With increasing concerns regarding environment protection, road congestion and traffic safety, transferring commodities from trucks to rails has been recognized as a potential solution. In contrast with the rapid development of Intelligent Transportation System (ITS) for roadway freight transportation, railway freight transportation lacks efficient technologies and tools to evaluate and predict its performance. In order to improve in this area, this paper proposes an analytical model to utilize sparse spatial-temporal data that was collected from five long-haul railway paths from the west coast of U.S. (Los Angeles or Seattle) to Columbus, Ohio, from March to December 2014. The analytical model consists of three data processing steps: complete path generation, complete path transformation, and distance and travel time disaggregation. It can automatically account for various time-consuming operations at railway stations in addition to segment travel times during long-haul freight shipments. Based on the results of the analytical model, different visualizations such as an average speed indicator and time-distance diagrams are introduced to evaluate performance under different conditions. They are intuitive and straightforward to use to evaluate performance bottlenecks. Example diagrams are shown for the Chicago area where freight trains can experience long delays. The results show that commuting passenger trains and inclement weather are two potential reasons behind the delays. A section-based model and path-based model are proposed for performance prediction to estimate the mean and variance of the arrival time to its destination for rail freight shipments. The experiments show the section-based model has an overall lower mean absolute percentage error (MAPE) and a higher reliability than the naïve path-based model.

CICTP