Evaluating Regional Rail Transit Safety: A Data-Driven Approach

To grasp the trend of safety conditions in the operation process of multisystem rail transit, and allocate transport capacity and maintenance resources reasonably, the managers should master accurate and comprehensive safety evaluation of regional rail transit system. A data-driven model for safety evaluation of regional rail transit system was proposed in this study. The deep autoencoder networks were employed to reduce the dimensions of the evaluation index system. The hybrid hierarchical k-means clustering method was applied to obtain the set of all possible safety status. The tree-augmented naïve Bayes algorithm was used to evaluate the overall safety. The validity and practicality of the model were verified using actual operations data from a rail transit network in regional urban agglomeration in China. A comparison with the actual situation shows that the proposed approach can evaluate the safety level of the network effectively.