[关键词]
[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。
[Key word]
[Abstract]
Random for est is an algo rithm w hich has o bv ious advantag es in dea ling w ith la rge data set based on classificatio n tr ee and pro po sed in this century In this paper, random for est w as applied to pr edict t he long2term precipitat ion, the Yangt ze River reg ioncs precipitatio n in January is taken as an ex ample, the random forest is used to select the im po rtant facto rs from the 74 at2 mo spheric circulation fact ors and t he precipitation mo nthly by The National Climate Center forecast as predictio n factor s and to predict the precipit ation. Besides, the neur al netw or k fo recast ing results ar e compared. Fo recasting results o f the two models, random for est mo del generalization erro r is 13%, forecast accuracy r ate is 75%, w hile the r ate o f neural netw ork accur acy is 67%. Besides, t his study also for ecasted the class of pr ecipitatio n of the flood season in the middle and low er reaches of the Yan2 g tze r iver region T he results showed that random forest is wo rthy of further research and applicat ion as the simulat ion and fo re2 casting of the lo ng2term pr ecipitation is r elatively go od.
[中图分类号]
[基金项目]
水利部公益性行业专项( 201201068) ; 水利公益性行业科研专项经费项目( 201301066)