摘要： 为准确估计堆石料力学本构模型参数，根据堆石料三轴压缩实验观测数据，提出一种基于神经网络的堆石料非线性本构模型参数反演方法。通过对三轴压缩实验轴向和径向应变的分段线性化处理，建立求解垂直荷载与应变之间关系的解析表达式。应用神经网络法对堆石料的力学模型参数进行反演，建立三轴压缩实验轴向和径向应变与模型参数之间的非线性映射关系，并据此进行堆石料模型参数估计。为验证反演方法的有效性，采用施工现场的堆石料进行三轴压缩实验，结果表明，与基于梯度优化搜索的参数估计方法相比，该方法具有更高的预测精度，最大相对误差降低了17.8%。

关键词: 神经网络, 参数反演, 混凝土面板堆石坝, 非线性本构模型, 最大相对误差

Abstract: Based on observed data for three-dimensional compressive test of rockfill material, a parameter inversion method of nonlinear constitutive model is proposed using neural network for estimating parameters of rockfill. The relationships between axial loads and strains are analytically expressed by linear approximating for axial and radial strains in three-dimensional compressive test. In order to validate the effectiveness of the proposed method, the three-dimensional compressive test of rockfill material is performed in laboratory. Experimental result shows that, compared with the estimation method for model parameter based on gradient optimization search, the proposed method provides higher prediction accuracy of the behavior of rockfill material tested, and the maximum relative error decreases by 17.8%.

Key words: neural network, parameter inversion, concrete-faced rockfill dam, nonlinear constitutive model, maximum relative error

中图分类号: 

• TP183