New Formulations and Solution Methods for the Dial-a-ride Problem [thesis]

Sharon Dong
The classic Dial-A-Ride Problem (DARP) aims at designing the minimum-cost routing solution that accommodates a set of user requests under constraints at the operations planning level. It is a highly constrained combinatorial optimization problem initially designed for providing door-to-door transportation for people with limited mobility (e.g. the elderly or disabled). It consists of routing and scheduling a fleet of capacitated vehicles to service a set of requests with specified pickup and
more » ... p-off locations and time windows. With the details of requests obtained either beforehand (static DARP) or en-route (dynamic DARP), dial-a-ride operators strive to deliver efficient and yet high-quality transport services that satisfy each passenger's individual travel needs. The goal of this thesis is threefold: (1) to propose rich DARP formulations where users' preferences are taken into account, in order to improve service quality of Demand-Responsive Transport (DRT) services and promote ridership strategically; (2) to develop novel and efficient solution methods where local search, column generation, metaheuristics and machine learning techniques are integrated to solve large-scale DARPs; and (3) to conduct real-life DARP case studies (using data extracted from NYC Yellow Taxi trip records) to test the practicality of proposed models and solution methods, as well as to emphasise the importance of connecting algorithms with real-world datasets. These aims are achieved and presented in the three core chapters of this thesis. In the first core chapter (Chapter 3), two Mixed Integer Programming (MIP) formulations (link-based and path-based) of DARP are presented, alongside with their objective functions and standard solution methods. This chapter builds the foundation of the thesis by elaborating the base models and algorithms that this thesis is based on, and by running benchmark experiments and reporting numerical results as the base line of the whole thesis. In the second core chapter (Chapter 4), two DARP models (one d [...]
doi:10.26190/unsworks/24100 fatcat:rku3v43l6rhpxaig77bhxfmo34