Demand-based operations of vehicle sharing systems

Vehicle sharing systems (VSSs) allow users to rent vehicles for a short period of time, in a more flexible and convenient manner compared to the traditional vehicle rental services. The long-term VSS subscription replaces the need for contract signing for each rental, while the vehicle stations are located more frequently than the rental offices. They are also convenient from the user perspective as they waive the fixed cost of owning a car as well as maintaining it. Furthermore, increasing global greenhouse gas emissions brings concerns about the mobility habits, such as inefficient usage of personalized transportation. Therefore, the literature focuses on these systems to optimize their operations to make them convenient for both users and operators. In this thesis, we focus on simulation-optimization frameworks that allow us to investigate the mutual influences between the system characteristics and operations of VSS. We first present a generalized and holistic VSS management framework that is applicable to a system using any vehicle type. We conduct a systematic and extensive literature review and we position the literature in line with the framework. We also report the possible research directions that are suggested by the authors of the reviewed papers and the studies that address those. This allows us to identify the gaps in the literature and interesting research avenues. Following the findings from the extensive literature review, we investigate the added value of data collection and demand forecasting in bike sharing systems. We design a simulation-optimization framework to account for both supply and demand sides of the system. In this scope, a discrete-event simulator to represent real-life is developed. We improve an optimization model from the literature, that solves routing of static rebalancing operations, and incorporate clustering to be able to solve large-size instances. With the developed framework, experiments are conducted on one synthetic with 35 stations and four real-life case studies of various sizes, with 21, 298, 681, and 1361 stations, respectively. We conclude that trip demand forecasting does not necessarily improve the level of service in smaller-size, whereas this becomes more significant in larger-size bike sharing systems. Finally, we enhance the simulation-optimization framework to support one-way car sharing systems and evaluate different rebalancing operations strategies. We also enrich the framework with the state-of-the-art simulation module, i.e., Multi-Agent Transport Simulation (MATSim), which allows us to include disaggregate demand in our framework. This way, we can investigate individualistic behavior in car sharing usage. The rebalancing operations are determined in the optimization module following a heuristic approach. We observe that conducting rebalancing operations increases the number of rentals under specific scenarios where agents follow similar activities every day. The level of service obta