处理非线性分类和回归问题的一种新方法(Ⅱ)——支持向量机方法在天气预报中的应用

A NEW METHOD FOR NON-LINEAR CLASSIFY AND NON-LINEAR REGRESSION Ⅱ: APPLICATION OF SUPPORT VECTOR MACHINE TO WEATHER FORECAST

将SVM (SupportVectorMachine)分类和回归方法首次应用于气象预报试验。利用1 990～ 2 0 0 0年 4～ 9月ECMWF北半球的 5 0 0hPa高度、85 0hPa温度、地面气压的 0 0 :0 0UTC分析场资料 ,建立四川盆地分区面雨量有无大于 1 5mm的SVM分类推理模型、四川盆地内单站气温的SVM回归推理模型 ,进行相应的预报试验 。

A novel weather forecast method using the support vector machine (SVM) is introduced. Both of SVM model of area rainfall categorical forecast of 15 mm excess and SVM model of single-station temperature regression in Sichuan basin are built upon ECMWF analysis fields of 500 hPa height, 850 hPa temperature, and sea level pressure from April to September through 1990-2000. Extensive experiments are performed with performances evaluated by the Threat Scores (TS) or Correlation Coefficient. Empirical results demonstrate much improved performance compared with those given by standard statistic analysis and forecast methods.