Résumé:
The dynamic ridesharing system is designed to match travelers with similar itineraries and schedules in a short period of time. These systems can provide significant social and environmental benefits by reducing the number of cars used. Effective and efficient optimization techniques that match drivers and passengers in real time are one of the necessary components of a successful dynamic ridesharing system, it is within this framework that our work falls.
In this work, we propose to solve the dynamic matching problem under constraints by reinforcement learning. We have proposed a new time-based modeling that aims to minimize passenger waiting time. Our framework is validated using spatiotemporal data of real rides from the New York City Taxi public dataset as well as a simulator we developed. The results obtained support our choices and have shown the effectiveness and robustness of our approach in real time.
