Elastic analytical method with machine learning for predicting the stratum displacement field induced by shallow tunneling

Abstract

Support vector regression (SVR) with sparrow search algorithm (SSA) is developed as the machine learning (ML) model to predict maximum surface settlement smax caused by tunneling. A novel method for calibrating boundary conditions of analytical solution is proposed, where the maximum surface settlement derived by the analytical method is equal to smax predicted by SSA-SVR method. The elastic analytical solution for stratum displacement of a shallow tunnel is presented by the complex variable method, when the calibrated nonuniform displacement function is applied as the tunnel displacement boundary condition. The proposed analytical solution-machine learning (AM) method can predict the stratum displacement field prior to the tunnel excavation. Seventy-three tunnel engineering cases are employed to verify the rationality of the proposed SSA-SVR method in predicting smax. The value of R2 in the training and test process is 0.894 and 0.877, respectively. Taking Heathrow Express Trial Tunnel as an example, the potential of the proposed AM method in predicting stratum displacement is presented where the influence of cohesion strength, internal friction angle, Young's elastic modulus of stratum, tunnel radius and depth are considered. The proposed AM method can well predict the stratum surface settlement trough curve, vertical and horizontal displacement at different positions.