图象识别法在磁暴预报中的应用

本文将所研究的全部太阳和地球物理资料时间划分为442个时间段,其中前195段为“下降期”,后247段为“低年期”.按一定的选暴标准将其分为有磁暴时间段“D”和无磁暴时间段“N”.以表的形式提出与磁暴的发生有一定相关的12个物理先兆.利用计算机进行“D”和“N”的识别.结果表明:1)在“下降期”中,综合耀斑指数CFI≥6、持续时间大于10分钟的IV型射电爆发和行星际磁场的扇形边界过地球等现象,是引起磁暴发生的主要原因;2)在“低年期”,当太阳上的耀斑及其相关活动现象不发生时,产生磁暴的主要条件是反映27天重现性的太阳活动源(冕洞发出的高速流)和行星际磁场的扇形边界过地球等先兆;3)在“试验时段”内,“D”和“N”类识别的识别率为73—82%,在“预报时段”内用此图象识别法识别率为73—80%,误识率均不大于30%.用多组阈值(K)进行了正交设计试验,并经过“显著性”检验,本方法成功率的置信度大于97.5%. 

Application of Pattern Recognition Method to the Prediction of Geomagnetic Storms

The solar and geophysical data in the period from January 1st, 1971 to January 18th, 1977 are devided into 442 intervals, each containing 5 consecutive days. The 195 intervals before September 1st, 1973 were difined as "declining phase" and the following 247 intervals as "minimum phase". Each interval can be characterized as geomagnetic disturbed "D" or calm "N" according to the. beginning of geomagnetic storm.On the other hand, we try to define by computer the disturbed days on the base of solar or geophysical features in association with the geomagnetic storms in the first two third of number of intervals (refered to as "learning period") in both declining and minimum phase in order to determine certain threshold values necessary for this method. Then we use this method to identify the "D" and "N" intervals in the remaining one third of number of intervals (refered to as "test period").The main results as follows:(1) In the declining phase the comprehensive flare index (CFI) of importance equal 6 or more, the radio bursts of Type IVwith duration 10 min, or more and the interplanetary magnetic field (IMP) sector boundary passage through Earth are most likely associated with geomagnetic storm.(2)In the minimum phase the solar active source exhibiting 27-day reoccurring tendency (i.e.. the high speed stream from the coronal hole) and the IMFsector boundary passage through Earth are most likely the cause of geomagnetic disturbances.(3)The recognition efficiency is 73-82% in the learning period and 73-80% in the test period with error < 30% in both eases.We also have made experiments of orthogonal design with a member of groups of threshold values and test of significance. The resulting confidence limit in this method is greater than 97.5%.