Optmal Jont Replenshment, Delvery and Inventory Management Polces for Pershable Products Leandro C. Coelho Glbert Laporte May 2013 CIRRELT-2013-32 Bureaux de Montréal : Bureaux de Québec : Unversté de Montréal Unversté Laval C.P. 6128, succ. Centre-vlle 2325, de la Terrasse, bureau 2642 Montréal (Québec) Québec (Québec) Canada H3C 3J7 Canada G1V 0A6 Téléphone : 514 343-7575 Téléphone : 418 656-2073 Télécope : 514 343-7121 Télécope : 418 656-2624 www.crrelt.ca
Optmal Jont Replenshment, Delvery and Inventory Management Polces for Pershable Products Leandro C. Coelho 1,2,*, Glbert Laporte 1,3 1 Interunversty Research Centre on Enterprse Networks, Logstcs and Transportaton (CIRRELT) 2 Department of Operatons and Decson Systems, Unversté Laval, 2325, de la Terrasse, Québec, Canada G1V 0A6 3 Department of Management Scences, HEC Montréal, 3000, Côte-Sante-Catherne, Montréal, Canada H3T 2A7 Abstract. In ths paper we analyze the optmal jont decsons of when, how and how much to replensh customers wth products of varyng ages. We dscuss the man features of the problem arsng n the jont replenshment and delvery of pershable products, and we model them under general assumptons. We then solve the problem by means of an exact branch-and-cut algorthm, and we test ts performance on a set of randomly generated nstances. Our algorthm s capable of computng optmal solutons for nstances wth up to 30 customers, three perods, and a maxmum age of two perods for the pershable product. For the unsolved nstances the optmalty gap s always small, less than 1.5% on average for nstances wth up to 50 customers. We also mplement and compare two suboptmal sellng prorty polces wth an optmzed polcy: always sell the oldest avalable tems frst to avod spolage, and always sell the fresher tems frst to ncrease revenue. Keywords. Pershable products, nventory control, replenshment, nventory-routng, vendor-managed nventory, branch-and-cut algorthm. Acknowledgements. Ths work was partly supported by the Natural Scences and Engneerng Research Councl of Canada (NSERC) under grant 39682-10. Ths support s gratefully acknowledged. We also thank Calcul Québec for provdng hgh performance parallel computng facltes. Results and vews expressed n ths publcaton are the sole responsblty of the authors and do not necessarly reflect those of CIRRELT. Les résultats et opnons contenus dans cette publcaton ne reflètent pas nécessarement la poston du CIRRELT et n'engagent pas sa responsablté. * Correspondng author: Leandro.Coelho@crrelt.ca Dépôt légal Bblothèque et Archves natonales du Québec Bblothèque et Archves Canada, 2013 Copyrght Coelho, Laporte and CIRRELT, 2013
1 Introducton Inventory control consttutes an mportant logstcs operaton, especally when products have a lmted shelf lfe. Keepng the rght nventory levels guarantees that the demand s satsfed wthout ncurrng unnecessary holdng or spolage costs. Several nventory control models are avalable [3], many of whch nclude a specfc treatment of pershable products [30]. Problems related to the management of pershable products nventores arse n several areas. Applcatons of nventory control of pershable products nclude blood management and dstrbuton [5, 9, 17, 18, 20, 25, 26, 33], as well as the handlng of radoactve and chemcal materals [1, 11, 36], of food such as dary products, fruts and vegetables [4, 12, 29, 31, 34, 35], and of fashon apparel [28]. Several nventory management models have been specfcally derved for pershable tems, such as the perodc revew wth mnmum and maxmum order quantty of Hajema [15], and the perodc revew wth servce level consderatons of Mnner and Transchel [24]. Revews of the man models and algorthms n ths area can be found n Nahmas [30] and n Karaesmen et al. [19]. A unfed analytcal approach to the management of supply chan networks for tme-senstve products s provded n Nagurney et al. [27]. Effcent delvery plannng can provde further savngs n logstcs operatons. The optmzaton of vehcle routes s one of the most developed felds n operatons research [21]. The ntegraton of nventory control and vehcle routng yelds a complex optmzaton problem called nventory-routng whose am s to mnmze the overall costs related to vehcle routes and nventory control. Recent overvews of the nventory-routng problem (IRP) are those of Andersson et al. [2] and of Coelho et al. [8]. The jont nventory management and dstrbuton of pershable products, whch s the topc of ths paper, gves rse to the pershable nventory-routng problem (PIRP). Nagurney and Masoum [25] and Nagurney et al. [26] studed the dstrbuton and relocaton of human blood n a stochastc demand context, consderng the pershablty and waste of CIRRELT-2013-32 1
blood related to age and to the lmted capacty of blood banks. Hemmelmayr et al. [16] studed the case of blood nventory control wth predetermned fxed routes and stochastc demand. The problem was solved heurstcally by nteger programmng and varable neghborhood search. Gumasta et al. [14] ncorporated transportaton ssues n an nventory control model restrcted to two customers only. Custódo and Olvera [10] proposed a strategcal heurstc analyss of the dstrbuton and nventory control of several frozen groceres wth stochastc demand. Mercer and Tao [23] studed the weekly food dstrbuton problem of a supermarket chan, wthout consderng product age. A theoretcal paper developng a column generaton approach was presented by Le et al. [22] to provde solutons to a PIRP. The optmalty gap was typcally below 10% for nstances wth eght customers and fve perods under the assumptons of fxed shelf lfe and flat value throughout the lfe of the product. Ths paper makes several scentfc contrbutons. We frst classfy and dscuss the man assumptons underlyng the management of pershable products. We then formulate the PIRP as a mxed nteger lnear program (MILP) for the most general case, and we also model t to handle the cases where retalers always sell older tems frst, and where they sell fresher tems frst. We devse an exact branch-and-cut algorthm for the soluton of the varous models. To the best of our knowledge, ths s the frst tme an IRP s modeled and solved exactly under general assumptons n the context of pershable products management. Our models do not requre any assumpton on the shape of the product revenue and nventory cost functons. We also establsh some relatonshps between the PIRP and the mult-product IRP recently studed by the authors [7]. The remander of the paper s organzed as follows. In Secton 2 we provde a formal descrpton of the PIRP. In Secton 3 we present our MILP model and ts two varants just descrbed, ncludng new vald nequaltes. Ths s followed by a decrpton of the branchand-cut algorthm n Secton 4. Computatonal experments are presented n Secton 5. Secton 6 concludes the paper. 2 CIRRELT-2013-32
2 Problem Descrpton The jont replenshment and nventory problem for pershable products s concerned wth the combned optmzaton of delvery routes and nventory control for products havng a transent shelf lfe. These products typcally have an expry date, after whch they are no longer ft for consumpton. Ths s the case of some law-regulated products such as food and drugs, but also of a wde varety of unregulated products whose qualty, appearance or commercal appeal dmnshes over tme, such as flowers, cosmetcs, pant, electronc products or fashon tems. In ths secton we dscuss four man assumptons underlyng the treatment of these knds of products, and we explan how we ncorporate them n our model. Specfcally, we dscuss the types of product pershablty n Secton 2.1, the assumptons governng the nventory holdng costs of these products n Secton 2.2, ther revenue as a functon of age n Secton 2.3, and the management of tems of dfferent ages held n nventory n Secton 2.4. 2.1 Types of product pershablty There exst two man types of pershable products accordng to how they decay [30]. The frst type ncludes products whose value does not change untl a certan date, and then goes down to zero almost mmedately. Ths s the case of products whose utlty eventually ceases to be valued by the customers, such as calendars, year books, electroncs or maps, whch quckly become obsolescent after a gven date or when a new generaton of products enters the market. However, ths s more a case of obsolescence than pershablty. Even though these tems may stll be n perfect condton, they are smply no longer useful. Wthn the same category, we fnd products wth an expry date, such as drugs, yogurt and bottled mlk. These products can be consumed whether they are top fresh or a few days old, but after ther expry date, they are usually deemed unft for consumpton. The second type ncludes products whose qualty or perceved value decays gradually over tme. Typcal examples are fruts, vegetables and flowers. The models ntroduced n CIRRELT-2013-32 3
Secton 3 can handle both types of products wthout any ad hoc modfcaton. 2.2 The mpact of tem age on nventory holdng costs As a rule, the unt nventory holdng cost changes wth respect to the age and value of a product. Ths general assumpton holds, for nstance, for nsurance costs whch are value related. All the varable costs related to the age of the product can be modeled through a sngle parameter, called the unt nventory holdng cost, whch depends on the age of the tem. In some contexts, all tems yeld the same holdng cost, regardless of ther age. Products wth a short shelf lfe usually ft n ths category. In ths case, the holdng cost, whch encompasses all other varable costs, can be captured by a unque nput parameter ndependent of the value and age of the product, whch s the case n most applcatons. 2.3 Revenue of the tem accordng to ts age A parameter that greatly affects the proft yelded by products of dfferent ages s ther perceved value by consumers. Brand new tems usually have a hgher sellng prce, whch decreases over tme accordng to some functon. In ths paper we do not make any specfc assumpton regardng the shape of ths functon. Rather, we assume that the sellng prce s known n advance for each product age. Note that the functon descrbng the relaton between prce and age can be non-lnear, non-contnuous or even non-convex, but t can stll be accommodated by our model, as wll be shown n Secton 3. 2.4 Inventory management polces The fnal assumpton we dscuss relates to the management of tems of dfferent ages held n nventory. It s up to the retaler to decde whch tems to offer to customers, whch wll nfluence the assocated revenue. In such a context, three dfferent sellng prorty polces can be envsaged. The frst one conssts of applyng a fresh frst (FF) polcy by whch 4 CIRRELT-2013-32
the suppler always sells the fresher tems frst. Ths polcy ensures a longer shelf lfe and ncreases utlty for the customers but, at the same tme, yelds a hgher spolage rate. The second polcy s the reverse. Under an old frst (OF) polcy, older tems are sold frst, whch generates less spolage, but also less revenue. The thrd polcy, whch we ntroduce n our model, s more flexble and general, and encompasses these two extremes. The optmzed prorty (OP) polcy lets the model determne whch tems to sell at any gven tme perod n order to maxmze proft. Ths means that dependng on the parameter settngs, one may prefer to spol some tems and sell fresher ones because they generate hgher revenues. In order to llustrate the FF and the OF polces, frst consder the case of bottled mlk havng a lmted shelf lfe. A retaler holds n nventory one unt of old mlk havng a remanng shelf lfe of one day, and one unt of one-day old mlk stll good for several days. The unt revenue s $2. If the retaler apples an FF polcy, he sells hs one-day old mlk today, makng $2 of revenue. Tomorrow, the remanng bottle wll be spoled and he wll make no revenue. The total revenue under the FF polcy s then $2. If, on the other hand, he apples an OF polcy, he sells hs old bottle today, and the newer one tomorrow, makng a total revenue of $4, or twce the revenue acheved under the FF polcy. Now consder the case of flowers, whose value declnes quckly from one day to the next. A one-day old bouquet of flowers generates a revenue of $10, whereas a two-day old bouquet yelds only $4. Under an FF polcy, she wll sell the one-day old flowers today, and nothng tomorrow, makng a total revenue of $10. Under an OF polcy, the retaler wll sell the older flowers today and the other ones tomorrow, achevng a smaller revenue of $8. Note that n these two examples, the OP polcy concdes wth ether the OF or the FF polcy. However, ths s not always the case, namely when the revenue functon s not monotonc wth respect to the age of the product. Consder for example the case of bananas, whch start ther shelf lfe as green products, not yet rpe for consumpton, then turn yellow when they reach ther peak value, and fnally become brown close to ther spolage date. Suppose there are two hands of bananas of each color n nventory. Let CIRRELT-2013-32 5
the revenue be $1.50 for a hand of green bananas, $2 for a yellow hand, and $0.50 for a brown hand. Note how the green product s valued hgher than the brown one, because t wll mature over tme and wll eventually become yellow. For a daly demand of one hand over two perods, the FF polcy yelds a revenue of $3, the OF polcy yelds only $1, but an OP polcy consstng of sellng yellow bananas each day yelds an optmal revenue of $4. If the nventory contans only green and yellow bananas, then the OF and OP polces concde; smlarly, f only yellow and brown bananas are consdered, then the FF and OP polces concde. Thus, the choce of whch of the FF or OF polcy to apply depends on the trade-off between the nventory level and the revenue functons of the product under consderaton. The advantage of the OP polcy s that t does not mpose any constrants on the age of the tems to sell and s able to generate the most general and proftable solutons. We mplement all three polces and we analyze ther trade-offs n the context of proft maxmzaton. 3 Mathematcal Formulatons We now formally descrbe the mathematcal formulaton of PIRP under the assumptons just presented for a sngle product and under the three nventory management polces just descrbed. The case of several products s conceptually smlar, but requres an addtonal ndex [7]. We assume that the routng cost matrx s symmetrc. Thus, we defne the problem on an undrected graph G = (V, E), where V = {0,..., n} s the vertex set and E = {(, j) :, j V, < j} s the edge set. Vertex 0 represents the suppler and the remanng vertces V = V \{0} correspond to n customers. A routng cost c j s assocated wth edge (, j) E. Because of the general assumptons presented n Secton 2, we consder that both the suppler and customers are fully aware of the number of tems n nventory accordng 6 CIRRELT-2013-32
to ther age. Ths s mportant because the sales revenue and nventory holdng costs are affected by the age of the product. The suppler has the choce to delver fresh or aged product tems, and each case yelds dfferent holdng costs. Each customer has a maxmum nventory holdng capacty C, whch cannot be exceeded n any perod of the plannng horzon of length p. At each tme perod t T = {1,..., p}, the suppler receves or produces a fresh quantty r t of the pershable product. We assume the suppler has suffcent nventory to meet the demand of ts customers durng the plannng horzon, and all demand has to be satsfed. At the begnnng of the plannng horzon the decson maker knows the current nventory level of the product at each age held by the suppler and by the customers, and receves nformaton on the demand d t of each customer for each tme perod t. Note agan that, as dscussed n the prevous secton, the demand can be equally satsfed by fresh or aged products, whch wll n turn affect the revenue. As s typcally the case n the IRP lterature [8], we assume that the quantty r t made avalable at the suppler n perod t can be used for delveres to customers n the same perod, and the delvery amount receved by customer n perod t can be used to meet the demand n that perod. A set K = {1,..., K} of vehcles are avalable. We denote by Q k the capacty of vehcle k. Each vehcle can perform at most one route per tme perod, vstng a subset of customers, startng and endng at the suppler s locaton. Also, as n other IRP papers, we do not allow splt delveres,.e., customers receve at most one vehcle vst per perod. The pershable product under consderaton becomes spoled after s perods,.e., the age of the product belongs to a dscrete set S = {0,..., s}. The product s valued accordng to ts age, and the decson maker s aware of the sellng revenue u g of one unt of product of age g. Lkewse, the nventory holdng cost h g n locaton V s a functon of the age g of the product. Ths general representaton allows for flat or varable revenues, and for flat or varable holdng costs dependng on the age and value of the product, thus coverng all stuatons descrbed n Secton 2. The nventory level I t held by customer n perod t comprses tems of dfferent ages. We CIRRELT-2013-32 7
break down ths varable nto I t = I gt, where I gt represents the quantty of product of g S age h n nventory at customer n perod t. Lkewse, we decompose the demand d t nto d gt. g S The am of the problem s to smultaneously construct vehcle routes for each perod and to determne delvery quanttes of products of dfferent ages for each perod and each customer, n order to maxmze the total proft, equal to the sales revenue, mnus the routng and nventory holdng costs. Ths problem s extremely dffcult to solve snce t encompasses several NP-hard problems such as the vehcle routng problem [21] and a number of varants of the classcal IRP [8]. Our MILP model works wth routng varables x kt j equal to the number of tmes edge (, j) s used on the route of vehcle k n perod t. We also use bnary varables y kt equal to one f and only f node s vsted by vehcle k n perod t. Formally, varables I t = g S denote the nventory level at vertex V at the end of perod t T, and d gt denotes the quantty of product of age g used to satsfy the demand of customer n perod t, and we denote by q gkt the quantty of product of age g delvered by vehcle k to customer n perod t. The problem can then be formulated under an OP polcy as follows: I gt (PIRP) maxmze g S t T u g dgt V g S t T h g Igt c j x kt j, (1) (,j) E k K t T subject to I gt 0 = I g 1,t 1 0 V k K q gkt g S\{0} t T (2) I 0t 0 = r t t T (3) I gt = I g 1,t 1 + k K q gkt d gt V g S\{0} t T (4) I 0t = k K q 0kt d 0t V t T (5) 8 CIRRELT-2013-32
g S k K q gkt j V,<j S j S,<j I gt C V t T (6) g S d t = g S q gkt V g S C g S q gkt d gt V t T (7) I g,t 1 V t T (8) C y kt V g S k K t T (9) Q k y0 kt k K t T (10) x kt j + j V,j< x kt j = 2y kt V k K t T (11) x kt j y kt ym kt S V k K t T m S (12) S k K y kt 1 V t T (13) I gt, d gt, qgkt 0 V g S k K t T (14) x kt 0 {0, 1, 2} V k K t T (15) x kt j {0, 1}, j V k K t T (16) y kt {0, 1} V k K t T. (17) The objectve functon (1) maxmzes the total sales revenue, mnus nventory and routng costs. Constrants (2) defne the nventory conservaton condtons for the suppler, agng the product by one unt n each perod. Constrants (3) ensure that the suppler always produces or receves top fresh products. Constrants (4) and (5) defne nventory conservaton and agng of the tems for the customers. Constrants (6) mpose a maxmal nventory capacty at each customer. Constrants (7) state that the demand of each customer n each perod s the sum of product quanttes of dfferent ages. Note that by desgn, any product whose age g s hgher than s s spoled, e.g., t no longer appears n the nventory nor can t be used to satsfy the demand. Constrants (8) and (9) lnk the quanttes delvered to the routng varables. In partcular, they only allow a vehcle to delver products to a customer f a vehcle has been assgned to t. Constrants (10) CIRRELT-2013-32 9
ensure that the vehcle capactes are respected. Constrants (11) and (12) are degree constrants and subtour elmnaton constrants, respectvely. Inequaltes (13) ensure that at most one vehcle vsts each customer n each perod, thus forbddng splt delveres. Constrants (14) (17) enforce ntegralty and non-negatvty condtons on the varables. Ths model can be strengthened through the ncluson of the followng famles of vald nequaltes [6]: y kt x kt 0 2y kt V k K t T (18) x kt j y kt, j V k K t T (19) y kt y kt 0 V k K t T (20) y kt 0 y k 1,t 0 k K\{1} t T (21) j< y k 1,t j V k K\{1} t T. (22) Constrants (18) and (19) enforce the condton that f the suppler s the mmedate successor of a customer n the route of vehcle k n perod t, then must be vsted by the same vehcle. A smlar reasonng s appled to customer j n nequaltes (19). Constrants (20) ensure that the suppler s vsted f any customer s vsted by vehcle k n perod t. When the vehcle fleet s homogeneous, one can break some of the vehcle symmetry by means of constrants (21), thus ensurng that vehcle k cannot leave the depot f vehcle k 1 s not used. Ths symmetry breakng rule s then extended to the customer vertces by constrants (22) whch state that f customer s assgned to vehcle k n perod t, then vehcle k 1 must serve a customer wth an ndex smaller than n the same perod. We also ntroduce addtonal cuts n order to strengthen ths formulaton. If the sum of the demands over [t 1, t 2 ] s at least equal to the maxmum possble nventory held, then there must be at least one vst to ths customer n the nterval [t 1, t 2 ]. Ths constrant can be strengthened by consderng that f the quantty needed to satsfy future demands 10 CIRRELT-2013-32
s larger than the maxmum nventory capacty, then several vsts are needed. Snce the maxmum delvery sze s the mnmum between the holdng capacty and the maxmum vehcle capacty, one can round up the rght-hand sde of (23). Makng the numerator tghter by consderng the actual nventory nstead of the maxmum possble nventory yelds nequaltes (24), whch cannot be rounded up because they would then become non-lnear due to the presence of the I t 1 varable n ther rght-hand sde: t 2 y kt k K t =t 1 t 2 d t C t =t 1 mn{max k {Q k }, C } V t 1, t 2 T, t 2 t 1 (23) t 2 y kt k K t =t 1 t 2 d t I t 1 t =t 1 mn{max k {Q k }, C } V t 1, t 2 T, t 2 t 1. (24) A dfferent verson of the same nequaltes can be wrtten as follows. It s related to whether the nventory hold at each perod s suffcent to fulfll future demands. In partcular, f the nventory held n perod t 1 by customer s not suffcent to fulfll future demands, then a vst to ths customer must take place n the nterval [t 1, t 2 ]. Ths condton can be enforced by the followng set of nequaltes: t 2 y kt k K t =t 1 t 2 d t I t 1 t =t 1 t 2 d t t =t 1 V t 1, t 2 T, t 2 t 1. (25) Even f these nequaltes are redundant for our model, they are useful n helpng CPLEX generate new cuts. It s relevant to note that ths model dstngushes tems of dfferent ages through the use of ndex g. The varables have a meanng smlar to those of the mult-product IRP [7]. In the case of a sngle pershable product, the model works as f products of dfferent ages are dfferent from each other (through ther ndex) and have dfferent profts, but contrary CIRRELT-2013-32 11
to what happens n the mult-product case, any of these products can be used to satsfy the same demand. Another partcularty of ths model s that at each perod, an tem transforms tself nto another one through the process of agng. Thus, our problem shares some features of the mult-product problem [7], but t s structurally dfferent from t. 3.1 Modelng an FF polcy We now show how the formulaton just descrbed can be used to solve the problem under an FF polcy under whch the retaler sells fresher tems frst. We add extra varables and constrants to the PIRP formulaton n order to restrct the choce of products age to be sold. We mplement ths dea as follows. We frst ntroduce new bnary varables L gt equal to one f and only f products of age g can be used to satsfy the demand of customer n perod t. The frst set of new constrants restrcts the use of varables d gt,.e., the use of products of age g to satsfy the demand of customer n perod t, only to those products allowed by the respectve L gt varables, that s: d gt U L gt V g S t T. (26) We also order the new varables n ncreasng order of age ndex. The followng set of constrants allows sellng products of age g + 1 only f products of age g have been used to satsfy the demand of customer n perod t: L gt L g+1,t V g S\{s} t T. (27) We then mpose the followng constrants to dsallow the use of older products f there exsts enough nventory of fresher products. The use of products of age g + 1 s allowed f and only f the total nventory of products of ages g, g 1,..., 0 s nsuffcent to satsfy the demand of customer n perod t. Ths can be enforced through the followng constrants: 12 CIRRELT-2013-32
U (1 L g+1,t ) g j=0 I jt + g q jkt d t + 1 V g S\{s} t T. (28) j=0 k K 3.2 Modelng an OF polcy It s straghtforward to model the OF polcy from the constrants developed for the FF case. Ths polcy can be enforced by consderng the same L gt three sets of constrants: varables and the followng d gt U L gt V g S t T. (29) Constrants (29) only allow the use of nventory of age g to satsfy the demand f ts assocated L gt varable s set to one. Then, we also rank the L gt varables n ncreasng order of age ndex. The followng set of constrants allow sellng products of age g 1 only f products of age g are beng used to satsfy the demand of customer n perod t: L g 1,t L g,t V g S\{0} t T. (30) Fnally, we force some of the L varables to take value zero by addng the followng constrants to the model. If the total nventory avalable of ages {g, g + 1,..., s} s suffcent to satsfy the demand of customer n perod t, then the rght-hand sde of nequaltes (31) s postve, whch n turn guarantees that L g 1,t wll take value zero: U (1 L g 1,t ) s j=g I jt + s q jkt d t + 1 V g S\{0} t T. (31) j=g k K 4 Branch-and-Cut Algorthm For very small nstances szes, the model presented n Secton 3 can be fully descrbed and all constrants and varables generated. It can then be solved by feedng t drectly nto an CIRRELT-2013-32 13
nteger lnear programmng solver. However, for nstances of realstc szes, the number of subtour elmnaton constrants (12) s too large to allow full enumeraton and these must be dynamcally generated throughout the search process. The exact algorthm we present s then a classcal branch-and-cut scheme n whch subtour elmnatons constrants are only generated and ncorporated nto the program whenever they are found to be volated. It works as follows. At a generc node of the search tree, a lnear program contanng a subset of the subtour elmnaton constrants s solved, a search for volated nequaltes s performed, and some of these are added to the current program whch s then reoptmzed. Ths process s reterated untl a feasble or domnated soluton s reached, or untl there are no more cuts to be added. At ths pont, branchng on a fractonal varable occurs. We provde a sketch of the branch-and-bound-and-cut scheme n Algorthm 1. 5 Computatonal Experments In order to evaluate the proposed algorthm, we have coded t n C++ and used IBM Concert Technology and CPLEX 12.5 runnng n parallel wth two threads. All computatons were executed on a grd of Intel Xeon processors runnng at 2.66 GHz wth up to 24 GB of RAM nstalled per node, wth the Scentfc Lnux 6.1 operatng system. 5.1 Instances generaton We have created randomly generated nstances to assess the performance of our algorthm on a wde range of stuatons. We have generated a total of 60 dfferent nstances whch vary n terms of the number of customers, perods, vehcles and maxmum age of the product. Our testbed s composed of nstances generated wth the followng parameters: Number of customers n: 10, 20, 30, 40, 50; Number of perods H: 3 for up to n = 50; 6 for up to n = 40; and 10 for up to n = 30; 14 CIRRELT-2013-32
Algorthm 1 Branch-and-cut algorthm 1: At the root node of the search tree, generate and nsert all vald nequaltes nto the program. 2: Subproblem soluton. Solve the LP relaxaton of the current node. 3: Termnaton check: 4: f there are no more nodes to evaluate then 5: Stop. 6: else 7: Select one node from the branch-and-cut tree. 8: end f 9: whle the soluton of the current LP relaxaton contans subtours do 10: Identfy connected components as n Padberg and Rnald [32]. 11: Determne whether the component contanng the suppler s weakly connected as n Gendreau et al. [13]. 12: Add all volated subtour elmnaton constrants (12). 13: Subproblem soluton. Solve the LP relaxaton of the current node. 14: end whle 15: f the soluton of the current LP relaxaton s nteger then 16: Go to the termnaton check. 17: else 18: Branchng: branch on one of the fractonal varables. 19: Go to the termnaton check. 20: end f CIRRELT-2013-32 15
Number of vehcles K: 1 for n = 10; 2 for n = 20 and 30; 3 for n = 40 and 50; Maxmum age of the products s: 2 for H = 3; 3 for H = 6; 5 for H = 10; Demand d t : randomly selected from the nterval [30, 210]; Poston (x, y) of the suppler and customers: randomly selected from the nterval [0, 1000]; Customers nventory capacty C : R max t {d t }, where R s randomly selected from the set {2, 3}; Intal nventory I 0 of fresh products: equal to C d 1 ; Revenue u g : equal to R 1 (R 1 R 2 ) g/s, where R 1 and R 2 are randomly selected from the ntervals [10, 20] and [4, 7], respectvely; Inventory holdng cost h g : equal to (R 1 + gr 2 / (1 + g)) /100, where R 1 and R 2 are randomly selected from the ntervals [0, 100] and [0, 70], respectvely; Vehcle capactes Q k : equal to 1.25 d t /(HK). V t T For each combnaton of the n, s, K and H parameters we have generated fve nstances, yeldng a total of 60 nstances. In what follows we provde average statstcs over fve nstances per combnaton. Detaled results are presented n Appendx A. These results along wth the nstances are also avalable for download from http://www.leandro- coelho.com. 5.2 Solutons for an OP polcy We provde n Table 1 average computatonal results for these nstances under the OP polcy. We have allowed the algorthm to run for a maxmum of two hours. When the tme lmt s reached, we report the best avalable lower and upper bound (soluton value) and 16 CIRRELT-2013-32
the optmalty gap. We report the nstance szes as (n-s-k-h), where n s the number of customers, s s the maxmum age of the product, K s the number of vehcles, and H s the length of the plannng horzon. The next columns report the average best soluton value obtaned, the average best bound, the average optmalty gap, the number of nstances out of the fve that were solved to optmalty, and the average runnng tme n seconds. Table 1: Summary of the computatonal results for the PIRP under the OP polcy Instance sze Best known Best known (n-s-k-h ) soluton value upper bound Gap (%) # solved Tme (s) PIRP-10-2-1-3 31529.90 31529.90 0.00 5/5 0.4 PIRP-10-3-1-6 61684.44 61684.44 0.00 5/5 2.4 PIRP-10-5-1-10 81094.96 81094.96 0.00 5/5 210.2 PIRP-20-2-2-3 62936.24 62936.24 0.00 5/5 27.8 PIRP-20-3-2-6 126736.20 128894.4 0 1.75 0/5 7200.6 PIRP-20-5-2-10 180919.00 186553.20 3.30 0/5 7201.4 PIRP-30-2-2-3 97580.90 97580.90 0.00 5/5 322.0 PIRP-30-3-2-6 192817.80 196322.20 1.79 0/5 7201.0 PIRP-30-5-2-10 294582.2 0 300742.00 2.17 0/5 7201.4 PIRP-40-2-3-3 127961.6 0 129832.00 1.45 0/5 7201.4 PIRP-40-3-3-6 250435.8 0 258103.4 0 3.10 0/5 7201.2 PIRP-50-2-3-3 177157.4 0 179724.40 1.46 0/5 7201.8 These results clearly ndcate that the performance of the algorthm s drectly related to the number n of customers and to the length H of the plannng horzon. For the nstances wth shorter plannng horzons (H = 3), the algorthm s always able to fnd optmal solutons wthn a few seconds of computatonal tme. Ths remans true even when the number of customers and vehcles ncreases, e.g., all fve nstances wth 30 customers and three perods were solved to optmalty, takng on average fve mnutes. Larger nstances wth up to 40 and 50 customers also wth three perods were solved wth a gap of less than 1.50% on average. CIRRELT-2013-32 17
5.3 Solutons for an FF and an OF polcy We also compare the soluton cost of the optmzed polcy wth respect to the age of the products sold wth the cost of the alternatve FF and OF polces. We frst consder the FF polcy whch maxmzes the revenue by always sellng fresher tems. Ths polcy, on the other hand, leads to more spolage, whch n turn ncreases the need for more delveres, thus ncreasng dstrbutons costs. The results are shown n Table 2 as percentages representng the proft decrease of the FF polcy wth respect to the OP polcy. We also report the optmalty gap, the number of nstances solved optmally, and the runnng tme n seconds. We note that the dffculty of solvng the PIRP under an FF polcy s smlar to that observed for the OP polcy, and the proft s only slghtly lower. Fnally, we provde the same comparson wth respect to the OF polcy. The summary of the results are shown n Table 3. Table 2: Summary of the computatonal results for the PIRP under an FF polcy Instance sze (n-s-k-h ) % decrease Opt gap (%) # solved Tme (s) PIRP-10-2-1-3 0.00 0.00 5/5 0.6 PIRP-10-3-1-6 0.17 0.00 5/5 3.2 PIRP-10-5-1-10 0.51 0.64 3/5 3251.0 PIRP-20-2-2-3 0.01 0.00 5/5 50.6 PIRP-20-3-2-6 0.14 1.97 0/5 7200.4 PIRP-20-5-2-10 0.10 3.42 0/5 7202.4 PIRP-30-2-2-3 0.12 0.33 4/5 1526.0 PIRP-30-3-2-6 0.01 1.83 0/5 7201.2 PIRP-30-5-2-10 0.35 2.45 0/5 7202.6 PIRP-40-2-3-3 0.14 1.62 0/5 7201.0 PIRP-40-3-3-6 0.25 3.38 0/5 7202.8 PIRP-50-2-3-3 0.35 1.89 0/5 7202.6 18 CIRRELT-2013-32
Table 3: Summary of the computatonal results for the PIRP under an OF polcy Instance sze (n-s-k-h ) % decrease Opt gap (%) # solved Tme (s) PIRP-10-2-1-3 13.91 0.00 5/5 0.2 PIRP-10-3-1-6 14.99 0.00 5/5 35.8 PIRP-10-5-1-10 10.43 0.85 3/5 3638.6 PIRP-20-2-2-3 18.84 0.00 5/5 6.0 PIRP-20-3-2-6 11.94 2.10 1/5 6622.0 PIRP-20-5-2-10 8.64 4.91 0/5 7201.6 PIRP-30-2-2-3 18.09 0.00 5/5 40.4 PIRP-30-3-2-6 9.97 1.76 0/5 7201.6 PIRP-30-5-2-10 7.95 3.22 0/5 7202.0 PIRP-40-2-3-3 16.09 0.60 2/5 6249.4 PIRP-40-3-3-6 9.56 3.36 0/5 7202.0 PIRP-50-2-3-3 16.53 1.38 0/5 7202.8 CIRRELT-2013-32 19
As was the case of the FF polcy, the dffculty of obtanng optmal and quas-optmal solutons s not affected by the ncluson of the new bnary varables and the new constrants. However, unlke the prevous polcy, the effect on cost of sellng older tems frst, thus dervng lower revenues, has a major effect on the total proft observed, whch decreases substantally over all nstances. 5.4 Solutons for alternatve revenue functons In order to assess the trade-off between the OP, FF and OF polces, we have changed how the product revenue vares lnearly as a functon of age. We have generated three varatons. In the frst mld scenaro, the dfference n cost between fresh and old products s reduced. In the second steep scenaro, the dfference s ncreased. Fnally, we have also created a flat scenaro case n whch the revenue of the product s constant as a functon of age. These three scenaros and the base case are depcted n Fgure 1. The slopes of the lnear functons n ncreasng order are equal to 2.4, 1.8 and 1.2. 18" Alterna2ve&revenue&func2ons& 15" Value&of&the&product& 12" 9" 6" 3" Flat" Mld" Base"case" Steep" 0" 0" 1" 2" 3" 4" 5" Age&of&the&product& Fgure 1: Four alternatve revenue functons We have desgned the followng experments n order to evaluate the mpact of these 20 CIRRELT-2013-32
changes n the trade-off between the dfferent polces. We have selected all 30 nstances contanng 10 and 20 customers. Each nstance was solved under the three polces and under the three alternatve revenue functons. In Table 4 we report the percentage decrease n proft wth respect to the optmzed polcy for each of the revenue functons consdered. Table 4: Percentage decrease n proft when usng alternatve revenue functons Instance sze FF polcy OF polcy (n-s-k-h ) Base case Mld Steep Flat Base case Mld Steep Flat PIRP-10-2-1-3 0.00 0.00 0.00 0.00 13.91 21.37 10.76 0.01 PIRP-10-3-1-6 0.17 0.00 0.00 0.02 14.99 19.69 10.52 0.46 PIRP-10-5-1-10 0.51 0.00 0.00 0.35 10.43 16.27 9.19 1.59 PIRP-20-2-2-3 0.01 0.00 0.00 0.00 18.84 25.02 14.01 0.11 PIRP-20-3-2-6 0.14 0.01 0.11 0.12 11.94 15.83 8.59 0.80 PIRP-20-5-2-10 0.10 0.05 0.03 0.40 8.64 11.22 7.44 1.58 Fnally, to better understand how dfferent revenues for products of dfferent ages affect the trade-off between each of the three polces, we have conducted the followng experments. We have selected one nstance (PIRP-10-5-1-10-1) and we have solved t usng the three polces for several slopes of the revenue functons. Specfcally, we have set the revenue of a fresh product to 20 and we have set the revenue of the oldest tem rangng from zero to 20, n steps of one unt. We have then plotted the values of the objectve functons of each one n the graph of Fgure 2. These new sets of experments confrm that on our data set the FF polcy provdes soluton values that are almost dentcal to the OP polcy. Note how the thn contnuous lne of the OP polcy s only slghtly hgher than the dotted lne of the FF polcy, but vsually ndstngushable from t. Ths mples that here the optmal polcy tends to favor the sale of fresher products. The OF polcy, on the other hand, provdes solutons whose cost s greatly affected by the revenue value of older products. The dfference between the polces s largest when these products are valued very low. When the revenue value for CIRRELT-2013-32 21
160000# Varable'revenue'func%ons' 150000# Solu%on'value' 140000# 130000# 120000# OP#polcy# OF#polcy# FF#polcy# 110000# 0# 0.1# 0.2# 0.3# 0.4# 0.5# 0.6# 0.7# 0.8# 0.9# 1# Revenue'for'one'unt'of'the'oldest'tem'as'a'frac%on'of'the'revenue'of'a'fresh'tem' (low'='steep'revenue'func%on,'hgh'='mld'revenue'func%on)' Fgure 2: Varable revenue functons. The horzontal axs ndcate the revenue for one unt of the oldest tem as a fracton of the revenue of a fresh tem. Low values on the horzontal axs ndcate a steep revenue functon wth respect to the age of the products. Hgh values on the horzontal axs ndcate a mld revenue functon wth respect to the age of the products. 22 CIRRELT-2013-32
older products ncreases, so does ncrease the proft of applyng an OF polcy, and the dfference between ths polcy and the other two tends to vansh. 6 Conclusons We have ntroduced the jont replenshment and nventory control of pershable products. We have modeled the problem under general assumptons as a MILP, and we have solved t exactly by branch-and-cut. We have also ntroduced, modeled and solved exactly two varants of the problem defned by applyng the OF and the FF sellng prorty polces, n whch the retaler sells wth hgher prorty older and fresher tems, respectvely. Our model remans lnear even when the product revenue decreases n a non-lnear or even n a non-convex fashon over tme. It keeps track of the number of tems of each age, and consders dfferent holdng costs for products of dfferent ages. The model optmally determnes whch tems to sell at each perod based on the trade-off between cost and revenue. The algorthm can effectvely compute optmal jont replenshment and delvery decsons for pershable products n an nventory-routng context for medum sze nstances. We have also shown that on our testbed, the proft s drastcally reduced when an OF polcy s appled, but the decrease s only margnal under an FF polcy. Extensve computatonal experments carred out on randomly generated nstances support these conclusons. Acknowledgments Ths work was partly supported by the Canadan Natural Scences and Engneerng Research Councl under grant 39682-10. Ths support s gratefully acknowledged. We also thank Calcul Québec for provdng hgh performance parallel computng facltes. CIRRELT-2013-32 23
Appendx A. Detaled computatonal results for the OP, FF and OF polces We present n Table 5 the detaled computatonal results for all nstances under the OP, the FF and the OF polces. References [1] V. Aelon, F. Castells, and A. Verouts. Lfe cycle nventory analyss of chemcal processes. Envronmental Progress, 14(3):193 200, 2006. [2] H. Andersson, A. Hoff, M. Chrstansen, G. Hasle, and A. Løkketangen. Industral aspects and lterature survey: Combned nventory management and routng. Computers & Operatons Research, 37(9):1515 1536, 2010. [3] S. Axsäter. Inventory control. Internatonal Seres n Operatons Research & Management Scence, volume 90. Sprnger, New York, 2006. [4] J. Blackburn and G. Scudder. Supply chan strateges for pershable products: The case of fresh produce. Producton and Operatons Management, 18(2):129 137, 2009. [5] E. Brodhem, C. Derman, and G. P. Prastacos. On the evaluaton of a class of nventory polces for pershable products such as blood. Management Scence, 21(11):1320 1325, 1975. [6] L. C. Coelho and G. Laporte. The exact soluton of several classes of nventory-routng problems. Computers & Operatons Research, 40(2):558 565, 2013. [7] L. C. Coelho and G. Laporte. A branch-and-cut algorthm for the mult-product mult-vehcle nventory-routng problem. Internatonal Journal of Producton Research, forthcomng, 2013. do: 10.1080/00207543.2012.757668. [8] L. C. Coelho, J.-F. Cordeau, and G. Laporte. Thrty years of nventory-routng. Transportaton Scence, forthcomng, 2013. [9] P. D. Cummng, K. E. Kendall, C. C. Pegels, J. P. Seagle, and J. F. Shubsda. A collectons plannng model for regonal blood supplers: descrpton and valdaton. Management Scence, 22(9):962 971, 1976. 24 CIRRELT-2013-32
Table 5: Detaled results of the computatonal experments for the PIRP OP polcy FF polcy OF polcy Instance sze Best known Best known Best known Best known Best known Best known Gap (%) Tme (s) Gap (%) Tme (s) (n-s-k-h ) soluton value upper bound soluton value upper bound soluton value upper bound Gap (%) Tme (s) PIRP-10-2-1-3-1 28863.4 28863.4 0.00 1 28863.4 28863.4 0.00 1 26221.0 26221.0 0.00 1 PIRP-10-2-1-3-2 34368.0 34368.0 0.00 1 34368.0 34368.0 0.00 1 29637.7 29637.7 0.00 0 PIRP-10-2-1-3-3 27895.7 27895.7 0.00 0 27895.7 27895.7 0.00 1 23062.5 23062.5 0.00 0 PIRP-10-2-1-3-4 33688.1 33688.1 0.00 0 33688.1 33688.1 0.00 0 30923.3 30923.3 0.00 0 PIRP-10-2-1-3-5 32834.3 32834.3 0.00 0 32834.3 32834.3 0.00 0 25874.1 25874.1 0.00 0 PIRP-10-3-1-6-1 67552.8 67552.8 0.00 1 67542.2 67542.2 0.00 1 60057.5 60057.5 0.00 5 PIRP-10-3-1-6-2 53367.7 53367.7 0.00 1 53330.5 53330.5 0.00 2 43119.8 43119.8 0.00 1 PIRP-10-3-1-6-3 67946.0 67946.0 0.00 3 67908.3 67908.3 0.00 4 59064.9 59064.9 0.00 7 PIRP-10-3-1-6-4 65375.6 65375.6 0.00 2 64918.2 64918.2 0.00 5 55391.3 55391.3 0.00 8 PIRP-10-3-1-6-5 54180.1 54180.1 0.00 5 54176.5 54176.5 0.00 4 44555.6 44555.6 0.00 158 PIRP-10-5-1-10-1 80471.9 80471.9 0.00 30 79740.7 79740.7 0.00 641 69734.9 69734.9 0.00 2000 PIRP-10-5-1-10-2 72194.8 72194.8 0.00 205 72149.4 72149.4 0.00 1018 66685.6 67738.8 1.57 7201 PIRP-10-5-1-10-3 101043.0 101043.0 0.00 427 100508.0 102792.0 2.27 7200 96025.7 96025.7 0.00 687 PIRP-10-5-1-10-4 82829.4 82829.4 0.00 28 82336.3 82336.3 0.00 196 73296.0 73296.0 0.00 1104 PIRP-10-5-1-10-5 68935.7 68935.7 0.00 361 68657.0 69301.6 0.93 7200 57416.6 58956.7 2.68 7201 PIRP-20-2-2-3-1 61780.2 61780.2 0.00 24 61780.2 61780.2 0.00 136 50548.4 50548.4 0.00 16 PIRP-20-2-2-3-2 75757.3 75757.3 0.00 1 75753.0 75753.0 0.00 2 64271.6 64271.6 0.00 2 PIRP-20-2-2-3-3 72546.5 72546.5 0.00 97 72546.5 72546.5 0.00 76 62656.7 62656.7 0.00 7 PIRP-20-2-2-3-4 52850.8 52850.8 0.00 14 52842.2 52842.2 0.00 36 42008.2 42008.2 0.00 4 PIRP-20-2-2-3-5 51746.4 51746.4 0.00 3 51746.4 51746.4 0.00 3 35895.6 35895.6 0.00 1 PIRP-20-3-2-6-1 110437.0 112517.0 1.88 7200 110343.0 112340.0 1.81 7200 87287.3 90277.8 3.42 7200 PIRP-20-3-2-6-2 133377.0 136382.0 2.25 7200 133126.0 136980.0 2.89 7201 117342.0 120768.0 2.91 7200 PIRP-20-3-2-6-3 106120.0 108651.0 2.38 7202 106033.0 108735.0 2.54 7200 91104.1 94204.4 3.40 7202 PIRP-20-3-2-6-4 135267.0 137210.0 1.43 7200 134850.0 137238.0 1.77 7201 122779.0 122779.0 0.00 4306 PIRP-20-3-2-6-5 148480.0 149712.0 0.83 7201 148395.0 149662.0 0.85 7200 139502.0 140613.0 0.79 7202 PIRP-20-5-2-10-1 200786.0 206053.0 2.62 7202 200646.0 206351.0 2.84 7201 183235.0 191178.0 4.33 7202 PIRP-20-5-2-10-2 152951.0 161040.0 5.28 7201 153008.0 160525.0 4.91 7201 132874.0 141960.0 6.83 7202 PIRP-20-5-2-10-3 182710.0 188156.0 2.98 7200 182683.0 188459.0 3.16 7202 172490.0 179249.0 3.91 7201 PIRP-20-5-2-10-4 146990.0 153093.0 4.15 7202 146316.0 153154.0 4.67 7204 133093.0 141996.0 6.68 7201 PIRP-20-5-2-10-5 221158.0 224424.0 1.47 7202 221003.0 224387.0 1.53 7204 204667.0 210435.0 2.81 7202 PIRP-30-2-2-3-1 85251.9 85251.9 0.00 1101 84740.0 86178.2 1.69 7202 71288.4 71288.4 0.00 137 PIRP-30-2-2-3-2 94711.4 94711.4 0.00 114 94633.3 94633.3 0.00 118 75580.3 75580.3 0.00 13 PIRP-30-2-2-3-3 99037.0 99037.0 0.00 46 99037.0 99037.0 0.00 41 77017.0 77017.0 0.00 24 PIRP-30-2-2-3-4 113737.0 113737.0 0.00 12 113737.0 113737.0 0.00 80 91090.1 91090.1 0.00 13 PIRP-30-2-2-3-5 95167.2 95167.2 0.00 337 95140.5 95140.5 0.00 189 84626.6 84626.6 0.00 15 PIRP-30-3-2-6-1 190666.0 196082.0 2.84 7201 190788.0 196539.0 3.01 7201 176515.0 181188.0 2.64 7202 PIRP-30-3-2-6-2 195358.0 196565.0 0.61 7200 195318.0 196563.0 0.63 7201 177796.0 179130.0 0.75 7201 PIRP-30-3-2-6-3 185507.0 188220.0 1.46 7200 185153.0 188306.0 1.70 7201 166860.0 169568.0 1.62 7202 PIRP-30-3-2-6-4 174064.0 176545.0 1.42 7201 174029.0 176540.0 1.44 7202 147141.0 149654.0 1.70 7201 PIRP-30-3-2-6-5 218494.0 224199.0 2.61 7203 218991.0 224188.0 2.37 7201 199631.0 203846.0 2.11 7202 PIRP-30-5-2-10-1 232289.0 238098.0 2.50 7201 230669.0 237852.0 3.11 7202 210635.0 219618.0 4.26 7201 PIRP-30-5-2-10-2 257061.0 263149.0 2.36 7201 256029.0 262717.0 2.61 7204 237088.0 245671.0 3.62 7202 PIRP-30-5-2-10-3 321116.0 325615.0 1.40 7202 320413.0 326010.0 1.74 7203 294469.0 300721.0 2.12 7202 PIRP-30-5-2-10-4 372682.0 377550.0 1.30 7201 372132.0 377642.0 1.48 7202 346923.0 354108.0 2.07 7203 PIRP-30-5-2-10-5 289763.0 299298.0 3.29 7202 288474.0 298017.0 3.30 7202 266616.0 277415.0 4.05 7202 PIRP-40-2-3-3-1 134602.0 136680.0 1.54 7200 133995.0 136662.0 1.99 7202 113171.0 113836.0 0.58 7200 PIRP-40-2-3-3-2 129497.0 132107.0 2.01 7205 129618.0 132099.0 1.91 7200 109140.0 109140.0 0.00 5353 PIRP-40-2-3-3-3 127505.0 129425.0 1.50 7202 127438.0 129524.0 1.63 7201 110936.0 111849.0 0.82 7201 PIRP-40-2-3-3-4 120444.0 121953.0 1.25 7200 120255.0 121919.0 1.38 7201 101223.0 101223.0 0.00 4292 PIRP-40-2-3-3-5 127760.0 128995.0 0.96 7200 127601.0 129128.0 1.19 7201 102356.0 103990.0 1.59 7201 PIRP-40-3-3-6-1 234408.0 244405.0 4.26 7201 233405.0 244450.0 4.73 7204 206077.0 216220.0 4.92 7202 PIRP-40-3-3-6-2 216410.0 223526.0 3.28 7200 215836.0 223805.0 3.69 7203 194909.0 201572.0 3.41 7202 PIRP-40-3-3-6-3 262138.0 268344.0 2.36 7201 260320.0 268184.0 3.02 7203 239755.0 246071.0 2.63 7201 PIRP-40-3-3-6-4 257099.0 265133.0 3.12 7202 257786.0 265014.0 2.80 7201 230334.0 237589.0 3.14 7203 PIRP-40-3-3-6-5 282124.0 289109.0 2.47 7202 281599.0 289187.0 2.69 7203 261346.0 268376.0 2.68 7202 PIRP-50-2-3-3-1 171373.0 174079.0 1.57 7201 170394.0 174429.0 2.36 7201 140191.0 142451.0 1.61 7203 PIRP-50-2-3-3-2 188028.0 189710.0 0.87 7202 187998.0 189718.0 0.91 7203 160553.0 162094.0 0.95 7202 PIRP-50-2-3-3-3 182640.0 184651.0 1.10 7202 182079.0 184624.0 1.39 7204 156807.0 158018.0 0.77 7202 PIRP-50-2-3-3-4 166408.0 170039.0 2.18 7202 166458.0 170129.0 2.20 7203 134964.0 137487.0 1.86 7204 PIRP-50-2-3-3-5 177338.0 180143.0 1.58 7202 175681.0 180209.0 2.57 7202 146842.0 149333.0 1.69 7203 CIRRELT-2013-32 25
[10] A. Custódo and R. Olvera. Redesgnng dstrbuton operatons: a case study on ntegratng nventory management and vehcle routes desgn. Internatonal Journal of Logstcs, 9(2):169 187, 2006. [11] H. Emmons. A replenshment model for radoactve nuclde generators. Management Scence, 14 (5):263 274, 1968. [12] P. Farahan, M. Grunow, and H.-O. Günther. Integrated producton and dstrbuton plannng for pershable food products. Flexble Servces and Manufacturng Journal, 24(1):28 51, 2012. [13] M. Gendreau, G. Laporte, and F. Semet. The coverng tour problem. Operatons Research, 45(4): 568 576, 1997. [14] K. Gumasta, F. T. S. Chan, and M. K. Twar. An ncorporated nventory transport system wth two types of customers for multple pershable goods. Internatonal Journal of Producton Economcs, 139(2):678 686, 2012. [15] R. Hajema. A new class of stock-level dependent orderng polces for pershables wth a short maxmum shelf lfe. Internatonal Journal of Producton Economcs, forthcomng, 2011. [16] V. Hemmelmayr, K. F. Doerner, R. F. Hartl, and M. W. P. Savelsbergh. Vendor managed nventory for envronments wth stochastc product usage. European Journal of Operatonal Research, 202(3): 686 695, 2010. [17] R. Jagannathan and T. Sen. Storng crossmatched blood: a pershable nventory model wth pror allocaton. Management scence, 37(3):251 266, 1991. [18] J. B. Jennngs. Blood bank nventory control. Management Scence, 19(6):637 645, 1973. [19] I. Z. Karaesmen, A. Scheller-Wolf, and B. Denz. Managng pershable and agng nventores: Revew and future research drectons. In K. G. Kempf, P. Kesknocak, and R. Uzsoy, edtors, Plannng Producton and Inventores n the Extended Enterprse, volume 151 of Internatonal Seres n Operatons Research & Management Scence, pages 393 436. Sprnger, New York, 2011. [20] K. E. Kendall and S. M. Lee. Formulatng blood rotaton polces wth multple objectves. Management Scence, 26(11):1145 1157, 1980. [21] G. Laporte. Ffty years of vehcle routng. Transportaton Scence, 43(4):408 416, 2009. 26 CIRRELT-2013-32
[22] T. Le, A. Dabat, J.-P. Rchard, and Y. Yh. A column generaton-based heurstc algorthm for an nventory routng problem wth pershable goods. Optmzaton Letters, 2012. do: 10.1007/ s11590-012-0540-2. [23] A. Mercer and X. Tao. Alternatve nventory and dstrbuton polces of a food manufacturer. Journal of the Operatonal Research Socety, 47(6):755 765, 1996. [24] S. Mnner and S. Transchel. Perodc revew nventory-control for pershable products under servcelevel constrants. OR Spectrum, 32(4):979 996, 2010. [25] A. Nagurney and A. H. Masoum. Supply chan network desgn of a sustanable blood bankng system. In T. Boone, V. Jayaraman, and R. Ganeshan, edtors, Sustanable Supply Chans, volume 174 of Internatonal Seres n Operatons Research & Management Scence, pages 49 72. Sprnger, New York, 2012. [26] A. Nagurney, A. H. Masoum, and M. Yu. Supply chan network operatons management of a blood bankng system wth cost and rsk mnmzaton. Computatonal Management Scence, 9(2):205 231, 2012. [27] A. Nagurney, M. Yu, A. H. Masoum, and L. S. Nagurney. Networks Aganst Tme Supply Chan Analytcs for Pershable Products. Sprnger, New York, 2013. [28] A. Nagurney, M. Yu, A. H. Masoum, and L. S. Nagurney. Fast fashon apparel supply chans. In Networks Aganst Tme Supply Chan Analytcs for Pershable Products, pages 117 139. Sprnger, New York, 2013. [29] A. Nagurney, M. Yu, A. H. Masoum, and L. S. Nagurney. Food supply chans. In Networks Aganst Tme Supply Chan Analytcs for Pershable Products, pages 65 88. Sprnger, New York, 2013. [30] S. Nahmas. Pershable Inventory Systems. Internatonal Seres n Operatons Research & Management Scence, volume 160. Sprnger, New York, 2011. [31] A. Osvald and L. Z. Strn. A vehcle routng algorthm for the dstrbuton of fresh vegetables and smlar pershable food. Journal of Food Engneerng, 85(2):285 295, 2008. [32] M. W. Padberg and G. Rnald. A branch-and-cut algorthm for the resoluton of large-scale symmetrc travelng salesman problems. SIAM Revew, 33(1):60 100, 1991. [33] C. C. Pegels and A. E. Jelmert. An evaluaton of blood-nventory polces: a Markov chan applcaton. Operatons Research, 18(6):1087 1098, 1970. CIRRELT-2013-32 27
[34] S. P. Sngh, V. Chonhenchob, and J. Sngh. Lfe cycle nventory and analyss of re-usable plastc contaners and dsplay-ready corrugated contaners used for packagng fresh fruts and vegetables. Packagng Technology and Scence, 19(5):279 293, 2006. [35] K. van Donselaar, T. van Woensel, R. A. C. M. Broekmeulen, and J. Fransoo. Inventory control of pershables n supermarkets. Internatonal Journal of Producton Economcs, 104(2):462 472, 2006. [36] X. Yang. Choosng transportaton alternatves for hghly pershable goods: a case study on nuclear medcne. PhD thess, Massachusetts Insttute of Technology, 2006. 28 CIRRELT-2013-32