超饱和设计数据新地分析方法.docxVIP

1 Introduction 1 Introduction Manypreliminaryindust rialscreeningexperimentstypicaUycontainlargenumber ofpotentiallyrelevantfactors．Among them，onlyfewbelievedtobeactive(haviIlg large efrects) The supersa七urated designs(SSDs)，nrst constructed systelnatically byBooth andCox(1962)，haSreceivedgreatdeal ofattention since the appearance of Lin(1993) After that，there several new methods of constructing ssDs(choosing the design matrix)．These include Lin(1995)，、Vu (1993)，Nguyen(1996)，cheng(1997)，Li and wu(1997)，Tang andⅥ～(1997)， Liu and zhang(2000)，Butler et a1． (2001)and on．Now it出so extends to mixed—level ssDs include Yamada and Lin(2002)，Yamada and Matsui(2002)， Fhng et al (2003)，Fhng et al (2004)，Li et a1．(2004)，1ramada et a1．(2006)and Liu et al_(2006)．In this paper，we focus on analyzing of two-level supersaturated designs Analyzing supersaturated designsischallengebecause che number ofperiments is less than the number of factors．‘rb丘nd the sparse active effbcts， variableselectionbecomes fundamentalin the analysisstageofsuchscreeningperiments．1tisas吼lmedthat there isinteractionbetween the factors andtheefrect is linear in the level of each factor．Hence the basic model for observations 玑is given by(assulne曲at the玑is centered) Ⅳ = 岛Z + 1，2，．．，n np， (1) ，∑芦 where p is the number of factors and礼is the number of observations(runs) Here z，，is the level of factor z，in i and is restricted to two values(+1 or一1)and ￡。～．Ⅳ(o：矿) The parameter J 0=1：2： ．：p)represents the efrect of factor J and it is assumed that onlv f色w of these Darameters different from O． Let Y=(Ⅳ1，驰，．．．：蜘)丁，x=(Xllx21．．．Ixp)be the design matr呔，where ■=(z1J，z巧，．．．，z哪)T，J=1：2，．．，p，and￡=(￡1，￡2，．．，E。)～Ⅳ(O，盯2I。) then f11 can be written Y=X口+￡ (2) Since the number of parameters p is more than the number of observations 凡，Ilsuauy least squares methods cannot be used with(2)and all p factors’rra， ditional approaches for supersa土urated designs，described in Lin(1995)，such