IET Intelligent Transport Systems

Research Article

ISSN 1751-956X
Vehicle routing problem solution considering Received on 12th February 2015
Accepted on 30th March 2015
minimising fuel consumption doi: 10.1049/iet-its.2015.0027
www.ietdl.org

Enjian Yao 1,2 ✉, Zhifeng Lang 3, Yang Yang 1, Yongsheng Zhang 1

1
MOE Key Laboratory for Urban Transportation Complex Systems Theory and Technology, Beijing Jiaotong University, Beijing,
People’s Republic of China
2
Center of Cooperative Innovation for Beijing Metropolitan Transportation, Beijing, People’s Republic of China
3
China Railway First Survey & Design Institute Group Co., Ltd., Xi’an, People’s Republic of China
✉ E-mail: [email protected]

Abstract: Recently, logistics is not only playing more and more important role in social and economic development, but
also caused serious energy consumption and environment pollution problems. Reasonable vehicle route planning is
viewed as an important solution to reduce logistics enterprises’ operation costs as well as alleviate the energy and
environmental problems. This study aims to propose a solution to time-dependent vehicle routing problem with time
windows (TDVRPTW) considering minimising fuel consumption. First, a mathematical TDVRPTW model with the
minimum fuel consumption as an objective function is established, in which the three-dimensional bin-packing
problem is considered as a sub-problem and the alternative stop point concept is newly proposed to reduce the
possible detouring distance and fuel consumption of logistic vehicle. Then, an ant colony algorithm is applied to solve
the problem, and the departure time optimisation is introduced to further improve the obtained results. Finally, the
proposed approach is evaluated with the real road network and traffic data of Beijing. The results show that the
method introduced in this study outperforms the existing approaches in reducing fuel consumption as well as route length.

1 Introduction delivery efficiency through minimising working time, route length

or vehicle number.
Logistics is playing a more and more important role in the Recently, with an increasing worldwide concern for the
development of social economy. However, logistics, especially the environment, more and more attention has been paid in dealing
city logistics delivery, is one of the main causes of energy with ‘green VRP’ [18–21]. In the existing ‘green VRP’ studies,
consumption and environment pollution problems in urban the vehicle route planning problem is optimised by minimising the
transportation system. The average profit of the logistics industry fuel consumption of delivery vehicles. However, vehicles were
in China now is about only 3% and the transportation cost assumed to stop only at the side of the road where the clients are
contributes most to the total logistics costs. According to statistics, located, delivery vehicles may run a roundabout route to serve
the fuel cost could take up about 50% of the total operation costs each client. In fact, delivery vehicle could temporarily stop at the
for any model of car and the ratio will intend to increase with the opposite side of the client and then deliveryman walks across the
climb of international oil price undoubtedly [1]. Therefore road to serve the client on the other side, thus vehicle’s detouring
minimising fuel consumption is viewed as an essential solution to could be avoided and then fuel consumption could be reduced. In
energy saving and environment protection as well as logistics addition, vehicle always leaves immediately when it has served
enterprises’ operation costs reduction [2–4]. one client in previous study, but none of these studies tried to
The vehicle routing problem (VRP) aims at finding an optimal set appropriately adjust the departure time from clients in order to
of routes for a fleet of vehicles initially located at a central depot on a further reduce the fuel consumption on the premise that the time
given network, such that the clients’ demands are satisfied and the windows of clients are satisfied. Therefore this study tries to
loading limitations of the vehicles are not exceeded [5–7]. There optimise the existing solution algorithms from these two aspects.
are many previous studies on the VRP and its variants, and the This paper is organised as follows. Section 2 presents a basic
most common one is the VRP with time windows (VRPTW), in integer programming model of TDVRPTW and introduces several
which the earliest and latest time limits are imposed on the relating issues to revise this model. In Section 3, the proposed
delivery time of each client [8–11], and the arrival at a client solution to solve the revised integer programming model is
outside the specified delivery time is either penalised (soft time presented in detail. Section 4 reports experimental results obtained
windows) or strictly forbidden (hard time windows). However, the by the new solution based on the real road network and traffic data
traditional VRPTW is designed on static network principle, that is, of Beijing and concluding remarks are given in Section 5.
the travel speed is considered constant throughout the day [12,
13]. In other words, this assumption may be far from the reality of
distribution problems under real road network traffic condition, for 2 Problem formulation
the travel time between two clients varies with the distance as well
as the departure time. Especially, in central urban areas, the travel 2.1 Basic integer programming model
speeds during the rush hours decrease substantially compared with
non-rush hours. Research relating to time-dependent VRPTW A central depot possesses K delivery vehicles with identical delivery
(TDVRPTW) has been done in the last few years and most of capacity to serve V clients. The depot is represented by node 0 and
them satisfy the important ‘non-passing’ property [14–17], which clients by nodes 1 to V. For each client, the service must be started
is also called the ‘first-in–first-out’ property. This property ensures during a specific time window [TEi, TLi], where TEi and TLi
that an earlier departure time produces an earlier arrival time, and denote the starting time and ending time of a time window,
vice versa [2]. However, the above studies have mainly focused on respectively. Then, the objective function of the TDVRPTW with

IET Intell. Transp. Syst., pp. 1–7

& The Institution of Engineering and Technology 2015 1
minimum fuel consumption can be described as follows 2.3 Revised fuel consumption model



The fuel consumption factor FCij depends largely on two key
min FCij × Lij × xkij (1)
k i[(0<V ) j[(0<V )
factors including travelling speed and loading weight. Fig. 1
illustrates the relationship between the fuel consumption factor and
the travelling speed for a mid-duty vehicle [25]. In the speed range
where FCij is the fuel consumption factor for the empty vehicle of 5–72 km/h, the fuel consumption factor declines gradually as
coming from location i to j, kg/100 km; Lij is the road length the travelling speed increases. A vehicle reaches its optimal fuel
between clients i and j, km; xkij is a binary decision variable, 1 if economy at around 72 km/h, above which the fuel usage tends to
node j is visited after node i by vehicle k (0 otherwise), and i ≠ j. rise with the increase in vehicle travelling speed.
Constraints (2)–(4) state that a client is visited only once by For any given vehicle, the vehicle engine operates differently
exactly one vehicle under different loads. The lower the vehicle load, the lesser the


load demanded from the engine, and vice versa. As a result, fuel
xkij = 1 (2) consumption factor depends on travelling speed as well as vehicle
k j load. Referring to the measurement of fuel consumption factor in


Suzuki’s study [26], the modified fuel consumption factor formula
xkij = 1 (3) considering vehicle load is expressed as
k i


FC′ij = l × M + FCij (9)
xkji = xkij ={0, 1}, ∀k [ K, ∀i, j [ (0 < V ) (4)
i i
where FCij′ is the fuel consumption factor corrected by the vehicle
Constraint (5) states that every vehicle should start from the depot load, kg/100 km; M is the weight of vehicle load, kg; and l is the
and finally return to it quality correction coefficient to be estimated.
Fig. 2 illustrates the relationship between travelling speed, loading


weight and fuel consumption factor when l equals 1.09 × 10−3/
xk0i = xki0 =1, ∀k [ K (5) 100 km. According to this figure, larger loading weight values
i i
would result in more fuel cost. As an example, for a vehicle
travelling with travelling speed of 100 km/h and loading weight of
Constraints (6) and (7) specify the time windows 2000 kg, the fuel consumption factor is 7.54 kg/100 km; but for
the same vehicle with travelling speed of 100 km/h and loading
xkij (ti + fij (ti )) ≤ TLj (6) weight of 500 kg, the fuel consumption factor is 5.90 kg/100 km.

TEi + Si ≤ ti (7)
2.4 Revised integer programming model
where ti is the departure time from client i; fij (ti) is the travel time On the basis of the analysis in Sections 2.2 and 2.3, the revised
from client i to j when the departure time is ti, min; and Si is the integer programming model for TDVRPTW which includes the
unloading time at client i, min. 3D bin-packing constrains and the revised fuel consumption factor
model can be expressed as follows
2.2 Three-dimensional (3D) bin-packing constraints



min FC′ij × Lij × xkij
To make the theoretical studies on VRP more practically applicable, k i[(0<V ) j[(0<V ) (10)
the vehicle capacity, in terms of weight capacity and volume, and the
physical arrangements of cargoes should be fully considered. Some s.t. constraints (2) − (8)
studies ensure that the cargoes assigned to a vehicle do not exceed
the vehicle capacity restriction in terms of weight but it is not
guaranteed that the cargoes can be physically loaded and arranged 3 Methodology
inside the vehicle, and vice versa [22, 23].
In fact, two combinatorial optimisation problems, the TDVRPTW To solve the mathematical model of TDVRPTW developed in
and the 3D bin-packing problem, are inherently related to each other Section 2, an improved ant colony algorithm is applied.
in a distribution and should be dealt simultaneously. In this study, the
3D bin-packing problem is processed as a sub-problem of the
TDVRPTW and the next client’s cargo demand will not be packed 3.1 Alternative stop points
until the previous one is totally packed. A packing is feasible only
when all the 3D bin-packing restrictions are satisfied: (i) each In the existing research, delivery vehicle was assumed to stop only at
client is served by exactly one vehicle; (ii) no vehicle carries a the side of the road where the client is located, thus vehicle may run a
total weight exceeding its weight capacity; (iii) for each vehicle
there exists a feasible orthogonal 3D loading of all the cargoes
demanded by all clients served by it, satisfying the constraints on
(iii.a) supporting area and (iii.b) last-in–first-out strategy [23].
A number of heuristic algorithms have been raised to solve this
problem. This sub-problem is simplified to constraint (8) and a
genetic algorithm described in detail in [24] is used in this paper.
Constraint (8) means that for a given vehicle, all the cargoes that
have been assigned to it can be loaded and satisfies all the 3D
bin-packing restrictions
 



w xi × xkij ≤ 1, ∀k [ K (8)
i[V j[(0<V )
Fig. 1 Fuel consumption factor and vehicle travelling speed for a mid-duty
where xi is the set of cargoes demanded by client i. truck

IET Intell. Transp. Syst., pp. 1–7

2 & The Institution of Engineering and Technology 2015
Fig. 2 Fuel consumption factor at various travelling speeds and loading weights

roundabout route to serve each client. However, for some client, the where TEB′ refers to the start time of the time window for the ASP
delivery vehicle could temporarily park at the opposite side and then B′; TLB′ refers to the end time of the time window for the ASP B′;
the deliveryman walks across the road to serve the client, thus and θ refers to the time that the deliveryman needed to cross the road.
vehicle’s detouring could be avoided and then fuel consumption
could be reduced. This kind of temporary parking point is defined 3.2 Key steps of ant colony algorithm
as alternative stop point (ASP). It means that a vehicle could stop
at ASP rather than the real stop point located at the same side of As ant colony algorithm is well documented in the literature, only the
client to provide service. route selection and pheromones update are briefly introduced in this
It is not always suitable for all clients to have an ASP. The setting study, and more details about ant colony algorithm could be found in
of ASP needs to meet three conditions: (i) the quantity of cargoes [8, 27].
demanded by the client cannot exceed the deliveryman’s carrying
capacity; (ii) the walk distance between a client and the nearest 3.2.1 Route generation: Each ant of the colony attempts to
pedestrian crossing facility should not exceed the deliveryman’s complete a solution using the following constructive procedure
tolerance; and (iii) temporary parking and deliveryman’ road until all the clients are serviced. The ant moves from client i to the
crossing should be allowed without violating the traffic rules in next j by choosing among the feasible clients’ original points or
terms of safety. the ASPs that have not been visited yet. The next client j is
If a client satisfies all the three conditions, an ASP could be set up selected using the following rule, called pseudo-random-
across the road accordingly. As shown in Fig. 3, a vehicle is assumed proportional action choice rule, determined by a fixed cut-off
to serve three clients (A, B and C ). On the basis of the traditional parameter q0 ∈ [0, 1] and a random number r ∈ [0, 1]. The smaller
TDVRPTW methods, an obviously roundabout route, route 1 will the q0 is, the higher the probability to make a random choice [26]
be provided for vehicle. On the other hand, if an ASP B′ is
considered for client B, route 2 rather than route 1 will be selected ⎧
⎨ arg max (taij × mbij ), r ≤ q0
for delivery vehicle, and the vehicle will stop at B′ and the j= Wi (13)
deliveryman could walk across the road to serve client B. ⎩
select j randomly from Wi , r . q0
Obviously, the length of route 2 is much shorter than route 1 and
therefore the fuel consumption will be effectively reduced.
The information of an ASP includes the position, the cargo where tij is the intensity of the pheromone on the road section
demands, the time windows and the unloading time. According to between client i and client j; μij is the visibility of the road section,
the definition of ASP, the time window and unloading time of an which is generally equal to the inverse of the distance between
ASP are modified from the corresponding client’s information using clients i and j; parameters α and β weigh the relative importance
of the pheromone values and the visibility values; Wi are the
feasible clients that have not been visited yet and that do not
[TEB′ , TLB′ ] = [TEB − u, TLB − u] (11)
violate any of the constraints.
SB′ = SB + 2 × u (12) For the client B and ASP B′ represent the same cargo demand, just
one of them needs to be selected.

3.2.2 Pheromones updating: Distinguished from most of

traditional studies, of which the goal is to minimise working time,
route length or vehicle number, the optimisation objective in this
paper is to minimise vehicles’ total fuel consumption. Thus,
pheromones update strategy should be adjusted accordingly.
Pheromones can be updated either locally or globally. Local
update is performed during the ant constructive procedure and
global update is performed once all the ants have finished their
iterations. In this paper, global update is adopted in the following
way

(1 − r) × tij + Q/E if vehicle pass road section ij
tij = (14)
(1 − r) × tij otherwise

where ρ ∈ [0, 1] is the evaporation coefficient; Q is a parameter; and

Fig. 3 Schema of ASP E is the fuel consumption of the vehicle route planning scheme, kg.

IET Intell. Transp. Syst., pp. 1–7

& The Institution of Engineering and Technology 2015 3
According to the formula above, it is clear that the less the fuel
consumption of a vehicle route planning scheme is, the more
likely this scheme be selected in the next iteration.

3.3 Departure time optimising

As the travel speeds on the route between a pair of clients varies at

different time in a day, and travel speed affects the fuel consumption
factor for a vehicle, changing the departure time from a client will
change the fuel consumption on the route. Certainly, the time
windows of clients must be satisfied while adjusting the departure
time.
Using the algorithms described in Sections 3.1 and 3.2, an initial
optimal distribution plan including vehicle routes, loading plan for
every vehicle and the departure time for each vehicle from each
client could be calculated. Then, through the adjustment of
departure time at each client, the initial optimal distribution plan
can be further improved.
The new departure time of vehicles from each client can be
obtained by postponing departure time for (n × t0) minutes as
shown in (15). In case the number of possible distributions is too
large, additional constraints have been added as follows
Fig. 4 Studying area in Beijing city
ti′ = ti + n × t0 (15)

s.t. xkij × (ti′ + fij (ti′ )) ≤ TLj (16) To validate the reliability of calculation results, a part of locations
are randomly selected out to form the experimental instance and five
TEi + Si ≤ ti′ (17) scenarios are considered for each instance. These five scenarios
differ from each other in terms of the objective function to be
where t′i is the new departure time from client i after adjustments; n optimised and the optimisation method; the details are shown in
is the number of adjustments at each client, a non-negative integer; t0 Table 1: scenario 1, the optimisation goal is to minimise the work
is the time-adjusting granularity. Constraints (16) and (17) mean that time without ASP setting and departure time optimising, it is
t′i must satisfy the time windows. As the time window lengths are viewed as the traditional method in this paper; scenario 2, the
∼30 min and the time-adjusting granularity t0 is equal to 5 min, optimisation goal is to minimise the fuel consumption without
the value range of n is set to [0, 6] to avoid going beyond the time ASP setting and departure time optimising; scenario 3, the
window. optimisation goal is to minimise the fuel consumption and with
After the adjustments of vehicles’ departure time at each client, the ASP setting only; scenario 4, the optimisation goal is to minimise
new distribution plans for each vehicle will be obtained. On the basis the fuel consumption and with departure time optimising only;
of the optimisation goal of Section 2, the final optimal distribution scenario 5, the optimisation goal is the minimum fuel consumption
plan with less fuel consumption than the original optimal with both ASP setting and departure time optimising.
distribution plan can then be obtained.

4.2 Computational results

4 Case study
To examine the validity, reliability and the calculation performance
To evaluate the performance of the proposed approach, total nine of the proposed approach under various client size conditions, total
computational experiments are carried out based on the real road nine instances’ analysis has been carried out. In detail, 10 clients
network of Beijing and the ASPs are taken into consideration. The
ant colony algorithm described in Section 3 was coded in
MATLAB and the experiments were executed on a personal
computer with an Intel i5, 4 GHz CPU and 8 GB of RAM.

4.1 Input data

The study area, a part of the central area of Beijing, is surrounded by

the red rectangle with each side about 10 km length (shown in
Fig. 4), and the total area is about 100 km2. Twenty-one locations
(1 depot and 20 clients) are randomly distributed throughout this
area (shown in Fig. 5) and each pair of them is connected by the
shortest path, which is calculated using a distance-based Dijkstra
algorithm in advance for the consideration of optimising
calculation time.
The dynamic traffic information is gathered by a floating car
system, which covers most of the roads in Beijing, and is updated
every 5 min. The coverage of real-time traffic information obtained
from floating car system over the road network is above 50%
between 8:30 and 22:30 (shown in Fig. 6), and the dynamic traffic
information is used for the real-time traffic information release of
Beijing. The fuel consumption for a vehicle running from one
location to another is calculated using the dynamic traffic
information with the method already illustrated in Section 2.3. Fig. 5 Location of depot and clients

IET Intell. Transp. Syst., pp. 1–7

4 & The Institution of Engineering and Technology 2015
 Table 2 Optimal results for five scenarios in instance 1
Scenarios Vehicle routes Route Fuel Travel
length, consumption, time,
km kg min

1 veh1: 1–3–18–1 73.27 11.06 356

veh2: 1–2–16–10–11–1
veh3: 1–9–19–17–8–1
2 veh1: 1–3–16–8–1 71.63 10.94 286
veh2: 1-2-18-10-11-1
veh3: 1–9–19–17–1
3 veh1: 1–3–18–24–25–1 71.24 10.63 276
veh2: 1-2-16-22-1
veh3: 1–9–19–17–1
4 veh1: 1–3–16–8–1 71.63 10.49 253
Fig. 6 Coverage of floating car over time in Beijing veh2: 1–2–18–10–11–1
veh3: 1–9–19–17–1
5 veh1: 1–3–16–22–1 69.98 10.30 241
veh2: 1–2–18–24–25–1
were randomly selected out from 20 clients in instances 1, 2, 3; 20 veh3: 1–9–19–17–1
clients were randomly selected out from 50 clients in instances 4,
5, 6; and 100 clients were randomly selected out from 120 clients Travel time refers to the time that vehicles spend on the road.
in instances 7, 8, 9.
The first instance is composed of locations 1, 2, 3, 8, 9, 10, 11, 16,
17, 18, 19 and location 1 is designated as the depot. Vehicles depart
from this depot at 8:00 in the morning to deliver goods to the clients.
Using the judgements introduced in Section 3.1, clients 8, 9, 10, 11
meet all the constraints and could set ASPs 22, 23, 24, 25,
respectively. The optimal solutions for the five scenarios are
summarised in Table 2 and the vehicle routes are shown in Fig. 7,
separately.
It is obvious that the distribution plan in scenario 1 consumes the
most fuel and scenario 5, which uses the approach (including both
ASP setting and departure time optimising) proposed in this paper,
consumes the least. Comparing the optimal solutions in scenarios
2 and 3, it can be found that three out of four ASPs are exploited
in scenario 3, thus it can be concluded that setting ASPs when
solving VRP can change the optimal vehicle route and reduce the
route length and fuel consumption, simultaneously. The
comparison of the optimal solutions in scenarios 2 and 4 shows
that the vehicle routes and the route lengths are the same, but the
fuel consumption and travel time in scenario 4 is less than that in
scenario 2. This result demonstrates that optimising the departure
time from clients can effectively reduce the fuel consumption even
if the vehicle route is fixed.
The optimal solutions’ evaluation indicators (i.e. fuel
consumption, rout length and travel time) and their decrements Fig. 7 Optimal vehicle routes for five scenarios in instance 1
compared with scenario 1 for nine instances are summarised
separately in Tables 3–5.
The experimental results for all instances indicate that the city, the vehicle speed on roads used in these instance analysis is
improved approach proposed in this paper (i.e. scenario 5) can almost <70 km/h. Therefore the optimal routes of the above nine
reduce fuel consumption, route length and travel time by average instances have less energy consumption as well as larger travel
13.87, 10.01 and 18.33%, respectively, compared with the speed (i.e. less travel time) by coincidence, but when the vehicle
traditional method in the distribution processes. speed keeps at a high level (i.e. above 72 km/h), the optimal route
However, it does not mean that minimising travel time will with less travel time will consume more fuel.
minimise fuel consumption definitely. As shown in Fig. 1, the fuel The calculating times are about 4 min for the instances 1, 2 and 3;
consumption factor declines gradually as the travelling speed 11 min for instances 4, 5 and 6; and 35 min for instances 7, 8 and 9,
increases in the speed range of 5–72 km/h, and tends to rise when respectively. It is viewed to be acceptable for practical application,
the travelling speed is above 72 km/h. Furthermore, based on the for the optimisation of distribution route, which belongs to route
traffic information gathered by a floating car system in Beijing planning rather than online optimisation, is processed before the
implementation of delivery. Meanwhile, the calculating time
depends largely on the performance of computer and the
Table 1 Experimental scenarios calculation performance of the algorithm is expected to be
improved continuously with the improvement of the computer
Scenarios Optimisation goal ASP Departure time technique.
setting optimising

1 minimum work time no no

2 minimum fuel no no 5 Conclusions
consumption
3 minimum fuel yes no
consumption This paper offers an efficient vehicle routing planning solution for
4 minimum fuel no yes TDVRPTW with the objective to minimise delivery vehicles’ fuel
consumption consumption. Different from the previous research, the 3D
5 minimum fuel yes yes
consumption
bin-packing problem is dealt as a sub-problem of the TDVRPTW
which makes the distribution solution more applicable to practice.

IET Intell. Transp. Syst., pp. 1–7

& The Institution of Engineering and Technology 2015 5
Table 3 Results of instances 1, 2 and 3
Instances Scenarios Indicators and decrements

FC, kg Decrements RL, km Decrements TT, min Decrements

1 1 11.06 — 73.27 — 356 —

2 10.94 1.08% 71.63 2.23% 286 19.66%
3 10.63 3.89% 71.24 2.77% 276 22.47%
4 10.49 5.15% 71.63 2.23% 253 28.93%
5 10.30 6.87% 69.98 4.49% 241 32.30%
2 1 12.58 — 83.64 — 308 —
2 9.84 21.78% 67.23 19.63% 284 7.79%
3 9.74 22.58% 66.47 20.53% 269 12.66%
4 9.54 24.17% 67.23 19.63% 250 18.83%
5 9.37 25.52% 64.26 23.17% 236 23.38%
3 1 11.25 — 72.36 — 296 —
2 10.42 7.38% 65.69 9.22% 273 7.77%
3 10.24 8.98% 64.94 10.26% 267 9.80%
4 9.91 11.91% 65.69 9.22% 251 15.20%
5 9.83 12.62% 64.94 10.26% 240 18.92%

Ten clients were randomly selected out from 20 clients in instances 1, 2 and 3. FC is the fuel consumption, kg; RL is the route length, km; and TT is the
travel time, min.
Bold values in this table are obtained using the improved approach proposed in this paper.

Table 4 Results of instances 4, 5 and 6

Instances Scenarios Indicators and decrements

FC, kg Decrements RL, km Decrements TT, min Decrements

4 1 14.61 — 89.87 — 581 —

2 13.88 5.01% 88.10 1.97% 546 6.02%
3 13.68 6.37% 84.60 5.87% 546 6.02%
4 13.31 8.89% 88.10 1.97% 516 11.19%
5 13.04 10.73% 84.60 5.87% 511 12.05%
5 1 17.05 — 109.41 — 670 —
2 14.33 15.99% 90.67 17.13% 564 15.82%
3 13.78 19.18% 88.51 19.11% 541 19.25%
4 13.67 19.87% 90.67 17.13% 528 21.19%
5 13.20 22.57% 88.51 19.11% 510 23.88%
6 1 15.95 — 102.17 — 629 —
2 15.98 –0.18% 103.71 –1.51% 625 0.64%
3 15.59 2.26% 101.89 0.27% 608 3.34%
4 15.20 4.71% 103.71 –1.51% 582 7.47%
5 14.81 7.14% 101.89 0.27% 564 10.33%

Twenty clients were randomly selected out from 50 clients in instances 4, 5 and 6. FC is the fuel consumption, kg; RL is the route length, km; TT is the
travel time, min.
Bold values in this table are obtained using the improved approach proposed in this paper.

Table 5 Results of instances 7, 8 and 9

Instances Scenarios Indicators and decrements

FC, kg Decrements RL, km Decrements TT, min Decrements

7 1 70.17 — 458.84 — 2744 —

2 65.47 6.70% 438.06 4.53% 2535 7.62%
3 62.30 11.22% 409.47 10.76% 2432 11.37%
4 63.68 9.25% 438.06 4.53% 2438 11.15%
5 60.19 14.22% 409.47 10.76% 2317 15.56%
8 1 67.58 — 431.20 — 2675 —
2 64.46 4.62% 414.65 3.84% 2545 4.86%
3 64.05 5.22% 411.35 4.60% 2528 5.50%
4 62.03 8.21% 414.65 3.84% 2413 9.79%
5 61.25 9.37% 411.35 4.60% 2375 11.21%
9 1 73.19 — 472.21 — 2879 —
2 65.45 10.58% 424.51 10.10% 2571 10.70%
3 64.11 12.41% 417.83 11.52% 2514 12.68%
4 62.85 14.13% 424.51 10.10% 2429 15.63%
5 61.66 15.75% 417.83 11.52% 2379 17.37%

One hundred clients were randomly selected out from 120 clients in instances 7, 8 and 9. FC is the fuel consumption, kg; RL is the route length, km; TT is
the travel time, min.
Bold values in this table are obtained using the improved approach proposed in this paper.

The ASP is proposed to reduce the possible detouring distance and The analysis using the real traffic data of Beijing shows that the
fuel consumption of logistic vehicle, and departure time optimising proposed solution outperforms the existing methods. The observed
is introduced to further improve the vehicle routing planning results. reductions in the fuel consumption, route length and travel time

IET Intell. Transp. Syst., pp. 1–7

6 & The Institution of Engineering and Technology 2015
from the proposed solution indicate the proposed approach is valid, 11 Qureshi, A.G., Taniguchi, E., Yamada, T.: ‘An analysis of exact VRPTW solutions
reliable and having good calculating performance. From the case on ITS data-based logistics instances’, Int. J. Intell. Transp. Syst. Res., 2012, 10,
(1), pp. 34–46
study, the proposed approach can reduce the fuel consumption by 12 Figliozzi, M.A.: ‘An iterative route construction and improvement algorithm for the
an average of 13.87% (up to a maximum of 25.52%) compared vehicle routing problem with soft time windows’, Transp. Res. C, 2010, 18, (5),
with the traditional method. pp. 668–679
13 Yu, B., Yang, Z.Z.: ‘An ant colony optimization model: the period vehicle routing
problem with time windows’, Transp. Res. E, 2011, 47, (2), pp. 166–181
14 Soler, D., Albiach, J., MartíNez, E.: ‘A way to optimally solve a time-dependent
6 Acknowledgments vehicle routing problem with time windows’, Oper. Res. Lett., 2009, 37, (1),
pp. 37–42
This research is supported by National 973 Program of China (no. 15 Lecluyse, C., Sörensen, K., Peremans, H.: ‘A network-consistent time-dependent
2012CB725403) and Research Fund for the Doctoral Program of travel time layer for routing optimization problems’, Eur. J. Oper. Res., 2013,
Higher Education of China (no. 20130009110002). 226, (3), pp. 395–413
16 Wang, H.F., Lee, Y.Y.: ‘Modeling of a time dependent alternative vehicle routing
problem with time windows’. Int. Conf. on Management and Service Science,
Wuhan, China, August 2011, pp. 1–4
7 References 17 Ichoua, S., Gendreau, M., Potvin, J.Y.: ‘Vehicle dispatching with time-dependent
travel times’, Eur. J. Oper. Res., 2003, 144, (2), pp. 379–396
1 Yu, L., Wang, D.: ‘Green logistics capacitated VRPSDP based on ant colony 18 Demir, E., Bektas, T.: ‘A review of recent research on green road freight
algorithm’, Comput. Syst. Appl., 2013, 22, (7), pp. 127–132 transportation’, Eur. J. Oper. Res., 2014, 237, (3), pp. 775–793
2 Kuo, Y.: ‘Using simulated annealing to minimize fuel consumption for the 19 Ubeda, S., Arcelus, F.J., Faulin, J.: ‘Green logistics at Eroski: a case study’,
time-dependent vehicle routing problem’, Comput. Ind. Eng., 2010, 59, (1), Int. J. Prod. Econ., 2011, 131, (1), pp. 44–51
pp. 157–165 20 Xiao, Y., Zhao, Q., Kaku, I., Xu, Y.: ‘Development of a fuel consumption
3 McIlroy, R.C., Stanton, N.A., Harvey, C.: ‘Getting drivers to do the right thing: a optimization model for the capacitated vehicle routing problem’, Comput. Oper.
review of the potential for safely reducing energy consumption through design’, Res., 2012, 39, (7), pp. 1419–1431
IET Intell. Transp. Syst., 2013, 8, (4), pp. 388–397 21 Kuo, Y., Wang, C.C.: ‘Optimizing the VRP by minimizing fuel consumption’,
4 Erdoğan, S., Miller-Hooks, E.: ‘A green vehicle routing problem’, Transp. Res. E, Manage. Environ. Qual.: Int. J., 2011, 22, (4), pp. 440–450
2012, 48, (1), pp. 100–114 22 Moura, A., Oliveira, J.F.: ‘An integrated approach to the vehicle routing and
5 He, J.: ‘An hybrid heuristic algorithm for the two-echelon vehicle routing container loading problems’, OR Spectrum, 2009, 31, (4), pp. 775–800
problem’. IET, Int. Conf. on Information Science and Control Engineering, 23 Gendreau, M., Iori, M., Laporte, G., Martello, S.: ‘A tabu search algorithm
Shenzhen, China, December 2012, pp. 1–5 for a routing and container loading problem’, Transp. Sci., 2006, 40, (3),
6 Szeto, W.Y., Wu, Y., Ho, S.C.: ‘An artificial bee colony algorithm for the
pp. 342–350
capacitated vehicle routing problem’, Eur. J. Oper. Res., 2011, 215, (1),
24 Bu, L., Yuan, X.J., Pu, Y., Chang, J.Q.: ‘Optimization based on genetic algorithm
pp. 126–135
for three-dimensional packing in a single container’, China Railw. Sci., 2004, 25,
7 Chen, H.K., Hsueh, C.F., Chang, M.S.: ‘The real-time time dependent vehicle
routing problem’, Transp. Res. E, 2006, 42, (5), pp. 383–408 (4), pp. 108–111
8 Yildirim, U.M.: ‘An ant colony algorithm for time-dependent vehicle routing 25 Yao, E.J., Song, Y.Y.: ‘Study on eco-route planning algorithm and environmental
problem with time windows’. Operations Research Proc. 2008, 2009, pp. 337–342 impact assessment’, J. Intell. Transp. Syst., 2013, 17, (1), pp. 42–53
9 Li, H., Lim, A.: ‘Local search with annealing-like restarts to solve the VRPTW’, 26 Suzuki, Y.: ‘A new truck-routing approach for reducing fuel consumption and
Eur. J. Oper. Res., 2003, 150, (1), pp. 115–127 pollutants emission’, Transp. Res. D, 2011, 16, (1), pp. 73–77
10 Pureza, V., Morabito, R., Reimann, M.: ‘Vehicle routing with multiple 27 Dorigo, M., Gambardella, L.M.: ‘A study of some properties of Ant-Q’, in Voigt,
deliverymen: modeling and heuristic approaches for the VRPTW’, Eur. J. Oper. H.M., Ebeling, W., Rechenberg, I., Schwefel, H.P. (Eds.): ‘Parallel problem
Res., 2012, 218, (3), pp. 636–647 solving from nature – PPSN IV’ (Springer, Berlin, Heidelberg, 1996), pp. 656–665

IET Intell. Transp. Syst., pp. 1–7

& The Institution of Engineering and Technology 2015 7