| Real-time estimates of traffic conditions are valuable information needed by operators of transportation facilities as well as travelers. This study aims to provide accurate travel time estimates using data collected by the electronic toll collection system instead of sensors and AVI readers specifically deployed for traffic monitoring. This dual use of toll readers for travel time estimation can be an attractive approach since it eliminates additional costs of deploying and maintaining sensors. However, this approach can present an important challenge in terms of accuracy of the estimates because readers are not located on the main roadway, but instead on the ramps, and the demand level associated with particular OD pairs is not always enough to obtain accurate average travel times. Therefore, two estimation methods based on universal kriging and mathematical programming are proposed to estimate single section travel times using vast amount of available data from the electronic toll collection system of NJ Turnpike.;To be valuable, travel time information must be updated continuously in real-time to provide not only estimates of current traffic conditions but also future projections. Time series models are commonly used in transportation area to obtain future traffic states. This thesis compares the prediction performance of a parametric model, ARIMA, a recently developed non-parametric model, SVR, a commonly used non-parametric model, ANN, and tests their performances under both typical and atypical traffic conditions. |
