Transportation Research Part D 23 (2013) 81–89

Contents lists available at SciVerse ScienceDirect

Transportation Research Part D

journal homepage: www.elsevier.com/locate/trd

Heterogeneous fixed fleet vehicle routing considering carbon

emission
Yong-Ju Kwon a, Young-Jae Choi a, Dong-Ho Lee b,⇑
a
Department of Industrial Engineering, Hanyang University, Seoul, Republic of Korea
b
Department of Industrial Engineering, Graduate School of Technology and Innovation Management, Hanyang University, Seoul, Republic of Korea

a r t i c l e i n f o a b s t r a c t

Keywords: The paper considers heterogeneous fixed fleet vehicle routing with carbon emission to
Heterogeneous vehicle routing minimizing the sum of variable operation costs. A cost-benefit assessment of the value
Carbon emission and trading of purchasing or selling of carbon emission rights, using a mixed integer-programming
Tabu search model to reflect heterogeneous vehicle routing, is incorporated. Essentially, the use of a
carbon market as a means of introducing more flexibility into an environmentally con-
strained network is considered. Tabu search algorithms are used to obtain solutions within
a reasonable amount of computation time. In particular, we show the possibility that the
amount of carbon emission can be reduced significantly without sacrificing the cost due
to the benefit obtained from carbon trading.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction

This study considers carbon emission in logistics systems, focusing on the vehicle routing problem. Most previous studies
on vehicle routing have focused on minimizing the distance traveled, the fleet size, and similar traditional parameters. Due to
the environmental concerns that have emerged over the last decades, there has been an increasing interest in bring environ-
mentally factors into the routing problem.1 This interest has more recently been compounded as new policies, such as carbon
trading, have been introduced for making industry more aware of their environmental cost. In particular, measures such a car-
bon market can note and stimulate a more environmentally efficient use of transportation, but in some cases, because it allows
holders of excessive carbon assets to sell their carbon emission rights, positively affect the financial performance of logistic ser-
vice providers.
The vehicle routing problems can be classified by vehicle types i.e., homogeneous and heterogeneous vehicle routing; we
focus on the latter. Given the diversity of the vehicle fleet and variations in carbon emissions, we look specifically at the car-
bon emission based heterogeneous vehicle routing problem (C-HVRP),2 and further narrow it down to the case of a fixed fleet
of vehicles and the problem of determining their optimal routing. To represent the problem mathematically, we adopt a mixed
integer-programming model for objective of minimizing the sum of variable operation costs, including a cost-benefit assess-
ment of acquiring carbon rights under a cap-and-trade regime.

⇑ Corresponding author.
E-mail address: [email protected] (D.-H. Lee).
1
Ericsson et al. (2006) identifies the impact of traffic disturbances on fuel consumption; Kara et al. (2007) defines an energy-minimizing vehicle routing
problem that minimizes the weighted load instead of a distance based objective function; Tavares et al. (2008) look at the effects of road inclination and load;
Bektasß and Laporte (2011) develop a model with an objective function that includes the travel distance as well as emission, operational and driver costs; and
Xiao et al. (2012) extended the capacitated vehicle routing problem taking into account travel distance and load impacts on fuel costs.
2
Pessoa et al. (2009) and Brandão (2011) provide recent studies of the traditional HVRP.

1361-9209/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.trd.2013.04.001
82 Y.-J. Kwon et al. / Transportation Research Part D 23 (2013) 81–89

Fig. 1. Heterogeneous vehicle routes: example.

Table 1
Calculating the amount of carbon emission using the fuel based method.
P
CO2 Emissions = j(FueljEFj)
Fuel consumption (Fuelj) Represented by fuel sold (TJ)
Activity data Calorific value 35.4 MJ/‘
Conversion factor 1 TJ = 106 MJ
Coefficient Emission factor (EFj) 74,100 kg CO2/TJ
Fuel type (j) Diesel (road transportation)

2. Problem description

Let G = (N, A) be a logistics network, where N = {0, 1, 2, . . . , n} is the set of nodes and A = {(i, j): i, j 2 N, i – j} is the set of arcs.
Node 0 denotes the depot, i.e., departure and arrival base of each vehicle, and the remaining nodes 1, 2, . . . , n represent cus-
tomers. A fleet K of heterogeneous vehicles of varying capacities and carbon emission levels is available at the depot. It is
assumed that the fleet size is fixed, and that each vehicle can travel at most one route. Each customer has a nonnegative
demand that must be satisfied and a distance between customers i and j is associated with each arc (i, j) 2 A. These distances
are symmetric and satisfy the triangle inequality.
The problem is to determine vehicle routes, given their capacities and customer demands, that minimize the sum of var-
iable operation costs and carbon emission trading net costs. Here consider a deterministic version of the problem; all data,
such as customer demands, vehicle capacities, and carbon emission amounts, are given in advance, and variable operation
cost is directly proportional to the distance traveled.
The decision variable is the same as that of the ordinary HVRP. More specifically, a solution can be represented as a set of
routes R1, R2, . . . , RV, where Rv = (0, iv1, iv2, . . . , 0) and V denotes the number of vehicles, with ivl being an index for the lth node
on route v. Fig. 1 shows an example involving three heterogeneous vehicle routes and three truck types; 1-, 2-, and 5-ton
vehicles.
Here, the cost benefit calculation associated with carbon trading is calculated from the difference between the actual car-
bon use in the network and the upper limit for the amount of carbon emission. In other words, if carbon emissions are great-
er than the upper bound, then carbon emission ‘‘cost’’ are incurred because additional carbon allowances have to be
purchased. The upper limit for carbon emissions is determined by a short-term cap-and-trade framework where the market
has cleared and thus at the margin for any small player the effective carbon price is known in advance. We use the fuel-based
method to determine this price in which carbon emissions from all sources of combustion are estimated on the basis of fuel
burned and average emission factors. (Eggleston et al., 2006) (Table 1)
The C-HVRP deployed involves two main constraints; customer demands must be satisfied and the demand over a vehi-
cle’s route must not exceed its capacity. Other assumptions that remain same as for the ordinary HVRP, i.e., all vehicles start
and end at the depot, a set of customers is visited in between, and each customer is visited once. To represent the problem, a
mixed integer programming model is used following that of Yaman (2006). The following notations are used.
 Y.-J. Kwon et al. / Transportation Research Part D 23 (2013) 81–89 83

Sets and indices

i, j nodes (customers), i, j 2 N|{0}
k vehicle types, k = 1, 2, . . . , mk, where mk is the number of type k vehicles

Parameters
qj demand at node (customer) j
dij distance of between nodes i and j (km)
vk variable operation cost of vehicle k per unit distance ($/km)
Qk capacity of type of vehicle k (ton)
ek amount of carbon emitted by vehicle k per unit distance (g/km)
AE upper limit for the amount of carbon emissions (g)
c net benefit of unit carbon emission trading (¢/g)

Decision variables
xijk is one if a type k vehicle travels from node i to j, otherwise zero
fij vehicle load from i to j
PC difference between the carbon consumption and the upper limit for carbon emissions

The mixed integer-programming model is given below. The set AK is defined as AK = {(i, j, k): (i, j) 2 A, k 2 Kij}, where Kij is
the set of vehicles that can serve both customers i and j without violating capacity constraints
XX
½P Minimize v k  dij  xijk þ c  PC
k2K ði;jÞ2A

subject to
XX
xijk ¼ 1 for all i ð1Þ
k2K i j2N

XX
xijk ¼ 1 for all j ð2Þ
k2K i i2N

X
x0jk 6 mk for all k ¼ 1; 2; . . . ; mk ð3Þ
j2Nnf0g

X X
fji  fij ¼ qi for all i 2 Njf0g ð4Þ
j2N j2N

X X
qj  xijk 6 fij 6 ðQ k  qi Þ  xijk for all ði; jÞ 2 A ð5Þ
k2K ij k2K ij

XX
dij  ek  xijk  AE ¼ PC ð6Þ
k2K i ði;jÞ2A

xijk 2 f0; 1g for all ði; j; kÞ 2 AK ð7Þ

fij P 0 for all ði; jÞ 2 A ð8Þ

The objective function denotes the sum of operation cost and carbon emission trading cost. Constraints (1) and (2) imply
that each customer is visited once by a vehicle, and constraint (3) limits the number of vehicles of each type. Constraint (4)
ensures that the demand of each customer be satisfied while removing sub-tours, and constraint (5) represents the vehicle
capacities, and constraint (6) balances actual and permitted carbon emission. Finally, constraints (7) and (8) represent the
conditions of decision variables.3

3. Tabu search algorithms

Tabu search (TS) is a meta-heuristic. Starting from an initial solution, the TS algorithm generates a new solution S0 in the
neighborhood of the original S, which is called a move. To escape from a local minimum in its search for the global minimum,

3
We see that the C-HVRP is NP-hard because its special case, the capacitated vehicle routing problem (CVRP) with homogeneous vehicles, is proved to be NP-
hard, and thus the tabu search algorithms offer reasonable solutions within a reasonable amount of computation time.
84 Y.-J. Kwon et al. / Transportation Research Part D 23 (2013) 81–89

Fig. 2. Neighborhood generation methods.

the move can be made to a neighbor solution even though it is worse than the current solution, and to avoid cycling on the
search process, the TS defines a set of moves that are tabu (forbidden), and are stored in a set K (the tabu list). The moves in
K define all tabu moves that cannot be applied to the current solution. The size of K is bounded by l; the tabu list size or tabu
tenure. If |K| = l, before adding a move to K, one must remove a move, usually the oldest in it. A tabu move can be chosen if it
creates a solution better than the current best solution: the aspiration criterion to revoke tabu status of some moves (Glover,
1989, 1990).
An application of TS is characterized by; solution representation method, initial solution method, neighborhood genera-
tion methods, definition of tabu moves with the tabu list size, and termination condition. First, a solution is represented as a
set of routes R1, R2, . . . , RV, where Rv = (0, iv1, iv2, . . . , 0). The initial solution is obtained as follows. First, from the set of unas-
signed nodes, the one with the greatest demand is selected with ties broken arbitrarily. Then, the selected node is added
to the route end of the vehicle with the largest capacity. If this capacity is inadequate, it is assigned to the vehicle with
the second largest capacity, and so on. Finally, after all nodes are assigned, the solution is improved using the 3-opt heuristic
(Lin and Kernighan, 1973).
The neighborhood generation structure determines the extent and the quality of the solution space explored. Here we
consider three neighborhood generation methods; insertion, swap and hybrid. Unlike the traditional approach to the vehicle
routing problem, these three methods evaluate the solutions include criterion involving carbon emissions.4 After the best
neighborhood solution is obtained from a neighborhood generation method, it is improved further by applying the 3-opt heu-
ristic to each of the changed routes. The move can be made to the improved neighbor solution, even though it may be worse
than the current solution.
The detailed descriptions of the insertion and the swap methods are given below, and Fig. 2 offers examples. In the figure,
it can be seen that the moves are made to neighbor solutions involving smaller amounts of carbon emission (E) without re-
gard to travel distance (D).

 Insertion method: This generates the neighborhood solutions by selecting Rp and Rq and then inserting a node on route Rp
to an arbitrary position on Rq.
 Swap method: This generates the neighborhood solutions by selecting a node on each Rp and Rq, and interchanging them.

4
See Laporte et al. (2000) and Toth and Vigo (2002) for details of approaches to the classical vehicle routing problems.
 Y.-J. Kwon et al. / Transportation Research Part D 23 (2013) 81–89 85

Table 2
Modified benchmark test instances.

Instances Number Upper Heterogeneous vehicle types

of nodes limit (g)
A B C D E F

QAa vAb nAc eAd QB vB nB e B QC vC nC eC QD vD nD eD QE vE nE e E QF vF nF e F

T13 50 510755 2 1.0 4 4.13 3 1.1 2 6.23 4 1.2 4 8.34 7 1.7 4 14.64 12 2.5 2 25.14 20 3.2 1 41.95
T14 50 398043 12 1.0 4 25.14 16 1.1 2 33.55 30 1.4 1 62.96
T15 50 435360 5 1.0 4 10.44 10 1.6 3 20.94 16 2.0 2 33.55
T16 50 453951 4 1.0 2 8.34 8 1.6 4 16.74 14 2.1 3 29.35
T17 75 666573 5 1.0 4 10.44 12 1.2 4 25.14 20 1.5 2 41.95 35 1.8 1 73.47
T18 75 653370 2 1.0 4 4.13 5 1.3 4 10.44 10 1.9 2 20.94 15 2.4 2 31.45 25 2.9 1 52.46 40 3.2 1 83.97
T19 100 850839 10 1.0 4 20.94 20 1.4 3 41.95 30 1.7 3 62.96
T20 100 874460 6 1.0 6 12.54 14 1.7 4 29.35 20 2.0 3 41.95
a
Vehicle capacity (unit: ton).
b
Variable transportation cost per unit distance (unit: $/km).
c
Number of available vehicles for each vehicle type.
d
Amount of carbon emission per unit distance (unit: g/km).

 Hybrid method: This generates the neighborhood solutions by selecting the insertion and the swap methods randomly at
an iteration and then taking the best as the move at the current iteration after evaluating all possible alternatives for
insertion or swap.

Regarding tabu moves in the TS algorithms, using the insertion method, if node i on route Rp is moved to route Rq at iter-
ation t, the move of node i from Rp to Rq is declared tabu during t + l. Similarly, in the swap method, the moves between two
routes, i.e., moves of node i (j) from Rp (Rq) to Rq (Rp), are declared tabu. The tabu list size l is determined using the simple
dynamic tabu term rule of Glover and Laguna (1993); i.e. l is set to an integer uniformly distributed over an interval [lmin,
lmax]. The oldest tabu move is removed before adding a new one if the tabu list is full. A tabu move can always be chosen
if it creates a solution better than the incumbent solution, i.e., the best objective value obtained so far. Finally, the TS algo-
rithms are terminated if there is no improvement for a certain number LTS after consecutive iterations.

4. Results

To test the performance of TS algorithms, computational experiments are applied to modified benchmark instances. We
compare TS-insertion, TS-swap and TS-hybrid algorithms, and report the costs, the distances traveled and the amounts of
carbon emission for each benchmark instance. Computer central processing unit (CPU) seconds are reported to show their
practical applicability.5 We performed sensitivity analyses on the effects the changes in the upper limit for the amount of car-
bon emitted and the net unit cost of traded carbon.
The test uses eight modified benchmark instances with 50, 75 and 100 nodes (Taillard, 1999). The demand (qj) at each
node is set to the original value of the benchmarks and the variable operation cost (vk) of each vehicle type was set to
the fuel cost per unit distance (km). Because the benchmarks were originally designed for the HVRP, we generate the amount
of carbon emission for each vehicle type and the net unit carbon cost (in all cases an actual benefit). Carbon emissions for
each vehicle type (ek) are estimated based on Xiao et al. (2012) framework, i.e., Y = 0.0000793X  0.026, where Y and X are
fuel the consumption rate (l/kg) and the weight of vehicle (kg). Here, X is replaced by vehicle capacity, Q. We assume that fuel
consumption increases linearly with distance traveled. For each vehicle type, carbon emissions are calculated using the data
in Table 1. The unit carbon emission cost is set at $25/ton. The upper limit (AE) for carbon emission for each instance is set to
the amount obtained from the initial solutions because estimate the exact value cannot be estimated (Table 2).
Preliminary calculations are conducted to refine appropriate parameters. First, lmin and lmax for the tabu list are set to 5
and 10, and LTS, the termination condition, to ten times the number of nodes. Finally, each test instance was solved ten times
due to the randomness of the TS algorithms, i.e., selecting the neighborhood solutions. The results are summarized in Table 3
where it can be seen that the hybrid neighborhood generation method betters the others. Although this may be expected
because the hybrid method includes the insertion and the swap methods, and hence searches a larger solution space, the
improvement is significant in terms of both financial cost and the amount of carbon emission, together with the percentage
reductions below the upper limit. Finally, although the TS-hybrid algorithm takes more computation time, it can give solu-
tions for the largest situations (with 100 nodes) within a minute.
Table 4 summarizes the results of the sensitivity analysis on changes in the net unit cost of carbon trading (which is gen-
erally a benefit, but because a cost when additional permits must be purchased), and the upper limit on the carbon emission
using the TS-hybrid algorithm. We see that carbon trading increases and hence the cost decreases as unit cost benefits of
trading increase. This indicates that the TS-hybrid algorithm can be used to maximize the benefit obtained from carbon

5
The TS algorithms were coded in C, and the test was done on a personal computer with an Intel core i7-2600 processor at 3.40 GHz clock speed.
 86
Table 3
Results of the TS algorithms.

Problem Nodes Carbon upper TS-Insertion TS-Swap TS-hybrid

ID limit

Y.-J. Kwon et al. / Transportation Research Part D 23 (2013) 81–89

Cost Distance Emission CPU Cost Distance Emission CPU Cost Distance Emission CPU
seconds seconds seconds
T13 50 510,755 2925.06a 1717.91 483425 (5.4%)c 0.1 2284.70 1253.23 368363 (27.9%) 0.3 2205.05 1228.52 357245 (30.1%) 0.3
2925.06b 1717.91 483425 (5.4%)d 0.1 2253.97 1214.94 361667 (29.2%) 0.4 1994.76 1108.27 323287 (36.7%) 0.5
T14 50 398,043 1069.45 964.31 379022 (4.8%) 0.2 1045.79 945.55 369964 (7.1%) 0.3 1087.85 982.55 385531 (3.1%) 0.2
994.56 889.98 352535 (11.4%) 0.3 1035.38 936.23 366139 (8.0%) 0.3 1007.63 902.91 357161 (10.3%) 0.3
T15 50 435,360 1737.26 1149.65 368619 (15.3%) 0.2 1569.60 1067.90 336085 (22.8%) 0.4 1315.78 916.12 284122 (34.7%) 0.4
1588.85 1051.51 338006 (22.4%) 0.2 1401.78 952.61 300447 (31.%) 0.7 1213.04 856.80 262869 (39.6%) 0.7
T16 50 453,951 1808.43 1088.44 344218 (24.2%) 0.3 2006.98 1220.59 383503 (15.5%) 0.2 1502.94 909.36 286630 (36.9%) 0.7
1542.53 906.34 291000 (35.9%) 0.4 1943.50 1182.92 371481 (18.2%) 0.3 1380.58 829.16 262577 (42.2%) 1.3
T17 75 666,573 1944.80 1467.31 548619 (17.7%) 0.6 1870.47 1427.63 526440 (21.0%) 1.9 1735.85 1310.29 488563 (26.7%) 1.6
1850.76 1391.50 521368 (21.8%) 0.6 1764.49 1344.00 496826 (25.5%) 2.5 1596.50 1210.82 446790 (33.0%) 2.6
T18 75 653,370 3668.25 1788.03 623168 (4.6%) 0.2 3484.57 1672.01 590418 (9.6%) 0.3 3268.97 1526.99 551060 (15.7%) 0.4
3609.65 1742.95 611991 (6.3%) 0.2 3419.40 1628.81 578619 (11.4%) 0.4 3065.19 1434.05 517619 (20.8%) 0.9
T19 100 850,839 2221.39 1439.63 660353 (22.4%) 4.8 2202.39 1437.51 657260 (22.8%) 13.4 1825.63 1189.22 544228 (36.0%) 12.4
2017.57 1304.58 599019 (29.6%) 5.4 2040.78 1344.21 612086 (28.1%) 19.7 1647.76 1070.75 490621 (42.3%) 29.2
T20 100 874,460 2945.23 1917.89 679713 (22.3%) 1.6 3300.83 2166.38 764390 (12.6%) 0.7 2304.74 1529.19 535076 (38.8%) 3.5
2592.43 1674.16 598976 (31.5%) 2.2 3300.83 2166.38 764390 (12.6%) 0.9 2091.12 1405.50 487362 (44.3%) 7.2
a
Average cost.
b
best cost (among 10 repetitions).
c
Average percentage of carbon emission reduction for the upper limit.
d
Best percentage of reduced carbon emission over the CO2 upper limit.
Table 4
Results of the sensitivity analysis.

Problem Nodes Carbon upper c = 0.001 (10 $/ton) c = 0.0025 (25 $/ton) c = 0.004 (40 $/ton)
ID limit
Cost without Emission Trading Total Cost without Emission Trading Total Cost without Emission Trading Total
trading benefit cost trading benefit cost trading benefit cost
T13 50 510,755a 2010.28 324,193 186.56 1823.72 1994.76 323,287 468.67 1526.09 2028.07 328,848 727.63 1300.44
459,680b 2021.52 324,979 134.70 1886.82 2073.72 335,334 310.87 1762.86 2093.73 336,829 491.40 1602.33

Y.-J. Kwon et al. / Transportation Research Part D 23 (2013) 81–89

357,529c 2018.69 327,119 30.41 1988.28 1994.89 324,751 81.95 1912.95 2125.69 341,650 63.52 2062.17
T14 50 398,043 1000.87 354,478 43.57 957.31 1007.63 357,161 102.21 905.43 911.18 322,514 302.12 609.06
358,239 1033.47 366,200 7.96 1041.43 1007.63 357,161 2.70 1004.94 1066.29 377,597 77.43 1143.72
278,631 954.46 337,532 58.90 1013.36 1083.81 383,985 263.39 1347.20 1083.81 383,985 421.42 1505.23
T15 50 435,360 1206.61 258,877 176.48 1030.13 1213.04 262,869 431.23 781.81 1197.14 257,960 709.60 487.54
391,824 1282.52 277,421 114.40 1170.07 1246.68 268,416 308.52 938.16 1233.55 266,208 502.46 731.09
304,752 1234.02 267,678 37.07 1196.95 1170.18 253,346 128.52 1041.67 1206.58 262,267 169.94 1036.64
T16 50 453,951 1361.78 258,657 195.29 1166.49 1380.58 262,577 478.44 902.15 1355.96 255,184 795.07 560.89
408,556 1418.19 267,937 140.62 1277.57 1413.56 270,672 344.71 1068.85 1438.34 270,978 550.31 888.03
317,766 1397.73 266,585 51.18 1347.38 1343.41 256,155 154.03 1189.38 1437.04 273,747 176.08 1260.96
T17 75 666,573 1575.41 442,142 224.43 1350.98 1596.50 446,790 549.46 1047.04 1587.35 445,298 885.10 702.25
599,916 1600.62 449,494 150.42 1450.20 1627.57 458,030 354.72 1272.86 1624.68 455,790 576.50 1048.46
466,602 1574.73 444,562 22.04 1552.69 1600.57 449,173 43.57 1557.00 1660.77 467,196 2.38 1663.15
T18 75 653,370 3058.46 512,982 140.39 2918.07 3065.19 517,619 339.38 2725.81 3191.18 536,965 465.62 2725.56
588,033 3177.43 534,223 53.81 3123.62 3142.13 535,082 132.38 3009.75 3170.47 537,808 200.90 2969.57
457,359 3007.89 507,195 49.84 3057.73 3047.27 515,347 144.97 3192.24 3171.40 533,858 306.00 3477.40
T19 100 850,839 1694.48 505,603 345.24 1349.24 1647.76 490,621 900.55 747.75 1689.01 506,724 1376.46 312.55
765,756 1621.20 481,557 284.20 1337.00 1704.45 509,788 639.92 1064.53 1690.89 502,977 1051.12 639.77
595,588 1663.55 496,810 98.78 1564.77 1703.91 509,103 216.21 1487.70 1601.59 477,479 472.44 1129.15
T20 100 874,460 2184.02 508,537 365.92 1818.10 2091.12 487,362 967.75 1123.38 2056.02 478,622 1583.35 472.67
787,014 2033.66 476,885 310.13 1723.53 1981.51 460,972 815.11 1166.41 2030.69 470,809 1264.82 765.87
612,122 2090.61 486,430 125.69 1964.92 2073.32 482,026 325.24 1748.08 2006.63 466,472 582.60 1424.03
a
0% Reduction in the initial upper limit for the amount of carbon emission.
b
10% Reduction in the initial upper limit for the amount of carbon emission.
c
30% Reduction in the initial upper limit for the amount of carbon emission.

87
88 Y.-J. Kwon et al. / Transportation Research Part D 23 (2013) 81–89

800
700
600
500
400
300
200 c = 0.004
100
c = 0.0025
0
(0%) c = 0.001
(10%)
(30%)
(a) T13 (n = 50)

500

400

300

200

100 c = 0.004
c = 0.0025
0
(0%)
c = 0.001
-100
(10%)
(30%)
-200

(b) T18 (n = 75)

Fig. 3. Trading benefit or cost for instances T13 and T18.

trading. For example, Fig. 3 shows the trading costs and benefits for T13 and T18. Costs increase as the upper limit on the
carbon emission decreases because of the need to purchase the carbon permits at the new lower upper limit for emissions.

5. Concluding remarks

This study considered the heterogeneous vehicle routing problem that determines a set of vehicle routes that satisfies
customer demands and vehicle capacities. We considered the problem with carbon emission for the objective of minimizing
the sum of variable operation costs. To represent the problem mathematically, an integer-programming model was used.
Then, due to the problem complexity, tabu search algorithms were deployed together with three neighborhood generation
methods. Computational experiments were done on modified benchmark instances, and the test results show that the tabu
search algorithm with hybrid neighborhood generation method performs better than the others. In particular, it is shown
from an additional test that the amount of carbon emission can be reduced significantly without sacrificing the cost due
to the benefit obtained from carbon trading.

Acknowledgements

The authors would like to appreciate for the valuable comments of the two anonymous referees.

References

Bektasß, T., Laporte, G., 2011. The pollution-routing problem. Transportation Research Part B: Methodological 45, 1232–1250.
Brandão, J., 2011. A tabu search algorithm for the heterogeneous fixed fleet vehicle routing problem. Computers and Operations Research 38, 140–151.
Eggleston, H.S., Buendia, L., Miwa, K., Ngara, T., Tanabe, K., 2006. IPCC Guidelines for National Greenhouse Gas Inventories. The National Greenhouse Gas
Inventories Programme, 2, Geneva.
Ericsson, E., Larsson, H., Brundell-Freij, K., 2006. Optimizing route choice for lowest fuel consumption: potential effects of a new driver support tool.
Transportation Research Part C: Emerging Technologies 14, 369–383.
 Y.-J. Kwon et al. / Transportation Research Part D 23 (2013) 81–89 89

Glover, F., 1989. Tabu Search – Part I. ORSA Journal of Computing 1, 190–206.
Glover, F., 1990. Tabu Search – Part II. ORSA Journal of Computing 2, 4–326.
Glover, F., Laguna, M., 1993. Tabu Search. In: Reeves, C. (Ed.), Modern Heuristic Techniques for Combinatorial Problems. Blackwell Scientific Publications,
Oxford.
Kara, I., Kara, B.Y., Yetis, M.K., 2007. Energy minimizing vehicle routing problem. Lecture Notes in Computer Science 4616, 62–71.
Laporte, G., Gendreau, M., Potvin, J.-Y., Semet, F., 2000. Classical and modern heuristics for the vehicle routing problem. International Transactions in
Operations Research 7, 285–300.
Lin, S., Kernighan, B.W., 1973. An effective heuristic algorithm for the traveling salesman problem. Operations Research 21, 503–511.
Pessoa, A., Uchoa, E., Aragao, P., 2009. A robust branch-cut and-price algorithm for the heterogeneous fleet vehicle routing problem. Networks 54, 167–177.
Taillard, E., 1999. A heuristic column generation method for the heterogeneous fleet VRP. RAIRO 33, 1–14.
Tavares, G., Zsigraiova, Z., Semiao, V., Grac, M., 2008. A case study of fuel savings through optimisation of MSW transportation routes. Management of
Environmental Quality: An International Journal 19, 444–454.
Toth, P., Vigo, D., 2002. The Vehicle Routing Problem: SIAM Monographs on Discrete Mathematics and Applications. SIAM, Philadelphia.
Xiao, Y., Zhao, Q., Kaku, I., Xu, Y., 2012. Development of a fuel consumption optimization model for the capacitated vehicle routing problem. Computers and
Operations Research 39, 1419–1431.
Yaman, H., 2006. Formulations and valid inequalities for the heterogeneous vehicle routing problem. Mathematical Programming 106, 365–390.