Probabilistic performance prediction of shield tunnels in operation through data mining

Abstract

To investigate the inherent uncertain and dynamic deterioration of metro shield tunnels in operation, the Bayesian ordered probit model, a data mining method, was used in this study. Through Markov Chain Monte Carlo (MCMC) simulation, the uncertainty in parameter estimation was significantly reduced, and the confidence in the results was improved. The effects of influencing variables on the deterioration process were evaluated. It was found that the tunnel sections with greater burial depths were more likely to deteriorate than the shallow ones. Crossing below a river or near a station or cross passage accelerated the deterioration rate. The deterioration probability increased as the tunnel became older. Finally, the model was applied to a tunnel section. It was shown that the probability of the best state decreased while that of the worst state increased with age. For states between the best and the worst, the probability would first increase, reach a peak, and then decrease. This study found that the ordered probit model with MCMC was a valuable probabilistic method for performance prediction, which is crucial for cost-effective decision-making in future work.