Link travel time estimation based on connected vehicle probe data
Metadata[+] Show full item record
[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] Connected vehicle technology is a wireless system to support communication for enhancing safety and mobility. The thesis proposes a single-link connected vehicle probe data (CVPD) model to estimate travel time on urban arterials. This research applies VISSIM to simulate probe vehicles that can generate and assemble snapshots via installed On-Board Unit (OBU). Each snapshot, which can include link identification, instantaneous speed, location and time, is generated by a generation protocol. Generation protocols can be changed by system operators to allow data collection to be tailored to specific situations. In theory, generated snapshots are added to the buffer in OBU and they will not be removed until the snapshots are transmitted to roadside equipment (RSE). However, actually, the space of buffer is limited. If the buffer became full and the probe vehicle does not travel into the range of a RSE, a recorded snapshot must be deleted to make room for a new snapshot. Under snapshots deletion condition, CVPD model may miss some useful data and hence estimate link travel time inaccurately. In order to save buffer's space, the number of recorded snapshots should be as small as possible. A new snapshot generation protocol, the protocol, is proposed in this thesis and its performance is compared with two existing protocols--the fixed observation interval protocol and the society of automotive engineers (SAE) protocol. The core of protocol is to only capture the vehicle snapshots when dramatic changes in acceleration take place. The dramatic changes are defined by predefined acceptable ranges based on different buffer space sizes. In this research, three kinds of buffer spaces are defined, which can provide enough space to save snapshots assembled by fixed observation protocols, including fixed-two-second (F2), fixed-four-second (F4) protocol, and SAE protocol, respectively. Two travel time estimation methods were used to estimate travel times from the generated snapshots--the time-speed plot method and the average speed method. The time-speed plot method relies on interpolation functions to fit time-dependent speed observations obtained from snapshots. Three piecewise different interpolation functions were investigated in this thesis--Linear interpolation, Cubic Spline interpolation and Cubic Hermite Spline interpolation. In this research, ten variations of protocol are proposed. Three of them are selected to compare the snapshots generated by F2, F4 and SAE protocol, respectively. The main objective of protocol selection is the number of collecting snapshots is equal to or less than F2, F4 and SAE protocol, respectively. Comparing F2, F4 and SAE, the number of snapshots recorded by the protocol reduces 30.11%, 26.02% and 4.39%, respectively. Two-tailed p-value and t-statistic are applied to analyze mean absolute relative error (MARE) of entire traffic network link travel time estimated by time-speed plot method and average speed method. Based on snapshots collected via fixed observation interval protocols, time-speed plot is always better than average speed method. However, according to snapshots recorded via SAE and protocol, only time-speed plot estimated by Cubic Hermite Spline interpolation is significantly better than average speed method. Therefore, Cubic Hermite Spline interpolation, which can be applied in a microscopic model, can construct smooth curves and effectively prevent overshoot or oscillation. In other words, Cubic Hermite Spline interpolation can avoid overestimating speed values and estimate time-speed plot more accurately than other two interpolation methods.
Access is limited to the University of Missouri - Columbia.