On frequent transit routes, there is a tendency for vehicles to bunch together, causing undue passenger waiting time. To avoid bus bunching, transit agencies strive to maintain even headways by holding vehicles at control points. The increasing availability and accuracy of real-time information enables transit agencies to apply dynamic holding methods that can stabilize their high-frequency routes based on the current operating conditions. This dissertation explores how real-time information can be used to stabilize high-frequency routes from theory to practice.
In the first part of this dissertation, a closed-form method to avoid bus bunching is derived by backward induction as a non-Markovian stochastic decision process. The method can minimize expected headway variance without using buffer time: the rate of vehicle dispatch is equal to the rate of arriving vehicles. The decision process is based on a set of arrival time probability distributions that get updated at each decision epoch. The method is therefore able to minimize the waiting time of passengers arriving at stops according to a Poisson Process. The closed-form method is verified by simulation.
In the second part of this dissertation, the method derived previously is compared to holding methods used in practice and recommended in the literature. The methods are tested on a simulation of Tri-Met Route 72 in Portland, OR, using historical data. A series of sensitivity analyses are conducted to evaluate the impact of parameter choice, control point selection, and prediction accuracy on the performance of each method. The methods based on predictions, and in particular the proposed method, were found to yield the best compromise between holding time and headway stability. We found that prediction errors had a marginal impact on the performance of these methods until they reached a breaking point, beyond which they had a disproportionate effect.
In the third part of this dissertation, the proposed holding method is implemented in three high-frequency transit routes: the Atlanta Streetcar and the Georgia Tech Red Stinger in Atlanta, GA, and the VIA Route 100 in San Antonio, TX. The performance of the proposed method is compared to the schedule currently used. The different institutional frameworks of the agencies that collaborated on this project and their impact on implementation are compared. In addition, the level of adoption by supervisors, dispatchers and operators and their effect on compliance are analyzed. Finally, the practical lessons learned from implementing a real-time dispatching method on high-frequency routes are discussed.
Dr. Kari Edison Watkins
Dr. Jorge Laval, Dr. Michael Rodgers, Dr. Alan Erera (ISyE), Dr. Michael Meyer (Parsons Brinckerhoff)