关键词：
烟草普通花叶病毒病;气象因子;预测模型;最大信息系数
摘 要：
研究烟草病害预测方法,可提前预警病害的发生,为制定综合治理方案以及农药合理施用方案等提供有效指导.研究首先引入非线性关联测度方法(Maximal information coefficient,MIC),并以此筛选与烟草普通花叶病毒病相关的气象因子;进一步以支持向量机(Support Vector Machine,SVM)多轮末位汰选进行自变量精细筛选;然后以地统计学(Geostatistics,GS)确定公用变程;对每一个预测样本都从训练集中找出距离小于公用变程的k个近邻,以SVR训练建模完成个体化预测.结果表明:基于k近邻的预测模型独立测试结果明显优于基于全部训练样本参与的预测模型的独立测试精度,且基于私有最近邻样本的个性化预测模型结果最优.
译 名：
A Nonlinear Prediction Model of Tobacco Mosaic Virus Based on Meteorological Factors
作 者：
GUO Sai;WU Ling;College of Information Science and Technology, Hunan Agricultural University;
关键词：
tobacco mosaic virus;;meteorological factors;;prediction model;;maximal information coefficient
摘 要：
Predication methods are very useful to the early warning and plan making for comprehensive treatment and reasonable chemical control of tobacco diseases. A nonlinear prediction model of tobacco mosaic virus based on meteorological factors was set in our study. Firstly, the maximal information coefficient(MIC) was introduced to select the meteorological factors that associated with tobacco mosaic virus. Secondly, the meteorological factors were removed nonlinearly using support vector machine(SVM). Thirdly, a common range was confirmed using geostatistics. Fourthly, K nearest neighbors of each test sample were found from the training set with their distances shorter than common range, then the models were constructed and the individualized prediction was feasible using SVM.The results showed that the KNN models performed better than the model using all training samples. Furthermore, the model based on individualized nearest neighbor sample performed best among the KNN models.
