Prediction Model of Train Fault Probability on Urban Rail Transit Main Line

Abstract:

A qualitative analysis was made to investigate the major influencing factors in predicting the probability of the train fault happening on urban rail main line. Then， a discrete dataset was collected about a single train’s fault in running for 120 000 km . Three alternative models were established on the basis of the data characteristics， Poisson distribution and zero-inflated Poisson distribution as well as the potential fault forms. According to the comparative study results，a Poisson distribution-based prediction model of train fault probability is finally proposed. Study results show that the train fault probability tends to increase with the increasing of train formation. It decreases first and then increases with the cumulative running kilometers，and the minimum train fault probability occurs in the fourth 120 000 km period， but the initial value is exceeded in the seventh 120 000 km period.