高面板堆石坝的有限元计算分析中本构模型与模型参数的选取至关重要,尤其模型参数的选取与实际参数存在一定的差异性。根据面板堆石坝的特点,结合实测资料,基于BP神经网络选用合理方法构造参数训练样本,并给出了参数反演步骤,对坝体主、次堆石区的邓肯E-B模型参数反演与三维有限元进行了比较分析,计算结果与实测资料较为吻合。 It is very important to select constitutive models and model parameters in finite element analysis of CFRD. In particular, there is some discrepancy between the selection of model parameters and the actual parameters. According to the characteristics of concrete face rockfill dams and the measured data, a reasonable method is adopted to construct training samples based on BP neural network. The parameters inversion steps are given. The parameters of Duncan EB model in main and secondary rockfill dams are retrieved and compared. Three-dimensional finite element analysis and comparison, the calculated results and measured data is more consistent.