基于有限元及神经网络的土体参数反分析

针对采用试验得到的土体力学参数进行模拟计算的结果与实测数据偏差较大的问题,应用有限元与BP神经网络相结合的位移反分析方法,把计算的桩体侧向位移作为神经网络的输入,对应的土体力学参数作为网络输出,对神经网络进行训练.训练成功后,再利用现场实测的桩体侧向位移带入训练好的网络进行反分析,从而得到符合实际的土性参数值,结合某深基坑工程,通过实测值与计算值的对比验证了该方法的科学性.

Inverse analysis of Soil Parameters Based on Finite ElementMethod and Neural Network

In order to bridge the gap between the calculated values by laboratory test and the measured values,the method of displacement inverse analysis combined with BP neural networks is introduced.BP neural network is trained through the input of the calculated displacement of piles and the output of the corresponding soil spatial parameters.After the success of training,these parameters are received through the network simulation of actual measured displacement of piles.A practical engineering illustrates the scientific nature of this method through a comparison of the measured and calculated values.