抄録
In industrial design optimization, the objectives and constraints are generally given as the implicit form of the design variables. In addition, the objectives and constraints are not explicitly known but can be evaluated through computationally intensive numerical simulation. Under this situation, the response surface methodology (RSM) is one of helpful approaches for design optimization. In recent, sequential approximate optimization (SAG) has gained its popularity. In this paper, a sequential approximate multi-objective optimization (SAMOO) with radial basis function (RBF) network is proposed. In the SAMOO, it is important to identify highly accurate pareto-frontier with a small number of function evaluations. In order to achieve this objective, a new function called the pareto-fitness function with the RBF network is developed. The pareto-fitness function can generate local minima around a set of pareto-optimal solutions, and the optimal solution of this function is taken as a new sampling point. Also, the density function to find the unexplored region is also employed. It is possible to add the new sampling points around a set of pareto-optimal solutions effectively by using both functions. Through numerical examples, some advantages employing both functions are well discussed.
