| Vehicle sharing systems have received continually growing interest in recent years, due in part to increasing energy and environmental concerns. Motivation for implementing these systems includes reductions in traffic congestion, noise and air pollution, and overall energy consumption. These systems are often extremely large-scale - for example, Zipcar maintains a fleet of over 7,000 vehicles in 3 countries, and Velib, a French bicycle sharing program, maintains over 20,000 bicycles across approximately 1,500 locations around Paris. In these systems, customers arrive to one of the rental stations in the network, rent a vehicle for some amount of time, and then return the vehicle to a station of their choosing. The underlying system behavior is both highly dynamic and subject to various types of uncertainty (e.g., customer demand, vehicle rental durations, location of vehicle inventory, etc.). Due to such complexities, effective management of these systems has proven very challenging.;Vehicle sharing systems face a number of important strategic planning and operational problems that are critical to their performance. Programs must determine the appropriate number of vehicles to maintain in their fleet, cognizant of factors such as service availability minimums, maintenance costs, revenue potential, etc. Programs face pricing and revenue management decisions such as determining rental rates for vehicles, customer fees, and dynamic pricing options. Policies for vehicle repositioning must be implemented to periodically move vehicles between stations in order to correct vehicle inventory imbalances that occur due to disparities in station demand. Each of these decision-making processes typically requires the solution of some difficult large-scale stochastic optimization problem. Additionally, it is important that solution methods are able to account for both temporal and spatial effects in the network, while also remaining structurally simple enough for real-world implementation.;In this dissertation, we propose a widely applicable closed queueing network framework for modeling general vehicle sharing systems and develop a queueing network theoretic-based approach for their design and management. We focus on extremely large-scale sharing systems, where the computational efficiency of solution methods becomes critical. This leads us to study system design decisions using an asymptotic approach in which fleet size grows large, resulting in solution methods that are both computationally efficient and extremely accurate for these real-world systems. The asymptotic analysis we develop is also broadly applicable to many other types of large-scale real-world systems, including communication networks, computer systems, and supply chains. We also study the dynamic control problem of vehicle repositioning and customer admission. It is well-known that centralized control methods become too data and computation-intensive for implementation on large-scale systems, and so we focus on decentralized approaches that are efficient and able to be implemented in an agent-based manner with low-complexity computations. |
