Computational intelligence method in multi-objective optimization

Decision makings may be formulated as optimization problem with multiple objectives, and a final decision is made from the set of Pareto optimal solutions which is called as Pareto frontier in the objective space. For searching Pareto frontier, so-called MOGA has been applied. On the other hand, the forms of objective functions in engineering design cannot be given explicitly in terms of design variable. In this situation, the values of objective functions can be evaluated by some analyses, which are usually very expensive. However, existing MOGAs need a large number of function evaluations for generating Pareto optimal solutions. Therefore, in order to decrease the number of function evaluations, this paper proposes a hybrid technique of MOGA introducing a prediction of objective function by support vector regression. Through the numerical examples, the effectiveness of the proposed method will be shown.