Dynamic ride-sourcing systems for city-scale networks, Part II: Proactive vehicle repositioning

Abstract

The emergence of ride-sourcing systems has unprecedentedly changed the market for on-demand mobility services. The service quality of the ride-sourcing systems and their effects on transportation networks rely on effective matching and redistributing of idle vehicles. In Part I, we introduced a vehicle–passenger matching method that jointly determines the maximum matching distance and matching intervals to minimize passengers waiting time. Moreover, a non-equilibrium macroscopic model is developed to predict the evolution of the ride-sourcing system states. In this paper, we propose a control method to transfer idle vehicles to balance the passengers’ demand and supply of ride-sourcing vehicles by repositioning the idle vehicles in locations where there is a higher possibility of faster pickups of new passengers. This paper develops a Nonlinear Model Predictive Controller (NMPC) that proactively transfers idle vehicles based on the predicted future state of the ride-sourcing system using the non-equilibrium model developed in Part I paper. The proposed NMPC implements an optimum rolling horizon strategy that solves a constrained optimization problem at each transferring sample time. The proposed method considers the effect of network congestion and impatient passengers and drivers. We show that the proactiveness of the NMPC incentivizes the drivers to follow the transferring commands instead of random cruising during idle periods. In microsimulation experiments, the designed controller improves the performance of the ride-sourcing system by reducing passengers’ average unassigned time (-20.4%) and waiting times (-12.4%), vehicles’ average waiting times (-8.8%), the fleet size (-18.6%), and increasing the number of the served trip requests (9.7%).