摘要
针对第二松花江流域中长期径流预测精度较低问题,为了分析物理预报因子的作用过程以提高汛期洪水预报精度,选取太阳黑子相对数为物理影响因素,进而识别其影响时滞,以影响时滞期内的太阳黑子相对数作为径流预报因子,以汛期(6—9月份)月平均径流为预报项目,采用BP神经网络识别映射关系,采用历史资料作为训练样本,完成网络训练和检验。以第二松花江干流控制性水利工程丰满水库为例,对2017年汛期月平均径流进行预报。结果表明:丰满水库汛期月平均入库流量为1 400 m^3/s,来水频率为11%,定性预报第二松花江流域2017年为丰水年;2017年丰满水库实际来水141.00亿m^3,为多年均值的112%,为偏丰来水年份,来水定性预报正确。该方法的创新点在于:采用全局敏感性分析方法识别太阳活动的影响时滞,以确定预报因子;采用BP网络模拟预报因子与预报项目的复杂非线性相关关系,以构建预报网络。研究成果为2017年吉林省水文预报和防汛决策提供了重要支撑。
Aiming at the problem of lower accuracy of the mid-long term runoff forecast of Second Songhuajiang River Basin, the relative sunspot number is selected as the physical influencing factor for analyzing the effect process of the physical forecast factor to enhance the accuracy of the flood forecast during flood season and then identify its influencing time delay, while the relative sunspot number in the period of the influencing time delay is taken as the runoff forecast factor and the monthly mean runoff (June to September ) is taken as the forecasting item, and then the network training and testing are completed through identifying the mapping relationship with BP neural network and taking the relevant historical data as the training samples. By taking Fengman Reservoir--the water control project of Second Songhuajiang River as the study case, the monthly mean runoff during the flood season of 2017 is forecasted. The result shows that the monthly inflow rate during flood season of Fengman Reservoir is 1 400 m 3/s with the water-incoming frequency of 11%, thus the year of 2017 is qualitatively forecasted to be a wet year for Second Songhuajiang River Basin. The actual inflow of Fengman Reservoir is 14.1 billion m 3 in 2017, which is 112% of the annual mean value and is close to a wet year, thus the qualitative forecast of the inflow is correct. The innovations of this method are as the followings: identifying the influencing time delay of solar activity for determining the forecast factor with the method of global sensitivity analysis and simulating the complicated nonlinear correlation between forecasting factor and forecasting item for establishing forecast network with BP neural network. The study result provides an important support for hydrological forecast and flood control decision-making of Jilin province in 2017.
作者
孙虹
李鸿雁
郭道华
鲍珊珊
SUN Hong;LI Hongyan;GUO Daohua;BAO Shanshan(Hydrology and Water Resources Bureau of Jilin Province (Water Environment Monitoring Center of Jilin Province ),Changchun 130033,Jilin,China;College of New Energy and Environment,Jilin University,Changchun 130021,Jilin,China)
出处
《水利水电技术》
北大核心
2019年第3期45-51,共7页
Water Resources and Hydropower Engineering
基金
国家自然基金委中韩合作项目(51711540299)
吉林省科技厅基础研究项目(20180101078JC)
关键词
太阳黑子
敏感性分析
非线性映射识别
长期径流预报
洪水预报
干旱洪涝灾害
防汛抗旱
sunspot
sensitivity analysis
non-linear mapping identification
long-term runoff forecast
flood forecast
drought and flood disasters
flood control and drought resistance
