Dynamic selection of optimal tunnel convergence prediction model for a probabilistic deformation prediction

Abstract

Objective In high geostress or complex geological conditions, tunnel convergence frequently exceeds the threshold, resulting in damage to support structures and, in extreme cases, tunnel collapse. Accurately predicting the deformation trend and convergence of surrounding rock during tunnel construction is crucial to ensuring the safety of workers and improving construction efficiency. Traditional single prediction models struggle to adapt to the dynamic nature of tunnel convergence, limiting their predictive accuracy. Methods To address this, this study introduces a dynamic prediction model for tunnel convergence based on continuous Bayesian updating and an optimal model selection strategy. Utilizing real-time monitoring data of tunnel convergence deformation, the parameters in three empirical models are continuously updated and refined. The optimal model is then selected to predict the final convergence deformation of the surrounding rock and quantify its associated uncertainty. Results The model was tested on 16 measurement points across 9 sections of the Baima Tunnel, achieving a mean relative error of only 3.22% between the predicted and monitored final convergence rates. Conclusion Additionally, with just 10 days of observed data, the model can forecast the final convergence deformation for up to 40 days post-excavation, offering valuable technical support for preventing squeezing disasters in the full-section tunnel excavation.

Keywords

• tunnel
• convergence
• dynamic forecasting
• model selection
• bayesian theory