Real-Time Estimation of Urban Street Segment Travel Time Using Buses as Speed Probes

AVL encompasses a variety of location and communication technologies. Archived AVL (or on-vehicle data recording) uses an onboard computer to record and store bus operation data and upload them to a data tank at a specific time (usually daily); real-time AVL (or offvehicle data recording) sends bus travel and operation data to a central computer in real time (usually at a frequency of 30 s to 5 min) (15, 16). The data generated by the two types of AVL systems are very different in content and have different applications in bus probe studies. For more information about the difference, see Pu and Lin (13). This paper focuses on the usability of bus travel information to infer general vehicle traffic conditions. The usability can be proven if two conditions are met: Condition 1, there are quantifiable relationships between bus travel and car travel; Condition 2, infrequent bus travel observations (constrained by the scheduled bus headway and AVL polling frequency) are sufficiently sensitive to infer real-time general vehicle traffic conditions (probably by means of the relationships identified in Condition 1). A large amount of historic data can be used to identify possible bus–car relationships, as in past bus probe studies (1–3, 6). Historic relationships, however, do not guarantee the real-time sensitivity of bus probes to traffic conditions, as bus observations could be too sparse to draw a reliable conclusion in a short time. Thus, Condition 2 needs to be satisfied. Previously, Pu and Lin identified statistically significant relationships between bus and car speeds with historic real-time AVL bus data and test car data on a signalized urban street in Chicago (17, 18). This paper is designed to examine whether real-time AVL bus data can be used for real-time estimation of car speeds and travel times. The real-time AVL system considered in both studies polls buses about every 30 s and obtains bus speed, location, time stamp, and other identity information. The study is structured as follows. A generic framework of realtime estimation is proposed, followed by a field study in which light and congested traffic conditions are observed and then a simulation study in which unexpected passenger demand surge is created. Conclusions are presented at the end of the paper.