With the popularity of Big Data and urban informatics, there is increasing interest in ways to use real time data to improve transportation system operations. In many real-wold applications, demand is revealed dynamically over time, and consequently the routes are determined dynamically as well. In this thesis, contributions are made to several key components of a “smart” transit system framework where dynamic operations are driven by real time information.
The first component is in dynamic routing and pricing of a fleet of vehicles. A new dynamic dial-a-ride policy is introduced that features non-myopic pricing based on optimal tolling of queues to fit with the multi-server queueing approximation method. By including social optimal pricing, the social welfare of the resulting system outperforms a pricing policy based on the marginal cost increase of a passenger over a range of test instances. In the examples tested, improvements in social welfare of the non-myopic pricing over the myopic pricing were in the 20% - 31% range.
The second component is in the informatics. Effective dynamic optimization of a system (routing, scheduling, fare setting, etc.) requires effective short term prediction of traveler/customer arrival using real-time data. Several recent methods for arrival process prediction, both offline and online, are investigated using real taxi data from New York. An experiment is conducted using the same data set to draw comparisons for arrival process modeling, suggesting that the temporal seasonal factors method from Ihlers et al. (2006) is more effective as an offline approach and the IntGARCH method from Matteson et al. (2011) is more effective as an online approach.
The third component investigated is in the prepositioning of idle vehicles. Vehicles that are positioned at locations that take into account future demand can lead to reduced wait times for passengers and improved level of service. A dynamic relocation model is proposed that includes queueing delay to approximate the congestion effect of future demand. A linear problem is formulated based on Marianov and Serra’s (2002) work. By varying customer arrivals, the approach provides a new managerial tool to find the optimal service level.