西江年最高水位的神经网络预报模型

对西江洪水发生的特征进行分析表明，洪水发生频率高，具有明显阶段性特征，并与流域面雨量密切相关。利用前期环流场、海表温度（SST）场及环流特征量资料选择初选预报因子，然后对初选预报因子作EOF分解构造综合预报因子，结合人工神经网络方法建立了西江年最高水位预报模型，并对预报模型进行独立样本试验。结果表明，该预报模型对历史样本拟合精度高，试报效果明显好于传统的逐步回归模型，可在汛期预测业务中应用。

A neural network prediction model of annual highest water level in Xijiang River

The characteristic analysis of flood data of the Xijiang River shows that variation of flood occurrence contain some obvious features,such as high frequency,distinct phases and close correlation to the areal precipitation in the Xijiang River drainage area.In this article,comprehensive predictors are presented by using former atmospheric circulation,sea surface temperature(SST),circulation characteristic data and EOF decomposition method.A neural network prediction method is combined to create an annual highest water level prediction model.The model is tested and result shows that this prediction model is significantly better than traditional regression model,and it has a good prospect in business application.