Algoritmo GVNS Híbrido Aplicado ao Problema das p-Medianas Capacitado

ABSTRACT

The Capacitated p-Median Problem (CPMP) consists of locating p facilities in a network composed of n clients and deciding which facility will serve each client to minimize the sum of all distances from each facility to each client and meet the facility capacity constraints. Since the CPMP belongs to the NP-hard class, in this work, four variants of the General Variable Neighborhood Search (GVNS) metaheuristic, named G-VND, G-RVND, GG-VND, and GG-RVND, are implemented for treating the CPMP. They differ concerning the method used to build an initial solution and the one used for local search. In the first two variants, the initial solution is generated randomly, and the local search method is performed via Variable Neighborhood Descent (VND) or Random Variable Neighborhood Descent (RVND), respectively. In turn, in the last two, the initial solution is executed via the construction phase of the Greedy Randomized Adaptive Search Procedure (GRASP) method. Using benchmark instances from the literature, we initially showed that the GG-VND variant performed better than others. Then, we show that this variant has an equivalent or superior performance when compared with the literature algorithms.

Keywords:
Capacitated p-Median Problem; General Variable Neighborhood Search; GRASP; metaheuristics