Bus-Arrival-Time Prediction Models: Link-Based and Section-Based

Abstract

Bus-arrival-time information service is a key component of an advanced transit information system. Instantaneous and accurate prediction of bus arrival time can help improve the quality of service, and attracts additional ridership. On the bases of analyses of bus running processes, bus arrival time was divided into interzone link travel time and section travel time. A self-adaptive exponential smoothing-based algorithm was proposed for interzone link travel time prediction, whereas link-based and section-based algorithms were proposed for section travel time prediction. With the automatic vehicle location data collected from an actual bus route, an experiment was conducted to measure the performance (accuracy and precision in prediction) of link-based and section-based models with respect to three dominant factors: day of week, time of day, and length of segment. The research results show that: (1) the overall performance of section-based models is superior to that of link-based models on all weekdays but Monday, and few differences exist on weekends between link-based and section-based models; (2) section-based models perform significantly better than link-based models during periods when traffic conditions are relatively better; (3) as the vehicle approaches the target stop, the accuracy of prediction decreases and the precision of prediction increases for both link-based and section-based models. In addition, the performance of section-based models is superior to that of link-based models, but the gap narrows gradually with decreases length of segment.