黑龙江干流水位预报方法研究

The method of water-level forecasting for the Heilongjiang mainstream

选用基于混合线性回归模型的黑箱模型方法对黑龙江干流中上游段包括漠河、鸥浦、三道卡、长发屯、乌云、勤得利6个站点进行水位过程模拟,探讨我国江河河道的水位预报问题。模型结构采用AIC准则和DW检验综合确定,参数通过最小二乘估计,模型以2006-2010年汛期(5-9月)日时间尺度的水文观测资料进行参数率定,以2011年汛期日时间尺度的水文观测资料进行模拟验证;并对模型输出不确定性,参数敏感性进行分析研究。结果表明:1所提出的模型方法在黑龙江干流地区应用较好,6个站点的预报合格率均达到85%以上,确定性系数也都为0.97以上,预报方案精度均能达到甲等,可以用于今后干流作业预报。2通过统计Monte Carlo模拟的输出变量标准差量化了模型输出不确定性,标准差越大,说明预报的洪水过程线越不确定,从而模型不确定性也就越大;运用标准回归系数敏感性分析得出,模型中干流入流不同时间阶数中最小的参数最为敏感,支流入流不同时间阶数中最小的参数次之,时间阶数大的参数一般不太敏感。

The black-box model based on multivariate hybrid linear regression is applied for water-level simulation of the upper and middle reaches of the Heilongjiang mainstream,including the Mohe station,the Oupu station,the Sandaoka station,the Changfatun station,the Wuyun station,and the Qindeli station. The model structure is determined by AIC and DW method; parameters of the model are estimated by the least square method. The models are calibrated with day time-scale hydrologic data of 2006-2010 in the flood season（ May to September）,and verified with day time-scale hydrologic data in the flood season（ May to September） in 2011. And the model＇s uncertainty and parameters sensivitity are analyzed. The results are as follows： 1 The model is applied well in the Heilongjiang mainstream and accurate within acceptable range. Forecast schemes＇ pass rate reaches more than 85% and coefficients of determination are above 0. 97 for six stations. The model can be used for water-level forecasting. 2 The paper quantifies uncertainty of the Multivariate hybrid linear regression model results using the Monte Carlo methods. The larger the standard deviation,the more uncertain the flood hydrograph forecasting,and thus model is more uncertain. The parameters＇ sensitivity analysis based on standardized regression coefficients shows that the parameters with minimum values in time order of mainstream are the most sensitive,the parameters with minimum values in time order of tributary take the second place,and the parameters with higher values in time order are less sensitive.