Quantitative analysis of risk propagation in urban rail transit: A novel ensemble learning method based on the structure of Bayesian Network

The quantification of risk propagation in urban rail transit systems is a critical task to ensure the safe operations. In this study, a novel ensemble learning method based on Bayesian network structure learning is developed to describe the risk propagation mechanisms. The proposed model addresses the reliance problem on ordering and is capable to quantify more complex risk propagation paths. First, an information-based criterion and a score function are proposed to construct the initial propagation structure. Second, a structure constructing algorithm is introduced to generate multiple Bayesian networks, forming a Bayesian Forest. Finally, three applications of the Bayesian Forest are introduced: scenario inference, sensitivity analysis and risk propagation chain evaluation. Additionally, a case study is made on the application of the proposed model to the Shanghai metro system to verify its effectiveness. The results validate the rationality of the ensemble learning method by analyzing multiple risk propagation paths. The interaction characteristics are explicitly described by sensitivity of risk factors and the significance of the risk propagation chain is accurately evaluated.