基于CEEMDAN-TCN模型的河南省月降水量预测  

Prediction of Monthly Precipitation in Henan Province Based on CEEMDAN-TCN Model

作　　者：王硕 陈中举[1] 许浩然 黄小龙 WANG Shuo;CHEN Zhong-ju;XU Hao-ran;HUANG Xiao-long(School of Computer Science,Yangtze University,Jingzhou 434023,Hubei Province,China)

机构地区：[1]长江大学计算机科学学院,湖北荆州434023

出　　处：《节水灌溉》2023年第8期26-33,共8页Water Saving Irrigation

基　　金：湖北省教育厅科学研究项目(B2021052)。

摘　　要：针对水文时间序列非线性难以预测的特性,为进一步提高降水量的预测精度,提出一种基于自适应噪声的完备经验模态分解(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)和时间卷积网络(Temporal Convolutional Network,TCN)的耦合模型,使用河南省1960年1月-2000年7月的月降水量数据,对2000年8月-2017年12月降水量进行预测。模型使用CEEMDAN将原始不平稳的降水量序列分解为一组相对平稳的子序列分量,再利用TCN对各子序列分别进行预测,将各子序列分量的预测结果叠加得到最终结果。为验证模型的有效性,将该模型与LSTM、TCN、CEEMDAN-LSTM模型进行对比。结果表明,CEEMDAN-TCN模型预测精度最高,相较于3种对比模型RMSE分别减少了42.82%、35.65%、18.12%,MAE分别减少了37.75%、27.53%、19.39%。在空间分布上,使用普通克里金插值法得到的CEEMDAN-TCN预测值与实际值的空间分布接近。综上,CEEMDAN方法可以有效降低月降水量数据的不平稳性,耦合CEEMDAN方法的组合模型较单一模型预测精度更高;CEEMDAN-TCN模型相较3种对比模型的预测精度均有不同程度提升,该方法将CEEMDAN信号分解技术、深度学习模型与降水量预测领域相结合,有效地提升了月降水量预测精度。In order to further improve the prediction accuracy of precipitation in response to the non-linear and difficult-to-predict nature of hydrological time series,the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and Temporal Convolutional Network(TCN)coupling model was proposed.Based on the monthly precipitation data of Henan province form January 1960 to July 2000,the monthly precipitation from August 2008 to December 2017 was predicted.The model used CEEMDAN to decompose the original unstationary precipitation sequence into a set of relatively stationary subsequence components,then predicated each subsequence component with TCN model.Finally the prediction result of subsequence components were superimposed to obtain the final results.In order to verify the effectiveness of the model,the model was compared with LSTM,TCN,and CEEMDAN-LSTM models.The results show that the CEEMDAN-TCN model proposed in this paper has the highest prediction accuracy,with 42.82%,35.65%,and 18.12%reduction in RMSE and 37.75%,27.53%,and 19.39%reduction in MAE,respectively,compared with the three comparison models.In the space distribution,the predicted values of the CEEMDAN-TCN model by using Ordinary Kriging interpolation are similar to the space distribution of the actual values.In summary,the CEEMDAN method can effectively reduce the nonstationarity of monthly precipitation data,and the combined model coupled with the CEEMDAN method has higher prediction accuracy than a single model;the CEEMDAN-TCN model proposed in this paper has different degrees of improvement in prediction accuracy compared with the three comparison models,and the method combines the CEEMDAN signal decomposition technology,deep learning model and the field of precipitation prediction,which effectively improves the monthly precipitation prediction accuracy.

关 键 词：降水量预测 模型精度比较 CEEMDAN-TCN 自适应噪声的完备经验模态分解 时间卷积网络 河南省 克里金插值法 

分 类 号：S27[农业科学—农业水土工程] P332[农业科学—农业工程]