A New Bayesian Network Model for the Risk Assessment of Water Inrush in Karst Tunnels

Abstract

Water inrush seriously restricts the safe construction of a karst tunnel. Once it occurs, it will cause serious consequences such as economic loss and casualties. Due to the complexity of an underground environment, it is difficult to calculate the probability of karst tunnel water inrush. Therefore, it is of great engineering significance to establish an effective risk assessment model. Based on the Bayesian theory, interpretation structure model, and generative adversarial network, a Bayesian network risk assessment model is established. The results show that firstly, twelve indexes selected can not only characterize the karst tunnel water inrush but also are easy to be counted, which effectively improves the accuracy of the Bayesian risk assessment model. Secondly, the Bayesian network risk assessment model overcomes the shortcomings of other risk assessment models that rely too much on geological data and improves the accuracy through massive data training. Thirdly, the corresponding noninrush samples are generated by the generative adversarial network and analytic hierarchy process, which effectively solve the problem of an unbalanced database. Finally, the Bayesian network risk assessment model is applied to the DK490+373 section of the Shangshan Tunnel. The assessment model is operable, effective, and practical, and it is also suitable for the situation of incomplete index statistics.