Estimation of Short-Term Bus Travel Time by Using Low-Resolution Automated Vehicle Location Data

Abstract

Estimation of short-term bus travel time is an essential component of effective intelligent transportation systems (ITS), including traveler information systems and transit signal priority (TSP) strategies. Several technologies, such as automated vehicle location (AVL) systems, can provide real-time information for estimation of bus travel time. However, low resolution of data from such technologies presents a challenge to accurate estimation of travel time. Several models for estimation of bus travel time at signalized urban arterials were developed and tested. These models used low-frequency AVL data and required only knowledge of network specifications such as locations of bus stops and intersections. First, a linear regression model was developed; it decomposed total travel time into its components, including running travel time, dwell time at bus stops, and delay at signalized intersections. Second, various machine learning models, including support vector regression (SVR) with nonlinear kernel, ridge, Lasso, decision tree, and Bayesian ridge were trained by using Python libraries such as scikit-learn and evaluated. A segment of Washington Street in Boston, Massachusetts, was selected as the study site. The results indicate that the SVR model outperformed other regression models in generalized error measures, in particular those of mean absolute error and root mean square error. The findings of this study can lead to improved traveler information systems and more-efficient TSP strategies and, overall, can contribute to better transit quality of service.