摘要
为提高河川径流的中长期预报精度并延长其预见期,采用小波分析充分提取有用信息,基于BP神经网络和GRNN神经网络,构建了两种小波神经网络耦合模型,测试了Daubechies族中9种母波函数对模型模拟效果的影响,并采用合格率(Q_(QR))、平均相对误差(M_(MPRE))、均方根误差(R_(RMSE))和确定性系数(N_(NSE))等指标评价了模型精度。将该模型应用于金沙江流域向家坝水文站未来1~5个月的径流预报,结果显示,相比于传统BP和GRNN模型,耦合模型具有明显优势,且基于小波分析的BP模型预报结果更接近实测值,预报精度更高,其未来4个月的平均相对误差在±20%以内。表明小波分析方法能充分挖掘隐藏在原始数据中的有用信息,可有效提高耦合模型的预报精度延长预见期,在径流预测方面有明显的优越性。
In order to improve the precision of mid-long term runoff forecasting and prolong forecast period,proposed hybrid models based wavelet analysis and two different neural network models including back propagation neural network（BP）and generalized regression neural network（GRNN）.Moreover,the influence of nine kinds of mother wave functions in Daubechies family on the prediction accuracy was also investigated.Qualified rate（QQR）,mean percent relative error（MMPRE）,root mean square error（RRMSE）and Nash-Sutcliffe efficiency（NNSE）were selected as criteria to evaluate the precision of these models.The hybrid models were carried out to predict 1-to 5-month-ahead runoff in Xiangjiaba station of Jinsha River Basin.The results indicate that the DWT-BP hybrid model performs better than other models and it could reach better accuracy even for 4-month-ahead prediction,and the MMPRE was within the±20%.In summary,the wavelet analysis can extract the information hidden in the original runoff series,which can improve the prediction accuracy and prolong forecast period.Therefore,the hybrid models are promising in monthly runoff prediction.
作者
孙娜
周建中
朱双
李薇
彭甜
SUN Na;ZHOU Jian-zhong;ZHU Shuang;LI Wei;PENG Tian(School of Hydropower and Information Engineering, Huazhong University of Science and Technology, Wuhan 430074, Chin)
出处
《水电能源科学》
北大核心
2018年第4期14-17,32,共5页
Water Resources and Power
基金
国家自然科学基金面上项目(51579107)
国家自然科学基金重大研究计划重点项目(91547208)
国家自然科学基金重点项目(51239004)
关键词
神经网络模型
水文预报
小波分析
月径流预报
artificial neural network model
hydrological forecasting
wavelet analysis
monthly runoff forecasting
