Appariement multi-objectif dans les systèmes de covoiturage dynamique

Abstract

Dynamic ridesharing represents an innovative solution to the challenges of modern urban mobility. This system allows users to find rides in real-time by connecting drivers and passengers sharing a similar route. Users benefit from shared transportation costs and it reduces the use of individual cars, thus contributing to an overall decrease in vehicle usage. In this work, we focused on the complex issue of dynamic matching. To address this challenge, we opted for a multi-objective reinforcement learning method. This solution takes into account a set of crucial constraints: spatiotemporal constraints, capacity constraints, waiting time constraints, distance constraints, and detour time constraints. Our model aims to achieve three main objectives: minimize passenger waiting time, reduce driver detour time, and maximize vehicle utilization. To validate our model, we used real data from the public New York City taxi dataset. Additionally, we developed a simulator to evaluate the performance of our approach. The results obtained are promising. These positive results highlight the potential of our dynamic ridesharing system to enhance the experience of both drivers and passengers, providing a flexible and robust solution.