Machine Learning in Conventional Tunnel Deformation in High In Situ Stress Regions

Abstract

Deformation prediction of extremely high in situ stress in soft-rock tunnels is a complex problem involving many parameters, and traditional analytical solutions and numerical simulations have difficulty achieving satisfactory results. This paper proposes the MIC-LSTM algorithm based on machine learning methods to predict the deformation of soft-rock tunnels under extremely high in situ stress conditions caused by construction. The study first analyzed the difficulties of engineering construction and the construction plan; then, numerical simulation was used to verify the modified construction plan. To prove that the construction plan was reasonable, machine learning was used to analyze the correlation of the various parameters that cause tunnel deformation; then, the future deformation of the tunnel was predicted. The study found that: (1) the new construction scheme contains symmetrical arrangement of bolts and two support structures along the tunnel vault can effectively control the deformation of the tunnel, and meet the requirements of the specification; (2) the rock uniaxial compressive strength had the greatest impact on tunnel deformation, and the rock humidity had the least influence on tunnel deformation; and (3) the prediction curve based on the deep learning model had a higher similarity to the monitoring curve compared with the traditional numerical analysis software. The MIC-LSTM machine algorithm provides a new approach to predicting the deformation of extremely high in situ stress soft-rock tunnels.