Impact of Parallel Memetic Algorithm Parameters on Its Efficacy

The vehicle routing problem with time windows (VRPTW) is an NP-hard discrete optimization problem with two objectives—to minimize a number of vehicles serving a set of dispersed customers, and to minimize the total travel distance. Since real-life, commercially-available road network and address databases are very large and complex, approximate methods to tackle the VRPTW became a main stream of development. In this paper, we investigate the impact of selecting two crucial parameters of our parallel memetic algorithm—the population size and the number of children generated for each pair of parents—on its efficacy. Our experimental study performed on selected benchmark problems indicates that the improper selection of the parameters can easily jeopardize the search. We show that larger populations converge to high-quality solutions in a smaller number of consecutive generations, and creating more children helps exploit parents as best as possible.