基于神经网络的大连东北海岸带碎石土地基强夯加固研究

Study on dynamic compaction of gravelly soil foundation in northeast coast of Dalian based on neural network

以大连某实际工程作为研究场地,室内试验与原位测试所得碎石土地基物理力学参数与实测所得强夯处理沉降量作为样本,通过BP神经网络对样本的训练、学习,建立地基土力学参数与强夯处理的沉降量之间的映射关系,利用所得映射关系对场地实测的沉降量进行物理力学参数的反演分析。结果表明:经过训练的神经网络模型可快速得出所需参数,利用flac3d以反演所得参数进行计算,模拟沉降量与实测沉降量的误差为4.87%,在可接受的范围之内;基于神经网络的位移反分析方法可以省去繁琐的测试工作,但该方法的实现需要有充足的样本数据作为支撑。

Taking a practical project in Dalian as the research site,the sample data were obtained from the laboratory test and in-situ test of the physical and mechanical parameters of gravelly soil foundation and the measured settlement of dynamic compaction.By training and learning the samples using BP neural network,the mapping relationship between the mechanical parameters of foundation soil and the settlement amount from dynamic compaction was established,which was then used for the inversion analysis of physical and mechanical parameters of measured settlement.The results showed that the trained neural network model can quickly output required parameters.Flac3d was used to calculate the parameters obtained by inversion.The difference between simulated settlement and measured settlement was 4.87%,which is within the acceptable limit.The displacement inverse analysis based on neural network can save tedious testing workflow,but the implementation of this method needs the support of sufficient sample data.