Mitigating subway construction collapse risk using Bayesian network modeling

Abstract

This paper develops a Bayesian network-based approach for risk modeling of collapses and derivative accidents to systematically avoid subway construction collapse events and mitigating the loss from derivatives. Given it was not feasible for risk quantification based upon current safety standards, a risk-based method was proposed for numerical computation of severity values from the perspective of workers' safety. In order for dimensionality reduction, each node's importance in the network was estimated through the variation of Gini index in random forest algorithm. The Bayesian network of subway construction collapse risk (SCCR-BN) contained two sub-graphs. Sensitivity analyses were performed to determine critical causal factor events for each collapse or non-collapse event. Countermeasures can be tailored for better response to these causal factors for risk controlling. Three representative scenarios were selected for case studies with the objective of demonstrating the applicability of SCCR-BN for dynamic investigation and prediction of subway construction collapse risks.