均匀设计抽样及其优良性质

Uniform Design Sampling and Its Fine Properties

抽样和设计是计算机试验的一个重要研究课题。本文提出了一种新的抽样方法—均匀设计抽样,研究了它的一些基本性质,并将其应用于数值积分近似计算,这种抽样是王元和方开泰(1981)均匀设计思想的一个发展,也是对Latin Hypercube抽样的一个重要改进,通过与Monte Carlo方法,Latin Hypercube抽样(包括OA-Based Latin Hypercube抽样)和均匀设计的比较,表明了这种抽样的优越性,最后还讨论了在一般分布情形下如何应用这种抽样。

This paper propose a new sampling method, uniform design sampling (UDS), which is a development of uniform design (UD) proposed by Wang anf Fanf (1981) and an improvement for Latin hypercube sampling (LHS) proposed by Mckay, conover and Beckman (1979). We study the properties of this sampling and apply it to computations of numerical integrals, and compare it with Monte-Carlo method, UD (including URD), and LHS (including OALHS). The results show that UDS is more powerful than the other methods. Also, we give a discussion of how to use the UDS to general distribution cases.