关键词: 径流量预测, EMD, BP神经网络, 昭通地区, 云南省

Abstract: In order to improve the prediction accuracy of river runoff,the Empirical Mode Decomposition(EMD)method was used to build a runoff prediction model,combined with Back-Propagation Artificial Neural Network.The monthly runoff data from Doushaguan hydrological station in Zhaotong City,Yunnan Province from 1959 to 2015 was used,and four schemes were set to construct a runoff prediction model.In Scheme 1,the runoff of the first month was used to predict the runoff of the next month.In Scheme 2,the runoff of the first two months was used to predict the runoff of the next month.In Scheme 3,the runoff of the first three months was used to predict the runoff of the next month.In Scheme 4,the EMD was used to decompose the original runoff sequence into high frequency item,low frequency item and trend item,then the decomposition data of the previous month were used to predict the decomposition data of the three items in the next month,and finally superimposed as the predicted next month runoff.The results showed that the R2 of Scheme 4(EMD based model)was 0.86,which was the highest among all the four schemes.The research result can provide technical support for the construction of runoff prediction model and the improvement of prediction accuracy.

Key words: runoff prediction, EMD, BP Neural Network, Zhaotong City, Yunnan Province,