基于改进BP神经网络的隧道位移反分析研究

Research on Back Analysis of Tunnel Displacement Based on Improved BP Neural Network

通过LM算法改进的BP神经网络位移反分析理论及数值模拟获得隧道位移反分析模型,基于现场勘测位移数据反演出围岩关键物理力学参数,将反演结果与该隧道区段的岩石试验所得参数进行对比,符合度高达97.3%,说明LM-BP神经网络位移反分析模型能较好地反映隧道中围岩物理力学参数与隧体形变之间的非线性模糊特性。此外,将该反演模型运用到与样本位置相距5m、10m、15m的区段进行力学参数反演,得出模型可靠性较高的区段范围,可在一定程度上通过反分析模型及该区段的岩石力学参数反推工程后续施工的围岩位移变化特征。

The tunnel displacement back analysis model is obtained through the BP neural network displacement back analysis theory and numerical simulation improved by the LM algorithm,and the key physical and mechanical parameters of the surrounding rock are inverted based on the field survey displacement data.The inversion results are compared with the rock test parameters of the tunnel section,the agreement is as high as 97.3%,indicating that the LM-BP neural network displacement back analysis model can better reflect the nonlinear fuzzy characteristics between the physical and mechanical parameters of the surrounding rock in the tunnel and the deformation of the tunnel.In addition,the inversion model is applied to the sections 5m,10m,and 15maway from the sample location to perform mechanical parameter inversion,and the section range with higher model reliability is obtained,which to a certain extent,the surrounding rock displacement change characteristics of the subsequent construction of the project can be reversed through the inverse analysis model and the rock mechanical parameters of the section.