基于支持向量机回归分析的降水量预测研究

Precipitation Forecast Based on Support Vector Regression Technique

在已有的降水量资料基础上,利用支持向量机回归分析方法建立了降水量预测模型。由于在利用降水量资料进行预测时特征值的选取没有统一标准,且目前鲜有文章涉及相关内容,因此提出了5种不同的特征值选取方法,并建立了相应的模型。通过评价对比,发现选取某月前10年同月的降水量作为特征值的预测模型输出结果精度最高,与实际情况更为接近,能很好地反映降水量的变化趋势。进而对该站未来5年的降水量进行预报。研究表明,在特征值选取合适的情况下,采用支持向量机回归分析对降水量进行预测,结果可靠、方法可行;预报结果和现有资料显示,该站所处地区近15年(2001-2015年)的降水量较早年偏少,处于降水量变化周期的枯水期。

This study employs the support vector regression technique to build precipitation forecast model based on the existing data. Since there is no uniform standards in selecting the feature values when forecasting with precipitation data as well as few papers have relevant contents, this paper puts forward 5 different methods of selecting feature values and builds corresponding models. Through evaluating the outputs of each model, it could be seen that the outputs of the model that selects the precipitation of the same month of the previous 10 years in a row are highly accurate, and the outputs can well reflect the variation tendency of precipitation and con form to the actual condition better. Further, future 5 years＇ precipitation of the station is forecasted. The study shows that under the condition of suitable feature values, using the method of support vector regression technique to forecast precipitation turns out to be reliable and feasible. The forecasted results and existing data indicate that the precipitation of the region where the station lies in is lower in recent 15 years（2001-2015） compared with early years, which means the region will be in the dry stage of the precipitation variation cycle.