The rich heterogeneous dial-a-ride problem with trip time prediction

Abstract

The dial-a-ride problem (DARP) involves designing vehicle routes to fulfill the door-to-door transportation requests of users where the goal is to minimize costs while satisfying transportation requests. In this paper, we introduce the rich heterogeneous DARP, which extends the generalized heterogeneous DARP to consider a fleet of buses and taxis, multiple depots, time windows at pickup and delivery locations, maximum ride and waiting times, and the possibility of an accompanying person. Our approach is based on a real service in Barcelona, and we also consider the variation in trip duration based on the time of day and day of the week. A predictive model is developed using machine learning techniques to estimate trip durations accurately. We apply our proposal to the daily door-to-door transportation of people with reduced mobility in Barcelona and demonstrate its superiority in terms of costs and quality of service by using the Gurobi optimizer. Additionally, we provide an analysis of the consequences of varying certain features on the costs and quality of service.