湖泊流域汇水径流过程的模拟预测是一种复杂系统中的时间序列分析问题。模型选择上,现有的机理模型法与辨识模型法各有利弊。同时,现有的模型多采用静态数据驱动模拟,不能有效利用传感网实时观测数据来改善模拟不确定性的问题。本文基...湖泊流域汇水径流过程的模拟预测是一种复杂系统中的时间序列分析问题。模型选择上,现有的机理模型法与辨识模型法各有利弊。同时,现有的模型多采用静态数据驱动模拟,不能有效利用传感网实时观测数据来改善模拟不确定性的问题。本文基于深度循环神经网络技术,提出一种适应动态数据驱动的模式,可融合遥感数据与原位传感器站点数据的DTSM(Dynamic Data Driven Time Series Model)时序模拟预测模型,并在观测值与数值模拟之间建立了一种能动态反馈、自适应调整的模拟框架,解决了传统辨识模型法对时序信息挖掘较弱导致模拟精度较低的问题。通过在鄱阳湖多个子流域入湖径流的案例中验证,显示静态数据驱动模式下,以不同数据源作为输入模拟时,本文DTSM模型的纳希效率系数Ens精度比机理模型提高10个百分点以上;相比静态模式,动态数据驱动模式的模拟精度有进一步提高,尤其是对于静态模式精度较低的流域,提高更为明显。展开更多
In the near-shore waters, the actual flow is mainly induced by tide, wind and salinity, and the river water runoff should also be included as a component in the estuary waters. The interactions among these major compo...In the near-shore waters, the actual flow is mainly induced by tide, wind and salinity, and the river water runoff should also be included as a component in the estuary waters. The interactions among these major components are very complicated. Many approaches were proposed to study isolated tide and wind-driven currents or run-off based on the measured velocity, with all its components taken as a whole. In this article, firstly, based on the actual hydrodynamic characteristics of estuarine and coastal waters, an approach is proposed to separate the measured velocity by considering the theoretical current velocity profiles and using the least squares method. The vertical structures of tidal, wind-driven currents, density current and runoff can be obtained as well as their proportions in the measured velocity. Then, this approach is applied to the analysis of velocity data obtained in the North Branch of Yangtze River estuary and of laboratory test data. The results are found to be satisfactory. Finally, this approach is used to separate the measured velocity in the South Branch of Yangtze River estuary, to determine not only the bed friction velocity and roughness height, but also the surface wind stress, and to estimate the wind velocity data above the water surface. The results show that this method is simple in principle, practical in use, and reasonable in obtained results. So it can be used to effectively analyze the field data.展开更多
文摘湖泊流域汇水径流过程的模拟预测是一种复杂系统中的时间序列分析问题。模型选择上,现有的机理模型法与辨识模型法各有利弊。同时,现有的模型多采用静态数据驱动模拟,不能有效利用传感网实时观测数据来改善模拟不确定性的问题。本文基于深度循环神经网络技术,提出一种适应动态数据驱动的模式,可融合遥感数据与原位传感器站点数据的DTSM(Dynamic Data Driven Time Series Model)时序模拟预测模型,并在观测值与数值模拟之间建立了一种能动态反馈、自适应调整的模拟框架,解决了传统辨识模型法对时序信息挖掘较弱导致模拟精度较低的问题。通过在鄱阳湖多个子流域入湖径流的案例中验证,显示静态数据驱动模式下,以不同数据源作为输入模拟时,本文DTSM模型的纳希效率系数Ens精度比机理模型提高10个百分点以上;相比静态模式,动态数据驱动模式的模拟精度有进一步提高,尤其是对于静态模式精度较低的流域,提高更为明显。
基金supported by the Public Welfare Projects of Ministry of Water Resources (Grant No.200701026)the National Natural Science Foundation of China (Grant Nos. 49971064, 50339010)
文摘In the near-shore waters, the actual flow is mainly induced by tide, wind and salinity, and the river water runoff should also be included as a component in the estuary waters. The interactions among these major components are very complicated. Many approaches were proposed to study isolated tide and wind-driven currents or run-off based on the measured velocity, with all its components taken as a whole. In this article, firstly, based on the actual hydrodynamic characteristics of estuarine and coastal waters, an approach is proposed to separate the measured velocity by considering the theoretical current velocity profiles and using the least squares method. The vertical structures of tidal, wind-driven currents, density current and runoff can be obtained as well as their proportions in the measured velocity. Then, this approach is applied to the analysis of velocity data obtained in the North Branch of Yangtze River estuary and of laboratory test data. The results are found to be satisfactory. Finally, this approach is used to separate the measured velocity in the South Branch of Yangtze River estuary, to determine not only the bed friction velocity and roughness height, but also the surface wind stress, and to estimate the wind velocity data above the water surface. The results show that this method is simple in principle, practical in use, and reasonable in obtained results. So it can be used to effectively analyze the field data.
