Abstract:

In order to enhance the level of railway safety management, Bayesian network and ISM were combined to establish a warning model for railway accidents. Firstly, the item point system was established, and the accident matrix was determined. Secondly, the causal effect algorithm was combined with ISM to calculate the node influence ranking. The ranking and the accident matrix were substituted into K2 algorithm to calculate the Bayesian network structure. Meanwhile, the expectation-maximization algorithm was used to calculate the parameters of Bayesian network. The relationship between the item point frequency and the accident probability was calculated using the backward inference of Bayesian network. The accident warning threshold could be obtained by using historical data. Finally, Chengdu bureau shunting accidents were taken for case analysis. The results show as follows: main factors affecting the derailment accidents are not implementing the system of request to open turnout and horn feedback in the non-centralized area, not strictly implementing the ″four must″ system for shunting service, illegal interventions of hump slip operation, and improper following or standing position in the shunting service. The warning threshold of the derailment accident is 0.036 1.

Key words: railway transportation accidents, Bayesian network, interpretative structural modeling(ISM), derailment accident, backward inference