Developing a deep learning method in safety prediction model in subway system (Case study: Tabriz subway)

Predicting the safety of the rail transport system is a fundamental problem in modeling and managing rail transport. In this paper, a profound learning-based safety prediction model for rail transport safety is proposed. Data were collected from the subway operating company and decisions were made on the factors of the forecasting model. These factors were used as DBN input. The structure of the input data determines the number of nodes in each layer. Sample data collected include types of accident-prone events, basic train information, and company operating information. Predictive factors were selected by analyzing these collected data. The DBN was then created based on the processed data. To show the accuracy of this model, a data set (Tabriz subway) has been investigated as a case study. Experiments on the data set show good predictive performance of the present model. These results show that deep learning is useful in learning the patterns of a rail system. Prediction and reality error between 0.08 and -0.08 are introduced. This is an acceptable error and cannot cause an incorrect level of security in the system. The dense safety range, both in the current situation and in the forecast model, is at the second level, i.e. relatively safety. This range is in the range of 0.8 to 0.89. Of course, the tendency of density after the second level is towards the first level to the third level. It is noteworthy that the forecast model (at high levels) shows higher values ​​than the existing safety.