基于支持向量机的干旱预测研究  

Study of Drought Prediction Based on Support Vector Machine

作　　者：樊高峰[1] 张勇[2] 柳苗[1] 毛燕军[1] 

机构地区：[1]浙江省气候中心,杭州310017 [2]国家气候中心,北京100081

出　　处：《中国农业气象》2011年第3期475-478,共4页Chinese Journal of Agrometeorology

基　　金：国家科技支撑计划课题(2008BAK50B07)

摘　　要：支持向量机(SVM)是基于统计学习理论的一种智能学习方法,可以用来解决样本空间的高度非线性的模式识别等问题。干旱是气候因子非线性复杂关系相互作用造成水分严重亏缺的一种气候异常反映,本文选择SVM方法,利用8月南方涛动指数、副高强度指数、极涡强度指数等15项因子,基于径向基核函数建立浙江省秋季的干旱预测模型,应用交叉验证方式确定最优模型参数,并进行了预测,对模型的检验结果表明,建立的干旱预测模型能直接对秋季干旱进行预测,并且有较高的准确率,可为气候预测从气候要素预测到气象灾害预测提供一种有效途径。Support Vector Machine（SVM）is an intellectual learning method based on the statistics theory.The SVM can solve problems of complicated non-linear pattern recognition of spatial samples.Drought is a respond of water deficit that resulted from the complicated non-linearity interrelationship of climate factors.Examined fifteen climate factors（southern oscillation index,subtropical high strength index and polar vortex strength index,etc.） by using the SVM method,this study had developed Zhejiang autumn drought prediction model which was based on the RBF kernel function of SVM.The best parameters of SVM for Zhejiang were determined by applying the cross-validation method.The autumn droughts predicted by the model of this study agreed well with the truth facts.These results demonstrated the autumn drought prediction model with a better accuracy rate could act as an effective approach of switching climate factor prediction to meteorological hazard prediction.

关 键 词：支持向量机 模式识别 干旱预测 

分 类 号：S423[农业科学—植物保护]