摘要
本文以黄河利津站和浙江省白溪水库的月径流水文序列为例,在自相关分析的基础上,建立自回归autoregression模型,并参照其结构建立了相应的resilient back propagation神经网络预报模型.比较结果显示:(1)resilient back propagation模型的模拟预报结果与序列的自相关性有密切关系;(2)当序列有较好的自相关性时,可参照autoregression模型建立相应的resilient back propagation模型;(3)与传统autoregression模型相比,resilient back propagation模型能取得更高的预报精度;且随着预报步长增加,resilient back propagation模型的优势更加明显.
Based on two examples of the Lijin station in the Yellow River and the Baixi Reservoir in Zhejiang province, it is verified that reliable and effective models could be established to model and forecast monthly runoff series, when the series is auto-correlative. Two types of modeling methods, the resilient back propagation modeling and the traditional autoregression modeling, are used in the Lijin station to model and forecast the monthly runoff series. The results show that： （1） the performance of Resilient Back Propagation model depends on the autocorrelativity of the original series; （2） only when the series has large value of autocorrelation coefficient, which means it can be simulated by autoregression model, could resilient back propagation model obtain reasonable results; （3） when both autoregression model and resilient back propagation model are used in the same series, resilient back propagation model brings out better performance in absolute relative error and absolute average relative error, especially when the length of forecasting step increases.
出处
《南京大学学报（自然科学版）》
CAS
CSCD
北大核心
2008年第6期666-673,共8页
Journal of Nanjing University（Natural Science）
基金
国家自然科学基金(40725010、40730635、40671030)
