天空云量预报及支持向量机和神经网络方法比较研究

THE STUDY ON FORECAST OF CLOUD AMOUNT WITH SVM AND ANN METHODS

使用支持向量机和人工神经网络两种方法,分别建立了天空云量的预报模型。利用2001年5月1日～2004年12月31日的武汉市地面、高空观测值及欧洲中心的24小时预报场等资料,通过按不同比例随机抽取样本进行交叉验证的方法,分析了SVM和ANN模型的预报能力和鲁棒性;然后再用全部样本资料建立预报模型,来预报2005年1月1日～5月31日武汉市天空云量。交叉验证和实例预报的结果显示:虽然SVM和ANN模型都表现了较好的预报能力,但SVM的预报能力高于ANN方法,且在计算速度上有ANN无法比拟的优势。

The cloud amount forecast is carried out based on SVM and ANN methods, and comparisons are made between the two methods. Using the surface and high-level observations in Wuhan and numerical forecast field from ECMWF during May 1^st 2001 to December 31^st 2004, cross-validations are performed with random samples in different proportion. It shows that both models are able to forecast cloud amount with robustness. And then the models are rebuilt with the whole sample data to forecast the cloud amount in Wuhan from January 1^st to May 31^st 2005. Cross-validations and real time forecast reveal that SVM is better than ANN in the ability of forecasting cloud amount, especially in computing speed.