摘要
针对传统的中长期水文预测方法由于缺乏对水文要素本身内部结构和变化特性的描述,往往导致建模过程中确定模型结构、参数等存在盲目性,而以往常用预测模型收敛速度较慢、模型结构及参数优化复杂等问题,将小波分析(WA)和GRNN神经网络联合使用,建立了中长期水文预测模型:即先应用WA揭示水文序列内部结构及变化特性,从而将原序列分为确定性成分和随机成分两部分,然后利用GRNN神经网络对确定性成分和随机成分分别进行模拟预测,最后将两部分结果叠加作为最终预测值。将该模型用于沱江中上游三皇庙水文站年径流的预测,并与传统方法进行对比。结果显示该模型预测效果较传统方法更好,能有效地揭示序列的时频结构和变化特性,对于生产应用具有较强的实际意义。
Because of the traditional medium-and long-term prediction method used to select parameters without reasonable basis when modeling, and the slow convergence speed, complex structure and parameter optimization process of the models we commonly used, an stochastic medium-and long-term prediction method was put forward based on WA and GRNN. The main idea of this model was as fol- lows ： First, the multi-time scale characters of hydrologic time series were analyzed with WA method. Then GRNN was used to predict the deterministic component and periodic component, respectively. Finally the two components were stacked as final prediction results. The proposed method was used to predict the data of annual runoff series for Tuojiang River. The results showed that the model was of higher accuracy and better qualified rate than traditional prediction models which indicated the validity and applicability for analyzing and modeling hydrological series.
出处
《四川大学学报（工程科学版）》
EI
CAS
CSCD
北大核心
2013年第6期39-46,共8页
Journal of Sichuan University (Engineering Science Edition)
基金
国家"973"重点基础研究发展计划资助项目(2013CB036401)
关键词
GRNN神经网络
小波分析
年径流
中长期预测
水文时间序列
GRNN neural network
wavelet analysis
annual runoff
medium-and long-term hydrologic forecasting
hydrologic time series
