应用机器学习方法的太阳质子事件短期预报模型  

SPE short-term prediction model with machine learning method

作　　者：李蓉[1] 崔延美[2] 

机构地区：[1]北京物资学院,北京101149 [2]中国科学院空间科学与应用研究中心,北京100090

出　　处：《中国科学：物理学、力学、天文学》2010年第7期938-944,共7页Scientia Sinica Physica,Mechanica & Astronomica

基　　金：国家自然科学基金重点项目(批准号:10973020);北京市属高等学校人才强教计划资助项目(编号:PHR200906210);北京市教育委员会科研基地建设项目(编号:WYJD200902);"十一五"国家科技支撑计划重点项目课题:区域性国际物流综合服务系统与应用示范北京市哲学社会科学规划项目(编号:09BaJG258)资助

摘　　要：本文选取三个描述太阳活动区磁场复杂性和非势性的特征物理量纵向磁场最大水平梯度|-hBz|m,强梯度中性线长度L,孤立奇点数目η.对这三个参量统计计算后结果作为预报因子,应用支持向量机作为预报方法建立一个基于磁场特征物理量的太阳质子事件短期预报模型,该模型可以预报活动区未来24小时是否爆发太阳质子事件.2002和2003年连续两年的样本检测并和基于传统预报因子的模型进行了比对,结果显示预报模型具有较高的准确率和较低的虚报率,从而验证了太阳光球磁场参量作为太阳质子事件预报因子的有效性.Solar proton events （SPEs） prediction is an important part of the space weather forecasts.The forecast method and predictor are two important elements in the research of proton event.Neural network method and traditional sunspot parameters are mainly taken in existed forecast models.Studies suggested that solar flares and coronal mass eruptions （CME） are closely correlated to the complexity of solar magnetic fields.In this paper,three magnetic physical properties （the maximum horizontal gradient of longitudinal magnetic field ｜△↓-hBz｜m,the length of neutral line with the strong gradients L,the number of singular points η） are selected and computed as predictors.The support vector machine （SVM） are applied as forecast method to constructed a SPE short-term prediction model,which predict the occurrence or not of a SPE in the future 24 hours.The samples of the year 2002 and 2003 were used to test the model and the results were compared with the model based on the traditional predictor.The results show that this prediction model has a higher accuracy rates and a lower false alarm rate,which demonstrated that solar photospheric magnetic properties are very available in the SPE forecast models.

关 键 词：太阳光球磁场 磁场复杂性和非势性 支持向量机 太阳质子事件预报 

分 类 号：P182.9[天文地球—天文学]