作　　者： ; ; ;

机构地区： 清华大学土木水利学院水利水电工程系

出　　处： 《清华大学学报（自然科学版）》 2003年第11期1538-1541,共4页

摘　　要： 挑流冲刷的深度直接关系到大坝的安全,是泄洪消能工设计的首要依据。为此以200多组原型观测资料为依据,建立了旨在预测冲坑深度的改进向后传播(BP)神经网络模型和广义回归神经网络(GRNN)模型,并对影响BP模型精度的网络拓扑结构、数据处理方式以及网络学习算法进行了分析。利用这两种模型对10个工程的冲坑深度进行了预报,并与传统预报公式的计算结果作了比较。结果表明:这两种模型都能比较准确地对冲刷进行预报,并各自在一定范围内占优;如果将二者联合使用,则预测结果明显优于传统公式。 The scour depth is always an important factor in designing discharge structures, since the scour by free jets on river beds can threaten dam safety. Two artificial neural networks were developed based on more than 200 sets of field data. One was a modified back propagation (BP) neural network while the other was a generalized regression neural network (GRNN). Both were used to predict the maximum scour depth. The factors that affect the BP model precision, such as network topology, preprocessing of training data, and the transfer functions, were analyzed to improve the results. The predicted results from the two models for ten examples were compared with that of traditional prediction equations. The results show that each model can accurately predict the scour depth and a combination of the two models gives a better accuracy than the existing formulae. 

关 键 词： 人工神经网络 挑流冲刷预报 挑流消能 冲坑深度 大坝

领　　域： [水利工程]