Advancement Through Research
Bus dwell time (DT), a time gap between a door opening and closing to serve passengers at bus stop, is one of the key elements for transit service route planning in urban area. Transit agencies have been asked for additional attentions to estimate an adequate DT of bus stop along a planning route because DT would account for a significant portion of service operation cost and effort. Despite the wealth of literatures regarding the bus DT estimation, little attention has been paid to the one for bus rapid transit (BRT). This article proposes a theoretical framework regarding DT estimation for BRT routes using real-time general transit feed specification (GTFS) data. This study investigated real-time GTFS data, collected from five different transit lines across the United States. The data contains either Automatic Vehicle Location (AVL) data or bus arrival/departure time information, queried in every 5 to10-second for five weekdays (Monday to Friday) via each transit agency’s application programming interface (API). The five transit lines are sbX Green Line (San Bernardino, CA), Orange Line (Los Angeles, CA), San Pablo Rapid Line (Oakland, CA), MAX (Kansas City, MO), and A Line (Minneapolis, MN). This study also used Google Map API travel time data to develop three DT estimation models, including two machine learning models, Random forest (RF) and Backpropagation (BP), and a simple regression model. The results show that RF and BP showed a relatively higher accuracy than the regression model in terms of performance measurements.
BRT, dwell time estimation, real-time GTFS, Google Map API, machine learning
