的基础上加入Kp现报，提前1～3．5h预报Kp；模型3输入9小时延迟的开磁通生成率和粘滞作用项，当前的太阳风速度、密度，行星际磁场总强度、By分量、Bz分量，提前3小时预报Kp。对1998年、2002年和2006年的测试结果表明：三个模型的预测值与实测值之间的相关系数分别为0．88、0．90、0．85，预测的均方根误差分别为0．65、0．62、0．72。As a global geomagnetic disturbance index, Kp is difficult to predict. Especially when Kp reaches 5, disturbance will reach the level of geomagnetic storm, which may cause spacecrafts and power system anomaly. Many studies have shown that there exists a high correlation between pairs of solar wind-magnetosphere coupling function and Kp index, and the linear predictor of Kp with merging rate and viscous term has been proved to be valuable. By taking advantage of coupling function and handful interplanetary data provided by ACE satellite since 1998, three Kp forecast models have been developed based on artificial neural networks. For different operational needs, the data is constructed to set in an innovative way to train the networks, and Kp is predicted with a lead time up to 3.5 hours. Model 1 inputs merging rate, viscous term, solar wind velocity, density, and interplanetary magnetic field B, By component and Bz component; model 2 takes the same inputs as model 1 in addition to nowcast Kp; model 3 inputs 9-hr delayed merging rate and viscous term, 3-hr solar wind and IMF parameters. The evaluation with the test data set over year 1998, 2002 and 2006 shows that the three models yield the correlation coefficients of 0.88, 0.90, and 0.88 , and the root mean square error （rinse） of 0.65, 0.62, and 0.72,respectively.

关 键 词：KP指数 神经网络 能量耦合函数 开磁通生成率 粘滞作用项 

分 类 号：V520.1[航空宇航科学与技术—人机与环境工程] P353.1[天文地球—空间物理学]