PhD Thesis Proposal: Pushpendra Singh

Optional ZOOM LINK
Meeting ID: 979 0416 5400
Passcode: 341663

"Transportation Network Design Under Demand Endogeneity and Uncertainty"

Abstract

The design of transportation networks such as roads, railways, transit, micromobility, and flight networks is challenging due to the many variables, uncertainties, and intricate demand-supply dynamics. This thesis develops new modeling techniques and solution algorithms to address these challenges. Specifically, it optimizes node selection, arc selection, and path selection for a variety of transportation systems.

Optimization of shared micromobility systems, such as bike sharing and e-scooter sharing, is beneficial not only in the first- and last-mile transportation context, but also to replace shorter car trips, thus improving the efficiency of transit systems while being environmentally friendly. The problem is challenging due to the interdependencies between the strategic and operational decisions, as well as those between the passenger demand and transportation supply. To overcome this challenge, we designed an integrated mathematical and computational framework that captures these interactions. We formulate a non-convex mixed-integer non-linear optimization model, and a fast iterative heuristic, offering near-optimal solutions in significantly shorter runtimes compared to state-of-the-art methods.

A major challenge in the design of transportation networks involves arc selection decisions. Embedded passenger choice models make this problem highly complex and difficult to solve, and is usually tackled with models and algorithms that are application-specific. We provide a general framework for this class of problems based on a general mixed-integer non-convex optimization model. We develop a solution approach based on the reweighted L-1 norm minimization method, shown to substantially outperform all algorithmic benchmarks.

Finally, we address the Tail Assignment Problem in airline operations, which involves assigning aircraft to scheduled flights under both hard and soft constraints. This in turn is modeled as a path selection. Two modeling approaches are provided, an arc-based one and a string-based one. We propose an accelerated column generation approach to solve the string-based formulation and plan to enhance the arc-based approach with a combination of optimization and machine learning techniques.

In summary, this thesis addresses the node selection (e.g., station locations), arc selection (e.g., roads, rail line), and path selection (e.g., sequence of flights) problems, combining exact and heuristic techniques using concepts drawn from operations research, signal processing, and machine learning.

Thesis Committee

• Prof. Vikrant Vaze (Chair
• Prof. Geoffrey Parker
• Prof. Wesley Marrero

Contact

For more information, contact Thayer Registrar at thayer.registrar@dartmouth.edu.