基于C4.5算法的长江中下游地区夏季降水预测模型研究及应用

为了对长江中下游夏季降水进行短期气候预测,利用国家气候中心提供的74项环流指数和NOAA整编的西太平洋型WP指数、MEI指数、ENSO指数等多种全球环流指数资料,归纳整理了影响长江中下游夏季降水的34个前期春季因子,讨论了前期春季因子与夏季降水的关系,并利用这34个前期春季因子通过数据挖掘中的C4.5算法对1951—2013年(63 a)长江中下游夏季降水,建立判别降水偏多以及偏少的两类决策树预测模型,并分别得到5条和7条综合判别规则。随机选取80%左右历史年份数据作为模型的训练集,两模型的训练集准确率分别为94.12%和93.88%,剩余20%年份数据作为模型测试集,模型的测试预测准确率分别达91.67%和85.71%。模型预测应用也显示结果正确。模型研究和应用显示,基于C4.5算法的长江中下游夏季降水预测模型具有较高的预测准确率,模型构建合理有效,判别规则依据大数据理论,广泛考虑相关因子以及因子的排列组合,智能化选择关键因子,易于客观化、自动化实施,为长江流域汛期降水的短期气候预测提供了新的思路与方法。

Research and application of summer rainfall prediction model in the middle and lower reaches of the Yangtze River based on C4.5 algorithm

In order to conduct the short-term climate prediction on summer precipitation in the middle and lower reaches of the Yangtze River, 74 circulation index data from National Climate Center and then global circulation index data such as the Western Pacific (WP) index, Multivariate ENSO Index(MEI), ENSO index and so on from NOAA were used to sum up 34 early spring factors affecting the summer precipitation in the middle and lower reaches of the Yangtze River and to analyze the relationship between early spring factors and summer rainfall. Furthermore, two kinds of summer rainfall decision tree prediction models are constructed for the middle and lower reaches of the Yangtze River during 1951-2013 based on the C4.5 algorithm of data mining. Five discriminant rules for "whether too much" and seven discriminant rules for "whether too little" were obtained. In whole data, about 80% history data are randomly selected as training set for the models with the training set accuracy of 94.12% and 93.88%, respectively. The remaining 20% data are used as the test set for the models with the prediction accuracy of 91.67% and 85.71% respectively. Results show that the summer rainfall prediction models in the middle and lower reaches of the Yangtze River based on C4.5 algorithm have higher prediction accuracy with reasonable and effective models construction. The models'' discriminant rules are based on large data theory, and consider a wide range of related factors and factor permutation and combination. The models are processed by intelligent selection of key factors, and objective and automatic implement is good enough to provide a new idea and method for the short-term climate prediction on precipitation in flood season of the Yangtze River Basin.