Application of Genetic Algorithms for the DARPTW Problem

Abstract

On the Dial-a-Ride with time windows (DARPTW) customer transportation problem, there is a set of requests from customers to be transported from an origin place to a delivery place through a locations network, under several constraints like the time windows. The problem complexity (NP-Hard) forces the use of heuristics on its resolution. In this context, the application of Genetic Algorithms (GA) on DARPTW was not largely considered, with the exception of a few researches. In this work, under a restrictive scenario, a GA model for the problem was developed based on the adaptation of a generic GA model from literature. Our solution applies data pre-processing techniques to reduce the search space to points that are feasible regarding time windows constraints. Tests show competitive results on Cordeau & Laporte benchmark datasets while improving processing times.

Solving the Dial-a-Ride Problem Using Agent- Based Simulation

“…When Zidi et al [10] considered soft time windows, their vehicle ride times were largely similar to those obtained by Cubillos et al [8], and significantly outperformed Cordeau and Laporte's (2003) customer ride times [7].…”

“…Solution methods may be designed for hard or soft time windows. For example, for the DARP, Cordeau and Laporte [7] use soft time windows to achieve low route costs, whereas Cubillos et al [8] use hard time windows to achieve good customer service.…”

“…A genetic algorithm approach was used by Cubillos et al [8], where time windows were considered as hard constraints. When Zidi et al [10] considered soft time windows, their vehicle ride times were largely similar to those obtained by Cubillos et al [8], and significantly outperformed Cordeau and Laporte's (2003) customer ride times [7].…”

“…Both Jorgensen et al [15] and Cubillos et al [8] apply their own version of a genetic algorithm to the same dataset of the DARP, using a cluster-first route-second approach. In both cases a genetic algorithm is used for the clustering, with Cubillos et al [8] using a preprocessing step before commencing with the clustering.…”

“…In both cases a genetic algorithm is used for the clustering, with Cubillos et al [8] using a preprocessing step before commencing with the clustering. This involved constructing a preliminary precedence table to reduce the search space of feasible solutions.…”