RBFネットワークによる多目的逐次近似最適化(機械力学,計測,自動制御)

Abstract

In this paper, a sequential approximate multi-objective optimization procedure by the Radial Basis Function (RBF) network with the Satisficing Trade-Off Method (STOM) is proposed. The sampling strategy is an important issue in the sequential approximate optimization. In this paper, the density function and the pareto fitness function are proposed. The objective of the density function is to find the sparse region in the design variable space. New samplings point are obtained by optimizing the density function. The objective of the pareto fitness function is to find the approximate set of pareto optimal solutions from the given data. New sampling point is obtained by optimizing the pareto fitness function. Both functions are constructed by the RBF network. By using both functions, the approximate set of pareto optimal solutions can be found effectively even when the set of pareto optimal solutions are separeted. Through simple numerical examples, the validity of proposed sampling strategy is examined.