Stochastic process model of the evolution of providers and users in a shared on-demand ride service system

Shared use of on-demand ride service can only reduce traffic congestion and its concomitant problem if its popularity exceeds a non-shared use which is often provided through the same platform and may dominate. Endogeneity among users (i.e. availability of sharing partners) and between users and providers (i.e. feedback loop between them) embeds uncertainty in the system. It complicates predicting what level of service users will experience. Several studies modelled day-to-day dynamics in an on-demand ride service system considering a feedback loop between providers and users, yet, it is often modelled with a deterministic approach. This thesis aims to develop a stochastic process model to investigate the impact of variability on the evolution of a system attribute to the feedback loop between users and providers and the endogeneity among users.
A stochastic process model represents the day-to-day learning and decision-making process of users and providers. Users (providers) reconsider a service to request (offer) every day by comparing their experienced utility (profit) to use (provide) each service with the collective average utility (profit) of unselected service between non-shared and shared service. The collective average utility (profit) is estimated every day as the weighted sum of the mean utility (profit) among today’s users (drivers) and the collective average utility (profit) of the previous day. Service shift occurs for only a proportion of those who consider a change, which reflects a range of choice inertia and hesitation towards change. Changes in fleet size and demand give rise to a new experience for users and drivers on the next day. A queueing model is utilized to model an on-demand ride service system, which provides variables to estimate utility and profit. For a shared service, a fair cost split among users is modelled with a modified Shapley Value, which is newly proposed in this research.
With the developed model, the two types of numerical experiments have been conducted with different parameter settings; 1) those with a fixed fleet size and 2) those with variable fleet size. The former experiments aimed to understand the attributes of the proposed model. The results suggest that the service network geometry is the main determinants of the stationary distribution of mode share. The experiment with unfixed fleet size showed that the proposed stochastic process consists of three regimes, the pseudo stable, the pseudo periodic, and the swan regime. It is discovered that, depending on the parameter setting, the frequency and length for each regime changes, which results in changing the stationary distribution of mode share and the proportion of fleet.