支持向量机方法在山东山洪地质灾害预报中的应用试验

Application of Support Vector Machine Method to Mountain-Flood Disaster Forecast in Shandong Province

用支持向量机(SVM)方法根据T213数值资料和济南、淄博、泰安、莱芜4站的降水实况资料对山东山洪灾害多发的鲁中山区进行了降水分类预报试验。结果表明:多项式核和径向基核函数建立的模型较好地提炼了降水信息,都具有很高的预报技巧,客观性和实用性强,有很强的推广能力;用径向基核函数建的非线性降水分类模型优于用多项式核函数建立的线性降水分类预报,特别是资料减少时,非线性降水分类预报明显优于线性降水分类预报;低层大气湿度可能对线性降水分类有重要影响;建模时用的资料数据格式与实际业务中获得的数据格式应尽量保持一致。

The precipitation classification forecast model of flooding for the mountain areas of central Shandong Province, was built based on T213 data and precipitation data from Jinan, Zibo, Taian and Laiwu by applying the Support Vector Machine （SVM） method. The results indicate that the model based on the polynomial kernel and the model based on radial basic function kernel can extract well precipitation information, and both have good forecast skill and prediction capabilities, but the later is better especially when the data is insufficient. The lower-layer moisture has a great influence on the model based on the polynomial kernel. The data used in modeling should keep as consistent as possible with the data used in operation in data format.