| Popis: |
Connected and autonomous vehicle (CAV) technologies can revolutionize the way we transport people and goods and may soon be publicly available, however proper planning for these technologies is crucial to their successful integration into our transportation systems. CAVs can reduce following headways and increase roadway capacity and stability, as well as allow for new, more efficient intersection controls with wireless communication capabilities. This work is twofold: (1) evaluating the traffic congestion impacts of AVs and reservation-based intersection control on real large-scale city networks in Texas using DTA and (2) developing methods to find optimal configurations of reservations and signals in a city network. The first part of this thesis evaluates CAV behavior impacts by simulating different mixed CAV and human vehicle (HV) demand scenarios. Results show improvements in network efficiency with increases in CAV penetration. Reservations were observed to perform better than signals in most scenarios. Namely, the Austin downtown network resulted in a 78% reduction in travel time. However, signals outperformed reservations in some high demand cases on arterial networks due to the reservation's first-come-first-serve (FCFS) policy allocating more capacity to local roads, resulting in arterial progression interruption and queue spillback onto close-proximity streets. The discovered paradoxical effects imply that some intersections are better suited for reservation control than others. The second part of this thesis finds and characterizes favorable mixed-configurations of reservation-based controls and signalized controls in a large city network which minimize total system travel times. As this optimization problem is bi-level and challenging, we propose three different methods to heuristically find effective mixed-configurations. The first method is an intersection ranking method uses simulation to assign a score to each intersection in a network based on localized potential benefit to system travel time under reservation control and then ranks all intersections accordingly. The second is another ranking method, however uses linear regression to predict an intersection's localized score. Finally, we present a genetic algorithm which iteratively approaches high-performing network configurations yielding minimal system travel times. We test the methods on the downtown Austin network and find configurations which are less than half controlled by reservation intersections that improve travel times beyond an all-reservation controlled network. Overall, our results show that the genetic algorithm finds the best performing configurations with the initial score-assigning ranking method performing similarly but much more efficiently. We finally find that favorable reservation placement is in consecutive chains along highly trafficked corridors. |
