Dynamic Supply-Demand Balancing Policy for CMoD Fleet

Autonomous Coordinated Mobility on Demand (CMoD) System is expected to revolutionize the contemporary point-to-point transportation system. One major advantage of the CMoD system is its ability to control the supply distribution over the operational area using vehicle rebalancing. Realtime implementation of this strategy comes under the broader category of sequential decision-making under uncertainties. In this paper, we elaborate on the critical model parameters such as the balancing frequency and the prediction horizon and present our dynamic supply-demand balancing policy that outputs the optimal set of parameters for a given state of the system, represented by the initial spatial distribution of the taxis and a partially observable demand pattern. We show that using the optimal parameters learned with this policy can improve the system performance by a striking 150% using the same fleet size w.r.t a naively chosen parameter set. This broad gap in the system performance shows that the supply-demand balancing systems are highly sensitive to these parameters. Finally, we present three case studies to elaborate on how the performance of the vehicle rebalancing system is affected by the model parameters as well as the demand market and the network structure.