Algorithmes hybrides pour l’optimisation globale

We introduce two evolutionnary hybrid optimizers, based on surrogate models which use a limited prescribed number of exact evaluations of the criterion and its gradient. The first algorithm uses a discontinuous ansatz with a clustering technique. The second one uses a Liszka-Orkisz interpolation scheme, and keeps memory of the exactly evaluated individuals of previous generations. These two methods are applied to a 3D forging shape optimization problem. The considered objective combines the total energy cost and a defect criterion. We present numerical results which illustrate the efficiency of the developped algorithms.