Real-time safety and mobility optimization of traffic signals in a connected-vehicle environment

Adaptive traffic signal control (ATSC) strategies are a promising approach to improving the efficiency of signalized intersections, especially in the era of connected vehicles (CVs) where real-time information on vehicle positions and trajectories is available. Recently, numerous ATSC algorithms have been proposed to accommodate real-time traffic conditions and optimize traffic efficiency. The common objective of these algorithms is to minimize total delays or maximize vehicle throughputs. Despite their positive impacts on traffic mobility, existing ATSC algorithms do not consider optimizing traffic safety. This is most likely due to the lack of tools to evaluate the safety of signalized intersections in real time. This thesis presents several advances toward the real-time safety and mobility optimization of traffic signals in a connected-vehicle environment. First, new models for the real-time safety evaluation of signalized intersections were developed and validated, using traffic video-data of six locations in two Canadian cities. The developed models relate the number of rear-end traffic conflicts, as a surrogate safety measure, to dynamic traffic parameters at the signal cycle level. Several traffic conflict indicators and multiple conflict severity levels were considered. The transferability of the developed models was also investigated and confirmed using additional traffic datasets for two corridors in the United States. Second, a new procedure to integrate the developed real-time safety models with traffic microsimulation models was proposed. The procedure was validated using real-world traffic video data of two signalized intersections in British Columbia. The results showed that the proposed models can predict traffic conflicts from traffic simulation with reasonable accuracy and subsequently can be used to investigate the safety impact of various CVs-based applications before field implementation. Third, a novel self-learning ATSC algorithm to optimize traffic safety using real-time CVs data was proposed. The algorithm was developed using the Reinforcement Learning approach, trained using a microsimulation model, and validated using real-world traffic data of two signalized intersections in British Columbia. Superior to the traditional actuated signal control system, the proposed algorithm showed positive safety and mobility impacts. The proposed ATSC algorithm was also found to be effective and feasible even under low market penetration rates of CVs.Applied Science, Faculty ofCivil Engineering, Department ofGraduat