Estimating Uncertainty of Bus Arrival Times and Passenger Occupancies
mode - bus, technology - automatic vehicle monitoring, technology - passenger information, technology - ticketing systems, technology - intelligent transport systems, ridership - attitudes, ridership - modelling, operations - reliability, planning - service quality, planning - signage/information, ridership - perceptions
Bus travel time models, Bus passenger occupancy models, Survival models, Linear regression, Count regression
Travel time reliability and the availability of seating and boarding space are important indicators of bus service quality and strongly influence users’ satisfaction and attitudes towards bus transit systems. With Automated Vehicle Location (AVL) and Automated Passenger Counter (APC) units becoming common on buses, some agencies have begun to provide real-time bus location and passenger occupancy information as a means to improve perceived transit reliability. Travel time prediction models have also been established based on AVL and APC data. However, existing travel time prediction models fail to provide an indication of the uncertainty associated with these estimates. This can cause a false sense of precision, which can lead to experiences associated with unreliable service. Furthermore, no existing models are available to predict individual bus occupancies at downstream stops to help travelers understand if there will be space available to board. The purpose of this project was to develop modeling frameworks to predict travel times (and associated uncertainties) as well as individual bus passenger occupancies. For travel times, accelerated failure-time survival models were used to predict the entire distribution of travel times expected. The survival models were found to be just as accurate as models developed using traditional linear regression techniques. However, the survival models were found to have smaller variances associated with predictions. For passenger occupancies, linear and count regression models were compared. The linear regression models were found to outperform count regression models, perhaps due to the additive nature of the passenger boarding process. Various modeling frameworks were tested and the best frameworks were identified for predictions at near stops (within five stops downstream) and far stops (further than eight stops). Overall, these results can be integrated into existing real-time transit information systems to improve the quality of information provided to passengers.
Permission to publish the abstract has been given by Mineta Transportation Institute, copyright remains with them.
Gayah, V.V., Yu, Z., & Wood, J.S. (2016). Estimating Uncertainty of Bus Arrival Times and Passenger Occupancies. Mineta National Transit Research Consortium, MNTRC Final Report 12-56, pp. 69.