Developing Sampling Strategies and Predicting Freeway Travel Time Using Bluetooth Data

Abstract

Accurate, reliable, and timely travel time is critical to monitor transportation system performance and assist motorists with trip-making decisions. Travel time is estimated using the data from various sources like cellular technology, automatic vehicle identification (AVI) systems. Irrespective of sources, data have characteristics in terms of accuracy and reliability shaped by the sampling rate along with other factors. As a probe based AVI technology, Bluetooth data is not immune to the sampling issue that directly affects the accuracy and reliability of the information it provides. The sampling rate can be affected by the stochastic nature of traffic state varying by time of day. A single outlier may sharply affect the travel time. This study brings attention to several crucial issues - intervals with no sample, minimum sample size and stochastic property of travel time, that play pivotal role on the accuracy and reliability of information along with its time coverage. It also demonstrates noble approaches and thus, represents a guideline for researchers and practitioner to select an appropriate interval for sample accumulation flexibly by set up the threshold guided by the nature of individual researches’ problems and preferences. After selection of an appropriate interval for sample accumulation, the next step is to estimate travel time. Travel time can be estimated either based on arrival time or based on departure time of corresponding vehicle. Considering the estimation procedure, these two are defined as arrival time based travel time (ATT) and departure time based travel time (DTT) respectively. A simple data processing algorithm, which processed more than a hundred million records reliably and efficiently, was introduced to ensure accurate estimation of travel time. Since outlier filtering plays a pivotal role in estimation accuracy, a simplified technique has proposed to filter outliers after examining several well-established outlier-filtering algorithms. In general, time of arrival is utilized to estimate overall travel time; however, travel time based on departure time (DTT) is more accurate and thus, DTT should be treated as true travel time. Accurate prediction is an integral component of calculating DTT, as real-time DTT is not available. The performances of Kalman filter (KF) were compared to corresponding modeling techniques; both link and corridor based, and concluded that the KF method offers superior prediction accuracy in link-based model. This research also examined the effect of different noise assumptions and found that the steady noise computed from full-dataset leads to the most accurate prediction. Travel time prediction had a 4.53% mean absolute percentage of error due to the effective application of KF.