中文摘要：

河道径流预报过程可以认为是一种复杂非线性函数关系的逼近过程。BP网络具有表达任意非线性映射的特性，因此建立了基于MATLAB神经网络工具箱的BP网络的径流预报模型。其中采用自相关函数确定网络输入层的神经元数，通过比较样本均方误差值来确定隐含层的神经元数。利用清江渔峡口以上流域1989—1995年的径流量资料对该模型进行了训练和检验,从而完成了该流域年径流量的预报，并且用多项精度评定指标对其进行了精度定量评价。结果表明：所建模型对所选流域的径流预报精度达到了乙等以上水平，具有一定的实用性。

英文摘要：

We established a runoff forecasting model based on the MATLAB neural network toolbox and BP network using auto - correlation function to determine the number of neurons of network input layer and compared the mean square error of values of samples to determine the number of neurons of the hidden layer. The model was trained and tested using the runoff data in the upstream basin of Yuxiakou of Qingjiang River during 1989 - 1995. We completed annual runoff prediction of the basin, and carried out quantitative evaluation on its accuracy based on multiple accuracy assessment indexes. The results show that the accuracy of annual runoff prediction of the model for the selected basin reaches level B or above.