Adaptive Multivariate Statistical Process Control for Monitoring Time-varying Processes 1

Size: px
Start display at page:

Download "Adaptive Multivariate Statistical Process Control for Monitoring Time-varying Processes 1"

Transcription

1 Adapve Mulvarae Sascal Process Conrol for Monorng me-varyng Processes Sang Wook Cho *, Elane. Marn *, A. Julan Morrs *, and In-eum Lee + * Cenre of Process Analycs and Conrol echnology, School of Chemcal Engneerng and Advanced Maerals, Unversy of Newcasle upon yne, NE 7RU, UK + Deparmen of Chemcal Engneerng, Pohang Unversy of Scence and echnology, San 3 Hyoja Dong, Pohang, , KOREA Absrac: An adapve mulvarae sascal process monorng MSPC approach s descrbed for he monorng of a process whch ncurs changes n he operang condons. Sample-wse and block-wse recursve formulae for updang he mean and covarance marx are derved. y ulsng, he curren model and he updaed mean and covarance srucures, a new model s derved recursvely. ased on he updaed PCA represenaon, wo monorng mercs, Hoellng s and he Q-sasc are calculaed and her conrol lms are updaed. For more effcen model updang, forgeng facors, whch can change wh me, for he updang of he mean and covarance are consdered. Furhermore, he updang scheme proposed s robus n ha no only reduces he false alarm rae n he monorng chars bu n ha he model s nsensve o oulers. he adapve MSPC approach developed s appled o a mulvarae sac sysem and a connuous srred ank reacor process wh he resuls beng compared o hose from he applcaon of sac MSPC. he revsed approach s shown o be effecve for he monorng of processes where fas or slow changes are ncurred. Keywords: me-varyng process, Adapve PCA, Varable forgeng facor, Robus parameer esmaon Submed o Indusral & Engneerng Chemsry Research o whom all correspondence should be addressed. el: Fax: E-mal: e.b.marn@newcasle.ac.uk

2 . Inroducon Mulvarae sascal projecon mehods such as prncpal componen analyss PCA and paral leas squares PLS combned wh sascal monorng chars have been wdely appled for on-lne connuous and bach process monorng. -3 ypcally monorng chars based on Hoellng and he Q-sasc 4,5 are used o deec fauls wh he conrbuon plo 6, whch shows he conrbuon of each measured varable o he monorng sasc, beng adoped o asss n he denfcaon of a possble assgnable cause of he faul. Mulvarae sascal process conrol MSPC schemes based on PCA or PLS are formulaed on he assumpon ha he process varables are ndependen, dencally dsrbued and lnearly correlaed. Furhermore, f he underlyng dsrbuon of he process varables s mulvarae normal, hen he process can be descrbed oally by he mean and covarance of he varables. hus, due o he lmaons mposed by he underlyng assumpons, MSPC s usually performed under seady sae condons. Occasonally, problems assocaed wh he effcency of he process monorng scheme can arse because he underlyng assumpons of convenonal MSPC are volaed. For example due o non-lnear process behavour, he presence of me-dependency auocorrelaon, process dynamcs or changes n operang condons. o address hese lmaons, a number of alernave monorng schemes have been proposed such as dynamc, non-lnear or adapve MSPC. In hs paper, an adapve PCA based MSPC scheme s proposed o monor processes where operang condon changes are common. Indusral processes usually exhb me-varyng behavor due o nenonal dsurbances such as se-pon changes or unwaned dsurbances such as he ageng of equpmen, foulng and caalys degradaon. here have been several papers presened on adapve PCA and PLS. For example Wold 7 proposed a scheme whch ulzed he exponenally weghed movng concep for he updang of boh PCA and PLS models. However, one lmaon wh hs algorhm s ha requres all he hsorcal daa o be used n he updang of he model every me a new sample becomes avalable. Dayal and MacGregor 8 descrbed a recursve exponenally weghed PLS algorhm, usng he PLS kernel algorhm, based on recursvely updang he covarance marces as opposed o usng he oal hsorcal daa se. Qn 9 developed an alernave sample-wse and block-wse recursve PLS algorhm ha only ulzes he loadng marx and he regresson coeffcens beween he npu and oupu score vecor pars n he PLS model, nsead of he covarance marces, maeralzng n more rapd updang of he model. More recenly, L e al. proposed wo effcen learnng algorhms for recursve PCA for he monorng of me-varyng processes. he sudy repored n hs paper focuses on developng a new adapve monorng scheme for me-varyng processes wh a number of conrbuons beng presened. Frs sample-wse and bach-wse recursve formulae are derved for updang he sample mean, varance, and covarance pror

3 o an effcen and fas learnng algorhm beng proposed for calculang a new PCA represenaon hrough eher he sample or block-wse approach. Secondly, an overall model updang and process monorng mehod s presened. More specfcally a new calculaon procedure for he varable forgeng facors s developed for updang he mean and covarance. Fnally, a robus model updang scheme s descrbed o reduce he effec of oulers on he model updang procedure. he remander of hs paper s srucured as follows. Secon presens he recursve updang forms of he sample mean and covarance wh he focus n Secon 3 beng on an alernave and effcen way by whch o fnd he egenvalue egenvecor pars n he updaed covarance or correlaon marx. In Secon 4, an overall adapve sascal process monorng sraegy s proposed. wo specfc aspecs are consdered whch mpac on he monorng sascs, oulers, and he selecon of he forgeng facors one for he mean and he oher for he covarance. Fnally, n Secon 5, he resuls of he proposed mehodology are assessed hrough smulaed examples where a number of dfferen ypes of changes are consdered ncludng drf, se-pon changes, a ramp and a change n he underlyng dmensonaly of he process. Fnally a case sudy on a connuous srred ank reacor process s consdered.. Recursve Form of he Gaussan Densy Funcon of a Random Vecor In MSPC, measuremens obaned durng normal operang condons are regarded as beng colleced from a saonary Gaussan process,.e. x ~ N µ, Σ. he maxmum lkelhood esmae of µ and Σ s gven by samples. ˆ m n x µ and Σ ˆ S x m x m n, respecvely, based on n. Sample wse Updang When he process operang condons change eher gradually or abruply, he mean vecor and covarance marx wll no be consan and wll need o be updaed. Each me a sample or a sample block become avalable, boh he mean and covarance are updaed wh he degree of change n he model srucure beng dependen on he magnude of he forgeng facors. Consequenly he PCA represenaon should accordngly be updaed o allow he monorng of such a me-varyng process, x N µ, Σ. he esmaed mean vecor and covarance marx a me pon are gven n equaons ~ and, respecvely, : m α x α x + αx + L+ α + α + L+ α x

4 and β x m x m β S, he sample mean a me pon can be denoed more smply as a weghed sum of he sample mean a me pon, m -, and he sample a me pon, x : α α x + α m m 3 α α Lkewse an alernave form of he sample covarance s gven by: β ~ ~ β xx + β S S 4 β β where, ~ x x m s he mean-cenered sample vecor a me pon. As becomes large, Eqs. 3 and 4 can be furher smplfed o he followng forms: m 5 α x + αm and ~ ~ β xx + βs S 6 β dag ~ x ~ x + βd D, From Eq. 6, f only he sample varance s beng consdered, hen where D s a dagonal marx whose dagonal elemens are dencal o hose of S. Also, he correlaon marx s esmaed as: ~ D / R 7 / β D x ~ x β As becomes large, Eq. 7 can be smplfed: 3

5 4 / / ~ ~ + β β R D x x D R 8. lock wse Updang When a process changes slowly and he samplng me s small compared wh he process me consan, s neffcen o updae he model each me a new sample becomes avalable, as descrbed n he prevous secon, snce s compuaonally nensve. Insead he PCA represenaon can be updaed afer a block of samples have been colleced. he block-wse updang approach has wo aracve feaures, frs has a low compuaonal cos and secondly reduces he rsk of updang he model based on a false alarm, snce s possble o decde wheher a deeced alarm s false pror o updang he model. he block-wse approach s dencal o he sample-wse updang procedure excep ha a sample block s used o adap he model nsead of a sngle sample. he recursve equaons for he esmaed mean and covarance are gven by: + α α α α m X X m 9 and ~ ~ ~ ~ + β β β β n S X X X X S where m N R X s he h sample block m s he number of measuremens, [ ] N R L, n s he sample number n he h block, and ~ m X X. he recursve calculaon for he sample correlaon marx s gven by: / / ~ ~ + β β β β n R R D X X D R

6 where / ~ ~ / D X X D R s he correlaon marx of he curren sample block and ~ ~ D. D s a dagonal marx whose dagonal elemens are β dag X X + βd dencal o hose of S. 3. Prncpal Componen Analyss PCA Model Updang When a new sample or block becomes avalable, he egenvalues and egenvecors loadng vecors of he newly updaed correlaon marx are calculaed o oban a new PCA represenaon. A number of approaches have been proposed o calculae he egenvalues and egenvecors. One of he smples approaches s o perform a sngular value decomposon SVD on he curren correlaon marx. More recenly, L e al. proposed wo recursve PCA algorhms based on rank-one modfcaon and Lanczos rdagonalzaon, whch are compuaonally more effcen han SVD. he precedng mehods requre he sorage and updang of he correlaon marx o consruc an updaed PCA represenaon. When he number of measured varables s farly large, e.g. over a hundred, whch s common n many ndusral processes, he procedure s me consumng. Consequenly, a revsed and more effcen algorhm was proposed o updae he egenvalues and egenvecors and hs dea s exended o he bach-wse case. he block-wse updang procedure s frs consdered snce sample-wse updang s a specal case of he block-wse case. ased on he mean vecor a me pon -, m, he prncpal loadng marx m a P R and he dagonal marx of egenvalues a dag λ, λ, L, λ a a Λ R a me pon, he newly updaed values are obaned by usng hese values and he new sample block X a me pon. a s he approxmaed rank of R. Frs, he correlaon marx of X s calculaed, / ~ ~ / D X X R n D and SVD s appled gvng an approxmaon of R : P Λ P R he columns of m b P R are he prncpal egenvecors and b dag λ, λ, L, λ b b Λ R s he dagonal marx whose elemens are he sgnfcan egenvalues of R, where b s he approxmaed rank of R marx R can be approxmaed as:. Lkewse, he prevous correlaon 5

7 6 P Λ P R 3 Subsung Eqs. and 3 no Eq., R may be approxmaed as: ˆ + β β P Λ P P Λ P R 4 Combnng he wo erms on he rgh sde of Eq. 4, he equaon can be rewren as: [ ] ˆ K K P P Λ Λ P P R β β 5 where b a m b b a a + R βλ βλ λ β λ β p p p p K L L. he updaed loadng vecors and egenvalues can be obaned by performng SVD on K K : p p K K λ 6 Alernavely, he loadng vecors and egenvalues can be calculaed usng b a b a + + R K K nsead of m m R K K,.e. q q K K γ. Mulplyng boh sdes of hs relaonshp by K δ gves: q K K q K K δ γ δ 7 Comparng Eq. 7 wh Eq. 6, can be deduced ha λ γ and q p δk, where / λ δ o ensure δ q K p. Sample-wse updang s a specal case and hence K becomes:.5 ~ + R β λ β λ β a m a p x D p p K L 8 If > a + m n he sample-wse updang procedure or f b a m + > n he block-wse updang procedure, hen s compuaonally effcen o use K K, nsead of K K o calculae P and Λ be-

8 cause K K makes he egenvalue problem more smple and rapd o solve. I should, however, be noed ha here are wo sources of runcaon error n calculang P and Λ n he block-wse updang procedure. hese maeralse from dscardng he nsgnfcan egenvecors and egenvalues pars n Eqs and 3. In he case of he sample-wse updang procedure, runcaon error may occur when approxmang he correlaon marx as defned n Eq. 3. In he above updang algorhm, he egenvecors and egenvalues, whch are consdered o be nsgnfcan, are dscarded o approxmae and R. R herefore he dmensonaly of he wo marces R and R, a and b, requre o be deermned. In each recursve model updang sep, here are several ways o selec a and b 3 bu he approaches are dependen on he praccal applcaon. hree mehods can be consdered fxng he parameer a or b as consan values; reanng hose egenpars whose egenvalues are larger han a predefned hreshold, or 3 reanng hose egenpars such ha a specfed fracon of he oal sum of egenvalues s ncluded, e.g Adapve Process Performance Monorng 4.. Adapve Monorng Sascs In general, a PCA-based MSPC scheme ulses wo monorng sascs, Hoellng s : p Λ p x P Λ P x 9 where p p p Λ R s he leadng prncpal marx of Λ, and he Q-sasc: Q x I PP x whch are expressed n erms of he Mahalanobs and Eucldean dsances, respecvely. Hoellng s represens he dsance n he PCA model space, whereas he Q-sasc ndcaes a dsance from he model space. Under he assumpon of mulvarae normaly of he observaons and emporal ndependence, he δ % conrol lm for Hoellng s s calculaed by means of a lm α χ approxmaon,.e. χ p, whch s generally used n qualy conrol because of s smplcy. In hs case δ denoes he sgnfcance level. hs mehod s favourable snce requres only small compuaonal me n each model updang sep. Also he δ % conrol lm for he Q-sasc s gven by: 7

9 Q lm z θ h θ + + θh h / h θ δ θ a j j Λ k p kk where θ + for j,,3, h θθ3 3 θ, and z δ s he normal devae. 4 In adapve PCA monorng, he wo conrol lms requre o be calculaed every me he PCA model s updaed snce he number of prncpal componens, p and he dagonal marx whose elemens are he egenvalues of he covarance marx, Λ are me-varyng. 4.. Varable Forgeng Facor As presened n Eqs. and, calculaon of he weghed mean and covarance requres he weghng parameers ermed forgeng facors α and β, o be deermned. If boh forgeng facors are uny, hey are he maxmum lkelhood esmaes calculaed from all he daa. y usng a forgeng facor ha s less han one, prevous samples are auomacally weghed ou whou deleng daa from he process model. As he value ges close o uny, he process has a long-erm memory, ha s, he number of prevous samples ha have an effec on he curren model ncreases. o dae mos model updang approaches have used an emprcal consan forgeng facor. However he opmal value of he forgeng facor vares sgnfcanly dependng on he rae of process change. When he process changes rapdly, he updang rae should be hgh, whereas when he change s slow and hus he essenal process nformaon s vald for a long perod, should be low. However, s lkely ha he rae of process change or varaon n real processes vary wh me. If hs s he case, he forgeng facor should be deermned accordng o he underlyng objecve of he process monorng scheme. o cope wh hs suaon, nsead of usng a fxed forgeng facor, can be adjused accordng o he curren process condons. Dayal and MacGregor 8 used he algorhm developed by Forescue a al. 5 o adjus he forgeng facor n her recursve PLS algorhm. he same concep of he varable forgeng facor has been appled n recursve PCA. 6 In hs sudy, a new algorhm for adapng he forgeng facor s proposed. Fundamenally dffers from he prevous algorhm proposed by Forescue a al. 5 n wo aspecs. Frs, he varable forgeng facor used for updang he sample mean and he covarance marx can ake dfferen values, enablng greaer flexbly and generaly. Secondly, he wo forgeng facors drecly depend on he change n he mean and covarance srucures, unlke he prevous approaches n whch boh depended on Hoellng s and he Q-sasc. In he proposed updang algorhm, wo forgeng parameers α and β are used o updae he sample mean vecor and covarance or correlaon marx, respecvely. he forgeng facor α for updang he mean vecor s calculaed as: 8

10 [ exp{ m }] n nor α αmax αmax αmn k m where α max and α mn are he maxmum and mnmum forgeng values, respecvely, k and n are funcon parameers defned below, and m s he Eucldean vecor norm of he dfference beween wo consecuve mean vecors. Here, mnor s he averaged m obaned usng hsorcal daa as wll be explaned n Secon 4.4. Smlarly, he forgeng facor β for updang he covarance or correlaon marx s gven by: [ exp{ R }] n nor β βmax βmax βmn k R 3 where β max and β mn are he maxmum and mnmum forgeng values, respecvely, and R s he Eucldean marx norm of he dfference beween wo consecuve correlaon marces. here are four funcon parameers ha requre o be deermned, α max or β max, α mn or β mn, k, and n. he nfluence of hese four parameers on he forgeng facor value s shown n Fg.. he change n he paerns of he forgeng facor were nvesgaed accordng o he magnude of x x, nor where x m or R, by varyng one of he four parameers and fxng he oher hree parameers. hs procedure was repeaed eravely. he defaul values of he four funcon parameers were aken as α max.99, α mn. 9, k.693 and n. Here, he defaul k value,.693, corresponds o he case where α αmax + αmn when n and x x nor. Fgures a and b show he effec of he maxmum and mnmum values, respecvely, of α on he forgeng facor. oh α mn and α max affec he range of he change,.e. he forgeng facor vares whn he lms of α mn and α max. Also, α mn and α max regulae he maxmum and mnmum rae of model adapaon, respecvely. he parameers k and n conrol he sensvy of he change n α. he larger he value of k, he faser α decreases as x x ncreases Fg. c, ndcang ha he model can be updaed more rapdly by a change n he mean or covarance f k s large. In Fg. d, he races converge a,.945 snce k s se such ha α when x xnor. As n ncreases, α changes rapdly abou he pon,.945. In oher words, he model adapaon shows nor more rapd change resulng n greaer sensvy o x x abou uny. As a lmng case, α nor 9

11 becomes a bnary varable, α s equal o α max and α mn, as n becomes suffcenly large. hen, f x xnor >, α max, else, α s equal o α mn. Even hough hese parameers vary dependng on he process characerscs, here needs o be praccal gudance o selec hem, hus s proposed ha an emprcal parameer selecon procedure s adoped: Selec α max and α mn ypcally, α max.999 ~. 99 and α mn.95 ~. 9, Deermne k such ha α µ αmax αmn + αmn when x xnor, ndcang ha when he curren mean change s he same as he normal mean change.e. as per nomnal daa, α s se o a value beween α mn and α max. For example, α.5 αmax + αmn when µ.5 and α.67α α when µ.67. max mn 3 Selec n from beween and 3 by consderng he sensvy of α o x xnor. hs gudance s equally applcable n he deermnaon of β Robus Adapaon usng Oulyng Samples Oulyng observaons can be recorded durng he normal operaon of a process. here are wo sraeges when dealng wh oulers n adapve modellng, frs, oulers can be gnored and a model s held whou updang unl he nex sample or block s consdered, alernavely, a robus parameer esmaor can be used o updae he model usng he oulyng samples so ha he effec of oulyng samples on model updang can be suppressed. One common robus esmaor s he M-esmaor proposed by Huber. 7 hs approach has been used for regresson and dmensonaly reducon problems. 8,9 Mos robus mehods are used o denfy a relable model where a daa se conans oulers. herefore, modellng and ouler denfcaon are performed eravely unl a robus model s denfed snce he presence of oulers s unknown. However, n he adapve modellng approach, a relable PCA model wll have been developed n he prevous sep and hus may be possble o decde f a new sample s an ouler by examnng he monorng chars. herefore, he am s o denfy unusual measured varables n he new sample and correc for hem. Once an oulyng sample, x, s deeced by usng he monorng chars, s value s replaced wh a robus esmae, z, whch s hen used n he updang algorhm nsead of x o updae he curren PCA represenaon. Here, z s represened as a weghed x: z Wx 4

12 where W dag w, w, L, w s a wegh marx. he wegh of each varable s calculaed accordng m o s relably whch s a funcon of he varable resdual obaned usng he curren model: x ˆ k xk w + k 5 ck where c k s a unng parameer, whch conrols he sharpness of he wegh funcon he covarance of he resdual, and assumng Σ x I, k e a me pon,.e. Σ I P P Σ I P P I P P p c can be calculaed as k z pk e c α x w k. Consderng, where z α s a normal devae. As he varable resduals e k ncrease, he wegh w k decreases and hus he dfference beween he orgnal and correced values ncreases. Oulers may rgger a warnng sgnal n he monorng char n a smlar manner o ha of a faul. herefore here s a rsk ha a curren model adaps o a faul or a dsurbance as well as an ouler when a robus updang mehod s ulsed. Hence, s mporan o dscrmnae beween oulers and process fauls. Oulers are unlkely o be generaed consecuvely, whereas fauls and dsurbances las for a leas a ceran perod. In hs respec, f monorng chas show an ou-of-conrol sgnal for several consecuve samples, e.g., hree samples, a process s consdered o be abnormal. Oherwse, he curren model s updaed usng he robus updang mehod Overvew of he Monorng Procedure he overall sraegy for me-varyng monorng usng he proposed mehod s as follows: Off lne learnng: Oban he nal values of he sample mean m, he varance D, he loadng marx P and he egenvalue marx Λ based on he ranng daa block. Oban he wo conrol lms lm, and Q lm,, and he number of prncpal componens, p. 3 Selec α max, α mn, β and max β. max 4 Calculae mnor and Rnor. Dvde he ranng daa no wo secons and calculae he nal m, D, and R usng he frs ranng daa se. hen, wh a fxed α and β e.g. se boh o.96, whch corresponds o a value of α.67αmax +. 33α mn, when α max. 99 and α mn. 9, updae m, D, and R and sore m and R usng he second ranng daa se. Calculae he average values mnor and Rnor.

13 5 Deermne he model parameers: n and k. 6 Deermne he nal values of he wo forgeng facors α and β Here, boh are se o.99. In effec, hey are no very crcal n on-lne model adapaon. On lne learnng and monorng: A me pon, usng he prevous values of m, D, P, Λ, lm,, Qlm,, p Calculae m and and Q for a new sample x or block afer mean cenerng and auoscalng usng D. lm, If > or Q > Q lm, go o sep 3 else go o sep 6. 3 Check f he new sample s an ouler eher auomacally or manually. For nsance, f s > lm, s or Q s > Q lm, s s,.e. f hree consecuve ou-of conrol sgnals have been generaed, he new sample s no an ouler. Oherwse, s an oulyng sample. 4 If s an ouler, go o sep 5. Oherwse, consder he curren condon o be abnormal. ypcally he curren process condon canno be deermned by examnng one sample. In hs case, he model s reaned whou updang and he curren sample s sored. hen, f he process condon s proven o be normal subsequenly, he model can be updaed n a block-wse manner. 5 Calculae he robus esmaed value, z, of he new sample. 6 Recalculae 7 Calculae m and and Q based on z. D 8 Calculae P and Λ by performng SVD on K K or K K. 9 Fnd he number of prncpal componens o rean, p, and he conrol lms, and Q lm,. lm, 5. Applcaon Sudes 5.. Illusrave Examples Four examples are consdered o llusrae he updang procedures proposed n he paper. I should be noed ha he followng scenaros are assumed o be reflecve of normal process changes. he followng mulvarae sac process s consdered: x Ξ, where [,, L, ] R s a Gaussan random vecor and 3 Ξ R s a ransformaon marx. Here, ~ N,. and Ξ j ~ N,. herefore he measuremen vecor 3 x R follows a normal dsrbuon wh µ x and Σ x. ΞΞ. samples were generaed ha were reflecve of normal process operaon. Four

14 ypes of process change were smulaed, each comprsng samples. able summarses hese process changes and he smulaon condons mplemened o smulae hese changes. In hs sudy, a block-wse adapve PCA modellng approach was adoped wh a block sze of fve samples. ha s, he PCA model was updaed every fve me pons. For he forgeng facors o be adjusable, he followng parameers: α β. 99, α β. 9, k. 455 and n max max mn mn were seleced. y seng k.455, f he mean change s equal o he average values obaned from he ranng daa se, hen he forgeng facor akes he value.96. he frs sep was o develop he baselne model usng he off-lne learnng procedure descrbed prevously. he ranng samples were ulzed o denfy he nal PCA model. he frs samples were used o denfy he nal mean and covarance srucures and he remanng samples were used for he deermnaon of he parameers for he forgeng facor updang Eqs. and 3. Afer denfyng he model and he parameers, he procedure for on-lne monorng descrbed n Secon 4.4 was appled o he samples aken from he slowly drfng process. In he frs case sudy P, hree measured varables, x, x, and x, slowly drf away from he normal operang condons hereby smulang a slow process change such as caalys deacvaon. he slow drfs of magnude.,.5, and. are nroduced o he hree varables a me pons 3, 5, and 7, respecvely, and are connued unl he end of he smulaon. As shown n Fgs. a, b, and c, he change n he mean of varables x, x, and x s racked closely by he model updang procedure wh lmed me delay. Durng he smulaon, he number of prncpal componens s fxed a egh snce he rank of he covarance marx does no change over me Fg. d. he forgeng facor α for he sample mean m connues o decrease from me pon wh akng a consan low value of.9 from me pon 7 unl he end of he smulaon hence he model s updaed rapdly as a resul of he mean change n x, x, and x Fg. e. On he oher hand, he forgeng facor β flucuaes around.96 wh no dsnc changes beng evden snce he sample covarance marx does no change sgnfcanly durng hs perod Fg. f. Hoellng s and he Q-sasc for adapve PCA and sac PCA are compared n Fg. 3. he PCA model whou updang s no longer vald afer he process begns o change a me pon 3 and hus resuls n he process appearng o beng ou-of-conrol n boh monorng chars. In conras, he monorng chars for he adapve PCA model shows ha he process remans n-conrol afer he changes have been nroduced. 3

15 he second case sudy P reflecs wo dfferen ypes of changes n wo of he measured varables. oh of hem descrbe se-pon changes n he process by assumng ha he wo varables x and x are measured conrolled oupus. he se-pon of x changes from o + wh frs order dynamcs and a me consan of samplng me pons a me pon, whle ha of x changes from o wh frs order dynamcs and a me consan of 3 samplng me pons a me pon 7. Fgures 4a and b show he acual values and updaed means of varables x and x, respecvely wh Fg. 4c capurng he dfference beween he heorecal and esmaed mean. As me progresses afer a process change has occurred, he updaed sample mean approaches s heorecal value. Smlar o he frs case, he number of prncpal componens does no change snce he underlyng dmensonaly of he process s consan Fg. 4d. As llusraed n Fg. 4e and 4f, he value of he forgeng facor α decreases sharply a around me pons and 5 o updae he model accordng o he change n he mean values, whereas he value of he forgeng facor, β, flucuaes around.96 whou any sgnfcan change beng ncurred. As shown n Fg. 5, he Q-sasc for he adapve PCA model does no show any sgnfcan abnormal paerns represenng ou-of-conrol behavour unlke ha for he sac PCA model. I can be observed ha Hoellng s does no pck up he change snce he drecon of hs change s nearly orhogonal o he model subspace and hus he process change does no sgnfcanly ncrease he value of Hoellng s. In he hrd case sudy P3, a ramp sgnal s nroduced wh magnude. o he, elemens of he ransformaon marx from me pon unl he end of he smulaon,.e. Ξ Ξ +.. As expeced, he covarance of x changes wh me afer he change n he ransformaon marx has been nroduced. Fgures 6a and b represen he heorecal and updaed varance of x, respecvely, and Fgure 6c represens he dfference beween hem. As for he frs wo case sudes, he number of prncpal componens remans consan durng he smulaon as shown n Fg. 6d. he value of he forgeng facor, α, does no show any sgnfcan change snce he parameer change n Ξ does no affec he mean of he varables. In conras, he forgeng facor, β, s slghly lower han.96 hus he covarance marx s connuously updaed afer me pon Fgs. 6e and f. Even hough Hoellng s for adapve PCA shows several false alarms, whose frequency s sascally accepable, he adapve PCA model gves relable adapve monorng chars for boh Hoellng s and he Q-sasc, for he process showng he slow covarance change compared wh he resuls for he sac PCA model Fg. 7. he las sudy P4 smulaes a change n he underlyng dmensonaly of he process whch gves rse o a change n he number of prncpal componens reaned n he PCA model. A me pon 5, he dmensonaly of changes from o 4 and hus he rank of Ξ also ncreases by he order of four. 4

16 Fgs. 8a-c show he cumulave percen varance capured by he PCA model consruced based on he ranng daa se, he frs 5 es daa pons colleced before he process change, and he fnal 5 es daa pons aken afer he process change, respecvely. In each fgure, he number of black bars ndcaes he number of prncpal componens reaned, whch s seleced as ha when he cumulave varance aans a value larger han 9%. efore he process change, he number of prncpal componens for adapve PCA model s he same as ha for he PCA model obaned usng he ranng daa. However, afer he change n he underlyng dmensonaly of he process, he number of prncpal componens reaned ncreases by wo and hus becomes as observed n Fg. 8d. As ndcaed n Fg. 8f, β rapdly decreases o.93 a me pon 5 o updae he model as a resul of hs abrup process change. he monorng resuls wh adapve PCA are llusraed n Fgs. 9a and b and hose wh sac PCA are dsplayed n Fgs. 9c and d. he char for Hoellng s for adapve PCA has a smaller false alarm rae han ha for sac PCA. I can be observed ha he conrol lm n hs char ncreases as he number of prncpal componens reaned n he PCA model ncreases. he Q-sasc whou model updang produces ou-of-conrol sgnals jus afer me pon 5, whereas ha wh model updang s sll vald afer ha me excep ha here are some false alarms jus afer he process changes. hs s because he model updang procedure requres me unl suffcen samples are colleced from he newly changed process o descrbe he new condon. 5.. Smulaed CSR Process A nonsohermal CSR process, s consdered for he applcaon of he adapve process monorng algorhm. A schemac of he process s gven n Fg.. In he reacor, reacan A was premxed wh a E / R solven and hen was convered no produc wh rae r β k e C. he dynamc behavour of he process s descrbed by he mass balance of reacan A and he oal energy balance for he reacng sysem: r dc V F C C Vr 6 d d UA Vρ C p ρc p F c + H r Vr d + UA F ρ C 7 c c pc b where he hea ransfer coeffcen UA s emprcally represened as UA β UA af c. he concenraon of he reacan mxure s calculaed from C F C + F C / F + F. he oule emperaure and he a a s s a s oule concenraon C are conrolled by manpulang he nle coolan flow rae F c and he nle reacan flow rae F a, respecvely. he model parameers used n he CSR modellng and he con- 5

17 roller parameers of he wo PI conrollers are gven n able. All process npus and dsurbances are generaed usng frs order auoregressve models or snusodal sgnals. In addon, all measured varables are conamnaed wh whe Gaussan nose o descrbe measuremen nose able 3. o evaluae he performance of he new adapve mehod, wo knds of process change are consdered: a slow decrease n he reacon rae, and a se-pon change n he reacor emperaure. For each smulaon, he CSR process was run for 5 mn, wh he process change beng nroduced a me pon 3 mn. Nne process varables were measured every mnue able 3. he samples obaned for me pons - 3 mn were used o buld an nal PCA model and he subsequen samples 3-5 mn were used o updae he model on-lne. In he frs case, a slow drf n β r of magnude -3 mn - was nroduced a me pon 3 mn and connued unl he end of he smulaon. he decrease n he reacon rae resuls n a decrease n he reacan flow rae o manan he requred oule concenraon and he decrease n he coolan flow rae wll manan he oule emperaure. Fgs. a and b represen he acual values and her esmaed mean values, hrough he adapve mehod, of F a and F c, respecvely. he proposed adapve scheme was able o closely rack he me rajecory of he varables wh a me delay of 5 mn. Durng he drf, he forgeng facor α decreases from.96 o around.93 o ake accoun of he mean change n he process Fg. d. he process monorng mehods based on sac PCA and wo adapve PCA models wh and whou ouler correcon are compared n Fg.. he sac PCA model becomes nvald jus afer he process change occurs Fgs. a and d, whereas he adapve PCA model adaps he Hoellng s and Q sascs and her conrol lms, makng he curren model vald. here are several oulers caused by he measuremen error as depced n Fgs. b and e. he presence of oulers deeroraes he performance of he model updang procedure. y usng he proposed robus updang mehodology when an ouler occurs, he false alarm rae n he monorng chars could be sgnfcanly reduced, especally wh respec o he Q-sasc Fgs. c and f. Unlke he frs smulaon sudy, he second process change caused by he se-pon change s very fas and predcable. Snce he me of he process change s known, may be sraghforward o dscrmnae beween he alarms from he process change and hose from oulers or fauls. In hs case, he se-pon of he oule emperaure changed from 368 o 37 a me pon 3 mn. he sample mean values of F c and calculaed usng he recursve scheme are llusraed n Fgs. 3a and b. he number of prncpal componens for he adapve model s sx excep for several me pons where ncreases o seven Fg. 3c. As shown n Fg. 3d, he forgeng facor, α, drops rapdly from.96 o.9 jus afer he se-pon change and hen reurns o s normal value.96 afer me pon 35 because he esmaed mean values have converged o her new values. 6

18 he monorng resuls are shown n Fg. 4. For me pons 3 35 mn, he Q-sasc for all he mehods show ou-of-conrol sgnals due o he se-pon change. Wh he excepon of hs me perod, he adapve PCA model combned wh he ouler correcon mehod gave he mos relable monorng resuls Fg. 4f, makng possble o suppress he effec of oulers on he updang of he model and he conrol lms of he monorng sascs. 6. Conclusons A new adapve MSPC mehod for me-varyng process monorng has been proposed. ased on he newly defned weghed mean and covarance, recursve forms of he mean and covarance are defned n a sample-wse and block-wse manner. hen, by usng hese recursve formulas, an effcen mehod for fndng a PCA model recursvely s developed. In hs approach, a loadng marx s sored nsead of a full covarance marx for he nex model updae, whch means ha hs mehod has lower sorage requremens compared wh he prevous approaches descrbed n he leraure. Furhermore, several mporan ssues for adapve process monorng are addressed. Frs, wo dfferen varable forgeng facors for mean and covarance updang are consdered and her calculaon based on he recen mean or covarance change s proposed. Secondly, oulyng samples, whch are recorded que ofen n real processes, are pre-reaed by means of a robus esmaor such ha he model s nsensve o he oulers and hus s correcly updaed. ased on he overall process monorng sraegy presened n hs sudy, wo applcaon examples were consdered: a smple sac mulvarae sysem and a CSR process. he ably of he adapve model o keep rack of he changes n he varable mean and covarance was nvesgaed. Also, a change n he number of prncpal componens and he forgeng facors accordng o process changes were consdered. Chars of Hoellng s and he Q-sasc based on he adapve monorng sascs and her conrol lms were observed o be relable for he monorng of boh slowly and abruply changng processes. In parcular, he correcon for oulers usng he robus esmaon mehod resuls n no only reducng he number of false alarms sgnfcanly bu also n prevenng he curren model from beng updaed by oulers. Acknowledgemens Dr. Cho acknowledges he fnancal conrbuon of he Cenre for Process Analycs and Conrol echnology. References. Kresa, J.; MacGregor, J.F.; Marln,.E. Mulvarae sascal monorng of process operang performance. Canadan Journal of Chemcal Engneerng 99, 69,

19 . Wse,.M.; Gallagher, N.. he process chemomercs approach o process monorng and faul deecon. Journal of Process Conrol 996, 6, MacGregor, J.F.; Kour,. Sascal process conrol of mulvarae processes. Conrol Engneerng Pracce 995, 34, racy, N.D.; Young, J.C.; Mason, R.L. Mulvarae conrol chars for ndvdual observaons, Journal of Qualy echnology 99, 4, Jackson, J.E. A user s gude o prncpal componens; John Wley & Sons: New York, Mller, P.; Swanson, R.; Heckler, C. Conrbuon Plos: he mssng lnk n mulvarae qualy conrol, Fall Conf. of he ASQC and ASA, Mlwaukee, WI, Wold, S., Exponenally weghed movng prncpal componens analyss and projecons o laen srucures. Chemomercs and Inellgen Laboraory Sysems 994, 3, Dayal,.S.; MacGregor, J.F. Recursve exponenally weghed PLS and s applcaons o adapve conrol and predcon. Journal of Process Conrol 997, 73, Qn, S.J., Recursve PLS algorhms for adapve daa monorng. Compuers & Chemcal Engneerng 998,, L, W.; Yue, H.H.; Valle-Cervanes, S.; Qn, S.J. Recursve PCA for adapve process monorng. Journal of Process Conrol,, Felz, C.J.; Shau, J. J. H. Sascal process monorng usng an emprcal bayes mulvarae process conrol char. Qualy and Relably Engneerng Inernaonal, 7, Lu, X.; Chen,.; hornon, S.M. Egenspace updang for non saonary process and s applcaon o face recognon. Paern Recognon 3, 36, Hall, P.; Marshall, D.; Marn, R. Mergng and splng Egenspace models. IEEE ransacons on Paern Analyss and Machne Inellgence, 9, Jackson, J.E.; Mudholkar, G. Conrol procedures for resduals assocaed wh prncpal componen analyss. echnomercs 979,, Forescue,.R.; Kershenbaum, L.S.; Ydse,.E. Implemenaon of self unng regulaors wh varable forgeng facors. Auomaca 98, 7, Lane, S.; Marn, E..; Morrs, A.J.; Gower, P. Applcaon of exponenally weghed prncpal componen analyss for he monorng of a polymer flm manufacurng process. ransacons of he Insue of Measuremen and Conrol 3, 5, Huber, P.J. A robus esmaon of a locaon parameer. Ann. Mah. Sa. 964, 35, Huber, P.J. Robus sascs; John Wley & Sons: New York, Lang, Y.-Z.; Kvalhem, O.M. Robus mehods for mulvarae analyss - a uoral revew. Chemomercs and Inellgen Laboraory Sysems 996, 3,.. Yoon, S.; MacGregor, J.F. Faul dagnoss wh mulvarae sascal models par I: usng seady sae faul sgnaures. Journal of Process Conrol,,

20 . Cho, S. W.; Marn, E..; Morrs, A.J.; Lee, I.-. Faul deecon based on a maxmum lkelhood PCA mxure. Indusral and Engneerng Chemsry Research 5, n press. 9

21 Ls of Fgures Fgure Effec of funcon parameers on varable forgeng facor. a Effec of α max, b Effec Fgure of α mn, c Effec of k, and d Effec of p on α. he defaul values of he four parameers are α. max 99, α. 9 mn, k.693, and n. me-seres plos of varable mean and model parameers for case P. a Measured varables x, x, and x, b Esmaed mean mx, c Dfference beween heorecal and esmaed mean values µ x m x, d Number of PCs for adapve PCA, e Forgeng facor α, f Forgeng facor β. Fgure 3 Process monorng chars for case P. a Hoellng s for adapve PCA, b Q-sasc for adapve PCA, c Hoellng s for PCA, d Q-sasc for PCA. Fgure 4 me-seres plos of varable mean and model parameers for case P. a Measured varables x and x, b Esmaed mean mx, c Dfference beween heorecal and esmaed mean values µ x m x, d Number of PCs for adapve PCA, e Forgeng facor α, f Forgeng facor β. Fgure 5 Process monorng chars for case P. a Hoellng s for adapve PCA, b Q-sasc for adapve PCA, c Hoellng s for PCA, d Q-sasc for PCA. Fgure 6 me-seres plos of varable varance and model parameers for case P3. a heorecal sandard devaon of varables x, b Esmaed sandard devaon, c Dfference beween heorecal and esmaed sandard devaon, d Number of PCs for adapve PCA, e Forgeng facor α, f Forgeng facor β. Fgure 7 Process monorng chars for case P3. a Hoellng s for adapve PCA, b Q-sasc for adapve PCA, c Hoellng s for PCA, d Q-sasc for PCA. Fgure 8 Cumulave percen varance CPV capured by PCA model and he model parameers for case P4. a CPV capured by he PCA model for he ranng daa, b CPV capured by he PCA model for he es daa for me pons 5, c CPV capured by he PCA model for he es daa for me pons 5, d Number of PCs reaned for adapve PCA, e Forgeng facor α, f Forgeng facor β. Fgure 9 Process monorng chars for case P4. a Hoellng s for adapve PCA, b Q-sasc for adapve PCA, c Hoellng s for PCA, d Q-sasc for PCA. Fgure Nonsohermal CSR process and nne measured varables. Coolan emperaure, Reacan mxure emperaure, 3 Reacan A concenraon, 4 Solven concenraon, 5 Solven flow rae, 6 Solue flow rae, 7 Coolan flow rae, 8 Oule concenraon, and 9 Oule emperaure.

22 Fgure Fgure Fgure 3 Fgure 4 me-seres plos of varable mean and model parameers for he frs case sudy reacon rae decrease n he CSR process. a Measured value of F a and s esmaed mean, b Measured value of F c and s esmaed mean, c Number of PCs for adapve PCA, d Forgeng facors. Process monorng chars for PCA, adapve PCA, and robus adapve PCA. Hoellng s for a PCA, b Adapve PCA, c Robus adapve PCA; Q-sasc for d PCA, e Adapve PCA, f Robus adapve PCA. me-seres plos of varable mean and model parameers for he second case sudy se-pon change n he CSR process. a Measured value of F c and s esmaed mean, b Measured value of and s esmaed mean, c Number of PCs for adapve PCA, d Forgeng facors. Process monorng chars for PCA, adapve PCA, and robus adapve PCA. Hoellng s for a PCA, b Adapve PCA, c Robus adapve PCA; Q-sasc for d PCA, e Adapve PCA, f Robus adapve PCA.

23 forgeng facor a ncreasng α max forgeng facor b ncreasng α mn x / x nor x / x nor forgeng facor c ncreasng k forgeng facor d ncreasng p x / x nor x / x nor Fg.. Effec of funcon parameers on varable forgeng facor. a Effec of α max, b Effec of α mn, c Effec of k, and d Effec of p on α. he defaul values of he four parameers are α. max 99, α.9 mn, k.693, and n.

24 x µx - mx mx b mx 5 mx mx c a x x x µx -mx - µx -mx µx - -mx Number of PCs α β d e f Fg.. me-seres plos of varable mean and model parameers for case P a Measured varables x, x, and x, b Esmaed mean mx, c Dfference beween heorecal and esmaed mean values µ x m x, d Number of PCs for adapve PCA, e Forgeng facor α, f Forgeng facor β. 3

25 a 3 Adapve PCA c PCA b 5 Adapve PCA d PCA 5 Q Q Fg. 3. Process monorng chars for case P. a Hoellng s for adapve PCA, b Q-sasc for adapve PCA, c Hoellng s for PCA, d Q-sasc for PCA. 4

26 x mx µx - mx a x x b mx mx c µx -mx µx -mx Number of PCs α β d e f Fg. 4. me-seres plos of varable mean and model parameers for case P. a Measured varables x and x, b Esmaed mean mx, c Dfference beween heorecal and esmaed mean values µ x m x, d Number of PCs for adapve PCA, e Forgeng facor α, f Forgeng facor β. 5

27 a 5 Adapve PCA c 3 PCA b Adapve PCA d 8 PCA 6 Q Q Fg. 5. Process monorng chars for case P. a Hoellng s for adapve PCA, b Q-sasc for adapve PCA, c Hoellng s for PCA, d Q-sasc for PCA. 6

28 σx sx σx -sx b c.6.4. a Number of PCs α β d e f Fg. 6. me-seres plos of varable varance and model parameers for case P3. a heorecal sandard devaon of varables x σ x, b Esmaed sandard devaon s x, c Dfference beween heorecal and esmaed sandard devaon σ x s, d Number of PCs for adapve PCA, e Forgeng facor α, f Forgeng facor β. x 7

29 a 4 Adapve PCA 3 c PCA b Adapve PCA d 4 PCA 5 3 Q Q Fg. 7. Process monorng chars for case P3. a Hoellng s for adapve PCA, b Q-sasc for adapve PCA, c Hoellng s for PCA, d Q-sasc for PCA. 8

30 Fg. 8. Cumulave percen varance CPV capured by PCA model and he model parameers for case P4. a CPV capured by he PCA model for he ranng daa, b CPV capured by he PCA model for he es daa for me pons 5, c CPV capured by he PCA model for he es daa for me pons 5, d Number of PCs reaned for adapve PCA, e Forgeng facor α, f Forgeng facor β. 9

31 4 a Adapve PCA 4 c PCA b Adapve PCA 5 d PCA Q Q Fg. 9. Process monorng chars for case P4. a Hoellng s for adapve PCA, b Q-sasc for adapve PCA, c Hoellng s for PCA, d Q-sasc for PCA. 3

32 CC 3 C a 6 F a 8 C 9 4 C s 5 F s C c 7 F c Fg.. Nonsohermal CSR process and nne measured varables. Coolan emperaure, Reacan mxure emperaure, 3 Reacan A concenraon, 4 Solven concenraon, 5 Solven flow rae, 6 Solue flow rae, 7 Coolan flow rae, 8 Oule concenraon, and 9 Oule emperaure. 3

33 F a a F a mf a b Number of PCs c d F c 5 F C mf C Forgeng facor α β Fg.. me-seres plos of varable mean and model parameers for he frs case sudy reacon rae decrease n he CSR process. a Measured value of F a and s esmaed mean, b Measured value of F c and s esmaed mean, c Number of PCs for adapve PCA, d Forgeng facors. 3

34 a PCA d PCA 5 Q b APCA e APCA Q c Robus APCA f Robus APCA Q Fg.. Process monorng chars for PCA, adapve PCA, and robus adapve PCA. Hoellng s for a PCA, b Adapve PCA, c Robus adapve PCA; Q-sasc for d PCA, e Adapve PCA, f Robus adapve PCA. 33

35 F c 5 F C mf C Number of PCs m Forgeng facor α β Fg. 3. me-seres plos of varable mean and model parameers n he second case sudy se-pon change n he CSR process. a Measured value of F c and s esmaed mean, b Measured value of and s esmaed mean, c Number of PCs for adapve PCA, d Forgeng facors. 34

36 PCA 4 PCA Q APCA APCA Q Robus APCA Robus APCA Q Fg. 4. Process monorng chars for PCA, adapve PCA, and robus adapve PCA. Hoellng s for a PCA, b Adapve PCA, c Robus adapve PCA; Q-sasc for d PCA, e Adapve PCA, f Robus adapve PCA. 35

37 Ls of ables able able able 3 Four ypes of process changes Model parameers used n he CSR modellng and he PI conrol parameers Paerns of process npus and dsurbances and measuremen nose 36

38 Paerns * P x x +. 3 > 3 * x x.5 5 > 5 Scenaro Change n he mean unl he end of smulaon me * x x +. 7 > 7 * P x x + exp > 3 Change n he mean * x x exp 7 3 > 7 P3 Ξ Ξ +. > Change n he covarance P4 Ξ R Ξ R > 5 Change n he number of PCs able : Four ypes of process changes nroduced no he smulaon sudy 37

39 Noaon Parameers and Consans Value V Volume of reacon mxure n he ank m 3 ρ Densy of reacon mxure 6 g/m 3 ρ c Densy of coolan 6 g/m 3 C p Specfc hea capacy of he reacon mxure cal/gk C pc Specfc hea capacy of he coolan cal/gk H r Hea of reacon cal/kmol k Pre-exponenal knec consan mn - E/R Acvaon energy / Ideal gas consan 833 K K c emperaure conroller gan -.5 τ I Inegral me 5 K c C Concenraon conroller gan.485 τ I C Inegral me able : Model parameers used n he CSR modellng and he PI conrol parameers. 38

40 Inpu and Dsurbance Measuremens Varable Mean φ σ v Varable F s.9 m 3 /mn..9 - c K c 365 K C a. - β r C s.5-5 β UA F s Varable Mean a b σ e 6 F a σ v 7 F c. - C a 9. kmol/m C.5-5 C s.3 kmol/m able 3. Paerns of process npus and dsurbances and measuremen nose C a and C s are modelled as x a sn b + v and he remanng npus and dsurbances are modelled as x φx + v, where he process nose v ~ N, σ v nose e ~ N, σ. e. All measured varables are conamnaed wh whe Gaussan 39

Solution in semi infinite diffusion couples (error function analysis)

Solution in semi infinite diffusion couples (error function analysis) Soluon n sem nfne dffuson couples (error funcon analyss) Le us consder now he sem nfne dffuson couple of wo blocks wh concenraon of and I means ha, n a A- bnary sysem, s bondng beween wo blocks made of

More information

Robustness Experiments with Two Variance Components

Robustness Experiments with Two Variance Components Naonal Insue of Sandards and Technology (NIST) Informaon Technology Laboraory (ITL) Sascal Engneerng Dvson (SED) Robusness Expermens wh Two Varance Componens by Ana Ivelsse Avlés avles@ns.gov Conference

More information

Variants of Pegasos. December 11, 2009

Variants of Pegasos. December 11, 2009 Inroducon Varans of Pegasos SooWoong Ryu bshboy@sanford.edu December, 009 Youngsoo Cho yc344@sanford.edu Developng a new SVM algorhm s ongong research opc. Among many exng SVM algorhms, we wll focus on

More information

John Geweke a and Gianni Amisano b a Departments of Economics and Statistics, University of Iowa, USA b European Central Bank, Frankfurt, Germany

John Geweke a and Gianni Amisano b a Departments of Economics and Statistics, University of Iowa, USA b European Central Bank, Frankfurt, Germany Herarchcal Markov Normal Mxure models wh Applcaons o Fnancal Asse Reurns Appendx: Proofs of Theorems and Condonal Poseror Dsrbuons John Geweke a and Gann Amsano b a Deparmens of Economcs and Sascs, Unversy

More information

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 4

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 4 CS434a/54a: Paern Recognon Prof. Olga Veksler Lecure 4 Oulne Normal Random Varable Properes Dscrmnan funcons Why Normal Random Varables? Analycally racable Works well when observaon comes form a corruped

More information

V.Abramov - FURTHER ANALYSIS OF CONFIDENCE INTERVALS FOR LARGE CLIENT/SERVER COMPUTER NETWORKS

V.Abramov - FURTHER ANALYSIS OF CONFIDENCE INTERVALS FOR LARGE CLIENT/SERVER COMPUTER NETWORKS R&RATA # Vol.) 8, March FURTHER AALYSIS OF COFIDECE ITERVALS FOR LARGE CLIET/SERVER COMPUTER ETWORKS Vyacheslav Abramov School of Mahemacal Scences, Monash Unversy, Buldng 8, Level 4, Clayon Campus, Wellngon

More information

Lecture 6: Learning for Control (Generalised Linear Regression)

Lecture 6: Learning for Control (Generalised Linear Regression) Lecure 6: Learnng for Conrol (Generalsed Lnear Regresson) Conens: Lnear Mehods for Regresson Leas Squares, Gauss Markov heorem Recursve Leas Squares Lecure 6: RLSC - Prof. Sehu Vjayakumar Lnear Regresson

More information

New M-Estimator Objective Function. in Simultaneous Equations Model. (A Comparative Study)

New M-Estimator Objective Function. in Simultaneous Equations Model. (A Comparative Study) Inernaonal Mahemacal Forum, Vol. 8, 3, no., 7 - HIKARI Ld, www.m-hkar.com hp://dx.do.org/.988/mf.3.3488 New M-Esmaor Objecve Funcon n Smulaneous Equaons Model (A Comparave Sudy) Ahmed H. Youssef Professor

More information

Lecture VI Regression

Lecture VI Regression Lecure VI Regresson (Lnear Mehods for Regresson) Conens: Lnear Mehods for Regresson Leas Squares, Gauss Markov heorem Recursve Leas Squares Lecure VI: MLSC - Dr. Sehu Vjayakumar Lnear Regresson Model M

More information

5th International Conference on Advanced Design and Manufacturing Engineering (ICADME 2015)

5th International Conference on Advanced Design and Manufacturing Engineering (ICADME 2015) 5h Inernaonal onference on Advanced Desgn and Manufacurng Engneerng (IADME 5 The Falure Rae Expermenal Sudy of Specal N Machne Tool hunshan He, a, *, La Pan,b and Bng Hu 3,c,,3 ollege of Mechancal and

More information

Linear Response Theory: The connection between QFT and experiments

Linear Response Theory: The connection between QFT and experiments Phys540.nb 39 3 Lnear Response Theory: The connecon beween QFT and expermens 3.1. Basc conceps and deas Q: ow do we measure he conducvy of a meal? A: we frs nroduce a weak elecrc feld E, and hen measure

More information

On One Analytic Method of. Constructing Program Controls

On One Analytic Method of. Constructing Program Controls Appled Mahemacal Scences, Vol. 9, 05, no. 8, 409-407 HIKARI Ld, www.m-hkar.com hp://dx.do.org/0.988/ams.05.54349 On One Analyc Mehod of Consrucng Program Conrols A. N. Kvko, S. V. Chsyakov and Yu. E. Balyna

More information

Clustering (Bishop ch 9)

Clustering (Bishop ch 9) Cluserng (Bshop ch 9) Reference: Daa Mnng by Margare Dunham (a slde source) 1 Cluserng Cluserng s unsupervsed learnng, here are no class labels Wan o fnd groups of smlar nsances Ofen use a dsance measure

More information

( t) Outline of program: BGC1: Survival and event history analysis Oslo, March-May Recapitulation. The additive regression model

( t) Outline of program: BGC1: Survival and event history analysis Oslo, March-May Recapitulation. The additive regression model BGC1: Survval and even hsory analyss Oslo, March-May 212 Monday May 7h and Tuesday May 8h The addve regresson model Ørnulf Borgan Deparmen of Mahemacs Unversy of Oslo Oulne of program: Recapulaon Counng

More information

Time-interval analysis of β decay. V. Horvat and J. C. Hardy

Time-interval analysis of β decay. V. Horvat and J. C. Hardy Tme-nerval analyss of β decay V. Horva and J. C. Hardy Work on he even analyss of β decay [1] connued and resuled n he developmen of a novel mehod of bea-decay me-nerval analyss ha produces hghly accurae

More information

Outline. Probabilistic Model Learning. Probabilistic Model Learning. Probabilistic Model for Time-series Data: Hidden Markov Model

Outline. Probabilistic Model Learning. Probabilistic Model Learning. Probabilistic Model for Time-series Data: Hidden Markov Model Probablsc Model for Tme-seres Daa: Hdden Markov Model Hrosh Mamsuka Bonformacs Cener Kyoo Unversy Oulne Three Problems for probablsc models n machne learnng. Compung lkelhood 2. Learnng 3. Parsng (predcon

More information

Ordinary Differential Equations in Neuroscience with Matlab examples. Aim 1- Gain understanding of how to set up and solve ODE s

Ordinary Differential Equations in Neuroscience with Matlab examples. Aim 1- Gain understanding of how to set up and solve ODE s Ordnary Dfferenal Equaons n Neuroscence wh Malab eamples. Am - Gan undersandng of how o se up and solve ODE s Am Undersand how o se up an solve a smple eample of he Hebb rule n D Our goal a end of class

More information

Robust and Accurate Cancer Classification with Gene Expression Profiling

Robust and Accurate Cancer Classification with Gene Expression Profiling Robus and Accurae Cancer Classfcaon wh Gene Expresson Proflng (Compuaonal ysems Bology, 2005) Auhor: Hafeng L, Keshu Zhang, ao Jang Oulne Background LDA (lnear dscrmnan analyss) and small sample sze problem

More information

Department of Economics University of Toronto

Department of Economics University of Toronto Deparmen of Economcs Unversy of Torono ECO408F M.A. Economercs Lecure Noes on Heeroskedascy Heeroskedascy o Ths lecure nvolves lookng a modfcaons we need o make o deal wh he regresson model when some of

More information

Machine Learning 2nd Edition

Machine Learning 2nd Edition INTRODUCTION TO Lecure Sldes for Machne Learnng nd Edon ETHEM ALPAYDIN, modfed by Leonardo Bobadlla and some pars from hp://www.cs.au.ac.l/~aparzn/machnelearnng/ The MIT Press, 00 alpaydn@boun.edu.r hp://www.cmpe.boun.edu.r/~ehem/mle

More information

Econ107 Applied Econometrics Topic 5: Specification: Choosing Independent Variables (Studenmund, Chapter 6)

Econ107 Applied Econometrics Topic 5: Specification: Choosing Independent Variables (Studenmund, Chapter 6) Econ7 Appled Economercs Topc 5: Specfcaon: Choosng Independen Varables (Sudenmund, Chaper 6 Specfcaon errors ha we wll deal wh: wrong ndependen varable; wrong funconal form. Ths lecure deals wh wrong ndependen

More information

Bayes rule for a classification problem INF Discriminant functions for the normal density. Euclidean distance. Mahalanobis distance

Bayes rule for a classification problem INF Discriminant functions for the normal density. Euclidean distance. Mahalanobis distance INF 43 3.. Repeon Anne Solberg (anne@f.uo.no Bayes rule for a classfcaon problem Suppose we have J, =,...J classes. s he class label for a pxel, and x s he observed feaure vecor. We can use Bayes rule

More information

Robustness of DEWMA versus EWMA Control Charts to Non-Normal Processes

Robustness of DEWMA versus EWMA Control Charts to Non-Normal Processes Journal of Modern Appled Sascal Mehods Volume Issue Arcle 8 5--3 Robusness of D versus Conrol Chars o Non- Processes Saad Saeed Alkahan Performance Measuremen Cener of Governmen Agences, Insue of Publc

More information

Polymerization Technology Laboratory Course

Polymerization Technology Laboratory Course Prakkum Polymer Scence/Polymersaonsechnk Versuch Resdence Tme Dsrbuon Polymerzaon Technology Laboraory Course Resdence Tme Dsrbuon of Chemcal Reacors If molecules or elemens of a flud are akng dfferen

More information

12d Model. Civil and Surveying Software. Drainage Analysis Module Detention/Retention Basins. Owen Thornton BE (Mech), 12d Model Programmer

12d Model. Civil and Surveying Software. Drainage Analysis Module Detention/Retention Basins. Owen Thornton BE (Mech), 12d Model Programmer d Model Cvl and Surveyng Soware Dranage Analyss Module Deenon/Reenon Basns Owen Thornon BE (Mech), d Model Programmer owen.hornon@d.com 4 January 007 Revsed: 04 Aprl 007 9 February 008 (8Cp) Ths documen

More information

FTCS Solution to the Heat Equation

FTCS Solution to the Heat Equation FTCS Soluon o he Hea Equaon ME 448/548 Noes Gerald Reckenwald Porland Sae Unversy Deparmen of Mechancal Engneerng gerry@pdxedu ME 448/548: FTCS Soluon o he Hea Equaon Overvew Use he forward fne d erence

More information

Fall 2010 Graduate Course on Dynamic Learning

Fall 2010 Graduate Course on Dynamic Learning Fall 200 Graduae Course on Dynamc Learnng Chaper 4: Parcle Flers Sepember 27, 200 Byoung-Tak Zhang School of Compuer Scence and Engneerng & Cognve Scence and Bran Scence Programs Seoul aonal Unversy hp://b.snu.ac.kr/~bzhang/

More information

RELATIONSHIP BETWEEN VOLATILITY AND TRADING VOLUME: THE CASE OF HSI STOCK RETURNS DATA

RELATIONSHIP BETWEEN VOLATILITY AND TRADING VOLUME: THE CASE OF HSI STOCK RETURNS DATA RELATIONSHIP BETWEEN VOLATILITY AND TRADING VOLUME: THE CASE OF HSI STOCK RETURNS DATA Mchaela Chocholaá Unversy of Economcs Braslava, Slovaka Inroducon (1) one of he characersc feaures of sock reurns

More information

Machine Learning Linear Regression

Machine Learning Linear Regression Machne Learnng Lnear Regresson Lesson 3 Lnear Regresson Bascs of Regresson Leas Squares esmaon Polynomal Regresson Bass funcons Regresson model Regularzed Regresson Sascal Regresson Mamum Lkelhood (ML)

More information

An introduction to Support Vector Machine

An introduction to Support Vector Machine An nroducon o Suppor Vecor Machne 報告者 : 黃立德 References: Smon Haykn, "Neural Neworks: a comprehensve foundaon, second edon, 999, Chaper 2,6 Nello Chrsann, John Shawe-Tayer, An Inroducon o Suppor Vecor Machnes,

More information

Comb Filters. Comb Filters

Comb Filters. Comb Filters The smple flers dscussed so far are characered eher by a sngle passband and/or a sngle sopband There are applcaons where flers wh mulple passbands and sopbands are requred Thecomb fler s an example of

More information

TSS = SST + SSE An orthogonal partition of the total SS

TSS = SST + SSE An orthogonal partition of the total SS ANOVA: Topc 4. Orhogonal conrass [ST&D p. 183] H 0 : µ 1 = µ =... = µ H 1 : The mean of a leas one reamen group s dfferen To es hs hypohess, a basc ANOVA allocaes he varaon among reamen means (SST) equally

More information

( ) () we define the interaction representation by the unitary transformation () = ()

( ) () we define the interaction representation by the unitary transformation () = () Hgher Order Perurbaon Theory Mchael Fowler 3/7/6 The neracon Represenaon Recall ha n he frs par of hs course sequence, we dscussed he chrödnger and Hesenberg represenaons of quanum mechancs here n he chrödnger

More information

January Examinations 2012

January Examinations 2012 Page of 5 EC79 January Examnaons No. of Pages: 5 No. of Quesons: 8 Subjec ECONOMICS (POSTGRADUATE) Tle of Paper EC79 QUANTITATIVE METHODS FOR BUSINESS AND FINANCE Tme Allowed Two Hours ( hours) Insrucons

More information

[Link to MIT-Lab 6P.1 goes here.] After completing the lab, fill in the following blanks: Numerical. Simulation s Calculations

[Link to MIT-Lab 6P.1 goes here.] After completing the lab, fill in the following blanks: Numerical. Simulation s Calculations Chaper 6: Ordnary Leas Squares Esmaon Procedure he Properes Chaper 6 Oulne Cln s Assgnmen: Assess he Effec of Sudyng on Quz Scores Revew o Regresson Model o Ordnary Leas Squares () Esmaon Procedure o he

More information

Dynamic Team Decision Theory. EECS 558 Project Shrutivandana Sharma and David Shuman December 10, 2005

Dynamic Team Decision Theory. EECS 558 Project Shrutivandana Sharma and David Shuman December 10, 2005 Dynamc Team Decson Theory EECS 558 Proec Shruvandana Sharma and Davd Shuman December 0, 005 Oulne Inroducon o Team Decson Theory Decomposon of he Dynamc Team Decson Problem Equvalence of Sac and Dynamc

More information

. The geometric multiplicity is dim[ker( λi. number of linearly independent eigenvectors associated with this eigenvalue.

. The geometric multiplicity is dim[ker( λi. number of linearly independent eigenvectors associated with this eigenvalue. Lnear Algebra Lecure # Noes We connue wh he dscusson of egenvalues, egenvecors, and dagonalzably of marces We wan o know, n parcular wha condons wll assure ha a marx can be dagonalzed and wha he obsrucons

More information

Additive Outliers (AO) and Innovative Outliers (IO) in GARCH (1, 1) Processes

Additive Outliers (AO) and Innovative Outliers (IO) in GARCH (1, 1) Processes Addve Oulers (AO) and Innovave Oulers (IO) n GARCH (, ) Processes MOHAMMAD SAID ZAINOL, SITI MERIAM ZAHARI, KAMARULZAMMAN IBRAHIM AZAMI ZAHARIM, K. SOPIAN Cener of Sudes for Decson Scences, FSKM, Unvers

More information

. The geometric multiplicity is dim[ker( λi. A )], i.e. the number of linearly independent eigenvectors associated with this eigenvalue.

. The geometric multiplicity is dim[ker( λi. A )], i.e. the number of linearly independent eigenvectors associated with this eigenvalue. Mah E-b Lecure #0 Noes We connue wh he dscusson of egenvalues, egenvecors, and dagonalzably of marces We wan o know, n parcular wha condons wll assure ha a marx can be dagonalzed and wha he obsrucons are

More information

Bayesian Inference of the GARCH model with Rational Errors

Bayesian Inference of the GARCH model with Rational Errors 0 Inernaonal Conference on Economcs, Busness and Markeng Managemen IPEDR vol.9 (0) (0) IACSIT Press, Sngapore Bayesan Inference of he GARCH model wh Raonal Errors Tesuya Takash + and Tng Tng Chen Hroshma

More information

Bernoulli process with 282 ky periodicity is detected in the R-N reversals of the earth s magnetic field

Bernoulli process with 282 ky periodicity is detected in the R-N reversals of the earth s magnetic field Submed o: Suden Essay Awards n Magnecs Bernoull process wh 8 ky perodcy s deeced n he R-N reversals of he earh s magnec feld Jozsef Gara Deparmen of Earh Scences Florda Inernaonal Unversy Unversy Park,

More information

[ ] 2. [ ]3 + (Δx i + Δx i 1 ) / 2. Δx i-1 Δx i Δx i+1. TPG4160 Reservoir Simulation 2018 Lecture note 3. page 1 of 5

[ ] 2. [ ]3 + (Δx i + Δx i 1 ) / 2. Δx i-1 Δx i Δx i+1. TPG4160 Reservoir Simulation 2018 Lecture note 3. page 1 of 5 TPG460 Reservor Smulaon 08 page of 5 DISCRETIZATIO OF THE FOW EQUATIOS As we already have seen, fne dfference appromaons of he paral dervaves appearng n he flow equaons may be obaned from Taylor seres

More information

GENERATING CERTAIN QUINTIC IRREDUCIBLE POLYNOMIALS OVER FINITE FIELDS. Youngwoo Ahn and Kitae Kim

GENERATING CERTAIN QUINTIC IRREDUCIBLE POLYNOMIALS OVER FINITE FIELDS. Youngwoo Ahn and Kitae Kim Korean J. Mah. 19 (2011), No. 3, pp. 263 272 GENERATING CERTAIN QUINTIC IRREDUCIBLE POLYNOMIALS OVER FINITE FIELDS Youngwoo Ahn and Kae Km Absrac. In he paper [1], an explc correspondence beween ceran

More information

In the complete model, these slopes are ANALYSIS OF VARIANCE FOR THE COMPLETE TWO-WAY MODEL. (! i+1 -! i ) + [(!") i+1,q - [(!

In the complete model, these slopes are ANALYSIS OF VARIANCE FOR THE COMPLETE TWO-WAY MODEL. (! i+1 -! i ) + [(!) i+1,q - [(! ANALYSIS OF VARIANCE FOR THE COMPLETE TWO-WAY MODEL The frs hng o es n wo-way ANOVA: Is here neracon? "No neracon" means: The man effecs model would f. Ths n urn means: In he neracon plo (wh A on he horzonal

More information

Computing Relevance, Similarity: The Vector Space Model

Computing Relevance, Similarity: The Vector Space Model Compung Relevance, Smlary: The Vecor Space Model Based on Larson and Hears s sldes a UC-Bereley hp://.sms.bereley.edu/courses/s0/f00/ aabase Managemen Sysems, R. Ramarshnan ocumen Vecors v ocumens are

More information

Let s treat the problem of the response of a system to an applied external force. Again,

Let s treat the problem of the response of a system to an applied external force. Again, Page 33 QUANTUM LNEAR RESPONSE FUNCTON Le s rea he problem of he response of a sysem o an appled exernal force. Agan, H() H f () A H + V () Exernal agen acng on nernal varable Hamlonan for equlbrum sysem

More information

Li An-Ping. Beijing , P.R.China

Li An-Ping. Beijing , P.R.China A New Type of Cpher: DICING_csb L An-Png Bejng 100085, P.R.Chna apl0001@sna.com Absrac: In hs paper, we wll propose a new ype of cpher named DICING_csb, whch s derved from our prevous sream cpher DICING.

More information

Video-Based Face Recognition Using Adaptive Hidden Markov Models

Video-Based Face Recognition Using Adaptive Hidden Markov Models Vdeo-Based Face Recognon Usng Adapve Hdden Markov Models Xaomng Lu and suhan Chen Elecrcal and Compuer Engneerng, Carnege Mellon Unversy, Psburgh, PA, 523, U.S.A. xaomng@andrew.cmu.edu suhan@cmu.edu Absrac

More information

Should Exact Index Numbers have Standard Errors? Theory and Application to Asian Growth

Should Exact Index Numbers have Standard Errors? Theory and Application to Asian Growth Should Exac Index umbers have Sandard Errors? Theory and Applcaon o Asan Growh Rober C. Feensra Marshall B. Rensdorf ovember 003 Proof of Proposon APPEDIX () Frs, we wll derve he convenonal Sao-Vara prce

More information

Learning Objectives. Self Organization Map. Hamming Distance(1/5) Introduction. Hamming Distance(3/5) Hamming Distance(2/5) 15/04/2015

Learning Objectives. Self Organization Map. Hamming Distance(1/5) Introduction. Hamming Distance(3/5) Hamming Distance(2/5) 15/04/2015 /4/ Learnng Objecves Self Organzaon Map Learnng whou Exaples. Inroducon. MAXNET 3. Cluserng 4. Feaure Map. Self-organzng Feaure Map 6. Concluson 38 Inroducon. Learnng whou exaples. Daa are npu o he syse

More information

Fault Diagnosis in Industrial Processes Using Principal Component Analysis and Hidden Markov Model

Fault Diagnosis in Industrial Processes Using Principal Component Analysis and Hidden Markov Model Faul Dagnoss n Indusral Processes Usng Prncpal Componen Analyss and Hdden Markov Model Shaoyuan Zhou, Janmng Zhang, and Shuqng Wang Absrac An approach combnng hdden Markov model (HMM) wh prncpal componen

More information

CHAPTER 10: LINEAR DISCRIMINATION

CHAPTER 10: LINEAR DISCRIMINATION CHAPER : LINEAR DISCRIMINAION Dscrmnan-based Classfcaon 3 In classfcaon h K classes (C,C,, C k ) We defned dscrmnan funcon g j (), j=,,,k hen gven an es eample, e chose (predced) s class label as C f g

More information

Panel Data Regression Models

Panel Data Regression Models Panel Daa Regresson Models Wha s Panel Daa? () Mulple dmensoned Dmensons, e.g., cross-secon and me node-o-node (c) Pongsa Pornchawseskul, Faculy of Economcs, Chulalongkorn Unversy (c) Pongsa Pornchawseskul,

More information

Comparison of Supervised & Unsupervised Learning in βs Estimation between Stocks and the S&P500

Comparison of Supervised & Unsupervised Learning in βs Estimation between Stocks and the S&P500 Comparson of Supervsed & Unsupervsed Learnng n βs Esmaon beween Socks and he S&P500 J. We, Y. Hassd, J. Edery, A. Becker, Sanford Unversy T I. INTRODUCTION HE goal of our proec s o analyze he relaonshps

More information

Chapter 6: AC Circuits

Chapter 6: AC Circuits Chaper 6: AC Crcus Chaper 6: Oulne Phasors and he AC Seady Sae AC Crcus A sable, lnear crcu operang n he seady sae wh snusodal excaon (.e., snusodal seady sae. Complee response forced response naural response.

More information

HEAT CONDUCTION PROBLEM IN A TWO-LAYERED HOLLOW CYLINDER BY USING THE GREEN S FUNCTION METHOD

HEAT CONDUCTION PROBLEM IN A TWO-LAYERED HOLLOW CYLINDER BY USING THE GREEN S FUNCTION METHOD Journal of Appled Mahemacs and Compuaonal Mechancs 3, (), 45-5 HEAT CONDUCTION PROBLEM IN A TWO-LAYERED HOLLOW CYLINDER BY USING THE GREEN S FUNCTION METHOD Sansław Kukla, Urszula Sedlecka Insue of Mahemacs,

More information

Forecasting customer behaviour in a multi-service financial organisation: a profitability perspective

Forecasting customer behaviour in a multi-service financial organisation: a profitability perspective Forecasng cusomer behavour n a mul-servce fnancal organsaon: a profably perspecve A. Audzeyeva, Unversy of Leeds & Naonal Ausrala Group Europe, UK B. Summers, Unversy of Leeds, UK K.R. Schenk-Hoppé, Unversy

More information

Relative controllability of nonlinear systems with delays in control

Relative controllability of nonlinear systems with delays in control Relave conrollably o nonlnear sysems wh delays n conrol Jerzy Klamka Insue o Conrol Engneerng, Slesan Techncal Unversy, 44- Glwce, Poland. phone/ax : 48 32 37227, {jklamka}@a.polsl.glwce.pl Keywor: Conrollably.

More information

Math 128b Project. Jude Yuen

Math 128b Project. Jude Yuen Mah 8b Proec Jude Yuen . Inroducon Le { Z } be a sequence of observed ndependen vecor varables. If he elemens of Z have a on normal dsrbuon hen { Z } has a mean vecor Z and a varancecovarance marx z. Geomercally

More information

Density Matrix Description of NMR BCMB/CHEM 8190

Density Matrix Description of NMR BCMB/CHEM 8190 Densy Marx Descrpon of NMR BCMBCHEM 89 Operaors n Marx Noaon Alernae approach o second order specra: ask abou x magnezaon nsead of energes and ranson probables. If we say wh one bass se, properes vary

More information

Single-loop System Reliability-Based Design & Topology Optimization (SRBDO/SRBTO): A Matrix-based System Reliability (MSR) Method

Single-loop System Reliability-Based Design & Topology Optimization (SRBDO/SRBTO): A Matrix-based System Reliability (MSR) Method 10 h US Naonal Congress on Compuaonal Mechancs Columbus, Oho 16-19, 2009 Sngle-loop Sysem Relably-Based Desgn & Topology Opmzaon (SRBDO/SRBTO): A Marx-based Sysem Relably (MSR) Mehod Tam Nguyen, Junho

More information

Sampling Procedure of the Sum of two Binary Markov Process Realizations

Sampling Procedure of the Sum of two Binary Markov Process Realizations Samplng Procedure of he Sum of wo Bnary Markov Process Realzaons YURY GORITSKIY Dep. of Mahemacal Modelng of Moscow Power Insue (Techncal Unversy), Moscow, RUSSIA, E-mal: gorsky@yandex.ru VLADIMIR KAZAKOV

More information

Graduate Macroeconomics 2 Problem set 5. - Solutions

Graduate Macroeconomics 2 Problem set 5. - Solutions Graduae Macroeconomcs 2 Problem se. - Soluons Queson 1 To answer hs queson we need he frms frs order condons and he equaon ha deermnes he number of frms n equlbrum. The frms frs order condons are: F K

More information

Notes on the stability of dynamic systems and the use of Eigen Values.

Notes on the stability of dynamic systems and the use of Eigen Values. Noes on he sabl of dnamc ssems and he use of Egen Values. Source: Macro II course noes, Dr. Davd Bessler s Tme Seres course noes, zarads (999) Ineremporal Macroeconomcs chaper 4 & Techncal ppend, and Hamlon

More information

THERMODYNAMICS 1. The First Law and Other Basic Concepts (part 2)

THERMODYNAMICS 1. The First Law and Other Basic Concepts (part 2) Company LOGO THERMODYNAMICS The Frs Law and Oher Basc Conceps (par ) Deparmen of Chemcal Engneerng, Semarang Sae Unversy Dhon Harano S.T., M.T., M.Sc. Have you ever cooked? Equlbrum Equlbrum (con.) Equlbrum

More information

Analysis And Evaluation of Econometric Time Series Models: Dynamic Transfer Function Approach

Analysis And Evaluation of Econometric Time Series Models: Dynamic Transfer Function Approach 1 Appeared n Proceedng of he 62 h Annual Sesson of he SLAAS (2006) pp 96. Analyss And Evaluaon of Economerc Tme Seres Models: Dynamc Transfer Funcon Approach T.M.J.A.COORAY Deparmen of Mahemacs Unversy

More information

THEORETICAL AUTOCORRELATIONS. ) if often denoted by γ. Note that

THEORETICAL AUTOCORRELATIONS. ) if often denoted by γ. Note that THEORETICAL AUTOCORRELATIONS Cov( y, y ) E( y E( y))( y E( y)) ρ = = Var( y) E( y E( y)) =,, L ρ = and Cov( y, y ) s ofen denoed by whle Var( y ) f ofen denoed by γ. Noe ha γ = γ and ρ = ρ and because

More information

WiH Wei He

WiH Wei He Sysem Idenfcaon of onlnear Sae-Space Space Baery odels WH We He wehe@calce.umd.edu Advsor: Dr. Chaochao Chen Deparmen of echancal Engneerng Unversy of aryland, College Par 1 Unversy of aryland Bacground

More information

2. SPATIALLY LAGGED DEPENDENT VARIABLES

2. SPATIALLY LAGGED DEPENDENT VARIABLES 2. SPATIALLY LAGGED DEPENDENT VARIABLES In hs chaper, we descrbe a sascal model ha ncorporaes spaal dependence explcly by addng a spaally lagged dependen varable y on he rgh-hand sde of he regresson equaon.

More information

ABSTRACT KEYWORDS. Bonus-malus systems, frequency component, severity component. 1. INTRODUCTION

ABSTRACT KEYWORDS. Bonus-malus systems, frequency component, severity component. 1. INTRODUCTION EERAIED BU-MAU YTEM ITH A FREQUECY AD A EVERITY CMET A IDIVIDUA BAI I AUTMBIE IURACE* BY RAHIM MAHMUDVAD AD HEI HAAI ABTRACT Frangos and Vronos (2001) proposed an opmal bonus-malus sysems wh a frequency

More information

Cubic Bezier Homotopy Function for Solving Exponential Equations

Cubic Bezier Homotopy Function for Solving Exponential Equations Penerb Journal of Advanced Research n Compung and Applcaons ISSN (onlne: 46-97 Vol. 4, No.. Pages -8, 6 omoopy Funcon for Solvng Eponenal Equaons S. S. Raml *,,. Mohamad Nor,a, N. S. Saharzan,b and M.

More information

Endogeneity. Is the term given to the situation when one or more of the regressors in the model are correlated with the error term such that

Endogeneity. Is the term given to the situation when one or more of the regressors in the model are correlated with the error term such that s row Endogeney Is he erm gven o he suaon when one or more of he regressors n he model are correlaed wh he error erm such ha E( u 0 The 3 man causes of endogeney are: Measuremen error n he rgh hand sde

More information

Density Matrix Description of NMR BCMB/CHEM 8190

Density Matrix Description of NMR BCMB/CHEM 8190 Densy Marx Descrpon of NMR BCMBCHEM 89 Operaors n Marx Noaon If we say wh one bass se, properes vary only because of changes n he coeffcens weghng each bass se funcon x = h< Ix > - hs s how we calculae

More information

Chapter 8 Dynamic Models

Chapter 8 Dynamic Models Chaper 8 Dnamc odels 8. Inroducon 8. Seral correlaon models 8.3 Cross-seconal correlaons and me-seres crosssecon models 8.4 me-varng coeffcens 8.5 Kalman fler approach 8. Inroducon When s mporan o consder

More information

DEEP UNFOLDING FOR MULTICHANNEL SOURCE SEPARATION SUPPLEMENTARY MATERIAL

DEEP UNFOLDING FOR MULTICHANNEL SOURCE SEPARATION SUPPLEMENTARY MATERIAL DEEP UNFOLDING FOR MULTICHANNEL SOURCE SEPARATION SUPPLEMENTARY MATERIAL Sco Wsdom, John Hershey 2, Jonahan Le Roux 2, and Shnj Waanabe 2 Deparmen o Elecrcal Engneerng, Unversy o Washngon, Seale, WA, USA

More information

Introduction to Boosting

Introduction to Boosting Inroducon o Boosng Cynha Rudn PACM, Prnceon Unversy Advsors Ingrd Daubeches and Rober Schapre Say you have a daabase of news arcles, +, +, -, -, +, +, -, -, +, +, -, -, +, +, -, + where arcles are labeled

More information

Mechanics Physics 151

Mechanics Physics 151 Mechancs Physcs 5 Lecure 9 Hamlonan Equaons of Moon (Chaper 8) Wha We Dd Las Tme Consruced Hamlonan formalsm H ( q, p, ) = q p L( q, q, ) H p = q H q = p H = L Equvalen o Lagrangan formalsm Smpler, bu

More information

A HIERARCHICAL KALMAN FILTER

A HIERARCHICAL KALMAN FILTER A HIERARCHICAL KALMAN FILER Greg aylor aylor Fry Consulng Acuares Level 8, 3 Clarence Sree Sydney NSW Ausrala Professoral Assocae, Cenre for Acuaral Sudes Faculy of Economcs and Commerce Unversy of Melbourne

More information

10. A.C CIRCUITS. Theoretically current grows to maximum value after infinite time. But practically it grows to maximum after 5τ. Decay of current :

10. A.C CIRCUITS. Theoretically current grows to maximum value after infinite time. But practically it grows to maximum after 5τ. Decay of current : . A. IUITS Synopss : GOWTH OF UNT IN IUIT : d. When swch S s closed a =; = d. A me, curren = e 3. The consan / has dmensons of me and s called he nducve me consan ( τ ) of he crcu. 4. = τ; =.63, n one

More information

This document is downloaded from DR-NTU, Nanyang Technological University Library, Singapore.

This document is downloaded from DR-NTU, Nanyang Technological University Library, Singapore. Ths documen s downloaded from DR-NTU, Nanyang Technologcal Unversy Lbrary, Sngapore. Tle A smplfed verb machng algorhm for word paron n vsual speech processng( Acceped verson ) Auhor(s) Foo, Say We; Yong,

More information

Mechanics Physics 151

Mechanics Physics 151 Mechancs Physcs 5 Lecure 9 Hamlonan Equaons of Moon (Chaper 8) Wha We Dd Las Tme Consruced Hamlonan formalsm Hqp (,,) = qp Lqq (,,) H p = q H q = p H L = Equvalen o Lagrangan formalsm Smpler, bu wce as

More information

P R = P 0. The system is shown on the next figure:

P R = P 0. The system is shown on the next figure: TPG460 Reservor Smulaon 08 page of INTRODUCTION TO RESERVOIR SIMULATION Analycal and numercal soluons of smple one-dmensonal, one-phase flow equaons As an nroducon o reservor smulaon, we wll revew he smples

More information

Standard Error of Technical Cost Incorporating Parameter Uncertainty

Standard Error of Technical Cost Incorporating Parameter Uncertainty Sandard rror of echncal Cos Incorporang Parameer Uncerany Chrsopher Moron Insurance Ausrala Group Presened o he Acuares Insue General Insurance Semnar 3 ovember 0 Sydney hs paper has been prepared for

More information

National Exams December 2015 NOTES: 04-BS-13, Biology. 3 hours duration

National Exams December 2015 NOTES: 04-BS-13, Biology. 3 hours duration Naonal Exams December 205 04-BS-3 Bology 3 hours duraon NOTES: f doub exss as o he nerpreaon of any queson he canddae s urged o subm wh he answer paper a clear saemen of any assumpons made 2 Ths s a CLOSED

More information

( ) [ ] MAP Decision Rule

( ) [ ] MAP Decision Rule Announcemens Bayes Decson Theory wh Normal Dsrbuons HW0 due oday HW o be assgned soon Proec descrpon posed Bomercs CSE 90 Lecure 4 CSE90, Sprng 04 CSE90, Sprng 04 Key Probables 4 ω class label X feaure

More information

Online Supplement for Dynamic Multi-Technology. Production-Inventory Problem with Emissions Trading

Online Supplement for Dynamic Multi-Technology. Production-Inventory Problem with Emissions Trading Onlne Supplemen for Dynamc Mul-Technology Producon-Invenory Problem wh Emssons Tradng by We Zhang Zhongsheng Hua Yu Xa and Baofeng Huo Proof of Lemma For any ( qr ) Θ s easy o verfy ha he lnear programmng

More information

ELASTIC MODULUS ESTIMATION OF CHOPPED CARBON FIBER TAPE REINFORCED THERMOPLASTICS USING THE MONTE CARLO SIMULATION

ELASTIC MODULUS ESTIMATION OF CHOPPED CARBON FIBER TAPE REINFORCED THERMOPLASTICS USING THE MONTE CARLO SIMULATION THE 19 TH INTERNATIONAL ONFERENE ON OMPOSITE MATERIALS ELASTI MODULUS ESTIMATION OF HOPPED ARBON FIBER TAPE REINFORED THERMOPLASTIS USING THE MONTE ARLO SIMULATION Y. Sao 1*, J. Takahash 1, T. Masuo 1,

More information

Anisotropic Behaviors and Its Application on Sheet Metal Stamping Processes

Anisotropic Behaviors and Its Application on Sheet Metal Stamping Processes Ansoropc Behavors and Is Applcaon on Shee Meal Sampng Processes Welong Hu ETA-Engneerng Technology Assocaes, Inc. 33 E. Maple oad, Sue 00 Troy, MI 48083 USA 48-79-300 whu@ea.com Jeanne He ETA-Engneerng

More information

F-Tests and Analysis of Variance (ANOVA) in the Simple Linear Regression Model. 1. Introduction

F-Tests and Analysis of Variance (ANOVA) in the Simple Linear Regression Model. 1. Introduction ECOOMICS 35* -- OTE 9 ECO 35* -- OTE 9 F-Tess and Analyss of Varance (AOVA n he Smple Lnear Regresson Model Inroducon The smple lnear regresson model s gven by he followng populaon regresson equaon, or

More information

EEL 6266 Power System Operation and Control. Chapter 5 Unit Commitment

EEL 6266 Power System Operation and Control. Chapter 5 Unit Commitment EEL 6266 Power Sysem Operaon and Conrol Chaper 5 Un Commmen Dynamc programmng chef advanage over enumeraon schemes s he reducon n he dmensonaly of he problem n a src prory order scheme, here are only N

More information

The Performance of Optimum Response Surface Methodology Based on MM-Estimator

The Performance of Optimum Response Surface Methodology Based on MM-Estimator The Performance of Opmum Response Surface Mehodology Based on MM-Esmaor Habshah Md, Mohd Shafe Musafa, Anwar Frano Absrac The Ordnary Leas Squares (OLS) mehod s ofen used o esmae he parameers of a second-order

More information

The Finite Element Method for the Analysis of Non-Linear and Dynamic Systems

The Finite Element Method for the Analysis of Non-Linear and Dynamic Systems Swss Federal Insue of Page 1 The Fne Elemen Mehod for he Analyss of Non-Lnear and Dynamc Sysems Prof. Dr. Mchael Havbro Faber Dr. Nebojsa Mojslovc Swss Federal Insue of ETH Zurch, Swzerland Mehod of Fne

More information

e-journal Reliability: Theory& Applications No 2 (Vol.2) Vyacheslav Abramov

e-journal Reliability: Theory& Applications No 2 (Vol.2) Vyacheslav Abramov June 7 e-ournal Relably: Theory& Applcaons No (Vol. CONFIDENCE INTERVALS ASSOCIATED WITH PERFORMANCE ANALYSIS OF SYMMETRIC LARGE CLOSED CLIENT/SERVER COMPUTER NETWORKS Absrac Vyacheslav Abramov School

More information

doi: info:doi/ /

doi: info:doi/ / do: nfo:do/0.063/.322393 nernaonal Conference on Power Conrol and Opmzaon, Bal, ndonesa, -3, June 2009 A COLOR FEATURES-BASED METHOD FOR OBJECT TRACKNG EMPLOYNG A PARTCLE FLTER ALGORTHM Bud Sugand, Hyoungseop

More information

Anomaly Detection. Lecture Notes for Chapter 9. Introduction to Data Mining, 2 nd Edition by Tan, Steinbach, Karpatne, Kumar

Anomaly Detection. Lecture Notes for Chapter 9. Introduction to Data Mining, 2 nd Edition by Tan, Steinbach, Karpatne, Kumar Anomaly eecon Lecure Noes for Chaper 9 Inroducon o aa Mnng, 2 nd Edon by Tan, Senbach, Karpane, Kumar 2/14/18 Inroducon o aa Mnng, 2nd Edon 1 Anomaly/Ouler eecon Wha are anomales/oulers? The se of daa

More information

CS286.2 Lecture 14: Quantum de Finetti Theorems II

CS286.2 Lecture 14: Quantum de Finetti Theorems II CS286.2 Lecure 14: Quanum de Fne Theorems II Scrbe: Mara Okounkova 1 Saemen of he heorem Recall he las saemen of he quanum de Fne heorem from he prevous lecure. Theorem 1 Quanum de Fne). Le ρ Dens C 2

More information

Tools for Analysis of Accelerated Life and Degradation Test Data

Tools for Analysis of Accelerated Life and Degradation Test Data Acceleraed Sress Tesng and Relably Tools for Analyss of Acceleraed Lfe and Degradaon Tes Daa Presened by: Reuel Smh Unversy of Maryland College Park smhrc@umd.edu Sepember-5-6 Sepember 28-30 206, Pensacola

More information

A Novel Iron Loss Reduction Technique for Distribution Transformers. Based on a Combined Genetic Algorithm - Neural Network Approach

A Novel Iron Loss Reduction Technique for Distribution Transformers. Based on a Combined Genetic Algorithm - Neural Network Approach A Novel Iron Loss Reducon Technque for Dsrbuon Transformers Based on a Combned Genec Algorhm - Neural Nework Approach Palvos S. Georglaks Nkolaos D. Doulams Anasasos D. Doulams Nkos D. Hazargyrou and Sefanos

More information

Appendix to Online Clustering with Experts

Appendix to Online Clustering with Experts A Appendx o Onlne Cluserng wh Expers Furher dscusson of expermens. Here we furher dscuss expermenal resuls repored n he paper. Ineresngly, we observe ha OCE (and n parcular Learn- ) racks he bes exper

More information

Appendix H: Rarefaction and extrapolation of Hill numbers for incidence data

Appendix H: Rarefaction and extrapolation of Hill numbers for incidence data Anne Chao Ncholas J Goell C seh lzabeh L ander K Ma Rober K Colwell and Aaron M llson 03 Rarefacon and erapolaon wh ll numbers: a framewor for samplng and esmaon n speces dversy sudes cology Monographs

More information