摘要
为提高洪水预报精度,以四川省寿溪流域和陕西省青阳岔流域为研究对象,利用径向基函数(RBF)神经网络和降雨方差(S^(2))对洪水进行分级,针对不同等级洪水建立多套流域参数进行洪水分级预报,减小由于全流域使用单一参数集而产生的预报误差,其次建立长短时记忆(LSTM)误差校正模型,对分级洪水预报结果进行校正。结果显示:利用洪水分级预报与LSTM集合校正的方法在寿溪流域应用,验证集场次洪水先后降低洪峰误差0.8%~7.8%和1.23%~4.30%,NSE分别提高0.014~0.053和0.06~0.09;在青阳岔流域应用,验证集场次洪水先后降低洪峰误差4.38%~16.8%和2.19%~8.14%,NSE分别提高0.011~0.053和0.06~0.12,且对峰现时差均有所改善。综合结果表明此误差集合校正方法对流域洪水预报减小预报误差,提高精度有一定的适用性。
In order to improve the accuracy of flood forecasting, the Shouxi basin in the middle reaches of the Yangtze River and the Qingyangcha Basin in the middle reaches of the Yellow River have been used as the research objects, and the radial basis function(RBF) neural network and rainfall variance(S^(2)) have been used to classify floods. Several sets of basin parameters have been established for flood forecasting to reduce forecast errors caused by a single set of parameters. In addition, a long and short-term memory(LSTM) error correction model has been established to correct the classified flood forecasting results. The results have shown that using the method of flood grading forecast and LSTM ensemble correction in the Shouxi watershed has reduced the flood peak error by 0.8%~7.8% and 1.23%~4.30%, and the NSE has increased by 0.014~0.053 and 0.06~0.09 respectively. Applied in the Qingyangcha Basin, the method has successively reduced the peak error by 4.38%~16.8% and 2.19%~8.14% for verification set floods. The NSE has been increased by 0.011~0.053 and 0.06~0.12, and the occurrence time difference of the peak has been improved. Comprehensive results have shown that this error ensemble correction method has certain applicability for reducing forecast errors and improving accuracy of flood forecasting.
作者
李志超
张怡雯
邬强
胡彩虹
LI Zhichao;ZHANG Yiwen;WU Qiang;HU Caihong(School of Hydraulic Science and Engineering,Zhengzhou University,Zhengzhou 450001,Henan,China;Ministry of Water Resources Information Center,Beijing 100053,China)
出处
《水利水电技术（中英文）》
北大核心
2022年第8期41-49,共9页
Water Resources and Hydropower Engineering
基金
国家重点研发计划项目(2019YFC1510703)
国家自然科学基金(51979250)。
关键词
洪水分级
误差校正
LSTM
洪水预报
flood classification
error correction
LSTM
flood forecasting
