Valid Inequalities Based on Demand Propagation for Chemical Production Scheduling MIP Models

Size: px
Start display at page:

Download "Valid Inequalities Based on Demand Propagation for Chemical Production Scheduling MIP Models"

Transcription

1 Vald Inequalte Baed on Demand ropagaton for Chemcal roducton Schedulng MI Model Sara Velez, Arul Sundaramoorthy, And Chrto Maravela 1 Department of Chemcal and Bologcal Engneerng Unverty of Wconn Madon 1415 Engneerng Dr., Madon, WI, 5376 Abtract. The plannng of chemcal producton often nvolve the optmzaton of the ze of the ta to be performed ubect to unt capacty contrant, a well a nventory contrant for ntermedate materal. Whle everal mxed-nteger programmng (MI) model have been propoed that account for thee feature, the development of tghtenng method for thee formulaton ha receved lmted attenton. In th paper, we develop a contrant propagaton algorthm for the calculaton of lower bound on the number and ze of ta neceary to atfy gven demand. Thee bound are then ued to expre three type of tghtenng contrant whch greatly enhance the computatonal performance of the MI chedulng model. Importantly, the propoed method are applcable to a wde range of problem clae and tme-ndexed MI model for chemcal producton chedulng. 1. Introducton Chemcal producton nvolve the converon of raw materal (feedtoc) nto fnal product, both of whch are mot often flud (gae or lqud). Flud can be mxed or plt n varable proporton and are often recycled (e.g., unreacted raw materal are recycled after they are eparated from the fnal product). roducton ta convert a et of nput materal nto a et of output materal, and are carred out n equpment unt that are capable of proceng dfferent amount of (flud) materal. In other word, the ze and therefore the number of ta to be executed can be an optmzaton decon. Whle there are chemcal faclte wth a wde range of proceng charactertc and contrant, there are two maor type of chemcal proceng, equental and networ. In equental proceng, t aumed that producton ta have a ngle nput and a ngle output materal; the nput materal of a ta produced by a ngle uptream ta; and the output of a ta conumed by a ngle downtream ta. Thu, materal from dfferent ta are not mxed, and the output of a ta cannot be plt to be conumed by multple ta. Alo, t mplctly aumed that each materal produced and conumed by at mot one ta. In networ proceng, ta may conume or produce multple materal, and a materal can be produced and conumed by dfferent type of ta. Alo, materal comng from dfferent ta can be mxed n a torage veel, the output of a ta can be conumed by multple downtream ta, and materal can alo be recycled. 1 Correpondng author; emal: maravela@wc.edu; Tel:

2 Interetngly, the method that have been developed to addre chemcal producton chedulng problem follow the aforementoned clafcaton. Early wor focued on equental faclte where a batch of raw materal ha to go through multple producton tage to be converted to the fnal product. In other word, the proceng of flud repreented a the proceng of a dcrete batch that ha to go through dfferent proceng tage; amount of materal are not montored and no materal balance are ncluded. We note that th type of modelng poble becaue plttng, mxng and recyclng are not allowed. We term thee approache a batch-baed. In mot cae, t alo aumed that the ze and thu the number of batche (for nown demand) were determned pror to optmzaton [elat, 212], whle recent approache conder multaneou batchng and chedulng [raad and Maravela, 28]. In the proce ytem engneerng (SE) lterature thee faclte are termed ether a mult-tage, f the routng through the tage the ame for all batche/product, or mult-purpoe, f there are batch/product-pecfc routng [Mendez et al., 26; Maravela, 212a]. Clearly, batch-baed method cannot be ued to addre problem n networ faclte where mxng, plttng, and recyclng are allowed; materal can be produced and conumed by dfferent ta; and the ze of the ta are optmzaton decon. To addre thee problem reearcher developed what we term materal-baed model, where the amount of materal proceed by a ta, a well a the nventore of materal, are explctly modeled. Another ey modelng apect of thee approache that a tme reference grd defned, and () the tart/end tme of ta are mapped onto th grd and () nventory contrant are expreed at the tme pont of the grd. The poneerng wor of antelde and co-worer employed a dcrete-tme grd, where the tme horzon dvded nto perod of equal and nown length and proceng tme are aumed to be contant [Kondl et al., 1993; Shah et al., 1993]. Snce then, a wde range of formulaton have been propoed ncludng contnuou-tme and mxed-tme grd, a well a unt-pecfc grd. egardle of the nature of the tme repreentaton and the detal of the formulaton, there are three proceng contrant that have to be enforced [Maravela, 212b]: a) Unt utlzaton: a unt cannot perform more than one ta at a tme. b) Materal balance: the nventory of a materal n a torage veel ha to be nonnegatve and le than the capacty of the veel. c) Batch-ze: the ze of a proceng ta (batch-ze) ha to be between a nonzero lower and an upper bound (capacty). The man dfference between the varou materal-baed formulaton le n the modelng of the frt contrant. In dcrete-tme model, t enforced through the wdely ued one-machne clque contrant, whle n contnuou-tme formulaton more complex et of contrant are ued. The econd contrant alway a flow balance contrant for a materal-tme node, mlar to thoe ued n producton plannng formulaton [ochet and Woley, 26]. Fnally, the thrd contrant reult n varable lower/upper bound contrant: f a ta executed, then the amount proceed hould be wthn a lower and upper bound; otherwe t zero. The ze of a ta hould be greater than or equal to the lower bound for afety and controllablty reaon, a well a due to recpe contrant. Mot of the aforementoned type of contrant have been tuded extenvely. Frt, vald nequalte and pecalzed oluton method have been developed for tme-ndexed ngle-machne problem [Woley, 2

3 199; van den Aer et al., 1999; van den Aer et al., 2]. Alo, many contrant propagaton algorthm have been developed for th contrant [Hooer, 2; Baptte et al., 21; Van Hentenryc and Mchel, 25; Hooer, 27]. Second, lot zng problem have materal balance that are mlar to thoe n networ chedulng problem, and addng vald contrant baed on thee materal balance and bll-of-materal nformaton reult n a gnfcantly tghter formulaton [ochet and Woley, 1993; Woley, 1997; Mller and Woley, 23].Fnally, vald nequalte have been derved from varable upper bound contrant n the context of fxed-charge networ problem [Nemhauer and Woley, 1988; Woley, 1998]. However, chedulng problem wth varable lower bound contrant have not been tuded uffcently. Burard and Hatzl (25) bound the number of batche of each ta and the number of batche produced of each materal by olvng a mall MI for every ta and an L for every materal, whle Jana and Flouda (28) propoe boundng the number of batche and cumulatve producton for each ta by olvng an MI boundng problem for every ta. Accordngly, n th paper, we tudy how the varable lower bound contrant can be ued to develop trong vald nequalte for MI chemcal producton chedulng problem. Specfcally, we frt develop an algorthm that allow u to calculate tght lower bound on the number of batche of each ta and the total amount of materal proceed by each ta that are neceary to atfy the gven demand. Thee lower bound, whch are calculated from ntance data wthn econd, are then ued to wrte two clae of vald nequalte whch greatly reduce the computatonal requrement. Our algorthm applcable to all type of networ faclte, ncludng faclte wth () multple ta producng or conumng the ame materal(), () multple unt capable of performng the ame ta(), and () recycle loop. Alo, they are applcable to all type of materal-baed MI formulaton that employ tmendexed varable. It mportant to note that materal-baed tme-ndexed MI model are ued a the bacbone n a wde range of chemcal producton chedulng tool, a well a n n-houe tool developed by chemcal compane [Wac, 29]. Alo, materal-baed model have been extended to addre problem n equental producton envronment a well a faclte that combne equental and networ envronment [Sundaramoorthy and Maravela, 211]. Neverthele, one of the lmtaton of both commercal product and pecalzed tool that they reman computatonally neffcent. Thu, our method, whch are applcable to all aforementoned model, have the potental to beneft a wde range of tool and model. The paper tructured a follow. In Secton 2, we ntroduce a networ-npred repreentaton of chemcal faclte, preent a formal tatement of the problem we conder and a wdely ued MI chedulng model, and cloe wth a motvatng example. In Secton 3, we preent the contrant propagaton algorthm for the calculaton of the lower bound on the number of batche and the total materal proceed. In Secton 4, we preent two type of tghtenng contrant, a well a extenon for problem wth ntermedate due tme. In Secton 5, we perform an extenve computatonal tudy ncludng more than 7 problem ntance. We ue lowercae talc letter for ndce, uppercae bold letter for et, lowercae Gree letter for parameter, and uppercae talc for optmzaton varable. 3

4 2. Bacground 2.1. Chemcal roducton Schedulng Notaton A chemcal faclty tranform a et of raw materal nto a et of valuable product through a equence of proceng ta carred out n equpment unt. A proceng ta convert a et of nput materal (raw materal or ntermedate) nto a et of output (ntermedate or fnal product) accordng to pecfed converon coeffcent. Note that the term ta ued to decrbe a type of operaton (e.g., the converon of A to B va reacton A B a ta), whch mple that a chedule may nclude multple executon of the ame ta. A ta can n general be carred out n multple unt of unequal capacty, whch mple that the number of batche of a ta neceary to atfy gven demand not nown pror to optmzaton. Thu, the term chedulng n the SE lterature ha been ued to decrbe a problem whch nclude three level of decon: () the determnaton of the number (and ze) of batche to be performed, () the agnment of batche to proceng unt, and () the tmng of batche o that at mot one batch executed on a unt. We aume that a chemcal faclty cont of: () proceng unt, where ta are executed; () torage veel, where materal are tored; and () ppng for materal tranfer. roceng unt and torage veel have capacty lmtaton, and each proceng unt and torage veel capable of proceng/torng a ubet of ta/materal. We do not conder other hared reource uch a utlte (e.g., electrcty) and labor. Fgure 1a how the proce flow dagram (FD) of a mple chemcal faclty, where raw materal converted nto ntermedate n unt U1, and the latter converted to ether product or n unt U2 or U3. Note that the FD how the phycal equpment of the faclty, but not necearly the ta that tae place on the unt. Alo, unt can perform multple ta; e.g., unt U2 and U3 can convert to or convert to. U2 Storage for Storage for U1 Storage for U3 Storage for {U1} {U2,U3} (a) roce Flow Dagram (b) State-Ta Networ Fgure 1. roce flow dagram (phycal layout of the faclty) and tate-ta networ repreentaton of a faclty roblem Statement To tudy chedulng problem, we have to ue a repreentaton that nclude not only the phycal reource, but alo the dfferent ta that can be carred out. In th paper, we adopt the tate-ta networ repreentaton, whch baed on the followng concept: a) State (materal): nclude feed, ntermedate, and fnal product and are repreented by crcle. b) Ta: are operaton that produce and conume tate (materal) and are depcted wth rectangle. c) Unt: unary reource for the executon of ta; multple unt may be able to proce a ngle ta, and a ngle unt may be able to proce multple ta. 4

5 State (crcle) and ta (rectangle) are connected by arrow, referred to a tream, howng the flow of materal. If a ta conume (produce) a tate, an arrow pont from the tate (ta) to the ta (tate). Fgure 1b how the STN repreentaton for a chedulng problem n the faclty hown n Fgure 1a. Note that unt are hown mplctly through the ta-unt compatblty nformaton gven by the dotted lne. The tructure of the networ defned n term of the followng et (ee Fgure 2a): Indce/Set, I J, S Subet I / I J S F ta proceng unt tate ta producng/conumng tate proceng unt that can proce ta S / S tate produced/conumed by ta fnal product We are alo gven the followng parameter: ϕ mn total cutomer demand for tate / mnmum/maxmum capacty of unt / τ ζ max fracton of tate produced/conumed by ta fxed proceng tme for ta n unt ntal nventory of tate In Fgure 2a, conumed by and and produced by ; therefore, I ={, } and I ={}. In Fgure 2b, produce and conume and, o S ={} and S ={,}. In general, a plu (+) upercrpt ndcate producton, and a mnu ( ) ndcate conumpton. If and can both be proceed n ether U2 or U3, then J I I I S S (a) Set I / I (b) Set S / S Fgure 2. Illutraton of general et Dcrete-tme Model S = J ={U1, U3}. To llutrate the underlyng concept and perform computatonal tude, we ntroduce the wdely-ued MI formulaton of Shah et al. (1993). We ue th model becaue a recent computatonal tudy howed that dcrete-tme model are more effectve for large-cale problem and, mot mportantly, can be ealy extended to account for a range of proceng retrcton and charactertc (e.g., ntermedate due date, holdng and utlty cot, varable reource requrement durng ta executon) at almot no computatonal cot [Sundaramoorthy and Maravela, 211b]. However, we 5

6 hghlght that our method are applcable to all tme-ndexed MI model for chemcal producton chedulng that account for varable batch-ze. Tme pont are ndexed by t. The formulaton nclude one famly of bnary varable: X t one f unt procee ta tartng at tme t and the followng non-negatve contnuou varable: B t S t t D t batch-ze of ta proceed n unt tartng at tme t nventory of tate at tme t amount of feed tate purchaed at tme t amount of fnal product delvered to cutomer at tme t The model cont of unt utlzaton (eqn. 1), unt capacty (eqn. 2), nventory capacty (eqn. 3), and materal balance (eqn. 4) contrant. I tt 1 t X t 1, t (1) S X B X, J, t (2) mn max t t t t, t (3) max S St 1 B B D t t t t t I J I J where D t fxed to the total amount of tate due at tme t; and t only non-zero for feed tate. Varable X t are fxed to zero for the τ -1 tme pont before the end of the tme horzon Motvatng Example We conder an ntance of the faclty ntroduced n Fgure 1, wth a demand for 9g of and 25g of. For each unt, the capacty (mn-max) and proceng cot of a ta n that unt are: 25-6g and $1 for U1; 4-5g and $25 for U2; and 35-45g and $3 for U3. Our obectve to mnmze cot. We generated a model for th ntance baed on formulaton preented n 2.3. The mnmum cot for the L-relaxaton $76.7, whle the number of batche of ta,, and are 1.9, 1.8, and.5, repectvely (ee Table 1). Table 1. Effect of tghtenng on the number of batche and cot for the networ n Fgure 1. # of Batche Mn. Cot ($) Optmal oluton Mnmum # of batche Burard method ropoed method L-elaxaton No tghtenng Burard method ropoed method (4) 6

7 However, f we tae nto account the capacte of the unt, we can nfer that more batche wll be requred n any feable oluton. mut produce 9g of to atfy demand; nce the maxmum batch-ze 5g, th requre at leat two batche. Smlarly, mut produce at leat 25g of, but, nce the mnmum batch-ze 35g, mut produce at leat 35g n at leat one batch. Together, and requre a mnmum of 125g (=9+35) of, whch produce n at leat three batche. Ung th contrant propagaton procedure, we calculate lower bound on the number of batche and cumulatve producton for each ta. Table 1 how the effect of tghtenng on the example n Fgure 1. Wthout tghtenng, the mnmum cot for the L-relaxaton $76.7. Wth the propoed method, the cot of the L relaxaton $15. Interetngly, the mnmum cot and number of batche of the L-relaxaton n th mple problem, match the optmal MI oluton. Our goal to develop a method that allow u to ytematcally calculate mlar bound n general producton networ, ncludng faclte wth recycle tream. Note that ung the tghtenng method propoed by Burard and Hatzl (25) mprove the cot of the Lrelaxaton to $96.7. Burard and Hatzl (25) do not conder mnmum unt capacte, o ther method requre to produce only 115g (=9+25) of n two batche. The tghtenng method of Jana and Flouda (28) fnd a lower bound on the number of batche whch a tght a our, but requre olvng a MI for every ta whch often a tme conumng a olvng the orgnal MI model. Alo, ther approach rele on a pecfc contnuou-tme model. Our propoed method do not requre olvng any auxlary MI, are vald for any networ, and can be ued wth any model employng a tme grd. 3. Contrant ropagaton Algorthm Our contrant propagaton algorthm calculate parameter whch are then ued to formulate the tghtenng contrant. The algorthm depend on the tructure of the proce networ. We clafy networ nto four categore accordng to Fgure 3: (1) networ wth no loop (Fgure 4a); (2) networ wth loop but no recycle tate (Fgure 4b); (3) networ wth recycle tate n a ngle loop (Fgure 4c); and (4) networ wth a recycle tate and multple neted loop (Fgure 4d). A recycle loop any cloed path (movng only n the drecton of the tream arrow) wthn the networ, and a recycle tate a tate n a loop that can be produced by multple ta. Each colored box n Fgure 3 ha t own pecfc tghtenng procedure that are decrbed n the ndcated ecton. All networ ue the tghtenng procedure for general networ ( 3.1). Networ are frt dvded nto networ wth and wthout recycle loop. If there are recycle loop, 3.2 decrbe the addtonal tghtenng method. ecycle loop are further dvded nto loop wth and wthout recycle tate, and 3.3 explan more tghtenng method for networ wth recycle tate. When a loop ha a recycle tate, t clafed a havng ether a ngle loop or a neted loop, whch decrbed n 3.4. In 3.5, we preent a general algorthm for applyng the tghtenng method to any networ. In 3.6, we decrbe an alternatve approach for networ wth recycle tate that alo wor well when multple ta can produce a ngle tate. 7

8 3.1 General Networ 3.2 Loop No Loop (Fg 4a) (a) General networ wthout loop (b) Networ wth a loop and no recycle tate 3.3 ecycle State No ecycle State (4b) 3.4 Neted Loop (4d) Sngle Loop (4c) Fgure 3. Clafcaton of networ General Networ Bacward ropagaton (c) Networ wth a recycle tate (d) Networ wth neted loop Fgure 4. Example of categore of networ. ω For bacward propagaton, we ntroduce the followng parameter: lower bound on the amount of tate requred to meet fnal demand μ / ν lower bound on the producton of ta requred to meet fnal demand, μ lower bound on the amount of tate that mut be produced by ta Frt, we need to fnd the mnmum amount requred for all tate, ω, and the mnmum producton for all ta, ; we calculate thee parameter equentally by bacward propagatng demand for fnal product. We etmate ω once nown for all ta conumng tate (all I ). Smlarly, we need to now ω for all tate produced by ta (all S ) before we calculate (ee Fgure 5). We calculate ω tartng wth the fnal product: S I F F S (5) In the HS of eqn 5, the top expreon the cutomer demand for fnal product mnu any ntal nventory whle the bottom expreon the amount of ntermedate requred by all ta conumng t mnu t ntal nventory. If the ntal nventory exceed the amount requred, ω wll be negatve, and the proce doe not need to produce tate. The mnmum amount of tate that mut be produced by ta, ν, ntroduced a an ntermedate parameter. Although ω can be negatve, ν mut be at leat zero. ST max, S, I MT S \ S, I (6) S5 S6 T4 (a) STN repreentaton of example S7 S5 T4 S6 (b) Sere of calculaton n bac-propagaton and T4 mut be etmated before ω S5 ω S5 and ω S5 mut be etmated before S7 Fgure 5. Example of bacward propagaton of demand. From demand for S6 and S7, we calculate ω S6 and ω S7 ; next, we calculate and T4, followed by ω S5 and ω. We cannot fnd yet becaue ω not nown, o we calculate, followed by ω and ω. Fnally, we calculate and ω. 8

9 # of Batche where S ST the et of tate produced by a ngle ta, S MT the et of tate that can be produced by multple ta, and S the et of recycle tate. Eqn. 6 doe not conder recycle tate, whch we dcu n 3.3. If multple ta can produce a tate, we aume that any ngle ta can meet the full demand; therefore, the mnmum producton of that tate by each ta zero, and, for thee tate, an alternatve approach that gve better reult decrbed n 3.6. After calculatng ν for all tate produced by a ta, we calculate μ, max S (7) The term nde the bracet the total amount ta mut proce to meet the demand for. We tae the maxmum over all tate produced by ta to enure the ta atfe demand for all tate. At th pont, we mut fnd the mnmum attanable producton amount, μ (8) where Δμ the mnmum amount that mut be added to μ to reach an attanable producton amount and dcued n arameter provde a tghter lower bound on the requred producton of ta. We bacward propagate demand untl ω and are nown for all tate and ta Attanable roducton Amount After fndng μ, we need Δμ to calculate. When only one unt can proce a ta, t traghtforward to fnd the range of attanable producton amount for any number of batche and to chec f the requred producton n one of thoe attanable range (ee example n Fgure 6). If μ fall n an attanable regon, demand can be met exactly, and Δμ zero; otherwe, Δμ the dtance between μ and the tart of the next attanable range. However, when multple unt can proce a ta, we mut fnd and chec attanable range for every poble combnaton of batche n unt. For example, f two unt can proce a ta, we need to chec 1 batch n U1, n U2; n U1, 1 n U2; 1 n U1, 1 n U2; etc. To reduce the number of max combnaton, we fnd an upper bound on the number of batche n a partcular unt,, by dvdng μ by the larget poble batch unt can proce and roundng up, max max J (9) roducton Amount Fgure 6. Attanable range for a unt wth a capacty of 3-4g are haded. 9

10 When the number of batche for a partcular unt at t upper bound, we are guaranteed the attanable range meet or exceed demand wth ut one unt, and the number of batche n all other unt et to zero to elmnate more combnaton. In total there are K range to chec, K max J J (1) For each range, {1,2,,K }, there a unque combnaton of, where gve the number of max batche n unt for range. The frt term n eqn. 1 the number of range wth =,1,2, ( - 1) for all max J, and the econd term the number of range wth = for one J and = for all other J. We loop over all {1,2,,K } and chec f fall nto an attanable range. If, for ta, J mn max J for ome, the mnmum requred producton can be met exactly, and Δμ zero. The LHS of eqn. 11 gve the lower bound of the attanable range, and the HS gve the upper bound. If μ not attanable for any range, we perform another loop over all to fnd the range that able to meet producton requrement and whoe lower bound cloet to μ. mn mn f J J mn f J The top lne et equal to the exce amount produced (the lower bound on the range mnu μ ) f the combnaton of unt at leat able to meet demand. The econd lne et to nfnty when the combnaton capacty too mall. Once all range have been checed, Δμ calculated: mn (13) A an example, we conder a ta that can be proceed n U1 or U2 and mut proce 55 g (ee max max Fgure 7). We fnd U1 3 and U2 2 from eqn. 9; and K =8 from eqn. 1. All for {1,2,,8} are hown on the left de of Fgure 7. The attanable range are haded (eqn. 11). Snce μ = 55 doe not fall n a gray range, eqn. 11 never atfed. The exce produced by each range the dtance between μ and the tart of the next haded range (eqn. 12). For th example, the mallet exce producton 5, capacte (mn/max): U1: 2/25 U2: 45/5 α U1 α U2 μ = 55 μ = 6 Eqn. 11 I μ n attanable regon? I demand met? Eqn. 12 Exce produton Δμ 1 no no 2 1 no no 3 2 no ye no no no ye no no no ye no ye 5 roducton Amount Fgure 7. Example of calculaton of the mnmum attanable producton amount. (11) (12) 1

11 o Δμ 5, and 6. The algorthm for fndng Δμ a follow: 1 Calculate and Κ and et α. max 2 For {1,2,, Κ } 3 If eqn. 11 true 4 Set Δμ = and top 5 end 6 end 7 For {1,2,, Κ } 8 Calculate (eqn. 12) 9 end 1 Calculate Δμ (eqn. 13) The attanable producton amount, 3.2. Networ wth Loop, now calculated ung eqn. 8. When there are loop n a networ, mple bacward propagaton wll not wor. We ue tear tream to brea the loop wth one tear tream n every loop. A tear tream cont of a ta, referred to a the tear ta, and a tate produced by that ta, referred to a the tear tate. Although any tream can be choen a the tear tream, ome choce are more convenent. If a loop ha a ta that produce a tate outde the recycle loop, th ta hould be ued a the tear ta; otherwe, any ta can be choen. The et I T and S T contan all tear ta and tate, and L I l and L S l gve all ta and tate n loop l. Fgure 8 provde an example for propagatng demand for a networ wth a loop. We wrte the value of μ nde the box repreentng ta, ω nde the crcle repreentng tate, and ν next to the tream connectng ta to tate. For mplcty, all example have one unt per ta, and capacte are gven for each ta. The recycle loop hown n bold. We choe a the tear ta becaue t produce outde the recycle loop, and S5 the tear tate. We ntalze ν,s5 for the tear tream to zero (Fgure 8b). Now, we can etmate a oon a ω nown. We bacward propagate demand a follow: ω =5 =1 ω =1 =1 ω =1 =11 (μ =1 and Δμ =1) ω =11-45=65 T4=65 ω S5 =65*.8=52. We top after calculatng ω S5 =52 for the tear tate (Fgure 8c). We need a mnmum of 52 g of S5, but only produce 5g (=1*.5). Therefore, we et ν,s5 =52, and delete ω and μ for all other tate and ta (Fgure 8d). We repeat bacward propagaton ung the new value ν,s5 =52 (Fgure 8e). Now, produce the requred 52g of S5. In general, we begn by ntalzng ν for all tear tream to zero. We bacward propagate demand untl ω etmated for a tear tate. We update ν for the tear tream ung eqn. 6 and compare t to the producton of the tear ta. If producton can meet demand (ν ), there no problem, and we contnue wth the bacward propagaton. If demand greater than producton (ν > ), we reet all other ω and, and retart the bacward propagaton ung the new ntal value for the tear tream. If, on the econd pa, demand for a tear tate tll exceed producton, the problem nfeable wth the gven ntal nventore. 11

12 S6 S5 8% T4 2% 5% 5% 5 S 1 45 L1: S5 T4 T T I S S5 L IL1,,,T4 L S,,,S5 L1 (a) Example STN and data Capacte -T4 mn 55 5 max 9 15 S5 S5 52 T4 T4 S6 S6 (b) Intalze tear tream (d) Chec tear tream and tart over T4 65 S T4 65 S S6 13 S6 13 (c) ropagate demand bacward Fgure 8. Networ wth a loop and no recycle tate. (e) Fnal mnmum producton amount 3.3. Networ wth ecycle State revouly, when multple ta could produce a ngle tate, we aumed that each ta wa capable of producng the full amount, and, therefore, each ta had a mnmum producton of zero (eqn. 6). However, when a tate S MT part of a loop, we can fnd a better etmate for ν. We refer to thee tate a recycle tate, S S MT. Note that the fracton of a tate that recycled le than one; n other word, f we tart out wth ome amount of the recycle tate and run only ta n the recycle loop, we wll never end up wth more of the recycle tate than what we had ntally. We ntroduce the followng new parameter for networ wth recycle tate: upper bound on the amount of tate that can be produced when only the mnmum amount of a recycle tate, ω, avalable upper bound on the producton of ta when only the mnmum amount of a recycle tate avalable ψ upper bound on the amount of recycle tate produced by ta when only the mnmum amount of tate avalable Example For the networ n Fgure 9, we tart by electng the tear tream n exactly the ame way a n 3.2. We chooe T4 a the tear tream and ntalze ν T4, to zero. The bacward propagaton contnue untl we calculate ω =98 n Fgure 9c. There a maxmum amount of that T4 () can produce wthout ncreang the requred producton of (T4) (ee ropoton 1); th amount gven by the parameter, whch we etmate by propagatng ω forward (Fgure 9d). In the forward propagaton, we tart out wth ω =98g of, whch mean can produce at mot =14g (=98/.7) of S5 (aumng there enough ). T5 need T5=2g of S5 (from Fgure 9c) and the remanng 12g (=14-2) of S5 are avalable for T4. Therefore, T4 produce up to ψ T4, =24g 12

13 (=12*.2) of. Snce ω =98g of are needed n all (from Fgure 9c), we conclude that mut produce the remanng 74g (=98-24), and ν, =74. capable of producng all 98g of (ψ, = wth unlmted ntal nventory of feed), o T4 not requred to produce any, and ν T4, =. Snce alo a tear tate, we mut enure that the producton of by T4 exceed ν T4, ; nce t doe (.2*12), we contnue wth the bacward propagaton untl we have calculated ω and for all tate and ta (Fgure 9e). 4% 6% 7% 3% S5 2% T5 T4 S7 8% S6 (a) Example STN and data S7 2 L1,T4 T L S S I L T S6 8 SL1,S5 I T4 L1: S5 T4 Capacte - T4 T5 mn max S5 (b) Intalze tear tream T5 T4 S7 S , S5, 74 T4, 24, (d) ropagate ω forward T5 T4 12 S7 S S5 T4 1 (c) ropagate demand bacward T S7 2 S T S5 (e) Fnal mnmum producton Fgure 9. Bacward propagaton n a networ wth a ngle recycle tate n a ngle loop. T4 1 8 S7 2 S General Method for ecycle State Bacward propagaton for networ wth recycle tate tart by ntalzng ν for tear tream to zero. A wth tandard loop, when ν calculated for a tear tream, we chec f producton exceed demand. After calculatng ω for a recycle tate, we label th tate * and fnd the maxmum amount that can be recycled. The mnmum amount of the labeled recycle tate that requred to meet demand, ω *, propagated forward around the recycle loop to fnd and for all tate and ta, repectvely. For all tate and ta outde the recycle loop, we et and to nfnty; th enure no feable oluton are cutoff and aume that thee tate are produced tartng from feed wth an unlmted upply. Once nown for all ta producng a tate ( I ), we calculate. We can calculate for the labeled recycle tate mmedately. for * L L f l and max 1 S l S I I S ll * I ll * (14) The frt expreon gve for the labeled recycle tate *. The econd expreon for all other tate n the recycle loop except tate produced by multple ta (other recycle tate) wthn any loop contanng *. The frt term n the condton n the econd lne gve all tate n any loop contanng *, and the econd term the number of ta wthn the loop that produce tate. If there another recycle 13

14 tate nde the loop, there are neted loop n the networ, and dcue how to fnd. The total amount of tate avalable,, the amount produced by all ta plu the ntal nventory. Once nown for all tate conumed by ta ( S ), we fnd. I, L L l l S l L l * L* mn f I and max S S 1 (15) Eqn. 15 gve for any ta nde the recycle loop except for ta that conume multple tate wthn any loop contanng *, whch are dcued n The frt term n the condton gve all ta n any loop contanng tate * and the econd term gve the number of tate wthn the loop that are conumed by ta. In the numerator, the mnmum amount of tate requred for other ta ubtracted to gve the maxmum amount of tate avalable for ta. Dvdng by gve the total amount ta need to proce to conume all of the avalable. Tang the mnmum over all tate conumed by ta enure there enough of every tate. We do not round to an attanable producton amount; roundng up wll reult n weaer bound, and roundng down may cutoff feable pont. Once nown for all ta producng the labeled recycle tate, we fnd the maxmum amount produced by each ta. * * * I (16) Fnally, we calculate the mnmum amount of tate * produced by each ta. * max, * * I * (17) I * \ The mot a ta can produce of a recycle tate ψ and the leat zero. If ν > ψ, there not enough ntal nventory, and the problem nfeable. If the demand cannot be atfed wthout a partcular ta, eqn. 17 gve the mnmum that ta mut produce. If a ngle ta or combnaton of ta can atfy the demand, all other ta do not need to upply any materal, and the fnal term n eqn. 17 le than zero. ropoton. There a maxmum amount of each recycle tate that can be produced by each ta,, wthout ncreang the mnmum requred producton of other ta producng that tate. roof: If ω the mnmum amount of recycle tate requred to meet fnal demand, and the maxmum amount of tate that can be produced by ta when tartng wth ω, then the amount of produced by ta and max, I. I, Let be the maxmum fracton of tate that recycled by ta. If an addtonal amount of tate,, produced by ta, the total amount of now requred at leat, and the maxmum amount produced by all ta I I. The total amount of that need to be produced by ta 14

15 max, I. I I, nce 1,, and. Therefore, all ta producng tate are at ther mnmum I I, producton when all other ta recycle at mot 3.4. Networ wth Neted Loop. Not all networ wth multple loop need the addtonal procedure decrbed n th ecton. There are two cae wth multple loop where the forward propagaton wll not wor Cae 1 When multple ta n a ngle loop produce another recycle tate, the forward propagaton decrbed prevouly fal when calculatng for that tate. For the networ n Fgure 1a, produced by and, whch both belong to loop L3 contanng recycle tate. We propagate the demand a decrbed prevouly untl t tme to propagate ω forward and calculate n Fgure 1b. We need and to calculate, but needed to calculate (Fgure 1c). Durng the forward propagaton, produce at mot =12g of. revouly (Fgure 1b), we determned that mut produce a mnmum of =112.5g of, meanng can now produce an extra 7.5g (= ). If procee all addtonal 7.5g of, t wll produce another.75g (=7.5*.1) of. roceng the addtonal.75g gve.75g (=.75*.1) more of ; th a geometrc ere that eventually converge to 8.33g (=7.5/(1-.1)) more of. We add the extra 8.33g to the orgnal 125g of needed to meet the demand to gve (Fgure 1c). Demand propagaton contnue untl ω and are nown for all tate and ta (Fgure 1d). The example generalzed to any proce networ. We wll refer to the frt labeled recycle tate a * (the one for whch we are tryng to fnd ν ) and the econd recycle tate a ** (the one for whch we are tryng to fnd ). The total amount of tate ** avalable 1 L L f l and max 2 S S I I (18) NC 1 S ll * I \I ll* T T L S, S, I SL1 L L L L L S S, I = I = I, L2 L3 L1 L2 L3 L1: L2: L3: (a) Example STN and data 1% 8% 1% 1 Capacte, mn 5 6 max 75 75,, (c) ropagate ω forward, 13.3, (b) Bacward propagaton (d) Fnal mnmum producton Fgure 1. Demand propagaton wth multple recycle tate n the ame loop

16 where I NC the et of ta for whch ha not been calculated, and ξ the maxmum fracton of tate that can be recycled by the loop contanng only ta n I NC. The algorthm provde a method for eepng trac of I NC. Eqn. 18 gve for any tate excluded from eqn. 14. The frt term the addtonal amount avalable durng the forward propagaton from all ta for whch ha already been calculated. Dvdng by 1-ξ gve the total addtonal amount of ** produced durng recyclng. Fnally the orgnal ω and the ntal nventory are added to gve. arameter ξ can be etmated when there only one loop contanng ** but not *;.e., l {l:** S,* S }. L l L L I l, S Sl for L L Il, S Sl L l S (19) The numerator gve the fracton of materal remanng n the loop after ta. The denomnator the fracton of the materal conumed by ta that already n the recycle loop. Eqn. 19 only vald when there only one loop contanng ** but not *. Eqn. 18 vald for any networ, but ξ need to be etmated Cae 2 When a ta nde a loop conume multple tate n a loop, the forward propagaton fal. In Fgure 11, conume and whch both belong to recycle loop L2. In Fgure 11b, demand ha been bacward propagated to calculate the demand for recycle tate. The demand need to be propagated forward, but we cannot calculate untl and are nown, and cannot be calculated untl nown (Fgure 11c). Durng the forward propagaton we have =18g of avalable. If there enough, can proce at mot 12g. We gnore n eqn. 15, and et 12. We contnue propagatng demand (Fgure 11d). The procedure ealy generalzed to any proce networ, I, L L NC l S l S \S S ll ll * * mn f I and max S S 2 (2) L1: 1% 9% L2: L3: 1% T T L S S I SL1 L L L L L S S, I = I = I, L2 L3 L1 L2 L3 9% 9 Capacte, mn 5 6 max (c) ropagate ω forward 18 (a) Example STN and data (b) Bacward propagaton (d) Fnal mnmum producton Fgure 11. Demand propagaton n a networ wth a recycle tate and multple loop

17 where S NC the et of tate for whch ha not been calculated; thee tate are gnored when calculatng. The algorthm provde a way to eep trac of S NC. Eqn. 2 gve for any ta n the loop excluded from eqn Complete Algorthm The complete algorthm combne the tghtenng procedure for all networ type. For each of the four categore, the relevant porton of the algorthm n Fgure 12 are colored a n Fgure 3. For convenence, we defne the new et: S tate n any recycle loop contanng tate, S SL SL I ta n any recycle loop contanng tate, I SL SL S L l S ll L I l S ll I NC /I NC ta for whch μ /μ not nown I A /I A ta avalable (ω /ω ha been calculated for all S / S ) for calculatng μ /μ S NC /S NC tate for whch ω /ω not nown S A /S A tate avalable (μ /μ ha been calculated for all I / I ) for calculatng ω /ω C S S C tate for whch ν ha been calculated for ta recycle tate for whch we need to perform a forward propagaton In the algorthm, we ue the parameter t to eep trac of how many tme tear tate S T ha been updated. A mentoned n 3.2, f the tear tate updated more than once, the ntance nfeable. General networ ecycle loop ecycle tate Neted loop NO Stop: Infeable YES I t 1 S T? 7. t =t +1 S T.t. ν >ρ I T I + 1. Set ν = S T, I T I + ; t = S T ; and S C ={: S T S + } 2. Set S NC =S, I NC =I, I A =, and S A =S F 3. Calculate ω S A (eqn. 5) I S A S =? YES 4. Calculate ν S A, I + (eqn. 6) NO I ν ρ S T, I T I +? YES NO 5. Set S NC =S NC \S A, S C =S C {: S + S A }, S A =, and I A ={: I NC, S C =S + } YES I S C =? NO 11. Calculate ψ S* (eqn. 16) and ν S* (eqn. 17) I S* + and remove * from S C YES 8. Set S C =S A S 9. Chooe an S C and label t * 1. Set μ = I * SL, ω = S * SL I NC =I * SL, S NC =S * SL, I A =, and S A ={*} I S NC = & I NC =? NO 12. Calculate ω S A (eqn. 1 or 14). Set S NC =S NC \ S A, S A =, and I A ={: I NC, S - S NC = } 13. Calculate μ I A (eqn. 11 or 16). Set I NC =I NC \I A, I A =, and S A ={: S NC, I + I NC = } Stop 6. Calculate μ (eqn. 7) and (eqn. 8) I A. Set I NC =I NC \I A. I A = and S A ={: S NC, I - I NC = } YES I I A = & S A =? Fgure 12. General algorthm for propagatng demand through a networ. NO I S A or S NC =? NO YES 14. Set S A ={: S S NC, (I + I * SL )\I NC } and I A ={: I NC, (S - S * SL )\S NC, S - S * SL >1}. Calculate ξ S A (eqn. 19) 17

18 3.6. State roduced by Multple Ta When multple ta can produce a tate, eqn. 6 gve ν = for each ta, whch may mean μ zero for uptream ta that mut produce ome materal. In Fgure 13, produced from or, o each ta ha a mnmum producton of zero. However, and both requre, whch produced by. If we ued eqn. 6 to fnd ν n tep 4 of the algorthm, we fnd that μ =, whch not a tght lower bound. Intead, we can olve a mple lnear program (L ) to fnd ν. mn Q.t. Q Q I I I Q S NC (L ) where Q a potve varable for the total amount of materal ta produce n a partcular oluton of (L ). The frt contrant requre that, for each tate, the amount produced plu any ntal nventory greater than the amount conumed. The econd contrant enforce that the amount produced of a tate mut exceed ω and only wrtten for tate for whch ω nown. The obectve to mnmze the amount produced by ta. When t tme to fnd ν n the algorthm, we olve (L ) for ta and then multply the optmal obectve value by to get ν. When a ta (other than a tear ta) produce multple tate, we only need to olve (L ) once and multply by the optmal value by for each tate S. For tear ta that produce multple tate, we olve (L ) twce, frt to calculate ν for any nontear tate produced by the ta and econd to update ν for the tear tate. When we ue th method, we olve (L ) to fnd ν for all ta n the networ. Th method alo provde an alternatve to the method for recycle tate decrbed n 3.3 and 3.4. For more complcated networ wth neted loop, th method may gve tghter bound and may be much mpler to ue. Demand tll bacward propagated accordng to the algorthm n Fgure 12 wth two change: (1) we calculate ν n tep 4 of the algorthm wth (L ) ntead of wth eqn. 6, and (2) we p (or anwer ye to) all tep nvolvng recycle tate (tep 8-14). 1% Method ω for ν 9% Eqn. 6 5 (L) Fgure 13. Compare the two method for fndng ν : (1) ung eqn. 6 and (2) olvng (L). Cutomer demand 5g of and all ta are proceed n a unt wth a capacty of -5g. 18

19 4. Vald Inequalte We wrte tghtenng contrant after calculatng and ω for all ta and tate. We fnd the mnmum number of batche proceed by a ta, λ, by dvdng the requred producton by the larget poble ze of a ngle batch of ta and roundng up. max J max (21) The mnmum number of batche provde a lower bound for the um of the agnment varable. J, t X t (22) When multple ta produce a tate, eqn. 22 may not provde a tght bound. Intead, we fnd bound for the mnmum number of batche from all ta producng a tate, κ. Agan, we dvde the requred amount of each tate by the larget poble amount of that tate that can be produced n a ngle batch and round up, max I, J max (23) Now, κ provde a bound for the agnment varable for all ta producng tate. I, J, t X t MT S (24) Eqn. 24 doe not provde new nformaton for tate produced by a ngle ta. Tghtenng contrant 22 and 24 have the ame form a thoe propoed by Burard and Hatzl (25) and Jana and Flouda (28), but the bound from the dfferent method may be dfferent. When a ta can be proceed n unt wth very dfferent capacte, eqn. 22 and 24 may not provde tght bound. Intead, we ue the maxmum batch-ze and mnmum producton requrement to fnd tronger general nequalte. J, t X ˆ max t mn where μ the tghtet bound for eqn. 25. In 3.1.2, we calculate baed on. Snce eqn. 25 baed on, doe not provde a tght bound. We fnd μ by performng another loop over all {1,2,,K } max wth the ame ued to calculate the attanable producton amount. f J J (26) otherwe max max ˆ (25) 19

20 The top expreon et ˆ to the producton amount at the end of attanable range f the range large enough to meet demand. The econd expreon et ˆ to nfnty when the attanable range too mall to meet demand. After checng all range, we et μ to the mallet of all ˆ. ˆ mn ˆ (27) The LHS of eqn. 25 can only tae on a dcrete et of value when X t bnary; thee value are gven by Σα where α an nteger and are the ame a the producton amount at the end of the attanable max range hown n Fgure 7. Therefore all poble value for the LHS occur at the end of an attanable range. To fnd the value of μ that gve the tghtet bound, we ue the producton amount at the end the attanable regon that cloet to but greater than (ee Fgure 14). In general, μ the mallet max producton amount at the end of an attanable range (Σα ) that at leat able to meet demand ( ). Eqn. 25 can alo be wrtten for tate produced by multple ta: I J, t X max t MT S (28) Fgure 15 llutrate the effect of the tghtenng contrant on the feable regon of the Lrelaxaton of the model for the example n Fgure 7 and 14. The total demand 55, μ =55, =6, and μ =75. The lne how eqn. 22 and eqn. 25 wth three HS value: μ,, and μ. Eqn. 22 requre at leat two batche. Improvng (ncreang) the HS of contrant 25 mprove the L-relaxaton of the model. When bacloggng not allowed, we ue due tme to tghten the formulaton further. Now, all parameter nclude a tme ndex (ϕ t, ω t, etc.) and are zero for tme when no order are due. For every due tme, we add all earler order to get ϕ t and calculate the other parameter a before. Intead of ummng the agnment varable over the entre horzon, we um t only over the tme avalable to tart the ta. t J t X t t, t (29) For tme when no order are due, all parameter are zero, and eqn. 29 not wrtten. We can alo wrte contrant 24, 25, and 28 to conder due tme by changng the tme over whch the agnment varable are ummed. For the example n Fgure 1, uppoe cutomer C1 demand 3g of at t=6 and 25g of at t=9, and cutomer C2 demand 6g of at t=9. The total fnal demand up to t=6 and up to t=9 are calculated. Table 2 lt order ze (ϕ t ), requred producton for tate (ω t ) and ta ( t ), and agnment bound (λ t ). Table 2. arameter value when due tme are ued to tghten the formulaton. Agnment t Order (ϕ t ) ω t t Bound (λ t )

21 # of Batche n U2 capacte (mn/max): U1: 2/25 U2: 45/5 α U1 α U2 μ = 55 μ î = 75 Eqn. 26 I demand met? μ î 1 no 2 1 no 3 2 ye no ye no ye ye roducton Amount Fgure 14. Example of μ calculaton. The LHS of eqn. 25 mut belong to the et {, 25, 5, 75 }. Snce the mnmum producton = 6, the LHS of eqn. 25 mut be at leat 75 n any feable oluton Integer ont Eqn. 22 Eqn. 25: Eqn. 25: Eqn. 25: # of Batche n U1 Fgure 15. Effect of tghtenng contrant on feable regon. 5. Computatonal Study 5.1. roblem Intance To determne how well the tghtenng contrant wor, we tet 72 ntance wth dfferent obectve, networ, problem feature, and tme horzon. Each ntance run wth dfferent tghtenng formulaton. Specfcally, we compare the two type of contrant baed on the number of batche (eqn. 22 and 24) and on the mnmum producton (eqn. 25 and 28). We alo loo at the effectvene of the extenon for procee wth due tme. We conder two obectve: mnmzaton of proceng cot and maepan. The proceng cot depend on the ta and unt, but not on the batch-ze. cot t,, J I X t t, MS X t J t (31) Table 3. The even contrant et. 21 (3)

22 Mnmum # of batche (eqn. 22, 24) Mnmum producton (eqn. 25, 28) Due Tme (eqn. 29) Set F1 F2 X F3 X F4 X X F5 X X F6 X X F7 X X X We ue four networ, N1-N4, to determne the effect of recycle tream and tate produced by multple ta on the effectvene of the tghtenng contrant (Fgure A1 n Electronc Companon). roce and order data are alo gven n the Electronc Companon (Table A1-A4). Networ N1 and N4 have no recycle tream, but networ N4 ha a tate produced by multple ta. Networ N2 and N3 have a recycle tream. Networ N2 taen from Kondl et al (1993). We conder three problem clae wth dfferent feature for each networ. roblem (a) ha all order due at the end of the tme horzon. To determne the mpact of ncludng due tme n the tghtenng contrant, problem (b) ha ntermedate due tme. To compare the two type of tghtenng contrant, unt capacte are changed for problem (c) o at leat one ta can be proceed by unt wth dfferent capacte. We olve each ntance wth three tme horzon: 4, 8, and 12 hour. The order ze and due tme cale wth the tme horzon to enure all tme horzon lead to chedule that are mlarly loaded. Startng wth a 4-hour horzon, the order ze and due tme are doubled for an 8-hour horzon and trpled for a 12-hour horzon. We conder even et of tghtenng contrant (Table 3). All formulaton contan eqn Formulaton F1 ha no tghtenng contrant. F2 nclude the contrant baed on the mnmum number of batche. Set F3 contan the contrant baed on the requred producton. F4 combne F2 and F3. Fnally, et F5-F7 are the ame a et F2-F4 but wth due tme. Only problem (b) run wth formulaton F5-F7, and the reult focu on formulaton F1-F4. All problem are olved ung GAMS 23.7/CLEX 12.3 on a computer wth 6 GB of AM and a 2.67 GHz Intel Core (7-92) proceor runnng on Wndow 7. We ue a reource lmt of 18. Model and oluton tattc for all problem are gven n the Electronc Companon (Table A5-A1). For the 36 ntance, mplementng the demand propagaton algorthm tae an average of.26 wth a maxmum tme of 4.3 when ung eqn. 6 to calculate ν, and an average of 2.4 wth a maxmum tme of 11.9 when ung (L ) to calculate ν eult Table 4 how aggregate reult for formulaton F1 and F4. roblem are grouped nto four categore: (1) thoe that are olved to optmalty by both formulaton, (2) thoe that are olved to optmalty only by F4, (3) thoe that are never olved to optmalty, and (4) thoe that are olved to optmalty only by F1. The average computatonal tme or optmalty gap gven for the problem n each of the four categore. For cot mnmzaton, 14 problem are n the frt category, and tghtenng 22

23 (r<) (r<) reduce the computatonal tme from 189 to 1.4. Category 2 contan 2 problem, and the tghtened formulaton olve thee problem n an average of 2.4 whle, wthout tghtenng, there an average optmalty gap of 1.3% after 18. For the two problem n category 3, tghtenng reduce the average optmalty gap from 2.1% to 1.5%. There are no problem n the fnal category. For maepan mnmzaton, 34 problem fall n the frt category, and the average computatonal tme 61 wthout tghtenng and 32 wth tghtenng. There one problem each n categore 2 and 4. Fgure 16 a performance profle for formulaton F1-F4. For each ntance, the computatonal tme for each formulaton dvded by the fatet tme over all formulaton for that ntance to gve r. The vertcal ax the probablty a formulaton wll olve a problem wthn (on the horzontal ax) tme the fatet formulaton. For cot mnmzaton, the untghtened formulaton olve only 17% of the problem wthn 15 tme the fatet formulaton, but F4 olve 94% of the problem wthn the ame tme. Ung the tghtenng contrant baed on the mnmum producton (n ether F3 or F4) gve the bet reult. For maepan mnmzaton, F3 perform the bet, but tghtenng generally le effectve and doe not alway mprove oluton tme. Fgure 17 compare the four formulaton for the two obectve. Each pont the average computatonal tme or optmalty gap over all networ, problem clae, and tme horzon (36 run per pont). The tghtenng contrant baed on the number of batche (F2) the leat effectve, and F3 and F4 are about equvalent. For cot mnmzaton, tghtenng ncreae the fracton of problem olved to optmalty and decreae the computatonal tme, whle for maepan mnmzaton, t doe not have a large mpact on the computatonal tme. Snce maepan mnmzaton appear to be an eaer problem and tghtenng not a effectve, we wll focu on cot mnmzaton for the remander of the ecton. Table 4. Summary of tghtenng reult wth average computatonal tme or optmalty gap. Cot Mnmzaton Maepan Mnmzaton Category # of problem F % 2.1% % F % % F1 F2 F3 F (a) Cot Mnmzaton (b) Maepan Mnmzaton Fgure 16. erformance profle comparng tghtenng formulaton F1-F4. 23

24 Fracton Solved to Optmalty Computatonal Tme () Optmalty Gap (%) Fracton Solved to Optmalty Computatonal Tme () Optmalty Gap (%) Fracton Solved to Optmalty Computatonal Tme () Optmalty Gap (%) Fgure 18 compare the four formulaton for the four networ averaged over the three problem type and three tme horzon (9 run per pont). All tghtened formulaton perform better than F1, and F3 and F4 are the mot effectve. When there are recycle tream, a n N2 and N3, F2 doe not perform a well. Fgure 19 how reult for the three problem clae averaged over the four networ and three tme horzon (12 run per pont). For all problem clae, tghtenng baed on the mnmum number of batche (F2 and F5) le effectve than ung the mnmum producton requrement (F3 and F6), but ung both type of nequalte n F4 and F7 ometme better. When the problem nclude due tme n type (b), addng the due tme to the tghtenng contrant (F5-F7) ha lttle mpact on the oluton tme. Fnally, n Fgure 2 we how reult for the dfferent tme horzon, where each pont an average over all problem clae and networ (12 run per pont). A expected, ncreang the tme horzon decreae the fracton of problem olved to optmalty, but ung tghtenng ncreae the fracton olved o only 8% of the problem olved wth a 4-hour horzon are not olved wth 8- or 12-hour horzon. We alo note that the average computatonal requrement and optmalty gap decreae notably F1 F2 F3 F4 F1 F2 F3 F4 F1 F2 F3 F4 (a) Formulaton (b) Formulaton (c) Formulaton Fgure 17. eult for the two obectve. Cot MS N1 N2 N3 N4 F1 F2 F3 F4 F1 F2 F3 F4 (a) Formulaton (b) Formulaton Fgure 18. eult for four proce networ (cot mnmzaton). F1 F2 F3 F4 (c) Formulaton a 9 b 6 c a b1.5 c a b c F1 F2 F3 F4 F5 F6 F7 F1 F2 F3 F4 F5 F6 F7 (a) Formulaton (b) Formulaton Fgure 19. eult for the three problem clae (cot mnmzaton). F1 F2 F3 F4 F5 F6 F7 (c) Formulaton 24

Solution Methods for Time-indexed MIP Models for Chemical Production Scheduling

Solution Methods for Time-indexed MIP Models for Chemical Production Scheduling Ian Davd Lockhart Bogle and Mchael Farweather (Edtor), Proceedng of the 22nd European Sympoum on Computer Aded Proce Engneerng, 17-2 June 212, London. 212 Elever B.V. All rght reerved. Soluton Method for

More information

Additional File 1 - Detailed explanation of the expression level CPD

Additional File 1 - Detailed explanation of the expression level CPD Addtonal Fle - Detaled explanaton of the expreon level CPD A mentoned n the man text, the man CPD for the uterng model cont of two ndvdual factor: P( level gen P( level gen P ( level gen 2 (.).. CPD factor

More information

Harmonic oscillator approximation

Harmonic oscillator approximation armonc ocllator approxmaton armonc ocllator approxmaton Euaton to be olved We are fndng a mnmum of the functon under the retrcton where W P, P,..., P, Q, Q,..., Q P, P,..., P, Q, Q,..., Q lnwgner functon

More information

Improvements on Waring s Problem

Improvements on Waring s Problem Improvement on Warng Problem L An-Png Bejng, PR Chna apl@nacom Abtract By a new recurve algorthm for the auxlary equaton, n th paper, we wll gve ome mprovement for Warng problem Keyword: Warng Problem,

More information

Pythagorean triples. Leen Noordzij.

Pythagorean triples. Leen Noordzij. Pythagorean trple. Leen Noordz Dr.l.noordz@leennoordz.nl www.leennoordz.me Content A Roadmap for generatng Pythagorean Trple.... Pythagorean Trple.... 3 Dcuon Concluon.... 5 A Roadmap for generatng Pythagorean

More information

Specification -- Assumptions of the Simple Classical Linear Regression Model (CLRM) 1. Introduction

Specification -- Assumptions of the Simple Classical Linear Regression Model (CLRM) 1. Introduction ECONOMICS 35* -- NOTE ECON 35* -- NOTE Specfcaton -- Aumpton of the Smple Clacal Lnear Regreon Model (CLRM). Introducton CLRM tand for the Clacal Lnear Regreon Model. The CLRM alo known a the tandard lnear

More information

Small signal analysis

Small signal analysis Small gnal analy. ntroducton Let u conder the crcut hown n Fg., where the nonlnear retor decrbed by the equaton g v havng graphcal repreentaton hown n Fg.. ( G (t G v(t v Fg. Fg. a D current ource wherea

More information

Chapter 6 The Effect of the GPS Systematic Errors on Deformation Parameters

Chapter 6 The Effect of the GPS Systematic Errors on Deformation Parameters Chapter 6 The Effect of the GPS Sytematc Error on Deformaton Parameter 6.. General Beutler et al., (988) dd the frt comprehenve tudy on the GPS ytematc error. Baed on a geometrc approach and aumng a unform

More information

m = 4 n = 9 W 1 N 1 x 1 R D 4 s x i

m = 4 n = 9 W 1 N 1 x 1 R D 4 s x i GREEDY WIRE-SIZING IS LINEAR TIME Chr C. N. Chu D. F. Wong cnchu@c.utexa.edu wong@c.utexa.edu Department of Computer Scence, Unverty of Texa at Autn, Autn, T 787. ABSTRACT In nterconnect optmzaton by wre-zng,

More information

Two Approaches to Proving. Goldbach s Conjecture

Two Approaches to Proving. Goldbach s Conjecture Two Approache to Provng Goldbach Conecture By Bernard Farley Adved By Charle Parry May 3 rd 5 A Bref Introducton to Goldbach Conecture In 74 Goldbach made h mot famou contrbuton n mathematc wth the conecture

More information

Root Locus Techniques

Root Locus Techniques Root Locu Technque ELEC 32 Cloed-Loop Control The control nput u t ynthezed baed on the a pror knowledge of the ytem plant, the reference nput r t, and the error gnal, e t The control ytem meaure the output,

More information

Improvements on Waring s Problem

Improvements on Waring s Problem Imrovement on Warng Problem L An-Png Bejng 85, PR Chna al@nacom Abtract By a new recurve algorthm for the auxlary equaton, n th aer, we wll gve ome mrovement for Warng roblem Keyword: Warng Problem, Hardy-Lttlewood

More information

Introduction to Interfacial Segregation. Xiaozhe Zhang 10/02/2015

Introduction to Interfacial Segregation. Xiaozhe Zhang 10/02/2015 Introducton to Interfacal Segregaton Xaozhe Zhang 10/02/2015 Interfacal egregaton Segregaton n materal refer to the enrchment of a materal conttuent at a free urface or an nternal nterface of a materal.

More information

Chapter 11. Supplemental Text Material. The method of steepest ascent can be derived as follows. Suppose that we have fit a firstorder

Chapter 11. Supplemental Text Material. The method of steepest ascent can be derived as follows. Suppose that we have fit a firstorder S-. The Method of Steepet cent Chapter. Supplemental Text Materal The method of teepet acent can be derved a follow. Suppoe that we have ft a frtorder model y = β + β x and we wh to ue th model to determne

More information

Structure and Drive Paul A. Jensen Copyright July 20, 2003

Structure and Drive Paul A. Jensen Copyright July 20, 2003 Structure and Drve Paul A. Jensen Copyrght July 20, 2003 A system s made up of several operatons wth flow passng between them. The structure of the system descrbes the flow paths from nputs to outputs.

More information

Kernel Methods and SVMs Extension

Kernel Methods and SVMs Extension Kernel Methods and SVMs Extenson The purpose of ths document s to revew materal covered n Machne Learnng 1 Supervsed Learnng regardng support vector machnes (SVMs). Ths document also provdes a general

More information

Foundations of Arithmetic

Foundations of Arithmetic Foundatons of Arthmetc Notaton We shall denote the sum and product of numbers n the usual notaton as a 2 + a 2 + a 3 + + a = a, a 1 a 2 a 3 a = a The notaton a b means a dvdes b,.e. ac = b where c s an

More information

Problem Set 9 Solutions

Problem Set 9 Solutions Desgn and Analyss of Algorthms May 4, 2015 Massachusetts Insttute of Technology 6.046J/18.410J Profs. Erk Demane, Srn Devadas, and Nancy Lynch Problem Set 9 Solutons Problem Set 9 Solutons Ths problem

More information

Module 5. Cables and Arches. Version 2 CE IIT, Kharagpur

Module 5. Cables and Arches. Version 2 CE IIT, Kharagpur odule 5 Cable and Arche Veron CE IIT, Kharagpur Leon 33 Two-nged Arch Veron CE IIT, Kharagpur Intructonal Objectve: After readng th chapter the tudent wll be able to 1. Compute horzontal reacton n two-hnged

More information

Module 9. Lecture 6. Duality in Assignment Problems

Module 9. Lecture 6. Duality in Assignment Problems Module 9 1 Lecture 6 Dualty n Assgnment Problems In ths lecture we attempt to answer few other mportant questons posed n earler lecture for (AP) and see how some of them can be explaned through the concept

More information

Amiri s Supply Chain Model. System Engineering b Department of Mathematics and Statistics c Odette School of Business

Amiri s Supply Chain Model. System Engineering b Department of Mathematics and Statistics c Odette School of Business Amr s Supply Chan Model by S. Ashtab a,, R.J. Caron b E. Selvarajah c a Department of Industral Manufacturng System Engneerng b Department of Mathematcs Statstcs c Odette School of Busness Unversty of

More information

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 30 Multcollnearty Dr. Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur 2 Remedes for multcollnearty Varous technques have

More information

Team. Outline. Statistics and Art: Sampling, Response Error, Mixed Models, Missing Data, and Inference

Team. Outline. Statistics and Art: Sampling, Response Error, Mixed Models, Missing Data, and Inference Team Stattc and Art: Samplng, Repone Error, Mxed Model, Mng Data, and nference Ed Stanek Unverty of Maachuett- Amhert, USA 9/5/8 9/5/8 Outlne. Example: Doe-repone Model n Toxcology. ow to Predct Realzed

More information

Not at Steady State! Yes! Only if reactions occur! Yes! Ideal Gas, change in temperature or pressure. Yes! Class 15. Is the following possible?

Not at Steady State! Yes! Only if reactions occur! Yes! Ideal Gas, change in temperature or pressure. Yes! Class 15. Is the following possible? Chapter 5-6 (where we are gong) Ideal gae and lqud (today) Dente Partal preure Non-deal gae (next tme) Eqn. of tate Reduced preure and temperature Compreblty chart (z) Vapor-lqud ytem (Ch. 6) Vapor preure

More information

No! Yes! Only if reactions occur! Yes! Ideal Gas, change in temperature or pressure. Survey Results. Class 15. Is the following possible?

No! Yes! Only if reactions occur! Yes! Ideal Gas, change in temperature or pressure. Survey Results. Class 15. Is the following possible? Survey Reult Chapter 5-6 (where we are gong) % of Student 45% 40% 35% 30% 25% 20% 15% 10% 5% 0% Hour Spent on ChE 273 1-2 3-4 5-6 7-8 9-10 11+ Hour/Week 2008 2009 2010 2011 2012 2013 2014 2015 2017 F17

More information

Single-Facility Scheduling over Long Time Horizons by Logic-based Benders Decomposition

Single-Facility Scheduling over Long Time Horizons by Logic-based Benders Decomposition Sngle-Faclty Schedulng over Long Tme Horzons by Logc-based Benders Decomposton Elvn Coban and J. N. Hooker Tepper School of Busness, Carnege Mellon Unversty ecoban@andrew.cmu.edu, john@hooker.tepper.cmu.edu

More information

Lecture 10 Support Vector Machines II

Lecture 10 Support Vector Machines II Lecture 10 Support Vector Machnes II 22 February 2016 Taylor B. Arnold Yale Statstcs STAT 365/665 1/28 Notes: Problem 3 s posted and due ths upcomng Frday There was an early bug n the fake-test data; fxed

More information

HMMT February 2016 February 20, 2016

HMMT February 2016 February 20, 2016 HMMT February 016 February 0, 016 Combnatorcs 1. For postve ntegers n, let S n be the set of ntegers x such that n dstnct lnes, no three concurrent, can dvde a plane nto x regons (for example, S = {3,

More information

Information Acquisition in Global Games of Regime Change (Online Appendix)

Information Acquisition in Global Games of Regime Change (Online Appendix) Informaton Acquton n Global Game of Regme Change (Onlne Appendx) Mchal Szkup and Iabel Trevno Augut 4, 05 Introducton Th appendx contan the proof of all the ntermedate reult that have been omtted from

More information

Winter 2008 CS567 Stochastic Linear/Integer Programming Guest Lecturer: Xu, Huan

Winter 2008 CS567 Stochastic Linear/Integer Programming Guest Lecturer: Xu, Huan Wnter 2008 CS567 Stochastc Lnear/Integer Programmng Guest Lecturer: Xu, Huan Class 2: More Modelng Examples 1 Capacty Expanson Capacty expanson models optmal choces of the tmng and levels of nvestments

More information

AP Statistics Ch 3 Examining Relationships

AP Statistics Ch 3 Examining Relationships Introducton To tud relatonhp between varable, we mut meaure the varable on the ame group of ndvdual. If we thnk a varable ma eplan or even caue change n another varable, then the eplanator varable and

More information

CHAPTER 9 LINEAR MOMENTUM, IMPULSE AND COLLISIONS

CHAPTER 9 LINEAR MOMENTUM, IMPULSE AND COLLISIONS CHAPTER 9 LINEAR MOMENTUM, IMPULSE AND COLLISIONS 103 Phy 1 9.1 Lnear Momentum The prncple o energy conervaton can be ued to olve problem that are harder to olve jut ung Newton law. It ued to decrbe moton

More information

Simultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals

Simultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals Smultaneous Optmzaton of Berth Allocaton, Quay Crane Assgnment and Quay Crane Schedulng Problems n Contaner Termnals Necat Aras, Yavuz Türkoğulları, Z. Caner Taşkın, Kuban Altınel Abstract In ths work,

More information

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming EEL 6266 Power System Operaton and Control Chapter 3 Economc Dspatch Usng Dynamc Programmng Pecewse Lnear Cost Functons Common practce many utltes prefer to represent ther generator cost functons as sngle-

More information

1 GSW Iterative Techniques for y = Ax

1 GSW Iterative Techniques for y = Ax 1 for y = A I m gong to cheat here. here are a lot of teratve technques that can be used to solve the general case of a set of smultaneous equatons (wrtten n the matr form as y = A), but ths chapter sn

More information

Note on EM-training of IBM-model 1

Note on EM-training of IBM-model 1 Note on EM-tranng of IBM-model INF58 Language Technologcal Applcatons, Fall The sldes on ths subject (nf58 6.pdf) ncludng the example seem nsuffcent to gve a good grasp of what s gong on. Hence here are

More information

Scattering of two identical particles in the center-of. of-mass frame. (b)

Scattering of two identical particles in the center-of. of-mass frame. (b) Lecture # November 5 Scatterng of two dentcal partcle Relatvtc Quantum Mechanc: The Klen-Gordon equaton Interpretaton of the Klen-Gordon equaton The Drac equaton Drac repreentaton for the matrce α and

More information

A Simple Inventory System

A Simple Inventory System A Smple Inventory System Lawrence M. Leems and Stephen K. Park, Dscrete-Event Smulaton: A Frst Course, Prentce Hall, 2006 Hu Chen Computer Scence Vrgna State Unversty Petersburg, Vrgna February 8, 2017

More information

Method Of Fundamental Solutions For Modeling Electromagnetic Wave Scattering Problems

Method Of Fundamental Solutions For Modeling Electromagnetic Wave Scattering Problems Internatonal Workhop on MehFree Method 003 1 Method Of Fundamental Soluton For Modelng lectromagnetc Wave Scatterng Problem Der-Lang Young (1) and Jhh-We Ruan (1) Abtract: In th paper we attempt to contruct

More information

Week3, Chapter 4. Position and Displacement. Motion in Two Dimensions. Instantaneous Velocity. Average Velocity

Week3, Chapter 4. Position and Displacement. Motion in Two Dimensions. Instantaneous Velocity. Average Velocity Week3, Chapter 4 Moton n Two Dmensons Lecture Quz A partcle confned to moton along the x axs moves wth constant acceleraton from x =.0 m to x = 8.0 m durng a 1-s tme nterval. The velocty of the partcle

More information

Chapter - 2. Distribution System Power Flow Analysis

Chapter - 2. Distribution System Power Flow Analysis Chapter - 2 Dstrbuton System Power Flow Analyss CHAPTER - 2 Radal Dstrbuton System Load Flow 2.1 Introducton Load flow s an mportant tool [66] for analyzng electrcal power system network performance. Load

More information

Assortment Optimization under MNL

Assortment Optimization under MNL Assortment Optmzaton under MNL Haotan Song Aprl 30, 2017 1 Introducton The assortment optmzaton problem ams to fnd the revenue-maxmzng assortment of products to offer when the prces of products are fxed.

More information

Start Point and Trajectory Analysis for the Minimal Time System Design Algorithm

Start Point and Trajectory Analysis for the Minimal Time System Design Algorithm Start Pont and Trajectory Analy for the Mnmal Tme Sytem Degn Algorthm ALEXANDER ZEMLIAK, PEDRO MIRANDA Department of Phyc and Mathematc Puebla Autonomou Unverty Av San Claudo /n, Puebla, 757 MEXICO Abtract:

More information

ECE559VV Project Report

ECE559VV Project Report ECE559VV Project Report (Supplementary Notes Loc Xuan Bu I. MAX SUM-RATE SCHEDULING: THE UPLINK CASE We have seen (n the presentaton that, for downlnk (broadcast channels, the strategy maxmzng the sum-rate

More information

728. Mechanical and electrical elements in reduction of vibrations

728. Mechanical and electrical elements in reduction of vibrations 78. Mechancal and electrcal element n reducton of vbraton Katarzyna BIAŁAS The Slean Unverty of Technology, Faculty of Mechancal Engneerng Inttute of Engneerng Procee Automaton and Integrated Manufacturng

More information

Design of Recursive Digital Filters IIR

Design of Recursive Digital Filters IIR Degn of Recurve Dgtal Flter IIR The outut from a recurve dgtal flter deend on one or more revou outut value, a well a on nut t nvolve feedbac. A recurve flter ha an nfnte mule reone (IIR). The mulve reone

More information

Variable Structure Control ~ Basics

Variable Structure Control ~ Basics Varable Structure Control ~ Bac Harry G. Kwatny Department of Mechancal Engneerng & Mechanc Drexel Unverty Outlne A prelmnary example VS ytem, ldng mode, reachng Bac of dcontnuou ytem Example: underea

More information

This appendix presents the derivations and proofs omitted from the main text.

This appendix presents the derivations and proofs omitted from the main text. Onlne Appendx A Appendx: Omtted Dervaton and Proof Th appendx preent the dervaton and proof omtted from the man text A Omtted dervaton n Secton Mot of the analy provded n the man text Here, we formally

More information

Outline and Reading. Dynamic Programming. Dynamic Programming revealed. Computing Fibonacci. The General Dynamic Programming Technique

Outline and Reading. Dynamic Programming. Dynamic Programming revealed. Computing Fibonacci. The General Dynamic Programming Technique Outlne and Readng Dynamc Programmng The General Technque ( 5.3.2) -1 Knapsac Problem ( 5.3.3) Matrx Chan-Product ( 5.3.1) Dynamc Programmng verson 1.4 1 Dynamc Programmng verson 1.4 2 Dynamc Programmng

More information

Preemptive scheduling. Disadvantages of preemptions WCET. Preemption indirect costs 19/10/2018. Cache related preemption delay

Preemptive scheduling. Disadvantages of preemptions WCET. Preemption indirect costs 19/10/2018. Cache related preemption delay 19/1/18 Preemptve cedulng Mot o wor on cedulng a been ocued on ully preemptve ytem, becaue tey allow ger reponvene: Preemptve Non Preemptve Dadvantage o preempton However, eac preempton a a cot: ontext

More information

Lecture Notes on Linear Regression

Lecture Notes on Linear Regression Lecture Notes on Lnear Regresson Feng L fl@sdueducn Shandong Unversty, Chna Lnear Regresson Problem In regresson problem, we am at predct a contnuous target value gven an nput feature vector We assume

More information

Estimation of Finite Population Total under PPS Sampling in Presence of Extra Auxiliary Information

Estimation of Finite Population Total under PPS Sampling in Presence of Extra Auxiliary Information Internatonal Journal of Stattc and Analy. ISSN 2248-9959 Volume 6, Number 1 (2016), pp. 9-16 Reearch Inda Publcaton http://www.rpublcaton.com Etmaton of Fnte Populaton Total under PPS Samplng n Preence

More information

Statistical Properties of the OLS Coefficient Estimators. 1. Introduction

Statistical Properties of the OLS Coefficient Estimators. 1. Introduction ECOOMICS 35* -- OTE 4 ECO 35* -- OTE 4 Stattcal Properte of the OLS Coeffcent Etmator Introducton We derved n ote the OLS (Ordnary Leat Square etmator ˆβ j (j, of the regreon coeffcent βj (j, n the mple

More information

On the SO 2 Problem in Thermal Power Plants. 2.Two-steps chemical absorption modeling

On the SO 2 Problem in Thermal Power Plants. 2.Two-steps chemical absorption modeling Internatonal Journal of Engneerng Reearch ISSN:39-689)(onlne),347-53(prnt) Volume No4, Iue No, pp : 557-56 Oct 5 On the SO Problem n Thermal Power Plant Two-tep chemcal aborpton modelng hr Boyadjev, P

More information

Generalized Linear Methods

Generalized Linear Methods Generalzed Lnear Methods 1 Introducton In the Ensemble Methods the general dea s that usng a combnaton of several weak learner one could make a better learner. More formally, assume that we have a set

More information

2.3 Least-Square regressions

2.3 Least-Square regressions .3 Leat-Square regreon Eample.10 How do chldren grow? The pattern of growth vare from chld to chld, o we can bet undertandng the general pattern b followng the average heght of a number of chldren. Here

More information

Computer Control Systems

Computer Control Systems Computer Control ytem In th chapter we preent the element and the bac concept of computercontrolled ytem. The dcretaton and choce of amplng frequency wll be frt examned, followed by a tudy of dcrete-tme

More information

and decompose in cycles of length two

and decompose in cycles of length two Permutaton of Proceedng of the Natona Conference On Undergraduate Reearch (NCUR) 006 Domncan Unverty of Caforna San Rafae, Caforna Apr - 4, 007 that are gven by bnoma and decompoe n cyce of ength two Yeena

More information

Introduction to Vapor/Liquid Equilibrium, part 2. Raoult s Law:

Introduction to Vapor/Liquid Equilibrium, part 2. Raoult s Law: CE304, Sprng 2004 Lecture 4 Introducton to Vapor/Lqud Equlbrum, part 2 Raoult s Law: The smplest model that allows us do VLE calculatons s obtaned when we assume that the vapor phase s an deal gas, and

More information

Problem #1. Known: All required parameters. Schematic: Find: Depth of freezing as function of time. Strategy:

Problem #1. Known: All required parameters. Schematic: Find: Depth of freezing as function of time. Strategy: BEE 3500 013 Prelm Soluton Problem #1 Known: All requred parameter. Schematc: Fnd: Depth of freezng a functon of tme. Strategy: In thee mplfed analy for freezng tme, a wa done n cla for a lab geometry,

More information

MMA and GCMMA two methods for nonlinear optimization

MMA and GCMMA two methods for nonlinear optimization MMA and GCMMA two methods for nonlnear optmzaton Krster Svanberg Optmzaton and Systems Theory, KTH, Stockholm, Sweden. krlle@math.kth.se Ths note descrbes the algorthms used n the author s 2007 mplementatons

More information

Linear Approximation with Regularization and Moving Least Squares

Linear Approximation with Regularization and Moving Least Squares Lnear Approxmaton wth Regularzaton and Movng Least Squares Igor Grešovn May 007 Revson 4.6 (Revson : March 004). 5 4 3 0.5 3 3.5 4 Contents: Lnear Fttng...4. Weghted Least Squares n Functon Approxmaton...

More information

NUMERICAL DIFFERENTIATION

NUMERICAL DIFFERENTIATION NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the

More information

More metrics on cartesian products

More metrics on cartesian products More metrcs on cartesan products If (X, d ) are metrc spaces for 1 n, then n Secton II4 of the lecture notes we defned three metrcs on X whose underlyng topologes are the product topology The purpose of

More information

MULTIPLE REGRESSION ANALYSIS For the Case of Two Regressors

MULTIPLE REGRESSION ANALYSIS For the Case of Two Regressors MULTIPLE REGRESSION ANALYSIS For the Cae of Two Regreor In the followng note, leat-quare etmaton developed for multple regreon problem wth two eplanator varable, here called regreor (uch a n the Fat Food

More information

OPTIMISATION. Introduction Single Variable Unconstrained Optimisation Multivariable Unconstrained Optimisation Linear Programming

OPTIMISATION. Introduction Single Variable Unconstrained Optimisation Multivariable Unconstrained Optimisation Linear Programming OPTIMIATION Introducton ngle Varable Unconstraned Optmsaton Multvarable Unconstraned Optmsaton Lnear Programmng Chapter Optmsaton /. Introducton In an engneerng analss, sometmes etremtes, ether mnmum or

More information

The multivariate Gaussian probability density function for random vector X (X 1,,X ) T. diagonal term of, denoted

The multivariate Gaussian probability density function for random vector X (X 1,,X ) T. diagonal term of, denoted Appendx Proof of heorem he multvarate Gauan probablty denty functon for random vector X (X,,X ) px exp / / x x mean and varance equal to the th dagonal term of, denoted he margnal dtrbuton of X Gauan wth

More information

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009 College of Computer & Informaton Scence Fall 2009 Northeastern Unversty 20 October 2009 CS7880: Algorthmc Power Tools Scrbe: Jan Wen and Laura Poplawsk Lecture Outlne: Prmal-dual schema Network Desgn:

More information

A Novel Approach for Testing Stability of 1-D Recursive Digital Filters Based on Lagrange Multipliers

A Novel Approach for Testing Stability of 1-D Recursive Digital Filters Based on Lagrange Multipliers Amercan Journal of Appled Scence 5 (5: 49-495, 8 ISSN 546-939 8 Scence Publcaton A Novel Approach for Tetng Stablty of -D Recurve Dgtal Flter Baed on Lagrange ultpler KRSanth, NGangatharan and Ponnavakko

More information

CHAPTER X PHASE-CHANGE PROBLEMS

CHAPTER X PHASE-CHANGE PROBLEMS Chapter X Phae-Change Problem December 3, 18 917 CHAPER X PHASE-CHANGE PROBLEMS X.1 Introducton Clacal Stefan Problem Geometry of Phae Change Problem Interface Condton X. Analytcal Soluton for Soldfcaton

More information

Synchronization Protocols. Task Allocation Bin-Packing Heuristics: First-Fit Subtasks assigned in arbitrary order To allocate a new subtask T i,j

Synchronization Protocols. Task Allocation Bin-Packing Heuristics: First-Fit Subtasks assigned in arbitrary order To allocate a new subtask T i,j End-to-End Schedulng Framework 1. Tak allocaton: bnd tak to proceor 2. Synchronzaton protocol: enforce precedence contrant 3. Subdeadlne agnment 4. Schedulablty analy Tak Allocaton Bn-Packng eurtc: Frt-Ft

More information

AGC Introduction

AGC Introduction . Introducton AGC 3 The prmary controller response to a load/generaton mbalance results n generaton adjustment so as to mantan load/generaton balance. However, due to droop, t also results n a non-zero

More information

Annexes. EC.1. Cycle-base move illustration. EC.2. Problem Instances

Annexes. EC.1. Cycle-base move illustration. EC.2. Problem Instances ec Annexes Ths Annex frst llustrates a cycle-based move n the dynamc-block generaton tabu search. It then dsplays the characterstcs of the nstance sets, followed by detaled results of the parametercalbraton

More information

An Interactive Optimisation Tool for Allocation Problems

An Interactive Optimisation Tool for Allocation Problems An Interactve Optmsaton ool for Allocaton Problems Fredr Bonäs, Joam Westerlund and apo Westerlund Process Desgn Laboratory, Faculty of echnology, Åbo Aadem Unversty, uru 20500, Fnland hs paper presents

More information

A NUMERICAL MODELING OF MAGNETIC FIELD PERTURBATED BY THE PRESENCE OF SCHIP S HULL

A NUMERICAL MODELING OF MAGNETIC FIELD PERTURBATED BY THE PRESENCE OF SCHIP S HULL A NUMERCAL MODELNG OF MAGNETC FELD PERTURBATED BY THE PRESENCE OF SCHP S HULL M. Dennah* Z. Abd** * Laboratory Electromagnetc Sytem EMP BP b Ben-Aknoun 606 Alger Algera ** Electronc nttute USTHB Alger

More information

Week 5: Neural Networks

Week 5: Neural Networks Week 5: Neural Networks Instructor: Sergey Levne Neural Networks Summary In the prevous lecture, we saw how we can construct neural networks by extendng logstc regresson. Neural networks consst of multple

More information

An Integrated OR/CP Method for Planning and Scheduling

An Integrated OR/CP Method for Planning and Scheduling An Integrated OR/CP Method for Plannng and Schedulng John Hooer Carnege Mellon Unversty IT Unversty of Copenhagen June 2005 The Problem Allocate tass to facltes. Schedule tass assgned to each faclty. Subect

More information

VQ widely used in coding speech, image, and video

VQ widely used in coding speech, image, and video at Scalar quantzers are specal cases of vector quantzers (VQ): they are constraned to look at one sample at a tme (memoryless) VQ does not have such constrant better RD perfomance expected Source codng

More information

For now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results.

For now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results. Neural Networks : Dervaton compled by Alvn Wan from Professor Jtendra Malk s lecture Ths type of computaton s called deep learnng and s the most popular method for many problems, such as computer vson

More information

find (x): given element x, return the canonical element of the set containing x;

find (x): given element x, return the canonical element of the set containing x; COS 43 Sprng, 009 Dsjont Set Unon Problem: Mantan a collecton of dsjont sets. Two operatons: fnd the set contanng a gven element; unte two sets nto one (destructvely). Approach: Canoncal element method:

More information

Difference Equations

Difference Equations Dfference Equatons c Jan Vrbk 1 Bascs Suppose a sequence of numbers, say a 0,a 1,a,a 3,... s defned by a certan general relatonshp between, say, three consecutve values of the sequence, e.g. a + +3a +1

More information

Affine transformations and convexity

Affine transformations and convexity Affne transformatons and convexty The purpose of ths document s to prove some basc propertes of affne transformatons nvolvng convex sets. Here are a few onlne references for background nformaton: http://math.ucr.edu/

More information

On the U-WPF Acts over Monoids

On the U-WPF Acts over Monoids Journal of cence, Ilamc Republc of Iran 8(4): 33-38 (007) Unverty of Tehran, IN 06-04 http://jcence.ut.ac.r On the U-WPF ct over Monod. Golchn * and H. Mohammadzadeh Department of Mathematc, Unverty of

More information

( ) 1/ 2. ( P SO2 )( P O2 ) 1/ 2.

( ) 1/ 2. ( P SO2 )( P O2 ) 1/ 2. Chemstry 360 Dr. Jean M. Standard Problem Set 9 Solutons. The followng chemcal reacton converts sulfur doxde to sulfur troxde. SO ( g) + O ( g) SO 3 ( l). (a.) Wrte the expresson for K eq for ths reacton.

More information

CHAPTER 17 Amortized Analysis

CHAPTER 17 Amortized Analysis CHAPTER 7 Amortzed Analyss In an amortzed analyss, the tme requred to perform a sequence of data structure operatons s averaged over all the operatons performed. It can be used to show that the average

More information

The optimal delay of the second test is therefore approximately 210 hours earlier than =2.

The optimal delay of the second test is therefore approximately 210 hours earlier than =2. THE IEC 61508 FORMULAS 223 The optmal delay of the second test s therefore approxmately 210 hours earler than =2. 8.4 The IEC 61508 Formulas IEC 61508-6 provdes approxmaton formulas for the PF for smple

More information

Lecture 20: Lift and Project, SDP Duality. Today we will study the Lift and Project method. Then we will prove the SDP duality theorem.

Lecture 20: Lift and Project, SDP Duality. Today we will study the Lift and Project method. Then we will prove the SDP duality theorem. prnceton u. sp 02 cos 598B: algorthms and complexty Lecture 20: Lft and Project, SDP Dualty Lecturer: Sanjeev Arora Scrbe:Yury Makarychev Today we wll study the Lft and Project method. Then we wll prove

More information

Superstructure Optimization of the Olefin Separation Process

Superstructure Optimization of the Olefin Separation Process Supertructure Optmzaton of the Olefn Separaton Proce Sangbum Lee, Jeffery S. Logdon*, Mchael J. oral*, and Ignaco. romann epartment of hemcal ngneerng, arnege Mellon Unverty, Pttburgh, PA15213;*BP, Napervlle,

More information

APPENDIX A Some Linear Algebra

APPENDIX A Some Linear Algebra APPENDIX A Some Lnear Algebra The collecton of m, n matrces A.1 Matrces a 1,1,..., a 1,n A = a m,1,..., a m,n wth real elements a,j s denoted by R m,n. If n = 1 then A s called a column vector. Smlarly,

More information

Calculation of time complexity (3%)

Calculation of time complexity (3%) Problem 1. (30%) Calculaton of tme complexty (3%) Gven n ctes, usng exhaust search to see every result takes O(n!). Calculaton of tme needed to solve the problem (2%) 40 ctes:40! dfferent tours 40 add

More information

Transfer Functions. Convenient representation of a linear, dynamic model. A transfer function (TF) relates one input and one output: ( ) system

Transfer Functions. Convenient representation of a linear, dynamic model. A transfer function (TF) relates one input and one output: ( ) system Transfer Functons Convenent representaton of a lnear, dynamc model. A transfer functon (TF) relates one nput and one output: x t X s y t system Y s The followng termnology s used: x y nput output forcng

More information

SOLUTION MANUAL ENGLISH UNIT PROBLEMS CHAPTER 9 SONNTAG BORGNAKKE VAN WYLEN. FUNDAMENTALS of. Thermodynamics. Sixth Edition

SOLUTION MANUAL ENGLISH UNIT PROBLEMS CHAPTER 9 SONNTAG BORGNAKKE VAN WYLEN. FUNDAMENTALS of. Thermodynamics. Sixth Edition SOLUTION MANUAL ENGLISH UNIT PROBLEMS CHAPTER 9 SONNTAG BORGNAKKE VAN WYLEN FUNDAMENTALS of Thermodynamc Sxth Edton CONTENT SUBSECTION PROB NO. Concept-Study Gude Problem 134-141 Steady Sngle Flow Devce

More information

Physics 111. CQ1: springs. con t. Aristocrat at a fixed angle. Wednesday, 8-9 pm in NSC 118/119 Sunday, 6:30-8 pm in CCLIR 468.

Physics 111. CQ1: springs. con t. Aristocrat at a fixed angle. Wednesday, 8-9 pm in NSC 118/119 Sunday, 6:30-8 pm in CCLIR 468. c Announcement day, ober 8, 004 Ch 8: Ch 10: Work done by orce at an angle Power Rotatonal Knematc angular dplacement angular velocty angular acceleraton Wedneday, 8-9 pm n NSC 118/119 Sunday, 6:30-8 pm

More information

Chapter Newton s Method

Chapter Newton s Method Chapter 9. Newton s Method After readng ths chapter, you should be able to:. Understand how Newton s method s dfferent from the Golden Secton Search method. Understand how Newton s method works 3. Solve

More information

SCENARIO SELECTION PROBLEM IN NORDIC POWER MARKETS. Juha Ojala. Systems Analysis Laboratory, Helsinki University of Technology

SCENARIO SELECTION PROBLEM IN NORDIC POWER MARKETS. Juha Ojala. Systems Analysis Laboratory, Helsinki University of Technology 2.2.2003 SCENARIO SELECTION ROBLEM IN NORDIC OER MARKETS MAT-2.08 INDEENDENT RESEARCH ROJECT IN ALIED MATHEMATICS Juha Oala oala@cc.hut.f Sytem Analy Laboratory, Heln Unverty of Technology ortfolo Management

More information

Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011

Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011 Stanford Unversty CS359G: Graph Parttonng and Expanders Handout 4 Luca Trevsan January 3, 0 Lecture 4 In whch we prove the dffcult drecton of Cheeger s nequalty. As n the past lectures, consder an undrected

More information

OPTIMAL COMPUTING BUDGET ALLOCATION FOR MULTI-OBJECTIVE SIMULATION MODELS. David Goldsman

OPTIMAL COMPUTING BUDGET ALLOCATION FOR MULTI-OBJECTIVE SIMULATION MODELS. David Goldsman Proceedng of the 004 Wnter Smulaton Conference R.G. Ingall, M. D. Roett, J. S. Smth, and B. A. Peter, ed. OPTIMAL COMPUTING BUDGET ALLOCATION FOR MULTI-OBJECTIVE SIMULATION MODELS Loo Hay Lee Ek Peng Chew

More information

8.6 The Complex Number System

8.6 The Complex Number System 8.6 The Complex Number System Earler n the chapter, we mentoned that we cannot have a negatve under a square root, snce the square of any postve or negatve number s always postve. In ths secton we want

More information

Superstructure Optimization of the Olefin Separation Process

Superstructure Optimization of the Olefin Separation Process Supertructure Optmzaton of the Olefn Separaton roce Sangbum Lee Jeffery S. Logdon* Mchael J. oral* and gnaco. romann epartment of hemcal ngneerng arnege Mellon Unverty ttburgh A15213 USA;*B Napervlle L6563

More information

A Hybrid MILP/CP Decomposition Approach for the Continuous Time Scheduling of Multipurpose Batch Plants

A Hybrid MILP/CP Decomposition Approach for the Continuous Time Scheduling of Multipurpose Batch Plants A Hybrd MILP/CP Decomposton Approach for the Contnuous Tme Schedulng of Multpurpose Batch Plants Chrstos T. Maravelas, Ignaco E. Grossmann Carnege Mellon Unversty, Department of Chemcal Engneerng Pttsburgh,

More information