Distributed evolutionary algorithms for simulation optimization

The optimization of such complex systems as manufacturing systems often nec essitates the use of simulation. Unfortunately, most current existing metho ds for simulation optimization present certain major inconveniences. In par ticular, they generally cannot take into account nonnumerical variables (e. g., dispatching rules), they may be sensitive to local extremums and they a re often time consuming. In this paper, the use of evolutionary algorithms is suggested for the optimization of simulation models. Several types of va riables are taken into account. The reduction of computing cost is achieved through the parallelization of this method, which allows several simulatio n experiments to be run simultaneously. Emphasis is put on a distributed ap proach where several computers manage both their own local population of so lutions and their own simulation experiments, exchanging solutions using a migration operator. After a first evaluation through a mathematical functio n with a known optimum, the benefits of this new approach are demonstrated through the example of a transport lot sizing and transporter allocation pr oblem in a manufacturing flow shop system, which is solved using a distribu ted software implemented on a network of eight Sun workstations.