河口网河区水文相关模拟的人工神经网络方法与应用

ANN Technique and Its Application in Hydrological Correlation Simulation for A River Network in Estuaries

珠江河口网河区的水文过程具有典型的非线性特征。采用人工神经网络（ＡＮＮ）的反误差传播模型，以西、北江径流水位及河口潮流水位作为输入，以顺德市主要水道控制站点水位变化作为响应输出，进行相关训练，模拟水位变化过程。通过１９９４年６～８月实际水位过程的可行性探讨和１９８１～１９９５年历史长序列的检验，相关验证成果精度达到水利部部颁标准，优于同时段多元线性回归的计算成果。在此基础上，对１９９６年７月、１９９７年７月两场洪水进行事后相关验证，成果精度均达到水利部部颁标准。并按网河区防洪与水利管理的要求，研制出ＡＮＮ方法的应用软件包。

The hydrological process of the river network in Pearl River Estuary has a typical characteristic of nonlinearity. A stage process is simulated by means of an counter - error propagation model adopting Artificial Neurolog Network (ANN). In the simulation, a correlation training is undertaken. The runoff stage in the West River and the North River and tidal current stage in the estuary is taken for input, and the stage variation at the control stations in the main waterways in Shunde for response. The feasibility analysis of the field stage data in June to August, 1994 and the verification with the long data series in 1981 - 1995 indicates that the result has an accuracy consistent with the standard issued by the Ministry of Water Resources, and higher than the result of multielement linear regression. Based on this, a correlation verification is carried out for the two flood processes in July, 1996 and July, 1997 after happening. The accuracy of the result also reaches the standard issued by the Ministry. An applicable soft ware of ANN technique has been developed.