基于BP神经网络的胶结砂砾石应力-应变关系预测

Stress-strain Relationship Prediction of Cemented Sand and Gravel Using BP Neural Network

在前期宏观试验基础之上,采用离散元模拟和BP神经网络相结合的方法获取不同胶凝材料掺量和围压下胶结砂砾石的应力-应变关系。根据前期胶凝材料掺量分别为20、40、60、80、100 kg/m3的胶结砂砾石三轴排水剪切试验结果,开展离散元数值模拟。以试验数据为学习样本,开展BP神经网络模型训练,预测胶凝材料掺量分别为30、50、70、90 kg/m3的胶结砂砾石应力-应变关系,并将预测结果和离散元模拟结果进行对比。研究结果表明,BP神经网络能够实现胶结砂砾石应力-应变关系的预测,并在较低围压下具有较好的精度。

Based on macroscopic tests,a combination of discrete element simulation and BP neural network is used to obtain the stress-strain relationships of cemented sand and gravel under different cementitious admixtures and confined pressure.The discrete element numerical simulation is carried out based on the results of triaxial drainage shear tests of cemented sand and gravel with cementation admixtures of 20,40,60,80 kg/m^(3)and 100 kg/m^(3),respectively.Taking the test data as leaning samples,the BP neural network model training is carried out to predict the stress-strain relationships of cemented sand and gravel with cementation admixtures of 30,50,70 kg/m^(3) and 90 kg/m^(3),respectively,and the predicted results are compared with these of discrete element simulations.The results show that the BP neural network can predict the stress-strain relationships of cemented sand and gravel with good accuracy under low confined pressures.