A prototype bus arrival prediction system using automatic vehicle location data

Abstract

Provision of expected arrival times of buses is regarded as very useful by bus passengers. The main advantages of such a system can be time savings, reduction in travel anxiety, and improvements in travel decisions. Several studies were performed in the past to predict the travel time of a transit vehicle to some or all stops on the transit route using data collected from various automatic vehicle location systems. This research examines two procedures to predict travel time to a stop for an on-campus route at Texas A&M University using data collected from a DGPS-based AVL system. An analysis of the accuracy of the AVL system indicated that 90 percent of data were within 11 meters from the center of the rightmost lane. Furthermore, it was discovered that 27 percent of data were not differentially corrected and the maximum error in the DGPS data was 7 meters. DGPS data collected over 45 trips were used to calibrate two models, namely, the time-based model and the distance-based model. Data pertaining to 10 trips were used for validation and comparison purposes. Predictions by the time-based model are based solely on the location of the bus on the route and the resulting time to the destination. This model takes into account the variation in average speeds of buses at different locations on the route. The distance-based model uses the distance-to-bus-stop value to estimate the travel time. This model takes the variation in speeds by the time of day into consideration. The predictions of both models are displayed on a kiosk and a variable message sign. The accuracy of predictions was evaluated using three measures, including the actual error in prediction, the absolute error in prediction, and the variation of error with the magnitude of the predicted travel time. The results indicated that the estimates of travel time by both models were reasonably accurate and not much different from one another. Additional research recommended in this area includes incorporating the schedule of buses directly into the models.