Effect of Parameters Tuned by a Taguchi Design L_9 3^4 in the GRASP Algorithm to Solve the Vehicle Routing Problem with Time Windows

Alma Danisa Romero Ocaño, Maria De los Angeles Cosío León, Víctor Manuel Valenzuela Alcaraz, Gener José Avilés Rodríguez, Anabel Martínez Vargas


Metaheuristic algorithms are black box procedures that analyze a subset of possible solutions to solve a problem or a set of instances. Before they are implemented, it is necessary to select an optimum parameter vector P ∗, a task known as tuning. The vector P ∗ affects the efficiency of metaheuristics in solving a given problem. In this paper, the impact of tuning parameters using the Taguchi L_9 3^4 statistical procedure is analyzed. The effect of this method is analyzed in the metaheuristic algorithm named Greedy Randomized Adaptive Search Procedure (GRASP), solving the problem of Vehicle Routes with Time Windows (VRPTW). The results offered by the algorithm in a subset of instances of 25 clients improve on average to those reported in the literature, using a P ∗ proposed by Taguchi calibration.


Tuning, Taguchi, metaheuristic, GRASP, VRPTW

Full Text: PDF (Spanish)