Search

Carsharing, as an alternative to private vehicle ownership, has spread worldwide in recent years due to its potential of reducing congestion, improving auto utilisation rates and limiting the environmental impact of emissions releases. Along with its growth, the flexibility of carsharing systems also brings complex problems to the operators. One dominant challenge in carsharing systems is to ensure the supply of vehicles can meet the demand of users in a cost-effective manner. This requires accurately predicting users' demand and optimally relocating vehicles in response to demand variations. The two principal areas of this thesis are methods to estimate demand and optimally relocate fleet. From the demand side, this study models users’ vehicle selection and utilisation patterns. Focusing on vehicle selection behaviour, a spatial hazard-based model (SHBM) is proposed to investigate the impacts of users’ socio-demographic attributes and fleet characteristics on their choice set formation behaviour in selecting vehicles. The modelling is achieved by regarding “distance to carsharing vehicle” as a random variable analogous to the duration in conventional hazard-based models. Data collected from the Australian carsharing company GoGet are utilised to calibrate the models. The accelerated failure time model with a log-logistic distribution is found to provide the best fit. Upon making a vehicle selection, users then decide the amount of consumption to allocate to each selected vehicle type. This process involves making multiple discrete choices of continuous amounts and is modelled by the multiple discrete-continuous extreme value (MDCEV) modelling framework. Three MDCEV models considering travel time, mileage, and monetary expenditure as the continuous consumption constraints are developed to estimate the impacts of a set of socio-demographic attributes on user’s vehicle choice and capture the satiation effect with increasing the consumption for each vehicle type. An efficient simulation procedure is applied to evaluate the performance of the three MDCEV models. The results indicate travel time, mileage and expenditure affect users’ vehicle usage pattern in the same way. The findings from these two demand models can be referred to by the operators when determining the most efficient allocation of resources within carsharing systems. From the operation side, the research develops and solves novel models for the vehicle stock imbalance problem in one-way carsharing systems. Previous studies have proposed relocation methods to handle it, but the interdependence between demand and supply has never been considered. The thesis proposes two relocation models to link demand and supply. Both incorporate a discrete choice model (DCM) in an integer linear programming (ILP) model to account for the interaction. The difference between them lies in the DCMs. In the first model, the DCM does not assume users’ demand to be elastic to vehicle availability. The ILP model solves optimal relocation decisions and updates vehicle availability for each station; the DCM then coupled with the updated vehicle availability changes users’ trip demand reciprocally. Built on the first model, the second model extends the DCM by including vehicle availability as a parameter directly affecting demand. In this new framework, demand and supply are linked by vehicle availability: it is the output of the ILP model and at the same time the input of the DCM. The nonlinearity of the DCM is further linearised through a linearisation approach. Both models are tested in the GoGet network. The results reveal if there is a strong interdependence between demand and supply, the supply has a critical impact on system profit. The core contribution of this thesis is to take the first attempt to understand and optimise carsharing systems considering the interdependency of demand and supply comprehensively.