基于CEEMDAN-SE-ARIMA组合模型的东北夏季降水预测

A hybrid CEEMDAN-SE-ARIMA model and its application to summer precipitation forecast over Northeast China

针对传统时间序列模型无法有效预测模态混叠数据的不足,本文提出了一种基于CEEMDAN-SE-ARIMA的组合模型,并且对东北地区2016—2020年夏季降水量进行了实证分析。首先,基于完全自适应集合经验模态分解方法,将降水时间序列分解为多个本征模态分量,并根据不同分量样本熵的计算结果进行分量序列重构。然后,针对每一个重构分量,构建自回归移动平均预测模型。最后,将各分量的预测值进行叠加,得到组合模型的预测值。此外,还构建了ARIMA单一模型和其他组合模型,旨在与CEEMDAN-SE-ARIMA组合模型对比。结果表明:CEEMDAN-SE-ARIMA组合模型考虑了时间序列的模态混叠特征,能有效提高东北地区夏季降水时序模型的预测能力,具有良好的预测应用价值。预测结果较单一模型和其他组合模型均有所提高,MASE降低了0.02~0.91 mm,RMSE降低了0.80~130.49 mm,MAE降低了2.52~129.84 mm,MAPE降低了1.08~35.53 mm。CEEMDAN-SE-ARIMA模型在降水变率较小的西北部区域预测效果更好,对东南部区域的极值分布中心预测较为准确。

This paper proposes a combination model based on CEEMDAN-SE-ARIMA that aims to address the shortcomings of traditional time series models that cannot effectively predict modal aliased data.The proposed modelcombines the advantages of the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN),the high short-term prediction accuracy of an auto-regressive integrated moving average model(ARIMA),and the fast efficiency of sample entropy(SE)reconstruction.The model is empirically analyzed for summer precipitation in Northeast China from 2016 to 2020.First,based on the fully adaptive ensemble empirical mode decomposition method,the precipitation time series is decomposed into multiple eigenmode components,and the component sequence is reconstructed according to the calculation results of the entropy of different component samples.Then,for each reconstruction component,an autoregressive moving average forecast model is constructed.Finally,the predicted value of each component is superimposed to obtain the predicted value of the combined model.Additionally,the ARIMA single model and other combined modelsare constructed to be compared with the CEEMDAN-SE-ARIMA combined model.The results show that the CEEMDAN-SE-ARIMA combined accounts for the time series’modal aliasing characteristics,effectively improves the forecasting ability of the summer precipitation time series model in Northeast China,and has good forecast application value.Compared with the single model and other combined models,the forecast results are improved.MASE decreases by 0.02—0.91 mm,RMSE decreases by 0.80—130.49 mm,MAE decreases by 2.52—129.84 mm,and MAPE decreases by 1.08—35.53 mm.The CEEMDAN-SE-ARIMA model has a better prediction effect in the northwest region,where the precipitation variability is small,and the prediction of the extreme value distribution center in the southeast region is more accurate.