摘要
为提高径流预测精度,采用径向基神经网络(RBFNN)数据延拓技术处理完全集合经验模态分解(CEEMDAN)方法中的端点效应问题,并根据分解结果特点构建RBFNN-ARIMA组合预测模型。以1957—2013年黄河源区唐乃亥水文站年径流数据为例,先将选定的序列采用RBFNN进行延拓,然后进行CEEMDAN分解,对得到的分解分量运用RBFNN-ARIMA组合模型进行预测重构得到年径流量预测结果。研究表明,原始序列经过RBFNN数据延拓后再进行CEEMDAN分解,其所得分量可以有效反映不同时间尺度上的波动特征;ARIMA模型对高频IMF1分量的拟合效果较差,对其他中低频分量拟合效果较好;RBFNN-ARIMA组合模型预测结果的平均相对误差为5.23%,相较于RBFNN模型和ARIMA模型预测精度分别提高了9.88%和5.62%。因此,运用基于CEEMDAN方法的"分解-预测-重构"模式进行水文预测,对原始序列进行合理延拓并针对各分量特点进行组合预测可有效提高预测精度。
In order to improve the accuracy of runoff prediction,the RBFNN data extension technique is used to suppress the endpoint effect in the CEEMDAN decomposition process,and the RBFNN-ARIMA model is constructed for the decomposition results.The model is applied to the data of Tangnaihai Hydrological Station from 1956 to 2013 to predict the annual runoff.The results show that CEEMDAN decomposition after the data extension of RBFNN could effectively reflect the fluctuation characteristics on different time scales.ARIMA model has poor fitting effect on high frequency IMF1 component,but good fitting effect on other middle and low frequency components.The average relative error of prediction results of the RBFNN-ARIMA model is 5.23%,which is 9.88%and 5.62%more accurate than that of the single RBFNN model and ARIMA model.When the“combination-prediction-reconstruction”model based on CEEMDAN method is used for hydrological prediction,the prediction accuracy can be effectively improved by extending the original sequence and applying the combinatorial prediction model according to the characteristics of each component.
作者
张金萍
靳有来
ZHANG Jinping;JIN Youlai(School of Water Conservancy Engineering,Zhengzhou University,Zhengzhou 450001,Henan,China;Yellow River Institute for Ecological Protection&Regional Coordinated Development,Zhengzhou University,Zhengzhou 450001,Henan,China)
出处
《水利水电技术（中英文）》
北大核心
2022年第1期55-62,共8页
Water Resources and Hydropower Engineering
基金
国家重点研发计划项目(2018YFC0406501)
2018年河南省高校科技创新人才支持计划项目(18HASTIT014)
河南省高等学校青年骨干教师培养计划项目(2017GGJS006)。
关键词
径流预测
完全集合经验模态分解
数据延拓
神经网络
黄河源区
runoff prediction
Complete Ensemble Empirical Mode Decomposition with Adaptive Noise
data extension
Radial Basis Function Neural Network
Yellow River source area
