Text

theses

Incidents are random, yet disruptive events that occur frequently on our roadways. Incidents also have a major impact on the other motorists, the environment and the economy. For example, as a result of incidents, the motorist can experience delays and delays may cause higher emission rates. Hence, without proper incident management strategies lives can be at risk and the motorist might experience long delays. Therefore, there is a need for better incident management strategies for improving the safety of the motorists, reduce congestion thus improving the productivity of our economy. In this thesis an integrated approach was used to mitigate the effects of incidents. The proposed had two components: (i) a real-time incident duration model designed to improve decision making during incident management operations, (ii) a novel statistical model for mapping incidents and their severity in space to detect the segments with higher accident risk with the goal of better planning for and responding to future incidents. The novelty of this approach is that it offers solutions for mitigating both short term and long term effects of incidents using a Bayesian framework. This probabilistic approach enables the representation of the probabilistic range of results rather than point estimates. Bayesian networks are used for predicting incident duration. The performance of the model was subsequently improved by introducing adaptive learning techniques to the model. A set of spatial models was estimated and risk maps were developed to investigate the locations with higher crash risk in order to facilitate incident management planning. For purposes of estimating the predictive power of the models, spatial data with different resolutions were used. The overall findings of this dissertation indicate that the integrated approach can be used for estimating incident duration in the long run as it adapts itself to minor and major changes to the system. It can also be used for effectively pinpointing the locations with higher crash risk since, unlike existing models, the model do not ignore the continuous nature of roadways.

ETD_5886

Ph.D.

Includes bibliographical references

by Sami Demiroluk

doi:10.7282/T3057DDT

ETD doctoral

The author owns the copyright to this work.