Research On Railway Accident Causation Modeling Analysis

With the development of economy and constant progress of technology, the railway industry, especially high-speed rail, has entered into a booming period. In the year of 2014, the production of new rail lines of China got the highest figure of record, with the length of railroad lines in service over 112 thousand kilometers. Moreover, the length of high-speed railroad lines in service was more than 16 thousand kilometers, which was almost the sum of the rest of the world. Obviously, the high-speed age has promoted the development of social economy. However, it also brings more security problems. In this context, it becomes necessary to took for some general rules from existing accident data, in order to provide the basis for railway safety supervision and management from a preventive point. Based on the theory of complex network, this paper built a railway accident causation model, analyzing the interreactions of railway accident factors. The work that has been done is listed as follows.(1) Built a railway accident causation model based on existing accident data. This paper built a railway accident causation network based on 2009-2013 FRA accident data.7 effective communities have been found with FN algorithm in community detection theory. The comparison on some basic topological parameters before and after the partition proved that it is more effective to analyze the interreactions of railway accident factors after the partition, laying the foundation for research on causation chains.(2) Built a railway accident causation model based on cascading failures. On the basis of railway accident causation model, in order to simplify and dig the relationship, this paper analyzed the relations between causation factors and built the railway accident causation associative network. With the ML model from cascading failures, the propagation process of danger was analyzed and causation chains from each community were found. After the statistical analyse of nodes in causation chains, it was proved that nodes with more degree had similar tend curves about frequency and amount.(3) Analyzed the emergence of the network after the change of weights. After the analysis of causation chains, a dynamic weighting model was built in order to investigate the emergence of the network made by the change of weights based on degree preference strategy. The result was shown with the length of causation chain, network efficiency and shortest path distribution. The result showed that it can stop the propagation of causation chains and improve the security level of the entire network to enhance the supervision and management of nodes with higher degree.