为了降低双屈服面数值建模的成本,提高模型的精度,提出了利用中密砂在等p路径下应力-塑性应变曲线具有归一化特征进行建模的方法。该方法选择合适的归一化指标对砂土三轴试验数据进行归一化,以归一化的试验数据为训练样本进行神经网络训练,得到了比较理想的砂土的双屈服面神经网络模型。模型的仿真值与试验值符合较好,表明该文所给出的建模方法是合理的。所提出的建模方法可以在所有试验数据的基础上自动实现概率寻优,能有效降低噪声信号的干扰、减小试验数据的分散造成的影响。 In order to reduce the cost of double-yield surface numerical modeling and improve the accuracy of the model, a method of modeling normalized stress-plastic strain curves using medium sand in equal p-path is proposed. In this method, the normalized index of sand is selected to normalize the triaxial test data of sand, and the normalized test data is used to train the neural network for training samples. The ideal model of double yield surface neural network for sand is obtained. The simulation value of the model is in good agreement with the experimental value, which shows that the modeling method given in this paper is reasonable. The proposed modeling method can automatically realize the probabilistic optimization based on all the experimental data, which can effectively reduce the interference of noise signals and reduce the influence of dispersion of test data.