Travel time estimation is important for a wide range of applications, including advanced traveler information systems (ATIS), sustainability analysis, and discrete choice modeling. Approaches to travel time estimation traditionally have been based on aggregate data sets that examine travel times over a number of days or travel times in previous time intervals. Automatic vehicle identification data make it possible to analyze travel time data at a totally disaggregate or individual commuter level. It is postulated in this research that the capability of modeling travel characteristics on a disaggregate level can improve the accuracy with which performance measures are quantified. The test beds examined are a 22-km section of the I-10 corridor and a 21-km section of the US-290 corridor in Houston, Texas. It was found that aggregation across days, which does not consider the effect of individual days, is 63 percent less accurate than aggregation by days, which does consider the effect of individual days. Even though the latter technique was found to be more accurate, it was illustrated that 40 percent of the regular commuters’ travel times are statistically different from these aggregate estimates. Similarly, for travel time variability, it was found that for approximately 20 percent of the cases the travel time standard deviations for regular commuters are statistically different from the aggregate estimates. These results illustrate the uniqueness of an individual commuter’s travel patterns and emphasize the benefit of conducting analyses at the level of the individual commuter for both ATIS and sustainable transportation.
Towards Attention-Based Convolutional Long Short-Term Memory for Travel Time Prediction of Bus Journeys
