Bayesian Network Modeling for Analysis and Prediction of Accidents in Railway Transportation of Dangerous Goods

The article is devoted to forecasting and preventing the development of accidents and expert assessment of accidents that have occurred on the example of railway transportation of dangerous goods. Bayesian belief networks are used, which make it possible to assess the uncertainty of the initial data, the causal relationships of events. Causal relationships are modeled using conditional probabilities that evaluate the degree of confidence in the truth of new incoming (causing) information based on previously received information. The Bayesian net method allows carrying out procedures that are not available for traditional quantitative risk assessment. Networks can be corrected by supplementing the constructed model with data on the failure rates of a real object. Computer modeling of the accident using a probabilistic graphical model was performed in the “GeNIe” software package with an analysis of its main factors. Accident modeling is accompanied by color thematic visualization of dependencies between random factors with the construction of a directed path in an acyclic graph, the vertices of which are the factors, and the edges determine the dependencies between them. On the example of a real catastrophe at a railway station, the effectiveness of the constructed model is confirmed. The analysis of the results obtained in the modes of sensitivity assessment and diagnostics was carried out, which made it possible to determine the main factors of the accident.