Practical Considerations in Implementing Reliability of Travel Time in Forecasting of Regionwide Travel

Abstract

The reliability of travel time has been shown to influence travelers’ decisions as to choice of destination and choice of route. This paper explores options for the inclusion of travel time reliability into a regional travel forecasting model and tests those options on a full-size network in the Wheeling, West Virginia, metropolitan area. Options include two path-building algorithms, parameters (locally derived versus borrowed) for the coefficient of variation of link travel time, and the size of the reliability ratio, which is the weight for the standard deviation of travel time when included in the path impedance function. The two path-building algorithms are the naïve algorithm, which does not find shortest paths but weights reliability the same on each link, and the shortest marginal algorithm, which does find shortest paths but emphasizes variations in delays on links earliest in a path. The tests are not considered definitive because of the relatively low congestion levels in Wheeling, but there were indications that the naïve algorithm and borrowed parameters performed best. The experience showed that including travel time reliability in regionwide travel forecasts is straightforward, there are ample resources to establish reasonable parameters, no compromises need to be made for implementation, and the computational burden is modest. However, it is important that any model have good travel time and delay estimates to begin with before reliability is layered on as another consideration for path selection.