基于近似支持向量机的能见度释用预报研究

利用2008—2010年逐年12月、次年1月的T511L61数值预报产品和单站观测资料，采用近似支持向量机方法，分别建立了南京、杭州和衢州站分类和回归结合的能见度释用预报模型（简称分类和回归结合模型）。利用2011年12月、次年1月资料作为独立样本，对模型进行试报检验，并与不分类条件下的纯回归模型进行对比。结果表明:分类和回归结合模型的预报效果好于纯回归模型，在24、36、48、60和72 h试报中，分类和回归结合模型的南京、杭州和衢州三站平均的准确率依次为75.5%、83.7%、72.1%、75.4%和78.0%，在除48 h的其余4个预报时次中，分类和回归结合模型的三站平均的准确率均高于纯回归模型。分类和回归结合模型在单站能见度预报中有较好的应用前景。 

VISIBILITY FORECAST BASED ON PROXIMAL SUPPORT VECTOR MACHINE

Based on T511L61 numerical prediction products and observed stations data of December and January from 2008 to 2010, a classification-regression forecast model is established for visibility at Nanjing, Hangzhou and Quzhou stations by using proximal support vector machine.The model is tested using independent samples of December and January in 2011, and compared with the regression one.The results indicate that the forecasting effect of the classification-regression model is better than that of the regression one.The average accuracy of classification-regression model for the three stations in 24 h, 36 h, 48 h, 60 h and 72 h is 75.5%, 83.7%, 72.1%, 75.4% and 78.0%, respectively. Its average accuracy is also higher than that of regression one.The classification-regression model is suitable for forecasting visibility at these stations.