随机森林在降水量长期预报中应用.docVIP

随机森林在降水量长期预报中应用 　　摘要：随机森林是21世纪提出的基于分类树的算法，在处理大数据集中具有明显优势，首度将其应用在降水长期预报中。以长江中下游地区1月份降水预报为例，运用随机森林模型构建原则，在74项大气环流因子以及前期月降水中筛选模型预报因子，进行长期降水量预报，并将其与神经网络模型预报效果进行对比，发现随机森林的泛化误差为13%，预报准确率达到75%，而神经网络的预报准确率仅为67%。此外，本研究还对长江中下游地区的汛期降水量进行了长期预报，结果表明，随机森林模型进行降水量长期预报中模拟和预报的效果令人满意，值得进一步研究和应用。 　　关键词：随机森林；长期降水预报；等级预报；泛化误差；重要性因子评价；决策树；神经网络 　　中图分类号：P457.6 文献标志码：A 文章编号2016 　　Longterm rainfall forecasting based on random forest 　　YU Shengnan1，CHEN Yuanfang1，GU Shenghua2，KANG You1，HE Ranran1 　　（1.College of Hydrology and Water Resources，Hohai University，Nanjing 210098，China； 　　2.Hydrological Station of Shanghai，Shanghai，200232，China） 　　Abstract：Random forest is an algorithm based on classification tree that was proposed in this century.It has obvious advantages in dealing with large data set.In this paper，random forest was applied to predict the longterm precipitation.The Yangtze River region′s precipitation in January was taken as an example，the random forest was used to select the important factors from 74 atmospheric circulation factors，and the precipitation monthly by The National Climate Center forecast was used as prediction factors to predict the precipitation.In addition，the neural network forecasting results were compared.The generalization error of random forest model is 13%，and the forecast accuracy rate is 75%，while the rate of neural network accuracy is 67%.Besides，this study also forecasted the class of precipitation of the flood season in the middle and lower reaches of the Yangtze River region.The results suggested that random forest is worthy of further research and application since the simulation and forecasting of the longterm precipitation is relatively good. 　　Key words：random forest；longterm rainfall forecasting；classification forecasting；generalization error；importance factors evaluation；decision tree；neural network model 　　降水长期预报通常是指预报时段为一个月以上的降水量预报[1]。降水预报一直以等级预报作为预报产品。近年来，随着区域经济发展和流域开发在国民经济中的地位日趋重要，特别需要进行区域及流域上的面平均降水量长期预报，以便为政府部门和水库