A hierarchical parallel strategy for aerodynamic shape optimization with genetic algorithm

An ecient parallel strategy is presented for optimization of the aerodynamic shapes using Genetic Algorithm (GA). The method is a hybrid Parallel Genetic Algorithm (PGA) that combines a multi-population PGA and master-slave PGA using Message Passing Interface. GA parameters are rstly tuned according to the fact that subpopulations evolve independently. The eect of the number of sub-population on the computational time is investigated. Finally, a new strategy is presented based on the load balancing that aims to decrease the idle time of the processors. The algorithm is used for optimization of a transonic airfoil. An unstructured grid nite volume ow solver is utilized for objective function evaluations. For the considered class of problems, the suggested Hierarchical Parallel Genetic Algorithm (HPGA) results in more than 30% reduction in optimization time in comparison to regular master-slave PGA. A semi-liner speed-up is also obtained which indicates that the model is suited for modern cluster work stations.