효율적인 제한조건경계 샘플링을 이용한 신뢰성 기반 순차적 근사 최적화

Abstract

There are two types of sequential approximate method: Sequential approximate optimization (SAO) is a global optimization that finds a global optimum using sequentially constructed surrogate model; and sequential approximate reliability analysis (SARA) is a method that sequentially generates sample points on constraint boundaries and performs reliability analysis using surrogate model. However, the optimums in SAO are likely to fall in failure region because the optimums are found by deterministic design optimization (DDO); and SARA does not guarantee that optimum is the global optimum. Because each method has the drawbacks individually, the method is necessary that complements the drawbacks while having the advantages of each method. In this paper, reliability-based SAO that apply efficient constraint boundary sampling (ECBS) to SAO is proposed to obtain the global optimum in RBDO problem. Reliabilitybased SAO generates sample points sequentially on the constraint boundaries that object function value is lower than the optimum and the region that is high probability of feasibility and far from existing sample points. Therefore, reliability-based SAO enhance not only the probability of finding the global optimum, but also reliability accuracy of the optimums. The accuracy of reliability-based SAO is verified by mathematical examples.