Estimation of run times in a freight rail transportation network

Abstract

The objective of this thesis is to improve the accuracy of individual freight train run time predictions defined as the time between departure from an origin node to arrival at a destination node not including yard time. A correlation analysis is conducted to identify explanatory variables that capture predictable sources of delay and influence run times for use in a regression model. A regression model is proposed utilizing the following explanatory variables: rolling historical average, congestion window, meets, passes, overtakes, direction, arrival headway, and departure headway to predict train run times. The performance of the proposed regression model is compared against a baseline simple historical averaging technique for a two year period of actual train operational data. The proposed regression model, though subject to specific limitations, offers substantial improvements in accuracy over the baseline technique and is recommended as justifying further exploration by the railroad to ultimately enable more accurate train schedules with subsequent improvements in railroad capacity, customer service, and asset utilization.