面板堆石坝沉降变形预测的神经网络方法

Neural networks method for predicting settlement deformation of concrete face rockfill dam

运用MATLAB语言提供的newff函数创建BP神经网络,研究面板堆石坝坝体最大沉降变形与坝高、坝体变形模量和河谷形态之间的内在联系.对样本集内的数据采用lnx函数进行非线性变换和比例归一化处理,使样本数据分布于[0.2,0.8]区间内,可改善网络的性能.对网络结构及训练样本的确定进行了分析,通过工程应用验证网络结构的可靠性及预测精度.结果表明,用试错法确定的网络结构是可靠稳定的,对坝体最大沉降变形预测的精度高于常用的工程类比法,此方法可应用于实际工程.

The relationships between the maximum settlement deformation of concrete face rockfill dam （CFRD） and dam height, deformation modulus of rockfill materials, and topographic conditions of dam site have been researched by constructing BP neural networks with Newff function of MATLAB language. The raw data in specimen set were taken nonlinear transformation using lnx function and normalization in proportion. It can improve the functions of networks to make the data in specimen set distribute in [0.2, 0.8]. Selections of the network structures and training specimen were discussed; the reliability and prediction accuracy of the network structures have been verified by applications of them to engineering. The results show that the network structures determined by trial and error procedure are of reliability and stability; the accuracy of prediction of the maximum settlement deformation of CFRD is better than one of engineering analogy method; the method proposed can be applied to engineering practice.