## Dissertations

Dissertation

Summer 8-31-2013

#### Degree Name

Doctor of Philosophy in Transportation - (Ph.D.)

#### Department

Civil and Environmental Engineering

I-Jy Steven Chien

Janice Rhoda Daniel

Lazar Spasovic

Jian Yang

#### Abstract

Congestion on freeways is increasing and a key source of it is non-recurring incidents. Accurate vehicle travel time predictions are needed during these incidents in order for roadway users to make informed trip decisions. Path based sensors are becoming a leading technology in gathering real-time travel time data. The data is used to make travel time predictions that are then provided through various means, such as dynamic message signs, to roadway users. These types of sensor are located at stationary points along a roadway and collect individual vehicle travel time data from vehicles as they drive pass the sensors.

The accuracy of the predictions, in terms of representing future travel times, is dependent on many factors including the sensor spacing along the roadway, the duration and location of a traffic incident, and the uncongested and congested traffic speeds and traffic flows. Understanding the relationship between the travel time prediction accuracy and the different variables is necessary to optimize sensor spacing. In addition, because incidents occur at different times of the day, have varying durations, occur at different locations, and cause different capacity reductions depending on the severity of the incident, the sensor spacing cannot be based on one incident scenario. Instead, multiple incident scenarios, along with the probability of each occurring, needs to be taken into account.

Path based sensor spacing during incidents on a freeway segment is optimized in this dissertation. In addition, the marginal benefit of additional sensors is calculated. A mathematical model and a solution methodology are developed. The mathematical model applies macroscopic traffic principles and shock wave theory. It calculates the travel time prediction error by sensor spacing during an incident on a freeway segment.

The solution algorithm consists of four main steps. First, historical incident data for the roadway are gathered. Second, the mathematical model is applied to determine the average travel time prediction error by sensor spacing for each of the historical incidents. Third, the weighted average travel time prediction error by sensor spacing is calculated, which considers all the possible incidents and the frequency of each occurring. Fourth, the optimal spacing is chosen which minimizes the weighted average error.

The applicability of the model and solution methodology is demonstrated through a case study of a ten mile freeway segment in Northern New Jersey.

COinS