A GRID-BASED HYBRID CELLULAR GENETIC ALGORITHM FOR VERY LARGE SCALE INSTANCES OF THE CVRP

This work presents a hybrid genetic algorithm (GA) for solving the largest existing benchmark instances of the capacitated vehicle routing problem (CVRP). The pop-ulation of the algorithm is structured by following two classical parallelization models for GAs: coarse- and fine-grained. Indeed, the proposed model is a distributed GA (coarse-grained) in which each island is a cellular GA (fine-grained). It has been called PEGA (Parallel cEllu-lar Genetic Algorithm). PEGA has been built on top of ProActive and it has been executed on a grid platform composed of more than 100 machines so as to reduce the computation time. The results show that, for many of the considered instances, PEGA improves the best results re-ported by any existing algorithm in the literature