- Email: [email protected]
The reduction of greenhouse gas emissions from freight transport by pooling supply chains
Int. J. Production Economics 143 (2013) 86–94
Contents lists available at ScienceDirect
Int. J. Production Economics journal homepage: www.elsevier.com/locate/ijpe
The reduction of greenhouse gas emissions from freight transport by pooling supply chains Shenle Pan a,b, Eric Ballot a,n, Fre´de´ric Fontane b a b
Mines ParisTech, CGS—Centre de Gestion Scientifique, 60 Boulevard Saint Michel 75272 Paris Cedex 06, France Mines ParisTech, CAOR—Centre de CAO-Robotique, Mathe´matiques et Syste mes, 60 Boulevard Saint Michel 75272 Paris Cedex 06, France
a r t i c l e i n f o
a b s t r a c t
Article history: Received 30 December 2009 Accepted 23 October 2010 Available online 28 October 2010
It is well known that freight consolidation is an effective way to improve the utilization of logistics resources. In fact today, this policy is locally and fragmentally implemented at the operational level. We propose here to explore the environmental impact of pooling of supply chains at the strategic level (merging supply chains). With real data from two main French retail chains and through an optimization model, we compute CO2 emissions for two transport modes, road and rail. As regards the general dependency of the emissions produced by the modes of transport on their loads, the emissions functions of the two modes are both piecewise linear and discontinuous functions. The supply network pooling proposed here is an efficient approach in reducing CO2 emissions. Even if the attention is focused on the emissions, the transportation costs are also studied and analyzed. & 2010 Elsevier B.V. All rights reserved.
Keywords: Transportation Pooling Supply chain CO2 emissions Mixed integer programming.
1. Introduction In a context of a global economy and fierce competition, companies made intense use of transport to cope with the demands from their customers. On the other hand, it is well known that the climate change problem and the increasing price of energy have received worldwide attention over the past decades. As a significant source of Greenhouse Gas (GHG) emissions, the transport sector is naturally concerned by global warming. As a result, improving transport efficiency is among the foremost concerns of supply chain management initiatives. Several studies show that today the lack of satisfaction regarding transport efficiency is based both on the use of vehicles and on GHG emissions, see Le´onardi and Baumgartner (2004) and McKinnon et al. (2003). From this standpoint, the consolidation of shipments was put forward and has been recently studied in several publications, see Bookbinder and Higginson (2002), Cheung et al. (2003), Ergun et al. (2007) and Tyan et al. (2003) and it has been shown that freight consolidation may be an efficient way to achieve lower costs, reduce inventories or increase service, see Hageback and Segerstedt (2004). In fact, at present this policy is locally and fragmentally implemented at the operational level using a 3PL for example, as it depends on the opportunities available to carriers to consolidate freight. As shown in Le´onardi and Baumgartner (2004) and
n
Corresponding author. Tel.: +33 1 40 51 90 97; fax: +33 1 40 51 90 65. E-mail addresses: [email protected] (S. Pan), [email protected] (E. Ballot), [email protected] (F. Fontane). 0925-5273/$ - see front matter & 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.ijpe.2010.10.023
McKinnon et al. (2003), despite the contribution of consolidation, the mean load factor of road transport is about 70% in general, which is consistent with our data, see Ballot and Fontane (2008). Thus, new approaches in logistical organization are called for. Furthermore, the achievement of economies of scale that are profitable for both suppliers and retail chains requires closer and longer term collaboration, see Groothedde et al. (2005). It is why, a new concept in this same area, the consolidation of freight between supply chains, namely pooling supply networks or supply chains, will be proposed and studied in this paper at the strategic level. From the perspective of sustainable development, this paper aims to evaluate the effect of this pooling on reducing GHG emissions, especially CO2 emissions. The air pollution is a known problem and a survey made by Cooper et al. (1997) shows the different mathematical programming techniques that could be used to manage the impact of emissions on the environment. However, the objective of reducing CO2 emissions by optimizing the supply chain has not yet been widely studied in logistics or in air pollution literature. Currently, much more attention is paid to new fuel economy technologies applied to vehicles to reduce the environmental impact of freight transport. Nevertheless, concern by the freight transport sector should justifiably have been warranted. The paper is organized as follows. In Section 2, we take a look at the pooling of supply networks of two large retail chains in France and we detail the preparation of data. Section 3 presents the methods to compute CO2 emissions from transport and the optimization model. In Section 4, we discuss the emissions results and economics results and associated analyses with the conclusions following in Section 5.
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
2. Methodology
87
Table 1 The number of warehouses (or plants) in each group under classification.
2.1. Pooling supply networks of retail chains in France This project was developed in collaboration with a French ´ter (www.club-demeter.fr) that proassociation called Club De´me vided us with all the data for our research. This support guarantees that the database we worked on was indeed in agreement with the real network flow of some of the companies in the association. The data mainly consists of the flows of goods of two major French retail chains in the first 12 weeks of 2006, and also of the geographic location of both distribution centers of the two retail chains and the warehouses of their common suppliers. Based on the database offered, we propose that the merging of two supply chains can be achieved as follows. As illustrated in Fig. 1, originally the flows from the suppliers’ (i, j and k) warehouse or plant (WH) are directly shipped to retailers’ (m, n and l) distribution centers (DC). To achieve the merging of the supply chains, we create two kinds of hub in this network, an upstream hub and a downstream hub. Further, two assumptions are made to simplify the problem: (1) the hubs are respectively among the set of WH or DC in the database; (2) goods are allowed to be transported from WH to DC directly or to be consolidated at upstream hubs, downstream hubs or both. Hence, the pooled network here is a three-echelon hub network, with the upstream path between WH and the upstream hub, the midstream path between the upstream hub and the downstream hub, and the downstream path between the downstream hub and DC. More details are discussed in the formulation section. The possibility of reducing CO2 emissions obviously depends on the consolidation of a mass of flows that will be shipped together between the different supply chains in order to increase the truck load factor of each shipment. 2.2. Construction of the database The database mentioned above consists of a large number of diverse goods’ flows (several thousands lines) concerning the most important 106 common suppliers of the two French retail chains. Initially, to be able to identically analyze the flows, they are all measured in number of pallets. This means that the flows are converted into quantities of equivalent full pallets. WH of Supplier i
WH of Supplier j
WH of Supplier k
DC of Retailer m
DC of Retailer n
DC of Retailer l
CARE GRO LIQ
Group A (pal/ weeko 200)
Group B (200 rpal/ week r 600)
Group C (pal/ week 4600)
8 25 21
9 29 25
13 27 34
First the flows are subdivided into three categories according to the different types of products involved: CARE (pharmaceuticals, cosmetics/perfume and hygiene), GRO (grocery) and LIQ (liquids). The main reason for this subdivision is that these classes of product require different handling through the transportation due to their varying logistic behavior. For example, a warehouse for liquid products will not have the same construction features as that for cosmetics. This classification in the French retail supply chains is maintained in this paper. According to the first study of this project by Ballot and Fontane (2008) using the same data, it was shown that mixing flows with very different sizes should be avoided. Moreover, after the classification of products, the suppliers’ warehouses in each class are divided into three groups based on the size of flows in the warehouse, since the flows of each warehouse (even of the same supplier) are independently dealt with in our case. In order to obtain a balanced amount of variables in each group, the boundaries of the groups are 0–200, 200–600 and 4600 pallet/week. Note that the flow per week involved here is an average of the flows covering the 12 weeks studied. This decomposition does not exactly follow the real volume breakdown, but help the optimization process. It could be seen as a limitation of this work. Finally the initial problem is broken down into 9 sub-problems in which the number of suppliers is showed in Table 1. As mentioned above, the procedure of classification and grouping to reduce the extent of the problem, which also determines the framework of the database, is important to decrease the computational time of an optimization model presented in next part.
3. Problem formulation The transportation problem is studied according to the emissions criteria but the cost is also assessed. The economic objective function is well known but to be able to calculate and minimize CO2 emissions from freight transport, we must first define an emissions function according to the mode of transport, as well as an optimization model.
3.1. Objective functions
After pooling
WH of Supplier i
WH of Supplier j / Upstream hub
WH of Supplier k
DC of Retailer m
DC of Retailer n / Downstream hub
DC of Retailer l
Fig. 1. Example of the merging supply chains.
Two modes of transport, road and rail, are taken into account in this research. The emissions function of their mode of transportation is based on several reports in this field, e.g. the MEET report (Hickman et al., 1999; Jorgensen and Sorenson, 1998) and the COST 319 project (Joumard, 1999). Furthermore, CO2 emissions factors in France applied to these functions can be found in some reports of the ADEME (Agency of Environment and Energy Management in France: www.ademe.fr), see De Boissieu (2006), Jancovici (2007). In these reports, CO2 emissions are as a rule measured in terms of tonkm freight transported, which is also adopted in our case, but the ton-km is changed into pallet-km due to the form of the database. In this section, we simply indicate the final formulas to calculate emissions that are applied to the optimization model.
88
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
3.1.1. Road transport emissions First of all, the mode of road transport here refers to transport by Heavy Duty Vehicle (HDV) only (32–40 ton for general merchandise). According to the emissions function for the HDV truck given by Hickman et al. (1999) and Jancovici (2007) some assumptions are made: (a) the average speed is 80 km/h; (b) the gradient of a road is not taken into account; (c) in general the truck considered here is fully loaded with 25 tons for weight or 33 pallets for volume. Particularly, for the care and grocery classes, it is assumed that the truck is fully loaded at the same time by weight and volume (33 pallets weighting 25 tons). However for the liquid class which is doubtless heavier than the others, the truck is fully loaded only with 23 pallets weighing 25 tons. As a result, the final CO2 emissions function with the variable of load is lxm x ð1Þ ev ðxÞ ¼ ðevf eve Þ þ eve c c where ev(x) is the CO2 emissions from a vehicle in kg/km with the variable of load x in pallet; evf is the CO2 emissions of a fully loaded (by weight) vehicle, which is evf ¼1.096 kg/km for HDV truck; eve is the CO2 emissions of an empty vehicle, which is eve ¼0.772 kg/km for HDV truck; c is the volume capacity of a vehicle; as said above, which is 33 pallet/vehicle for the care and grocery classes and 23 pallet/vehicle for the liquid class. 3.1.2. Rail transport emissions According to the data from Fret SNCF (the main freight rail company in France, www.fret.sncf.com), 90% of the freight train locomotives in France are electrically driven, thus only the electrically powered locomotive is considered here. The emissions of air pollutant related to rail transport are calculated in two steps. The first step is the estimation of the energy consumption of a train in kJ per ton-km, which is required to move the train. After this step, the amount of emissions can be calculated according to the energy required. Consequently, the pollutant emissions function related to energy consumption is as follows Ei ¼ WSEC
Tkm 1 BSEFi Tpt 3:6 106
ð2Þ
where Ei is the total emissions of air pollutant i in kg; WSEC is the weight specific energy consumption of the train in kJ/ton-km; Tkm is the amount of freight transported by the train in ton-km; Tpt is the load factor of the train, in ton-km/total train ton; BSEFi is the brake specific emissions factor in g/kWh of energy produced. Formula (2), which is corrected and validated by the main author of the MEET report (Jorgensen and Sorenson, 1998), is the general form of calculation of air pollutant emissions. In this study, we focus on CO2 emissions as the most representative of the emissions of electricity production. In France, according to the report of the ADEME (Jancovici, 2007), the BSEFco2 of EDF (French’s leading energy company, www.edf.com) in 2007 was 42 g/kWh. Some assumptions are also made here to simplify the problem: (a) the average speed is 100 km/h; (b) the mean distance between stops is 100 km; (c) a train is composed of 13 wagons, which is the minimum actual size for chartering a train (half train operation); (d) because of the large weight capacity of the wagon (56 ton/ wagon), a wagon is only fully loaded by a volume equivalent to 36 pallets. The assumptions (c) and (d) imply that the volume capacity of train here is 468 pallet/train. Similar to formula (1), formula (3) given below is used to calculate CO2 emissions from rail transport. lxm x ð3Þ et ðxÞ ¼ ðetf ete Þ þ ete c c where et(x) is the CO2 emissions from a train in kg/km with the variable of load x in pallet; etf is the CO2 emissions of a fully loaded (by volume) train, which is etf ¼0.96 kg/km for the care and grocery classes and etf ¼1.16 kg/km for the liquid class; ete is the CO2
emissions of an empty train, which is ete ¼0.498kg/km; c is the volume capacity of a train; as said above, which is c ¼468 pallet/ train. To compare the CO2 emissions of these two transport modes, illustrated in Fig. 2, it is assumed that one pallet weighs one ton, which means that the ton-km and pallet-km units are now equivalent. The conclusion, here very favorable to rail transport, should be considered with respect to the source of production of electricity that emits very little CO2/kWh in France. Due to the upper integer part of x/c, functions (1) and (3) are in fact piecewise linear and discontinuous functions. That is also the foundation of the optimization model, which will be presented in the next section.
3.1.3. Transport cost For the economic evaluation we consider a formula with a structure similar to (1) or (3), but based on the classical cost approach. As the economic data is not based on published fares but on real cost paid, the specific details are not given here. However on the big picture and from a transport point of view only, the train faces an important fixed cost and therefore is cheaper than truck only on trips greater than approximately 600 km with fully loaded transport means. The economic function, also a piecewise linear and discontinuous function, serves to evaluate the cost of the emissions optimized solution or to classically optimize cost to highlight the differences. 3.2. Optimization model Since the purpose of our optimization project is to minimize the CO2 emissions related to freight transport in two large supply chains, the emissions functions are adopted in the optimization model via an objective function. The discontinuity of the functions results in a Mixed Integer Linear Programming (MILP) problem in our case.
3.2.1. Formalization Croxton et al. (2003b) cites three models, namely the incremental model, the Multiple Choice Model (MCM) and the convex combination model to describe a formula for the piecewise linear function. Furthermore, according to this article and another one concerning the knapsack problem (Kameshwaran and Narahari, 2007), these three MILP models for nonconvex piecewise linear minimization problems are equivalent. The MCM model is adopted in our case to the fact that the emissions produced by freight transport in an arc depend on the sum of the flow of pallets on this
Fig. 2. Comparison of CO2 emissions between road and rail transport.
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
arc and noted x. The linearization of flow and emissions can be displayed as Fig. 3 and is modeled by Eqs. (5)–(9). As described in part 2, the pooling of supply chains in this case is carried out by way of a three-echelon network. Thus the problem actually involves minimizing the sum of CO2 emissions of the three sections of transport (upstream, midstream and downstream). To simplify the presentation of the problem, the flows would only be considered for one period and only one mode of transport would be involved. The extensions to several periods or the two transport modes represent no difficulty and therefore are not presented here as well as the classical cost objective function. The optimization problem is stated as follows (for notation refer to Fig. 3). X X X Min ½dom eðxom Þ þ ½dmn eðxmn Þ þ ½dnd eðxnd Þ ð4Þ om A Au
mn A Am
nd A Ad
with X
fas ysa þvsa zsa
ð5Þ
Piecewise linear constraints X xa ¼ zsa , 8a A A
ð6Þ
eðxa Þ ¼
s A Sa
s A Sa
lbsa ysa r zsa r ubsa ysa , X
ysa r 1,
8a A A,8s A Sa
ð7Þ
8a A A
ð8Þ
8a A A,8s A Sa
ð9Þ
s A Sa
ysa A f0,1g,
Flow-balance constraints xka Z 0, X
8a A A,8kA K
xkom ¼ Rko ,
8k A K,8o A O
ð10Þ ð11Þ
mAM
X
xknd ¼ Bkd ,
8k A K,8d A D
ð12Þ
nAN
X
xkom ¼
X
xkmn ,
8k A K,8m A M
ð13Þ
8k A K,8n A N
ð14Þ
nAN
oAO
X mAM
xkmn ¼
X
xknd ,
dAD
Hub selection constraints X pkom ¼ 1, 8k A K,8o A O
ð15Þ
mAM
pkom A f0,1g,
8k A K,8o A O,8m A M
ð16Þ
xkom rpkom Z,
8k A K,8o A O,8m A M
ð17Þ
8d A D
ð18Þ
X
qnd ¼ 1,
89
qnd A f0,1g,
8n A N,8d A D
ð19Þ
xknd rqnd Z,
8k A K,8n A N,8d A D
ð20Þ
where Z is a large enough constant; K the set of commodities; and each has at least one specific origin and destination node that are respectively called warehouse and distribution center; O, M, N, D denote respectively the set of WH (source nodes), candidate upstream hubs, candidate downstream hubs, destination nodes (DC); Au, Am, Ad the subsets of arc set A, of which Au is the arc set on the upstream flow, Am on the midstream and Ad on the downstream; A the set of all arcs and each arc aAA; dij the distance of arc ijAA; e(x) the piecewise linear function of CO2 emissions related to the flow x; Rko the quantity of commodity kAK supplied at WH oAO; Bkd the quantity of commodity kAK required at DC dAD; Sa represents the number of segments on each aAA; s the piecewise linear segment sASa on each aAA; zsa load of the segment sASa on each aAA; fas the nonnegative fixed value of the intercept of segment sASa on arc aAA, in our case the piecewise function is identical on each arc, so the intercept of the same segment on every arc is always the same according to the mode of transport, such as 0.772 for road transport and 0.498 for rail; vsa slope of segment sASa on each arc aAA. Because of the fact that all the segments are parallel, the slope here is constant with the same product and transport mode, such as vsa ¼ 0:00982 for the care and grocery classes and vsa ¼ 0:0141 for the liquid products in road transport. With regard to rail transport; similarly we have vsa ¼ 0:00099 for care and grocery, and vsa ¼ 0:0014 for liquid products; ysa binary variable on each arc aAA, with ysa ¼ 1if segment sASa contains a nonzero flow, and ysa ¼ 0 otherwise; lbsa and ubsa lower and upper bounds of total flow lie in segment sASa on each arc. In particular, for every arc aAA we assume that lb1a ¼ 0 and ub6a ¼ þ 1 and also for each segment s, ubsa lbsa ¼ wc (which represents the capacity of conveyance depending on the pallet’s density and area of the conveyance); xa the total P flow of all commodities on arc aAA xa ¼ k A K xka 8aAA and decision variables xka the quantity of shipment of commodity kAK on arc aAA; pkom the binary variable, with pkom ¼ 1 if commodity kAK supplied at WH oAO is converged at hub upstream mAM, and 0 otherwise; qnd the binary variable, qnd ¼1 if DC dAD is served by the downstream hub nAN, and 0 otherwise. Constraints (6)–(9) and Eq. (5) which come from the MCM in Croxton et al. (2003b) assure that the total product flow on arc a must lie on a single segment and e(0) ¼0. Eqs. (11)–(14) are the flow-balance constraints in the three-echelon multi commodity network. Constraints (15)–(20) correspond to the assumption that each supplier must have at most one related upstream hub to where his flow will be shipped and each DC must be served by at most one downstream hub. This assumption is to simplify flow management at each inventory point. In addition, if we take into consideration both transport modes (truck and train) in the model at the same time, the objective function, as well as the number of variables, is doubled, as a result of the selection between road and rail transport. Moreover, another additional constraint to bound the minimal flow by train exists only in the case of rail transport. 3.2.2. Approaches to reduce the optimization problem This MILP transportation problem is a NP-Hard problem, see Ahuja et al. (1993), so the computing time of this problem could be influenced by its size. As a representative factor of the size of an instance of a problem, the number of variables is about 23,000 in the biggest problem among the sub-problems presented. This leads us to explore the possibility of being able to reduce the size in order to solve the problem. Thus, in the model we made several assumptions as follows:
nAN
Fig. 3. Variables for each emissions segment.
a. As stated in part 3, the emissions functions are nonconvex piecewise linear and discontinuous, and the number of
90
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
segments depends on the size of flow. Concerning similar problems, some articles based on analyzing piecewise function efficacy indicate that some attributes of the function deeply influence computational time, e.g. the continuity and the number of segments, see Croxton et al. (2007), Kameshwaran and Narahari (2007) and Marins et al. (1997). Therefore, to simplify the function in road transport for example, we assume that a truck is fully loaded in an arc with a shipment above a volume equivalent to 5 trucks in one week. In other words, the maximum number of segments is set at 6 (so Sa ¼6 and s r6), in addition, the slope of the sixth segment is 0.3321 for the care and grocery classes, and 0.476 for liquids. Due to a significant difference in load capacity between trucks and trains, the function in rail transport is reduced to 2 continuous segments, which represent the case that trains are always fully loaded above one train. The slope of the second segment in rail transport is 0.002 for the care and grocery classes, and 0.00248 for liquids. b. According to Croxton et al. (2003a), with the data in our case we can tighten the LP relaxation of the model by limiting the maximum bound of each product’s flow in any arc in the network. c. Considering the fact that consolidation would not occur if the distance between WH and the upstream hub or between the downstream hub and DC is too great, a condition of distance in both upstream and downstream paths is introduced to filter candidate upstream and downstream hubs. In other words, the possible hubs are situated within a certain radius from WH or DC. We studied the sensibility of the solution according to radius variation, and 100 km was shown to be the best compromise between computing time and solution quality.
3.2.3. Lower limit of emissions Aside from the model presented to minimize CO2 emissions, it might also be valuable to calculate the lower limit of the emissions in transport, meaning that all shipments are direct from WH to DC with an absolutely fully loaded transport mean. As a result, Eq. (5) in the optimization model will be replaced by a completely linear objective function. With this limit we are not looking for an operational solution but a measure of the remaining potential of emissions saving with the same transport mean. The assumptions made allow all the sub-problems to be solved by coding and executing a computational program. The results obtained are discussed in the next part.
4. Results and discussions The model presented is coded in ILOG’s OPL 6.3 software with CPLEX 12.1 and run on Quad CPU Q6700 (2.66 GHz) hardware having 4 GB of RAM. The MILP problem is solved with OPL’s default settings, except Memory Emphasis parameter to true to deal with the large size of the problems, refer to ILOG (2008). Furthermore, all the results obtained showed a gap of less than 3% between the best integer solution and the upper bound. Note that all the values exhibited in the following tables are measured in tons. It should also be recalled that, as stated in the methodology part, the suppliers in each class are divided in 3 groups: group A o200 pallet/week, group B 200–600 pallet/week and group C 4600 pallet/week. 4.1. Road transport CO2 emissions optimization Initially, road transport is the exclusive transport mode involved in the optimization problem. In the framework of the 9 sub-problems
presented, the purpose of this section is to evaluate the performance of merging supply chains in terms of CO2 emissions (emissions/ pooled in the tables) using the optimization model, by comparing results with the emissions from the actual transport system (actual emissions in the tables). Additionally, the lower limit of emissions (minimum emissions in the tables) will also be introduced as the optimum situation in which either the trains or trucks are absolutely fully loaded (linear emissions function). The results in Table 2 show that the approach of merging supply chains significantly reduces CO2 emissions from transport, despite the fact that its gain in emissions is about half the maximum gain defined by the lower limit, namely the theoretical minimum emissions. Furthermore, it is obvious that the relative reduction in group C is much less than the other groups. This situation can be explained by the fact that at present the suppliers with massive flows logically make more efforts to saturate their means of transport and these efforts are more likely successful. Besides the reduction of CO2, it is also important to indicate that the number of transport paths falls because of the network pooling, as shown in Fig. 4. 4.2. Road and rail transport CO2 emissions optimization This section aims to determine the effect of joint road/rail transport in reducing CO2 emissions. This means that road and rail transport are alternative transport modes to be chosen according to the size of flows and their performance in terms of emissions. To reflect practical situations in transportation, a lower limit of flow for rail transport is fixed at 468 pallets, which is a volume equivalent to half of a train of 26 wagons. According to the results in Table 3, especially in group C, it can be concluded that joint road and rail transport is a significant way to reduce CO2 emissions, provided the electrically powered train generates low emissions in France thanks to the low emissions electricity Table 2 Emissions from road transport pooled network per week (ton CO2). Groups of suppliers CARE Actual emissions Emissions/pooled Reduction Absolute Relative (%) Minimum emissions Reduction Absolute Relative (%) GRO Actual emissions Emissions/pooled Reduction Absolute Relative (%) Minimum emissions Reduction Absolute Relative (%) LIQ Actual emissions Emissions/pooled Reduction Absolute Relative (%) Minimum emissions Reduction Absolute Relative (%)
A
B
C
S
51 28
93 62
319 290
463 380
23 45 11
31 33 45
29 9 266
83 18 322
40 78
48 52
53 17
141 30
132 67
309 219
670 628
1111 916
65 49 29
90 29 147
42 6 568
195 18 744
103 78
162 52
102 15
736 33
183 93
291 255
1257 1193
1731 1541
90 49 39
36 12 177
64 5 1134
190 11 1350
144 79
114 39
123 10
381 22
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
91
Fig. 4. Example of transportation network before and after merging by road transport (CARE group C).
Table 3 Emissions from road and rail transport pooled network per week (ton CO2). Groups of suppliers CARE Actual emissions Emissions/pooled Reduction Absolute Relative (%) GRO Actual emissions Emissions/pooled Reduction Absolute Relative (%) LIQ Actual emissions Emissions/pooled Reduction Absolute Relative (%)
A
B
C
S
51 28
93 44
319 157
463 229
23 45
49 53
162 51
234 50
132 67
309 210
670 257
1111 534
65 49
99 32
413 62
577 52
183 93
291 247
1257 485
1731 812
90 49
44 15
772 61
919 53
production mainly nuclear and hydro. However, it should be noted that the problem of rail transport here is to obtain a (consolidated) flow that simultaneously meets the weekly demand of retailers and the minimum flow for rail transport (468 pallets). That is to say that the weekly stock level of retailers cannot be challenged. Comparing to Fig. 4, we can easily remark in Fig. 5 that the introduction of the rail transport simplifies the transport network and indicates that it is the means of transport of massive flows.
4.3. Economic evaluation of the solutions optimized for the emissions In complement of the emissions optimization, the solutions are evaluated by the computation of the transportation cost of the different proposed supply chains. Table 4 shows the results of these evaluations for the 9 solutions (trucks only). These results display to a certain extent a correlation between emissions and costs. This correlation is not as perfect as for a single truck as the structure of the network evolves with the pooling of the supply networks. But from a global point of view, we can affirm that with a single transport mode the emissions reduction and the cost reduction are compatible, thanks to the consolidation of the flows resulting from the pooling of the supply networks. The cost evaluations of the road and rail solutions show a different situation when rail is selected by the optimization, as Table 5 illustrates. This occurs typically for group C and less for groups B. The cost of groups C, suppliers with high flows, are significantly more expensive due to rail cost but also have the best
emissions reduction in absolute. In total, the 52% CO2 emissions reduction is achieved with an increase of 85% of the total cost. Table 5 also demonstrates that there is no economic interest for a supply chain or even a pooled network of supply chains with important flows (group C) to prefer an emissions optimized solution with train operations as it is much more expensive with today costs in a country like France, despite the interest in GHG emissions reduction.
4.4. Emissions vs. cost optimizations All the results from Tables 2 to 5 are based on the minimization of the emissions function. To measure the difference with the economic approach, we also use the cost function as the objective function of the same transportation problem. The minimization of cost is given in Table 6 for the truck only. The results are very similar and the small differences are within the 3% gap between the solution and the lower bound and therefore can be seen as round-offs of calculation with no signification. The conclusion is a convergence between the economic and the emissions approaches with trucks only in the pooling of supply networks. The minimization of the cost for the truck and the train formulation lead also to a similar result as shown in Table 6 as train transportation is more expensive with the trains’ cost nowadays. In the near future, the cost of freight by rail may increase less than the cost of freight by road, due to the opening to market of the train operations and less taxation, thanks to reduced CO2 emissions. Or in another way we could see it as the trade-off between economy and environment, as proposed with a different heuristic method by Frota Neto et al. (2008) to find the Pareto efficiency frontier. In this perspective, we compute the impact of a relative variation of truck and train costs in the framework of a normalized total cost. This impact is implemented with a factor a applied to the truck cost and a factor b applied to the train cost and the optimization problem was solved several times with variation of a and b to determine frontier of efficiency. Fig. 6 shows the gap between the economic solution and the emissions solution for the high volume (group C) Care product class, as an example. It highlights the strong sensitivity on emissions of the economic solution under a variation of relative truck and train costs. If the train cost decreases more rapidly than the truck cost increase in the future, the pooling of supply chain offers a solution for both cost efficiency and emissions reduction for the important flows. But the current structure of the costs also reveals that the pooling is not an attractive solution from an economic standpoint to reach the solution of emissions minimization as it requests very drastic changes in this case (truck cost increased by more than 200% and train cost reduced by more than 70%).
92
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
Fig. 5. Example of transportation network before and after merging by joint transport (CARE group C).
Table 4 Transport cost for the emissions minimization of the pooled network (trucks only). Groups of suppliers
A
Optimized emissions/cost
TCOn2
103 h
TCOn2
51 28
43 32
93 62
23 45
11 26
132 67
LIQ Actual Emissions/pooled Reduction Absolute Relative (%)
C 103 h
S
TCOn2
103 h
TCOn2
103 h
115 103
319 290
546 536
463 380
704 671
31 33
12 10
29 9
10 2
83 18
33 5
114 84
309 219
375 353
670 628
1160 1159
1111 914
1649 1596
65 49
30 26
90 29
22 6
42 6
1 0
197 18
53 3
183 93
137 93
291 255
324 318
1257 1193
1737 1716
1731 1541
2189 2127
90 49
44 32
36 12
6 2
64 5
21 1
190 11
71 3
CARE Actual Emissions/pooled Reduction Absolute Relative (%) GRO Actual Emissions/pooled Reduction Absolute Relative (%)
B
Table 5 Transport cost for the emissions minimization of the pooled network (trucks and trains). Groups of suppliers Optimized emissions/cost CARE Actual Emissions/pooled Reduction Absolute Relative (%) GRO Actual Emissions / pooled Reduction Absolute Relative (%) LIQ Actual Emissions/pooled Reduction Absolute Relative (%)
A
B n
3
C n
3
S n
3
10 h
TCOn2
103 h
TCO2
10 h
TCO2
10 h
TCO2
51 28
43 32
93 44
115 145
319 157
546 1202
463 229
704 1379
23 45
11 26
49 53
30 26
162 51
656 120
234 51
675 96
132 67
114 84
309 210
375 450
670 257
1160 2890
1111 534
1649 3424
65 49
30 26
99 32
75 20
413 62
1730 149
577 52%
1775 108
183 93
137 93
291 247
324 351
1257 485
1737 3183
1731 825
2189 3627
90 49
44 32
44 15
27 8
772 61
1446 83
906 52
1429 65
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
93
Table 6 Transport emissions for the cost minimization of the pooled network (trucks only). Groups of suppliers
A
B
TCO2
103 h
CARE Emissions optimization Transport cost optimization
28 29
32 31
62 65
GRO Emissions optimization Transport cost optimization
67 70
84 86
LIQ Emissions optimization Transport cost optimization
93 93
93 92
103 h
TCO2
103 h
103 101
290 295
536 534
380 389
671 666
219 224
353 339
628 645
1159 1155
914 939
1496 1580
255 258
318 315
1193 1217
1716 1722
1541 1568
2127 2129
Apart from the environmental aspect, the economic aspect of transport was also taken into consideration and shows a strong convergence between transportation cost and emissions within the pooling framework with trucks only. This convergence is no more achieved with joint trucks and trains operations with the current cost structure and logistic organization on important flows. A significant part of the trade-off could be reached by more competitive rail freight operations and truck cost increases. However future research is still needed to generalize these results obtained in a specific context and in the logistic organization and transportation technology to overcome the greater cost of the environmental more friendly transportation means.
350
CO2 Emissions (ton)
Current solution Min Cost (+2%, -18%)
250
(+16%, -33%) (+60%, -60%)
200
(+200%, -70%) Min CO2
150 100
400
600
800
1000 1200 Transport Cost (k )
103 h
S
TCO2
TCO2
400
300
C
1400
Fig. 6. Evolution of truck and train costs relative variations on CO2 emissions (truck cost evolution, train cost evolution).
5. Conclusion This paper explored the effect of pooling supply chains’ networks on reducing CO2 emissions from transport with two possible modes, i.e. road and rail, in the context of a national distribution network of two major French retailer chains. We first pointed out the principal drawback of current freight consolidation, namely that it currently takes place in a local and fragmentary manner using carriers and 3PL. This conducted us to search for another approach to achieve a more efficient consolidation. In the aim of reducing CO2 emissions from freight transport, we propose the pooling of supply chains at the strategic level, an approach that demands further and long-term collaboration between the actors of supply chains (suppliers and retailers in this case). In this article geographical and products’ flows pooling were proposed, namely geographical consolidation among suppliers or retailers with similar flows. Moreover, an optimization model with a piecewise linear objective function was set up to quantitatively evaluate its impact on reducing CO2 emissions. Benefiting from our ´ter, we tested the model with the real association with Club De´me data of merchandise flows over 12 weeks in a large national distribution network composed of two French retailers’ supply chain. A split in size and type of flows according to real operations allows us to reduce the size of the problem and to obtain solutions with a gap less than 3%. Finally, the results in Tables 2 and 3 globally yield a relative reduction of CO2 emissions of 14% exclusively with road transport and of 52% with joint road and rail transport. It should be recalled that all the results obtained in this paper are conditional upon the same, even improved weekly service rate to retailers. Consequently, it can be concluded that pooling supply chains as studied here provides a solution to significantly reduce CO2 emissions from freight transport.
Acknowledgments Playing an indispensable role in our research, we gratefully ´ter Associaacknowledge the support provided by the Club De´me tion. The authors are also grateful to the R2DS program of the Ile-deFrance Region that provided us with funding for this project. References Ahuja, R.K., Magnanti, T.L., Orlin, J.B., 1993. Network Flows: Theory, Algorithms, and Applications. Prentice Hall, New Jersey. Ballot, E., Fontane, F., 2008. Reducing greenhouse gas emissions through the collaboration of supply chains: lessons from French retail chains. In: Proceedings of the International Conference on Information Systems, Logistics and Supply Chain Conference. Madison, Wisconsin. Bookbinder, J.H., Higginson, J.K., 2002. Probabilistic modeling of freight consolidation by private carriage. Transportation Research Part E: Logistics and Transportation Review 38 (5), 305–318. Cheung, R.K., Tong, J.H., Slack, B., 2003. The transition from freight consolidation to logistics: the case of Hong Kong. Journal of Transport Geography 11 (4), 245–253. Cooper, W.W., Hemphill, H., Huang, Z., Li, S., Lelas, V., Sullivan, D.W., 1997. Survey of mathematical programming models in air pollution management. European Journal of Operational Research 96 (1), 1–35. Croxton, K.L., Gendron, B., Magnanti, T.L., 2003a. Models and methods for merge-intransit operations. Transportation Science 37 (1), 1–22. Croxton, K.L., Gendron, B., Magnanti, T.L., 2003b. A comparison of mixed-integer programming models for nonconvex piecewise linear cost minimization problems. Management Science 49 (9), 1268–1273. Croxton, K.L., Gendron, B., Magnanti, T.L., 2007. Variable disaggregation in network flow problems with piecewise linear costs. Operations Research 55 (1), 146–157. De Boissieu, C., 2006. Rapport du Groupe de Travail ’’Division par Quatre des E´missions de Gaz a Effet de Serre de la France a L’horizon 2050’’ Sous la Pre´sidence de Christian de Boissieu. MINEFI, Paris, France. Ergun, O., Kuyzu, G., Savelsbergh, M., 2007. Reducing truckload transportation costs through collaboration. Transportation Science 41 (2), 206–221. Frota Neto, J.Q., Bloemhof-Ruwaard, J.M., van Nunen, J.A.E.E., van Heck, E., 2008. Designing and evaluating sustainable logistics networks. International Journal of Production Economics 111 (2), 195–208. Groothedde, B., Ruijgrok, C., Tavasszy, L., 2005. Towards collaborative, intermodal hub networks: a case study in the fast moving consumer goods market. Transportation Research Part E: Logistics and Transportation Review 41 (6), 567–583.
94
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
Hageback, C., Segerstedt, A., 2004. The need for co-distribution in rural areas: a study of Pajala in Sweden. International Journal of Production Economics 89 (2), 153–163. Hickman, J., Hassel, D., Joumard, R., Samaras, Z., Sorenson, S., 1999. Methodology for Calculating Transport Emissions and Energy Consumption. Transport Research Labotory, Crowthorne, UK. ILOG, 2008. O.P.L. Studio 6.1: the user’s manual. ILOG Inc. Jancovici, J.M., 2007. Bilan Carbones: Calcul des Facteurs D’e´missions et Sources Bibliographiques Utilise´es.. ADEME, Paris, France. Jorgensen, M.W., Sorenson, S.C., 1998. Estimating emissions from railway traffic. International Journal of Vehicle Design 20 (1), 210–218. Joumard, R., 1999. Methods of Estimation of Atmospheric Emissions from Transport: European Scientist Network and Scientific State-of-the-Art (Action COST 319 Final Report). INRETS, Lyon, France.
Kameshwaran, S., Narahari, Y., 2007. Nonconvex piecewise linear knapsack problems. European Journal of Operational Research 192 (1), 56–68. Le´onardi, J., Baumgartner, M., 2004. CO2 efficiency in road freight transportation: status quo, measures and potential. Transportation Research Part D 9 (6), 451–464. Marins, F.A.S., Senne, E.L.F., Darby-Dowman, K., Machado, A.F., Perin, C., 1997. Algorithms for network piecewise-linear programs: a comparative study. European Journal of Operational Research 97 (1), 183–199. McKinnon, A., Ge, Y., Leuchars, D., 2003. Analysis of Transport Efficiency in the UK Food Supply Chain.. Heriot-Watt University, Edinburgh, UK. Tyan, J.C., Wang, F.-K., Du, T.C., 2003. An evaluation of freight consolidation policies in global third party logistics. Omega 31 (1), 55–62.
Contents lists available at ScienceDirect
Int. J. Production Economics journal homepage: www.elsevier.com/locate/ijpe
The reduction of greenhouse gas emissions from freight transport by pooling supply chains Shenle Pan a,b, Eric Ballot a,n, Fre´de´ric Fontane b a b
Mines ParisTech, CGS—Centre de Gestion Scientifique, 60 Boulevard Saint Michel 75272 Paris Cedex 06, France Mines ParisTech, CAOR—Centre de CAO-Robotique, Mathe´matiques et Syste mes, 60 Boulevard Saint Michel 75272 Paris Cedex 06, France
a r t i c l e i n f o
a b s t r a c t
Article history: Received 30 December 2009 Accepted 23 October 2010 Available online 28 October 2010
It is well known that freight consolidation is an effective way to improve the utilization of logistics resources. In fact today, this policy is locally and fragmentally implemented at the operational level. We propose here to explore the environmental impact of pooling of supply chains at the strategic level (merging supply chains). With real data from two main French retail chains and through an optimization model, we compute CO2 emissions for two transport modes, road and rail. As regards the general dependency of the emissions produced by the modes of transport on their loads, the emissions functions of the two modes are both piecewise linear and discontinuous functions. The supply network pooling proposed here is an efficient approach in reducing CO2 emissions. Even if the attention is focused on the emissions, the transportation costs are also studied and analyzed. & 2010 Elsevier B.V. All rights reserved.
Keywords: Transportation Pooling Supply chain CO2 emissions Mixed integer programming.
1. Introduction In a context of a global economy and fierce competition, companies made intense use of transport to cope with the demands from their customers. On the other hand, it is well known that the climate change problem and the increasing price of energy have received worldwide attention over the past decades. As a significant source of Greenhouse Gas (GHG) emissions, the transport sector is naturally concerned by global warming. As a result, improving transport efficiency is among the foremost concerns of supply chain management initiatives. Several studies show that today the lack of satisfaction regarding transport efficiency is based both on the use of vehicles and on GHG emissions, see Le´onardi and Baumgartner (2004) and McKinnon et al. (2003). From this standpoint, the consolidation of shipments was put forward and has been recently studied in several publications, see Bookbinder and Higginson (2002), Cheung et al. (2003), Ergun et al. (2007) and Tyan et al. (2003) and it has been shown that freight consolidation may be an efficient way to achieve lower costs, reduce inventories or increase service, see Hageback and Segerstedt (2004). In fact, at present this policy is locally and fragmentally implemented at the operational level using a 3PL for example, as it depends on the opportunities available to carriers to consolidate freight. As shown in Le´onardi and Baumgartner (2004) and
n
Corresponding author. Tel.: +33 1 40 51 90 97; fax: +33 1 40 51 90 65. E-mail addresses: [email protected] (S. Pan), [email protected] (E. Ballot), [email protected] (F. Fontane). 0925-5273/$ - see front matter & 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.ijpe.2010.10.023
McKinnon et al. (2003), despite the contribution of consolidation, the mean load factor of road transport is about 70% in general, which is consistent with our data, see Ballot and Fontane (2008). Thus, new approaches in logistical organization are called for. Furthermore, the achievement of economies of scale that are profitable for both suppliers and retail chains requires closer and longer term collaboration, see Groothedde et al. (2005). It is why, a new concept in this same area, the consolidation of freight between supply chains, namely pooling supply networks or supply chains, will be proposed and studied in this paper at the strategic level. From the perspective of sustainable development, this paper aims to evaluate the effect of this pooling on reducing GHG emissions, especially CO2 emissions. The air pollution is a known problem and a survey made by Cooper et al. (1997) shows the different mathematical programming techniques that could be used to manage the impact of emissions on the environment. However, the objective of reducing CO2 emissions by optimizing the supply chain has not yet been widely studied in logistics or in air pollution literature. Currently, much more attention is paid to new fuel economy technologies applied to vehicles to reduce the environmental impact of freight transport. Nevertheless, concern by the freight transport sector should justifiably have been warranted. The paper is organized as follows. In Section 2, we take a look at the pooling of supply networks of two large retail chains in France and we detail the preparation of data. Section 3 presents the methods to compute CO2 emissions from transport and the optimization model. In Section 4, we discuss the emissions results and economics results and associated analyses with the conclusions following in Section 5.
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
2. Methodology
87
Table 1 The number of warehouses (or plants) in each group under classification.
2.1. Pooling supply networks of retail chains in France This project was developed in collaboration with a French ´ter (www.club-demeter.fr) that proassociation called Club De´me vided us with all the data for our research. This support guarantees that the database we worked on was indeed in agreement with the real network flow of some of the companies in the association. The data mainly consists of the flows of goods of two major French retail chains in the first 12 weeks of 2006, and also of the geographic location of both distribution centers of the two retail chains and the warehouses of their common suppliers. Based on the database offered, we propose that the merging of two supply chains can be achieved as follows. As illustrated in Fig. 1, originally the flows from the suppliers’ (i, j and k) warehouse or plant (WH) are directly shipped to retailers’ (m, n and l) distribution centers (DC). To achieve the merging of the supply chains, we create two kinds of hub in this network, an upstream hub and a downstream hub. Further, two assumptions are made to simplify the problem: (1) the hubs are respectively among the set of WH or DC in the database; (2) goods are allowed to be transported from WH to DC directly or to be consolidated at upstream hubs, downstream hubs or both. Hence, the pooled network here is a three-echelon hub network, with the upstream path between WH and the upstream hub, the midstream path between the upstream hub and the downstream hub, and the downstream path between the downstream hub and DC. More details are discussed in the formulation section. The possibility of reducing CO2 emissions obviously depends on the consolidation of a mass of flows that will be shipped together between the different supply chains in order to increase the truck load factor of each shipment. 2.2. Construction of the database The database mentioned above consists of a large number of diverse goods’ flows (several thousands lines) concerning the most important 106 common suppliers of the two French retail chains. Initially, to be able to identically analyze the flows, they are all measured in number of pallets. This means that the flows are converted into quantities of equivalent full pallets. WH of Supplier i
WH of Supplier j
WH of Supplier k
DC of Retailer m
DC of Retailer n
DC of Retailer l
CARE GRO LIQ
Group A (pal/ weeko 200)
Group B (200 rpal/ week r 600)
Group C (pal/ week 4600)
8 25 21
9 29 25
13 27 34
First the flows are subdivided into three categories according to the different types of products involved: CARE (pharmaceuticals, cosmetics/perfume and hygiene), GRO (grocery) and LIQ (liquids). The main reason for this subdivision is that these classes of product require different handling through the transportation due to their varying logistic behavior. For example, a warehouse for liquid products will not have the same construction features as that for cosmetics. This classification in the French retail supply chains is maintained in this paper. According to the first study of this project by Ballot and Fontane (2008) using the same data, it was shown that mixing flows with very different sizes should be avoided. Moreover, after the classification of products, the suppliers’ warehouses in each class are divided into three groups based on the size of flows in the warehouse, since the flows of each warehouse (even of the same supplier) are independently dealt with in our case. In order to obtain a balanced amount of variables in each group, the boundaries of the groups are 0–200, 200–600 and 4600 pallet/week. Note that the flow per week involved here is an average of the flows covering the 12 weeks studied. This decomposition does not exactly follow the real volume breakdown, but help the optimization process. It could be seen as a limitation of this work. Finally the initial problem is broken down into 9 sub-problems in which the number of suppliers is showed in Table 1. As mentioned above, the procedure of classification and grouping to reduce the extent of the problem, which also determines the framework of the database, is important to decrease the computational time of an optimization model presented in next part.
3. Problem formulation The transportation problem is studied according to the emissions criteria but the cost is also assessed. The economic objective function is well known but to be able to calculate and minimize CO2 emissions from freight transport, we must first define an emissions function according to the mode of transport, as well as an optimization model.
3.1. Objective functions
After pooling
WH of Supplier i
WH of Supplier j / Upstream hub
WH of Supplier k
DC of Retailer m
DC of Retailer n / Downstream hub
DC of Retailer l
Fig. 1. Example of the merging supply chains.
Two modes of transport, road and rail, are taken into account in this research. The emissions function of their mode of transportation is based on several reports in this field, e.g. the MEET report (Hickman et al., 1999; Jorgensen and Sorenson, 1998) and the COST 319 project (Joumard, 1999). Furthermore, CO2 emissions factors in France applied to these functions can be found in some reports of the ADEME (Agency of Environment and Energy Management in France: www.ademe.fr), see De Boissieu (2006), Jancovici (2007). In these reports, CO2 emissions are as a rule measured in terms of tonkm freight transported, which is also adopted in our case, but the ton-km is changed into pallet-km due to the form of the database. In this section, we simply indicate the final formulas to calculate emissions that are applied to the optimization model.
88
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
3.1.1. Road transport emissions First of all, the mode of road transport here refers to transport by Heavy Duty Vehicle (HDV) only (32–40 ton for general merchandise). According to the emissions function for the HDV truck given by Hickman et al. (1999) and Jancovici (2007) some assumptions are made: (a) the average speed is 80 km/h; (b) the gradient of a road is not taken into account; (c) in general the truck considered here is fully loaded with 25 tons for weight or 33 pallets for volume. Particularly, for the care and grocery classes, it is assumed that the truck is fully loaded at the same time by weight and volume (33 pallets weighting 25 tons). However for the liquid class which is doubtless heavier than the others, the truck is fully loaded only with 23 pallets weighing 25 tons. As a result, the final CO2 emissions function with the variable of load is lxm x ð1Þ ev ðxÞ ¼ ðevf eve Þ þ eve c c where ev(x) is the CO2 emissions from a vehicle in kg/km with the variable of load x in pallet; evf is the CO2 emissions of a fully loaded (by weight) vehicle, which is evf ¼1.096 kg/km for HDV truck; eve is the CO2 emissions of an empty vehicle, which is eve ¼0.772 kg/km for HDV truck; c is the volume capacity of a vehicle; as said above, which is 33 pallet/vehicle for the care and grocery classes and 23 pallet/vehicle for the liquid class. 3.1.2. Rail transport emissions According to the data from Fret SNCF (the main freight rail company in France, www.fret.sncf.com), 90% of the freight train locomotives in France are electrically driven, thus only the electrically powered locomotive is considered here. The emissions of air pollutant related to rail transport are calculated in two steps. The first step is the estimation of the energy consumption of a train in kJ per ton-km, which is required to move the train. After this step, the amount of emissions can be calculated according to the energy required. Consequently, the pollutant emissions function related to energy consumption is as follows Ei ¼ WSEC
Tkm 1 BSEFi Tpt 3:6 106
ð2Þ
where Ei is the total emissions of air pollutant i in kg; WSEC is the weight specific energy consumption of the train in kJ/ton-km; Tkm is the amount of freight transported by the train in ton-km; Tpt is the load factor of the train, in ton-km/total train ton; BSEFi is the brake specific emissions factor in g/kWh of energy produced. Formula (2), which is corrected and validated by the main author of the MEET report (Jorgensen and Sorenson, 1998), is the general form of calculation of air pollutant emissions. In this study, we focus on CO2 emissions as the most representative of the emissions of electricity production. In France, according to the report of the ADEME (Jancovici, 2007), the BSEFco2 of EDF (French’s leading energy company, www.edf.com) in 2007 was 42 g/kWh. Some assumptions are also made here to simplify the problem: (a) the average speed is 100 km/h; (b) the mean distance between stops is 100 km; (c) a train is composed of 13 wagons, which is the minimum actual size for chartering a train (half train operation); (d) because of the large weight capacity of the wagon (56 ton/ wagon), a wagon is only fully loaded by a volume equivalent to 36 pallets. The assumptions (c) and (d) imply that the volume capacity of train here is 468 pallet/train. Similar to formula (1), formula (3) given below is used to calculate CO2 emissions from rail transport. lxm x ð3Þ et ðxÞ ¼ ðetf ete Þ þ ete c c where et(x) is the CO2 emissions from a train in kg/km with the variable of load x in pallet; etf is the CO2 emissions of a fully loaded (by volume) train, which is etf ¼0.96 kg/km for the care and grocery classes and etf ¼1.16 kg/km for the liquid class; ete is the CO2
emissions of an empty train, which is ete ¼0.498kg/km; c is the volume capacity of a train; as said above, which is c ¼468 pallet/ train. To compare the CO2 emissions of these two transport modes, illustrated in Fig. 2, it is assumed that one pallet weighs one ton, which means that the ton-km and pallet-km units are now equivalent. The conclusion, here very favorable to rail transport, should be considered with respect to the source of production of electricity that emits very little CO2/kWh in France. Due to the upper integer part of x/c, functions (1) and (3) are in fact piecewise linear and discontinuous functions. That is also the foundation of the optimization model, which will be presented in the next section.
3.1.3. Transport cost For the economic evaluation we consider a formula with a structure similar to (1) or (3), but based on the classical cost approach. As the economic data is not based on published fares but on real cost paid, the specific details are not given here. However on the big picture and from a transport point of view only, the train faces an important fixed cost and therefore is cheaper than truck only on trips greater than approximately 600 km with fully loaded transport means. The economic function, also a piecewise linear and discontinuous function, serves to evaluate the cost of the emissions optimized solution or to classically optimize cost to highlight the differences. 3.2. Optimization model Since the purpose of our optimization project is to minimize the CO2 emissions related to freight transport in two large supply chains, the emissions functions are adopted in the optimization model via an objective function. The discontinuity of the functions results in a Mixed Integer Linear Programming (MILP) problem in our case.
3.2.1. Formalization Croxton et al. (2003b) cites three models, namely the incremental model, the Multiple Choice Model (MCM) and the convex combination model to describe a formula for the piecewise linear function. Furthermore, according to this article and another one concerning the knapsack problem (Kameshwaran and Narahari, 2007), these three MILP models for nonconvex piecewise linear minimization problems are equivalent. The MCM model is adopted in our case to the fact that the emissions produced by freight transport in an arc depend on the sum of the flow of pallets on this
Fig. 2. Comparison of CO2 emissions between road and rail transport.
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
arc and noted x. The linearization of flow and emissions can be displayed as Fig. 3 and is modeled by Eqs. (5)–(9). As described in part 2, the pooling of supply chains in this case is carried out by way of a three-echelon network. Thus the problem actually involves minimizing the sum of CO2 emissions of the three sections of transport (upstream, midstream and downstream). To simplify the presentation of the problem, the flows would only be considered for one period and only one mode of transport would be involved. The extensions to several periods or the two transport modes represent no difficulty and therefore are not presented here as well as the classical cost objective function. The optimization problem is stated as follows (for notation refer to Fig. 3). X X X Min ½dom eðxom Þ þ ½dmn eðxmn Þ þ ½dnd eðxnd Þ ð4Þ om A Au
mn A Am
nd A Ad
with X
fas ysa þvsa zsa
ð5Þ
Piecewise linear constraints X xa ¼ zsa , 8a A A
ð6Þ
eðxa Þ ¼
s A Sa
s A Sa
lbsa ysa r zsa r ubsa ysa , X
ysa r 1,
8a A A,8s A Sa
ð7Þ
8a A A
ð8Þ
8a A A,8s A Sa
ð9Þ
s A Sa
ysa A f0,1g,
Flow-balance constraints xka Z 0, X
8a A A,8kA K
xkom ¼ Rko ,
8k A K,8o A O
ð10Þ ð11Þ
mAM
X
xknd ¼ Bkd ,
8k A K,8d A D
ð12Þ
nAN
X
xkom ¼
X
xkmn ,
8k A K,8m A M
ð13Þ
8k A K,8n A N
ð14Þ
nAN
oAO
X mAM
xkmn ¼
X
xknd ,
dAD
Hub selection constraints X pkom ¼ 1, 8k A K,8o A O
ð15Þ
mAM
pkom A f0,1g,
8k A K,8o A O,8m A M
ð16Þ
xkom rpkom Z,
8k A K,8o A O,8m A M
ð17Þ
8d A D
ð18Þ
X
qnd ¼ 1,
89
qnd A f0,1g,
8n A N,8d A D
ð19Þ
xknd rqnd Z,
8k A K,8n A N,8d A D
ð20Þ
where Z is a large enough constant; K the set of commodities; and each has at least one specific origin and destination node that are respectively called warehouse and distribution center; O, M, N, D denote respectively the set of WH (source nodes), candidate upstream hubs, candidate downstream hubs, destination nodes (DC); Au, Am, Ad the subsets of arc set A, of which Au is the arc set on the upstream flow, Am on the midstream and Ad on the downstream; A the set of all arcs and each arc aAA; dij the distance of arc ijAA; e(x) the piecewise linear function of CO2 emissions related to the flow x; Rko the quantity of commodity kAK supplied at WH oAO; Bkd the quantity of commodity kAK required at DC dAD; Sa represents the number of segments on each aAA; s the piecewise linear segment sASa on each aAA; zsa load of the segment sASa on each aAA; fas the nonnegative fixed value of the intercept of segment sASa on arc aAA, in our case the piecewise function is identical on each arc, so the intercept of the same segment on every arc is always the same according to the mode of transport, such as 0.772 for road transport and 0.498 for rail; vsa slope of segment sASa on each arc aAA. Because of the fact that all the segments are parallel, the slope here is constant with the same product and transport mode, such as vsa ¼ 0:00982 for the care and grocery classes and vsa ¼ 0:0141 for the liquid products in road transport. With regard to rail transport; similarly we have vsa ¼ 0:00099 for care and grocery, and vsa ¼ 0:0014 for liquid products; ysa binary variable on each arc aAA, with ysa ¼ 1if segment sASa contains a nonzero flow, and ysa ¼ 0 otherwise; lbsa and ubsa lower and upper bounds of total flow lie in segment sASa on each arc. In particular, for every arc aAA we assume that lb1a ¼ 0 and ub6a ¼ þ 1 and also for each segment s, ubsa lbsa ¼ wc (which represents the capacity of conveyance depending on the pallet’s density and area of the conveyance); xa the total P flow of all commodities on arc aAA xa ¼ k A K xka 8aAA and decision variables xka the quantity of shipment of commodity kAK on arc aAA; pkom the binary variable, with pkom ¼ 1 if commodity kAK supplied at WH oAO is converged at hub upstream mAM, and 0 otherwise; qnd the binary variable, qnd ¼1 if DC dAD is served by the downstream hub nAN, and 0 otherwise. Constraints (6)–(9) and Eq. (5) which come from the MCM in Croxton et al. (2003b) assure that the total product flow on arc a must lie on a single segment and e(0) ¼0. Eqs. (11)–(14) are the flow-balance constraints in the three-echelon multi commodity network. Constraints (15)–(20) correspond to the assumption that each supplier must have at most one related upstream hub to where his flow will be shipped and each DC must be served by at most one downstream hub. This assumption is to simplify flow management at each inventory point. In addition, if we take into consideration both transport modes (truck and train) in the model at the same time, the objective function, as well as the number of variables, is doubled, as a result of the selection between road and rail transport. Moreover, another additional constraint to bound the minimal flow by train exists only in the case of rail transport. 3.2.2. Approaches to reduce the optimization problem This MILP transportation problem is a NP-Hard problem, see Ahuja et al. (1993), so the computing time of this problem could be influenced by its size. As a representative factor of the size of an instance of a problem, the number of variables is about 23,000 in the biggest problem among the sub-problems presented. This leads us to explore the possibility of being able to reduce the size in order to solve the problem. Thus, in the model we made several assumptions as follows:
nAN
Fig. 3. Variables for each emissions segment.
a. As stated in part 3, the emissions functions are nonconvex piecewise linear and discontinuous, and the number of
90
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
segments depends on the size of flow. Concerning similar problems, some articles based on analyzing piecewise function efficacy indicate that some attributes of the function deeply influence computational time, e.g. the continuity and the number of segments, see Croxton et al. (2007), Kameshwaran and Narahari (2007) and Marins et al. (1997). Therefore, to simplify the function in road transport for example, we assume that a truck is fully loaded in an arc with a shipment above a volume equivalent to 5 trucks in one week. In other words, the maximum number of segments is set at 6 (so Sa ¼6 and s r6), in addition, the slope of the sixth segment is 0.3321 for the care and grocery classes, and 0.476 for liquids. Due to a significant difference in load capacity between trucks and trains, the function in rail transport is reduced to 2 continuous segments, which represent the case that trains are always fully loaded above one train. The slope of the second segment in rail transport is 0.002 for the care and grocery classes, and 0.00248 for liquids. b. According to Croxton et al. (2003a), with the data in our case we can tighten the LP relaxation of the model by limiting the maximum bound of each product’s flow in any arc in the network. c. Considering the fact that consolidation would not occur if the distance between WH and the upstream hub or between the downstream hub and DC is too great, a condition of distance in both upstream and downstream paths is introduced to filter candidate upstream and downstream hubs. In other words, the possible hubs are situated within a certain radius from WH or DC. We studied the sensibility of the solution according to radius variation, and 100 km was shown to be the best compromise between computing time and solution quality.
3.2.3. Lower limit of emissions Aside from the model presented to minimize CO2 emissions, it might also be valuable to calculate the lower limit of the emissions in transport, meaning that all shipments are direct from WH to DC with an absolutely fully loaded transport mean. As a result, Eq. (5) in the optimization model will be replaced by a completely linear objective function. With this limit we are not looking for an operational solution but a measure of the remaining potential of emissions saving with the same transport mean. The assumptions made allow all the sub-problems to be solved by coding and executing a computational program. The results obtained are discussed in the next part.
4. Results and discussions The model presented is coded in ILOG’s OPL 6.3 software with CPLEX 12.1 and run on Quad CPU Q6700 (2.66 GHz) hardware having 4 GB of RAM. The MILP problem is solved with OPL’s default settings, except Memory Emphasis parameter to true to deal with the large size of the problems, refer to ILOG (2008). Furthermore, all the results obtained showed a gap of less than 3% between the best integer solution and the upper bound. Note that all the values exhibited in the following tables are measured in tons. It should also be recalled that, as stated in the methodology part, the suppliers in each class are divided in 3 groups: group A o200 pallet/week, group B 200–600 pallet/week and group C 4600 pallet/week. 4.1. Road transport CO2 emissions optimization Initially, road transport is the exclusive transport mode involved in the optimization problem. In the framework of the 9 sub-problems
presented, the purpose of this section is to evaluate the performance of merging supply chains in terms of CO2 emissions (emissions/ pooled in the tables) using the optimization model, by comparing results with the emissions from the actual transport system (actual emissions in the tables). Additionally, the lower limit of emissions (minimum emissions in the tables) will also be introduced as the optimum situation in which either the trains or trucks are absolutely fully loaded (linear emissions function). The results in Table 2 show that the approach of merging supply chains significantly reduces CO2 emissions from transport, despite the fact that its gain in emissions is about half the maximum gain defined by the lower limit, namely the theoretical minimum emissions. Furthermore, it is obvious that the relative reduction in group C is much less than the other groups. This situation can be explained by the fact that at present the suppliers with massive flows logically make more efforts to saturate their means of transport and these efforts are more likely successful. Besides the reduction of CO2, it is also important to indicate that the number of transport paths falls because of the network pooling, as shown in Fig. 4. 4.2. Road and rail transport CO2 emissions optimization This section aims to determine the effect of joint road/rail transport in reducing CO2 emissions. This means that road and rail transport are alternative transport modes to be chosen according to the size of flows and their performance in terms of emissions. To reflect practical situations in transportation, a lower limit of flow for rail transport is fixed at 468 pallets, which is a volume equivalent to half of a train of 26 wagons. According to the results in Table 3, especially in group C, it can be concluded that joint road and rail transport is a significant way to reduce CO2 emissions, provided the electrically powered train generates low emissions in France thanks to the low emissions electricity Table 2 Emissions from road transport pooled network per week (ton CO2). Groups of suppliers CARE Actual emissions Emissions/pooled Reduction Absolute Relative (%) Minimum emissions Reduction Absolute Relative (%) GRO Actual emissions Emissions/pooled Reduction Absolute Relative (%) Minimum emissions Reduction Absolute Relative (%) LIQ Actual emissions Emissions/pooled Reduction Absolute Relative (%) Minimum emissions Reduction Absolute Relative (%)
A
B
C
S
51 28
93 62
319 290
463 380
23 45 11
31 33 45
29 9 266
83 18 322
40 78
48 52
53 17
141 30
132 67
309 219
670 628
1111 916
65 49 29
90 29 147
42 6 568
195 18 744
103 78
162 52
102 15
736 33
183 93
291 255
1257 1193
1731 1541
90 49 39
36 12 177
64 5 1134
190 11 1350
144 79
114 39
123 10
381 22
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
91
Fig. 4. Example of transportation network before and after merging by road transport (CARE group C).
Table 3 Emissions from road and rail transport pooled network per week (ton CO2). Groups of suppliers CARE Actual emissions Emissions/pooled Reduction Absolute Relative (%) GRO Actual emissions Emissions/pooled Reduction Absolute Relative (%) LIQ Actual emissions Emissions/pooled Reduction Absolute Relative (%)
A
B
C
S
51 28
93 44
319 157
463 229
23 45
49 53
162 51
234 50
132 67
309 210
670 257
1111 534
65 49
99 32
413 62
577 52
183 93
291 247
1257 485
1731 812
90 49
44 15
772 61
919 53
production mainly nuclear and hydro. However, it should be noted that the problem of rail transport here is to obtain a (consolidated) flow that simultaneously meets the weekly demand of retailers and the minimum flow for rail transport (468 pallets). That is to say that the weekly stock level of retailers cannot be challenged. Comparing to Fig. 4, we can easily remark in Fig. 5 that the introduction of the rail transport simplifies the transport network and indicates that it is the means of transport of massive flows.
4.3. Economic evaluation of the solutions optimized for the emissions In complement of the emissions optimization, the solutions are evaluated by the computation of the transportation cost of the different proposed supply chains. Table 4 shows the results of these evaluations for the 9 solutions (trucks only). These results display to a certain extent a correlation between emissions and costs. This correlation is not as perfect as for a single truck as the structure of the network evolves with the pooling of the supply networks. But from a global point of view, we can affirm that with a single transport mode the emissions reduction and the cost reduction are compatible, thanks to the consolidation of the flows resulting from the pooling of the supply networks. The cost evaluations of the road and rail solutions show a different situation when rail is selected by the optimization, as Table 5 illustrates. This occurs typically for group C and less for groups B. The cost of groups C, suppliers with high flows, are significantly more expensive due to rail cost but also have the best
emissions reduction in absolute. In total, the 52% CO2 emissions reduction is achieved with an increase of 85% of the total cost. Table 5 also demonstrates that there is no economic interest for a supply chain or even a pooled network of supply chains with important flows (group C) to prefer an emissions optimized solution with train operations as it is much more expensive with today costs in a country like France, despite the interest in GHG emissions reduction.
4.4. Emissions vs. cost optimizations All the results from Tables 2 to 5 are based on the minimization of the emissions function. To measure the difference with the economic approach, we also use the cost function as the objective function of the same transportation problem. The minimization of cost is given in Table 6 for the truck only. The results are very similar and the small differences are within the 3% gap between the solution and the lower bound and therefore can be seen as round-offs of calculation with no signification. The conclusion is a convergence between the economic and the emissions approaches with trucks only in the pooling of supply networks. The minimization of the cost for the truck and the train formulation lead also to a similar result as shown in Table 6 as train transportation is more expensive with the trains’ cost nowadays. In the near future, the cost of freight by rail may increase less than the cost of freight by road, due to the opening to market of the train operations and less taxation, thanks to reduced CO2 emissions. Or in another way we could see it as the trade-off between economy and environment, as proposed with a different heuristic method by Frota Neto et al. (2008) to find the Pareto efficiency frontier. In this perspective, we compute the impact of a relative variation of truck and train costs in the framework of a normalized total cost. This impact is implemented with a factor a applied to the truck cost and a factor b applied to the train cost and the optimization problem was solved several times with variation of a and b to determine frontier of efficiency. Fig. 6 shows the gap between the economic solution and the emissions solution for the high volume (group C) Care product class, as an example. It highlights the strong sensitivity on emissions of the economic solution under a variation of relative truck and train costs. If the train cost decreases more rapidly than the truck cost increase in the future, the pooling of supply chain offers a solution for both cost efficiency and emissions reduction for the important flows. But the current structure of the costs also reveals that the pooling is not an attractive solution from an economic standpoint to reach the solution of emissions minimization as it requests very drastic changes in this case (truck cost increased by more than 200% and train cost reduced by more than 70%).
92
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
Fig. 5. Example of transportation network before and after merging by joint transport (CARE group C).
Table 4 Transport cost for the emissions minimization of the pooled network (trucks only). Groups of suppliers
A
Optimized emissions/cost
TCOn2
103 h
TCOn2
51 28
43 32
93 62
23 45
11 26
132 67
LIQ Actual Emissions/pooled Reduction Absolute Relative (%)
C 103 h
S
TCOn2
103 h
TCOn2
103 h
115 103
319 290
546 536
463 380
704 671
31 33
12 10
29 9
10 2
83 18
33 5
114 84
309 219
375 353
670 628
1160 1159
1111 914
1649 1596
65 49
30 26
90 29
22 6
42 6
1 0
197 18
53 3
183 93
137 93
291 255
324 318
1257 1193
1737 1716
1731 1541
2189 2127
90 49
44 32
36 12
6 2
64 5
21 1
190 11
71 3
CARE Actual Emissions/pooled Reduction Absolute Relative (%) GRO Actual Emissions/pooled Reduction Absolute Relative (%)
B
Table 5 Transport cost for the emissions minimization of the pooled network (trucks and trains). Groups of suppliers Optimized emissions/cost CARE Actual Emissions/pooled Reduction Absolute Relative (%) GRO Actual Emissions / pooled Reduction Absolute Relative (%) LIQ Actual Emissions/pooled Reduction Absolute Relative (%)
A
B n
3
C n
3
S n
3
10 h
TCOn2
103 h
TCO2
10 h
TCO2
10 h
TCO2
51 28
43 32
93 44
115 145
319 157
546 1202
463 229
704 1379
23 45
11 26
49 53
30 26
162 51
656 120
234 51
675 96
132 67
114 84
309 210
375 450
670 257
1160 2890
1111 534
1649 3424
65 49
30 26
99 32
75 20
413 62
1730 149
577 52%
1775 108
183 93
137 93
291 247
324 351
1257 485
1737 3183
1731 825
2189 3627
90 49
44 32
44 15
27 8
772 61
1446 83
906 52
1429 65
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
93
Table 6 Transport emissions for the cost minimization of the pooled network (trucks only). Groups of suppliers
A
B
TCO2
103 h
CARE Emissions optimization Transport cost optimization
28 29
32 31
62 65
GRO Emissions optimization Transport cost optimization
67 70
84 86
LIQ Emissions optimization Transport cost optimization
93 93
93 92
103 h
TCO2
103 h
103 101
290 295
536 534
380 389
671 666
219 224
353 339
628 645
1159 1155
914 939
1496 1580
255 258
318 315
1193 1217
1716 1722
1541 1568
2127 2129
Apart from the environmental aspect, the economic aspect of transport was also taken into consideration and shows a strong convergence between transportation cost and emissions within the pooling framework with trucks only. This convergence is no more achieved with joint trucks and trains operations with the current cost structure and logistic organization on important flows. A significant part of the trade-off could be reached by more competitive rail freight operations and truck cost increases. However future research is still needed to generalize these results obtained in a specific context and in the logistic organization and transportation technology to overcome the greater cost of the environmental more friendly transportation means.
350
CO2 Emissions (ton)
Current solution Min Cost (+2%, -18%)
250
(+16%, -33%) (+60%, -60%)
200
(+200%, -70%) Min CO2
150 100
400
600
800
1000 1200 Transport Cost (k )
103 h
S
TCO2
TCO2
400
300
C
1400
Fig. 6. Evolution of truck and train costs relative variations on CO2 emissions (truck cost evolution, train cost evolution).
5. Conclusion This paper explored the effect of pooling supply chains’ networks on reducing CO2 emissions from transport with two possible modes, i.e. road and rail, in the context of a national distribution network of two major French retailer chains. We first pointed out the principal drawback of current freight consolidation, namely that it currently takes place in a local and fragmentary manner using carriers and 3PL. This conducted us to search for another approach to achieve a more efficient consolidation. In the aim of reducing CO2 emissions from freight transport, we propose the pooling of supply chains at the strategic level, an approach that demands further and long-term collaboration between the actors of supply chains (suppliers and retailers in this case). In this article geographical and products’ flows pooling were proposed, namely geographical consolidation among suppliers or retailers with similar flows. Moreover, an optimization model with a piecewise linear objective function was set up to quantitatively evaluate its impact on reducing CO2 emissions. Benefiting from our ´ter, we tested the model with the real association with Club De´me data of merchandise flows over 12 weeks in a large national distribution network composed of two French retailers’ supply chain. A split in size and type of flows according to real operations allows us to reduce the size of the problem and to obtain solutions with a gap less than 3%. Finally, the results in Tables 2 and 3 globally yield a relative reduction of CO2 emissions of 14% exclusively with road transport and of 52% with joint road and rail transport. It should be recalled that all the results obtained in this paper are conditional upon the same, even improved weekly service rate to retailers. Consequently, it can be concluded that pooling supply chains as studied here provides a solution to significantly reduce CO2 emissions from freight transport.
Acknowledgments Playing an indispensable role in our research, we gratefully ´ter Associaacknowledge the support provided by the Club De´me tion. The authors are also grateful to the R2DS program of the Ile-deFrance Region that provided us with funding for this project. References Ahuja, R.K., Magnanti, T.L., Orlin, J.B., 1993. Network Flows: Theory, Algorithms, and Applications. Prentice Hall, New Jersey. Ballot, E., Fontane, F., 2008. Reducing greenhouse gas emissions through the collaboration of supply chains: lessons from French retail chains. In: Proceedings of the International Conference on Information Systems, Logistics and Supply Chain Conference. Madison, Wisconsin. Bookbinder, J.H., Higginson, J.K., 2002. Probabilistic modeling of freight consolidation by private carriage. Transportation Research Part E: Logistics and Transportation Review 38 (5), 305–318. Cheung, R.K., Tong, J.H., Slack, B., 2003. The transition from freight consolidation to logistics: the case of Hong Kong. Journal of Transport Geography 11 (4), 245–253. Cooper, W.W., Hemphill, H., Huang, Z., Li, S., Lelas, V., Sullivan, D.W., 1997. Survey of mathematical programming models in air pollution management. European Journal of Operational Research 96 (1), 1–35. Croxton, K.L., Gendron, B., Magnanti, T.L., 2003a. Models and methods for merge-intransit operations. Transportation Science 37 (1), 1–22. Croxton, K.L., Gendron, B., Magnanti, T.L., 2003b. A comparison of mixed-integer programming models for nonconvex piecewise linear cost minimization problems. Management Science 49 (9), 1268–1273. Croxton, K.L., Gendron, B., Magnanti, T.L., 2007. Variable disaggregation in network flow problems with piecewise linear costs. Operations Research 55 (1), 146–157. De Boissieu, C., 2006. Rapport du Groupe de Travail ’’Division par Quatre des E´missions de Gaz a Effet de Serre de la France a L’horizon 2050’’ Sous la Pre´sidence de Christian de Boissieu. MINEFI, Paris, France. Ergun, O., Kuyzu, G., Savelsbergh, M., 2007. Reducing truckload transportation costs through collaboration. Transportation Science 41 (2), 206–221. Frota Neto, J.Q., Bloemhof-Ruwaard, J.M., van Nunen, J.A.E.E., van Heck, E., 2008. Designing and evaluating sustainable logistics networks. International Journal of Production Economics 111 (2), 195–208. Groothedde, B., Ruijgrok, C., Tavasszy, L., 2005. Towards collaborative, intermodal hub networks: a case study in the fast moving consumer goods market. Transportation Research Part E: Logistics and Transportation Review 41 (6), 567–583.
94
S. Pan et al. / Int. J. Production Economics 143 (2013) 86–94
Hageback, C., Segerstedt, A., 2004. The need for co-distribution in rural areas: a study of Pajala in Sweden. International Journal of Production Economics 89 (2), 153–163. Hickman, J., Hassel, D., Joumard, R., Samaras, Z., Sorenson, S., 1999. Methodology for Calculating Transport Emissions and Energy Consumption. Transport Research Labotory, Crowthorne, UK. ILOG, 2008. O.P.L. Studio 6.1: the user’s manual. ILOG Inc. Jancovici, J.M., 2007. Bilan Carbones: Calcul des Facteurs D’e´missions et Sources Bibliographiques Utilise´es.. ADEME, Paris, France. Jorgensen, M.W., Sorenson, S.C., 1998. Estimating emissions from railway traffic. International Journal of Vehicle Design 20 (1), 210–218. Joumard, R., 1999. Methods of Estimation of Atmospheric Emissions from Transport: European Scientist Network and Scientific State-of-the-Art (Action COST 319 Final Report). INRETS, Lyon, France.
Kameshwaran, S., Narahari, Y., 2007. Nonconvex piecewise linear knapsack problems. European Journal of Operational Research 192 (1), 56–68. Le´onardi, J., Baumgartner, M., 2004. CO2 efficiency in road freight transportation: status quo, measures and potential. Transportation Research Part D 9 (6), 451–464. Marins, F.A.S., Senne, E.L.F., Darby-Dowman, K., Machado, A.F., Perin, C., 1997. Algorithms for network piecewise-linear programs: a comparative study. European Journal of Operational Research 97 (1), 183–199. McKinnon, A., Ge, Y., Leuchars, D., 2003. Analysis of Transport Efficiency in the UK Food Supply Chain.. Heriot-Watt University, Edinburgh, UK. Tyan, J.C., Wang, F.-K., Du, T.C., 2003. An evaluation of freight consolidation policies in global third party logistics. Omega 31 (1), 55–62.