摘要
鉴于山洪突发性强、历时短、陡涨陡落等致使在模拟预报过程中具有较大难度和不确定性问题,构建了基于深度学习的LSTM网络模型进行山洪确定性预报和概率预报,从精度和可靠度两方面研究其在西南山区的适用性。并以西南山洪易发区寿溪河流域为例进行模拟,结果显示LSTM网络模型更易发现暴雨洪水之间的深层规律,验证期平均纳什效率系数达0.954,与BP模型相比,显著提升了洪水预报精度,尤其是大洪水;概率预报有效降低了山洪预报的不确定性,洪峰附近的流量数据基本落入预报区间内,有效提高了预报可靠度。
Because flash flood has the characteristics of strong suddenness, short duration, and steep rise and fall, it has greater difficulty and uncertainty in the simulation and forecasting process. In this paper, a deep learning-based LSTM network model is constructed for deterministic and probabilistic forecasting of flash flood, and its applicability in southwest mountainous areas is studied in terms of accuracy and reliability. Taking the Shouxi River Basin in the southwest mountain torrent-prone area as the research area, the results show that the LSTM network model can better find the deep law between storms and floods. The average Nash efficiency coefficient during the verification period reached 0.954, which is a significant improvement compared with the BP model. The accuracy of flood forecasting is improved, especially for large floods. The probability forecast effectively reduces the uncertainty of flash flood forecasting, and the flow data near the flood peak basically falls within the forecast interval, which effectively improves the forecast reliability.
作者
尹兆锐
李红霞
唐萱
龚志惠
YIN Zhao-rui;LI Hong-xia;TANG Xuan;GONG Zhi-hui(College of Water Resource&Hydro power,Sichuan University,Chengdu 610065,China;State Key Laboratory of Hydraulics and Mountain River Engineering,Sichuan University,Chengdu 610065,China)
出处
《水电能源科学》
北大核心
2022年第2期88-91,共4页
Water Resources and Power
基金
国家重点研发计划(2019YFC1510703)
国家自然科学基金面上项目(51979177,51879172)。
关键词
山洪预报
寿溪河流域
深度学习
LSTM网络
区间预报
flash flood forecast
Shouxi River Basin
deep learning
LSTM network
interval forecast
