Solving the parameter setting in multi-objective evolutionary algorithms using grid:Cluster

Abstract

The parameter values of a Multi-objective Evolutionary Algorithm greatly determine the behavior of the algorithm to find good solutions within a reasonable time for a particular problem. In general, static strategies consume lots of computational resources and time. In this work, a tool is used to develop a static strategy to solve the parameter setting problem, applied to the particular case of the Multi-objective 0/1 Knapsack Problem. GRID:Cluster makes feasible a dynamic on-the-fly setup of a secure and fault-tolerant virtual heterogeneous parallel machine without having administrator privileges. In the present work is used to speed-up the process of finding the best configuration, through optimal use of available resources. It allows the construction of a driver that launches, in a systematically way, different algorithm instances. Computational results show that, for a particular problem instance, the best behavior can be obtained with the same parameter values regardless of the applied algorithm. However, for different problem instances, the algorithms have to be tuned with other parameter values and this is a tedious process, since all experiments have to be repeated, for each new set of parameter values to be studied. © 2010 Springer-Verlag Berlin Heidelberg.