摘要
为了提高太阳耀斑预报模型的预报精度,提出了一种结合支持向量机和近邻法(SVM-KNN方法)的太阳耀斑预报方法。将太阳耀斑预报问题看作一个模式识别问题,在此基础上建立新的预报方法。选择太阳活动区的特征参量作为预报因子,如果活动区未来48小时发生大于等于M级耀斑标识为正例样本,未发生耀斑为反例样本,由这些样本组成训练集代入SVM训练算法构造了耀斑预报模型。通过输入活动区的特征参量值,预报模型使用SVM-KNN分类算法预报该活动区未来2天内是否发生太阳耀斑。模拟预报结果表明,新方法比使用SVM方法具有较高的报准率,可以应用到其它太阳活动预报领域。
A prediction method of combining support vector machine （SVM） and nearest neighbors principle （called SVM-KNN method） is proposed in order to improve the prediction accuracy. Firstly, the solar flare prediction is taken as a pattern recognition problem, based on which the new classifying method is applied. The feature parameters of solar active region are selected as predictor. The sample is labeled positive sample ifa solar flare of large and equal than M class is busted in it in the future 48 hours. Otherwise, it is taken as negative sample. The training set composed using these samples is put into SVM training algorithm to construct prediction model. After inputting the parameters value of active region, the SVM-KNN method is applied to predict whether a solar flare will burst in the coming two days. The prediction results show that the new method has higher forecast accuracy than that of SVM method has. This promising method is expected to play an important role in other solar flare forecasting fields.
出处
《计算机工程与设计》
CSCD
北大核心
2009年第15期3605-3607,3611,共4页
Computer Engineering and Design
基金
国家自然科学基金重点项目(10673017)
北京市属高等学校人才强教计划基金项目(PHR200906210)
北京市教育委员会科研基地建设基金项目(WYJD200902)
北京市教育委员会科技计划基金项目(KM200810037001)
关键词
支持向量机
核函数
最优分类超平面
预报因子
代表点
support vector machine
kernel function
optimal separating hyperplane
predictor
representative point
