GA-BP模型在HSS模型参数取值中的应用

Application of GA-BP Model in Parameter Value of HSS Model

摘要:

小应变硬化土（HSS）模型可以有效反映土的压缩硬化特性和小应变特性，非常适合黄土基坑的数值模拟计算。但是，HSS模型包含了11个硬化土（HS）模型参数和2个小应变参数，而这2个小应变参数往往需要采用试验方法确定，获取过程复杂。为了探讨小应变参数的预测方法，采用经过遗传算法优化的BP神经网络模型，即GA-BP神经网络模型，首先根据预设的小应变参数水平经过数值模拟计算得到49组位移数据，然后将得到的数据用于GA-BP神经网络的训练，待GA-BP神经网络的预测误差达到要求之后，再使用实际的位移数据反演得到小应变参数，最后基于预测得到的小应变参数进行数值模拟。结果显示，GA-BP神经网络模型预测的小应变参数在基坑围护结构最大水平位移和地表最大沉降计算方面表现良好，可以应用于实际工程。

Abstract:

The small strain hardening soil （HSS) model can effectively reflect the compressive hardening characteristics and small strain characteristics of soil, which is very suitable for the numerical simulation calculation of loess foundation pit. But, the HSS model contains 11 parameters of the hardening soil （HS) model and 2 small strain parameters, and these two small strain parameters are often determined by experimental methods, and the acquisition process is complex. In order to discuss the prediction methods of small strain parameters, the BP neural network model optimized by genetic algorithm, namely GA-BP neural network model, is adopted. Firstly, 49 groups of displacement data are obtained by the numerical simulation calculation according to the preset small strain parameter level. Then the obtained data are used for the training of GA-BP neural network. After the prediction error of GA-BP neural network reaches the requirement, the small strain parameters are obtained by using the actual displacement data inversion. Finally, the numerical simulation is carried out for the small strain parameters according to the prediction. The result shows that the small strain parameters predicted by GA-BP neural network perform well in the calculations of the maximum level of foundation pit support structure and the maximum settlement of earth surface, which can be used for the practical projects.