粗粒土BP神经网络本构模型研究

为了描述具有应力应变关系非线性和剪胀性的粗粒土本构特性,利用改进的BP神经网络算法,通过优选网络结构和对粗粒土的大型三轴固结排水剪试验数据样本学习,建立了一个以平均主应力p和广义剪应力q作为网络输入向量、以体应变εv和剪应变εs作为网络输出向量的粗粒土BP神经网络本构模型。利用此模型对粗粒土的应力应变关系进行了预测,整体预测结果的最大误差均在10%内。预测表明本神经网络模型具有良好的预测精度和适用性。

A STUDY OF THE BP NEURAL NETWORK CONSTITUTIVE MODEL FOR GRANULAR SOILS

In order to describe the constitutive characteristics of non-linearity of stress-strain relation- ship and dilatancy of granular soils, a BP typed neural network constitutive model, which uses mean prin- cipal stress p and generalized shear stress q as inputs while the volumetric strain ev and shear strain es as the outputs, was suggested based on published experiment data from large scaled consolidated-drained tri- axial shear tests on some granular soils. The model was then used for predicting the stress-strain relations of the granular soils. Generally, the maximum relative errors of the results of prediction are less than 10% ,Which demonstrate the neural network constitutive model for granular soils are of good accuracy of prediction and serviceability.