Total deformation prediction of the typical loess tunnels

Abstract

The total deformation of loess tunnels is an essential basis for determining reasonable support timing, optimizing support parameters, and predicting collapses. Accurately predicting the total deformation is of considerable significance to the safe design and construction of loess tunnels. In this paper, a representative data sample set of the total deformation in loess tunnels was obtained using numerical simulations, and a back-propagation neural network (BPNN) was adopted. The input parameters were the bulk density, water content, elastic modulus, Poisson’s ratio, cohesion, internal friction angle of the loess, tunnel depth, and primary support strength. The output parameters were the total vertical and horizontal convergence deformation of the loess tunnel. The prediction results showed that the BPNN model could effectively predict the total deformation of a typical loess tunnel. Finally, the sensitivity analysis of the impact factors showed that the primary support strength and internal friction angle were parameters affecting the total deformation the most and least, respectively. This study can provide targeted guidance for the safe construction of loess tunnels.