Reducing Empty Truck Trips in Long Distance Network by Combining
Trips
B
´
arbara da Costa Rodrigues and Andr
´
e Gustavo dos Santos
Departamento de Inform
´
atica, Universidade Federal de Vic¸osa, Av. P. H. Rolfs s/n - 36570-900 Vic¸osa, MG, Brazil
Keywords:
Trips Combination, MILP, Heuristic, Combinatorial Optimization.
Abstract:
Brazilian import and export activities on ports are subject to considerable slow queues and congestion, revea-
ling a lack of medium and/or short-term logistic planning. One of the causes is the number of trucks traveling
with empty containers, performing one-way trips, from inland cities to the port or from the port to the cities.
This issue may be reduced by combining trips, i.e., after bringing goods to the port (export trip), a truck should,
when possible, carry goods from the port to the origin or a nearby city (import trip). In this paper we inves-
tigate a combinatorial optimization problem where a set of import/export/inland trips should be combined in
order to reduce total traveling time, which in turn reduces the number of empty trucks traveling to/from the
port. Individual trips and combined trips must obey national law regulation of resting time, as typical road trips
in Brazil covers hundreds, even thousands of kilometers. We also consider opening operation hours on each
location (time windows), which may force a driver to wait upon arriving. We test exact and heuristic appro-
aches, and present the total travel time and number of trucks needed for each solution, considering instances
based on real freight data.
1 INTRODUCTION
Due to the lack of infrastructure for grain storage,
most of the Brazilian production is destined to expor-
tation, leading to a high demand of freightage to the
ports, as well as import of fertilizers and other agri-
cultural inputs to the grain exporting regions (Cai-
xeta Filho, 2010). Most of the freight transport is
done by road modal, which may bring practicality,
agility and flexibility in cases of route exchanges. Ne-
vertheless, a number of disadvantages may be listed,
for example, high cost of fuel and tolls, poor condition
of roads in some regions, among others. Those dis-
advantages, allied to lack of good logistic planning,
cause an extra cost to the final price of the transpor-
ted goods, congestion in the port area, delivery delays
and also empty trucks, traveling without any freight.
Empty truck trips may be found specially on im-
port/export trips, when the carrier take care only of
incoming or outgoing trips. The result is a somehow
needless increasing in the number of trucks using the
roads and ports, consequently increasing congestion
and polluting gas emission generated by fuel combus-
tion (Schulte et al., 2015).
In the context of this work we have import trips, as
shown in Figure 1. The truck collects a freight (con-
tainer, bagged or bulk goods) at the port. Then travels
to a customer or depot that demands that freight. After
unloading, the truck travels back to the port unladen.
Figure 1: Import scenario.
An empty truck travel also occurs in export trips,
as shown in Figure 2. The truck collects goods at a
customer or a depot and travels to the port, where
sometimes faces a queue of trucks waiting for unlo-
ading, due to congestion or ships that are not ready
to receive the load. After unloading the truck comes
back to the customer empty.
A third situation of empty trucks traveling occurs
in inland trips, between cities. The truck travels from
one city to another to transport goods from customer
to customer and returns empty to the origin point, due
to lack of return demand or vice versa (Figure 3).
Those are, of course, inefficient ways to use the
da Costa Rodrigues, B. and Gustavo dos Santos, A.
Reducing Empty Truck Trips in Long Distance Network by Combining Trips.
DOI: 10.5220/0006709303190327
In Proceedings of the 20th International Conference on Enterprise Information Systems (ICEIS 2018), pages 319-327
ISBN: 978-989-758-298-1
Copyright
c
2019 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
319
Figure 2: Export scenario.
Figure 3: Inland trips.
trucks and the road network. It may be the case that
a customer that exports goods does not import any
goods, at least not from the same port. But generally
the customer demands goods as well, and the truck
may return loaded if it visits a company that sells the
goods the customer needs. Even more, from the port
to that company the truck may bring goods impor-
ted by the company that are on the port waiting to
be transported. Thus, an export trip may be combi-
ned to an import trip, or an import trip followed by
an inland trip, as shown in Figure 4. A perfect com-
bination would include a sequence of trips in which
the destination of a trip is the origin of the next one.
A good combination allows a repositioning trip bet-
ween the end of a trip and the start of the next one, as
long as those locations are not far away, i.e., when the
repositioning may be done by a small trip.
Figure 4: Combination of export, import and inland trips.
The combination of trips becomes an alternative
to avoid such problem. Import and export trips that
would be performed by different trucks may be per-
formed sequentially by the same truck, thus decrea-
sing the number of empty trucks traveling in the road
system, which contributes to the transportation net-
work as a whole (reduction of truck traffic and road
congestion) besides to the environment (reduction of
emission of polluting gases) (Islam, 2017a). Further-
more, Schulte et al. (Schulte et al., 2015) mention that
such combination contributes also to decrease opera-
tional costs and additional gains to drivers.
The objective of this work is to propose a mat-
hematical model to find a good (possible the best)
combination of trips (export, import and inland trips)
to reduce empty trucks travelling through the roads,
which besides the contributions aforementioned, may
reduce the waiting time in ports and congestion in
ports area. The model includes constraints to assure
drivers welfare by enforcing current laws regulation.
This paper is organized as follows: in Section 2
we present related works from the literature; a for-
mal definition of the problem is presented in Section
3; in the following, in Section 4, the proposed met-
hod is described, which consists of a mixed integer
linear programming, a heuristic for evaluating each
possible combination of trips, and an integer linear
programming for choosing a set of those combinati-
ons; experimental results are presented in Section 5
and conclusions and future works in Section 6.
2 LITERATURE REVIEW
The combination of trips to decrease empty trucks
traveling along the roads have already been addres-
sed in the literature. Gavish and Schweitzer (Gavish
and Schweitzer, 1974) are among the first to propose
such approach.
¨
Ozener and Ergun (
¨
Ozener and Ergun,
2008) study a logistic network in which shippers col-
laborate to share a common carrier. Their study has
identified routes in which a collaborative scheme may
reduce the shared costs among shippers. Audy et al.
(Audy et al., 2011) shows that in their context both
the cost and the delivery time may be reduced using
collaborative transportation.
Caballini et al. (Caballini et al., 2015) studied a
problem similar to the one we study here, with ex-
port, import and inland trips in a port context, where
all trucks may transport two 20 ft containers. They
proposed a Mixed Integer Linear Programming for-
mulation (MILP) to minimize the overall costs of trips
subject to time windows and time limit in the rou-
tes. They study the impact of trips combination on
real data in the port of Genoa, Italy, showing that trip
combination may reduce the costs and the number of
empty tricks traveling to/from the port.
ICEIS 2018 - 20th International Conference on Enterprise Information Systems
320
Schulte et al. (Schulte et al., 2015) developed si-
mulation models for coordinated truck appointments
and used the proposed approach to solve the problem
as a TSP with time windows allowing collaboration.
They used instances based on real data of the port
of Santo Antonio, Chile, and integrated their appro-
ach to a Truck Apointment System (TAS), a tool to
schedule and follow cargo arriving, allowing collabo-
ration among ports and transportation companies. As
a result, ports could reduce port-related polluting gas
emission using the model in real-time.
Islam (Islam, 2017b) also simulated the sharing of
trucks in a port environment. He compared two scena-
rio, the first considering sharing/collaboration and the
second without it. Using data of a local port, he sho-
wed that the collaboration between trucks increases
the use of the port capacity as well reduces polluting
gas emission and congestion around the port.
A more recent work of Caballini et al. (Cabal-
lini et al., 2017) proposes a model to reduce costs and
the number of trips that trucks travel empty, take ad-
vantage of the capacity of the trucks. They show the
efficacy of the proposed approach by computational
experiments.
The present work differs from the ones cited above
by including local transport regulation laws and time
windows, thus adapting previous ideas towards the
Brazilian exportation context. Caballini et al. (Ca-
ballini et al., 2015) is the one more similar, but here
we include a heuristic phase, due the complexity of
the resuling model, when all considered characteristis
are included.
3 PROBLEM DEFINITION
We are given as input a set T of trips. For each
trip i T we have its origin and destination location,
respectively O
i
and D
i
, and the distance d
i
between
those locations, in Km. We also have the distance e
il
between the destination of a trip i T and the ori-
gin of a trip l T \i, also in Km. This distance is
traveled by a truck covering two trips i, l sequentially
when D
i
6= O
l
(considered 0 when they coincide) and
is called repositioning trip. The duration of each trip
i is defined by t
i
= d
i
/speed, and the duration of re-
positioning trip is e
il
/speed. In all cases we consider
a constant speed of 80 km/h. This is in fact the max-
imum speed for heavy trucks in Brazilian roads, but
as most of the trips are long distance trips, trucks will
using this speed most of the time, and then may be
used as average speed. Moreover, there is a service
time S that is including in all trips, covering the loa-
ding and unloading service at cities and ports.
Trips are under transport regulation laws that im-
pose 30 min of rest after each 5,5 hours of travel, and
8 hours after each 12 hours. The first one imposes a
small rest during a trip, and the second one a long rest
(night/sleep resting, for example). Besides mandatory
resting times, drivers are subject to waiting times due
to opening and closing time of locations (deposit in
cities and ports). For each location there is a time win-
dow, and operations may be done only inside the time
window. This is represented in Figure 5; in this case,
the truck arrived within the time window. For each
trip i we know the time window of the origin location,
[P
O
i
,P
ˆ
O
i
], and of the destination location, [P
D
i
,P
ˆ
D
i
]. If
the truck arrives before the opening time, it must wait
until the window opens (see Figure 6). If the truck
arrives after the closing time, it must wait until the
next day, for the following time window, as represen-
ted in Figure 7. This may happen also when starting
a trip and in repositioning trips. Those waiting times
are added to the total duration of the combination of
trips. For some trips there is a good combination to
avoid those wasted time, but for some the total du-
ration time may include many hours due to waiting
times. The choice of combinations must be carefully
done to avoid or minimize that.
Figure 5: Trip arriving inside the time window.
Figure 6: Trip arriving before the time window.
Figure 7: Trip arriving after the time window.
We consider combinations of at most 3 trips. A
good example of combination of 3 trips is the one de-
picted in Figure 4 where an import trip is followed
by an inland trip and then an export trip. This case
happens when the destination of the first trip does not
have any goods to send to the port. Instead of co-
ming back to the port empty, the driver travels to a
nearby city and carries the truck with goods prepa-
red for exportation. Another case is the combination
of export/inland/inland trips: a rural producer exports
goods and imports agricultural inputs; after unloading
the goods at the port, the driver travels to a nearby city
to load the agricultural inputs to bring to the producer.
A combination of more than 3 trips would include a
Reducing Empty Truck Trips in Long Distance Network by Combining Trips
321
double trip to the port or from the port, which can be
modeled as two combinations.
The objective is then to find the best combinations
of trips in order to minimize the total duration time,
considering the duration of all combinations chosen
and trips performed as single trip, if a trip is chosen
not to be combined. The formal definition of the ob-
jective and the aforementioned constraints is detailed
in the MILP formulations proposed in the following
section.
4 SOLUTION METHOD
Our first attempt was to propose a MILP formula-
tion that includes all characteristics of the problem,
as the one proposed by Caballini et al. (Caballini
et al., 2015). The formulation had variables to cont-
rol which trips belongs to each combination and their
sequence in the combination. Moreover there were
a considerable number of variables to control res-
ting/waiting time within trips (following law regula-
tions) and between trips (due to time windows). Ho-
wever, we are working with a more complex problem,
mainly because we have to deal with some very long
distance trips, that may spam more than one day, even
when performed as a single trip. The control of the
starting and ending of each trip becomes more com-
plex due to different restring times that are manda-
tory along the way. Although we could indeed include
all characteristics in a MILP formulation, it turns out
to solve only instances with very small number of
trips, and using a long CPU time. Nevertheless we
can still solve the problem using exact MILP formu-
lation by decomposing the problem: firstly we do a
pre-processing to determine the best way to combine
each sequence of 2 or 3 trips, and then we search the
best set of combinations that covers all trips at a mi-
nimum cost.
For the first phase, we propose a MILP formula-
tion to determine the best duration of a given sequence
of trips. This includes determining the best starting
and ending time of each trip in order to reduce the
total duration time. The proposed formulation is pre-
sented in Section 4.1. Although the formulation can
handle at a reasonable time all the subsets of 2 or 3
trips, it may be impracticable to use it for large in-
stances because of the exponential number of possi-
ble combinations to be solved. We then propose, at
Section 4.2, a constructive heuristic to be used when
needed as an alternative in this phase. After the pre-
processing phase, independently of the method used,
a set-covering based formulation (Section 4.3) is used
to determine the best set to cover all trips.
4.1 Pre-processing Combinations
In order to choose the best combinations for a given
set of trips, we first do a pre-processing to define the
optimal cost (in terms of total duration) for each pos-
sible combination of 2 or 3 trips, i.e., we generate all
sequence of 2 and 3 trips, and for each one of them,
we evaluate its total duration time by deciding the
start and finish time of each trip, besides the resting
and waiting times. In this section we describe a MILP
formulation for this task, and in the following section
we describe a greedy constructive heuristic.
We propose the following MILP formulation to
define the minimal cost for the combination of trips
i,l,k T sequentially.
We use indexes i,il,ilk when describing variables
or data respectively for trip i, trip l (performed after
i) and trip k (performed after i and l). For example,
if q denotes the starting time of a trip, q
i
,q
il
and q
ilk
denote the starting time of trips i,l,k in the combina-
tion. Moreover, we use
0
to indicate a given time in a
day, i.e., remaining hours discounting full days. For
example, if q
ilk
= 83, the starting time of trip k is 83
hours after time 0 of the planning horizon, i.e., 3 days
and 11 hours, then q
0
ilk
= 11.
We use the following decision variables:
q: starting time of a trip;
q
0
: starting hour of a trip;
f : ending time of a trip;
f
0
: ending hour of a trip;
r
j
: resting time of a daily journey (8h every 12h);
r
t
: resting time during a trip (30’ every 5:30h);
w
q
: waiting time to start a trip;
w
f
: waiting time after finishing a trip;
b
f
: binary, if a trip ends before time window;
a
f
: binary, if a trip ends after time window;
b
q
: binary, if a trip is ready to start before time
window;
a
q
: binary, if a trip is ready to start after time win-
dow;
z: number of full days of a given duration time;
m
j
,m
t
: number of required resting periods;
The constraints below show how f
0
i
is defined
from f
i
(1)-(2), how binaries b
i
and a
i
are set when
trip i did not finish within the operation time [P
O
i
,P
ˆ
O
i
]
of the destination of trip i (3)-(6) and how the waiting
time is set in these cases (7): the remaining hours until
the opening time if arrived early, or the hours until the
ICEIS 2018 - 20th International Conference on Enterprise Information Systems
322
opening time of the next day if arrive tardy. Similar
constraints are defined for f
il
, f
ilk
,q
il
,q
ilk
.
z
i
= f
i
/24 (1)
f
0
i
= f
i
24bz
i
c (2)
f
0
i
b
i
P
O
i
(3)
f
0
i
P
O
i
(1 b
i
) (4)
(24 f
0
i
)a
i
(24 P
ˆ
O
i
) (5)
24 f
0
i
(24 P
ˆ
O
i
)(1 a
i
) (6)
w
i
= (P
O
i
f
0
i
)b
i
+ (24 f
0
i
+ P
O
i
)a
i
(7)
b
i
{0, 1} (8)
a
i
{0, 1} (9)
w
i
0 (10)
The integer value z
i
= bz
i
c may be defined for a
continuous value z
i
by the constraint z
i
1 z
i
z
i
.
The following expressions define the total resting
time of different types while performing trip i. Si-
milar expressions are used for trips l and k (variables
r
j
il
,r
t
il
,r
j
ilk
,r
t
ilk
).
m
j
i
= t
i
/12 (11)
m
t
i
= t
i
/5.5 (12)
r
j
i
= 8bm
j
i
c (13)
r
t
i
= 0.5bm
t
i
c (14)
The complete MILP formulation is then:
min B
ilk
= ( f
ilk
+ w
ilk
q
i
) (15)
subject to:
P
O
i
q
i
P
ˆ
O
i
(16)
f
i
= q
i
+t
i
+ r
j
i
+ r
t
i
(17)
q
il
= f
i
+ w
i
+ e
il
+ r
q
il
+ r
0q
il
+ s (18)
f
il
= q
il
+ w
il
+t
l
+ r
j
il
+ r
t
il
(19)
q
ilk
= f
il
+ w
il
+ e
lk
+ r
q
ilk
+ r
0q
ilk
+ s (20)
f
ilk
= q
ilk
+ w
ilk
+t
k
+ r
j
ilk
+ r
t
ilk
(21)
(22)
(1)-(9) for f
i
, f
il
, f
ilk
,q
il
,q
ilk
(11)-(14) for r
j
i
,r
t
i
,r
j
il
,r
t
il
,r
j
ilk
,r
t
ilk
f
i
, f
il
, f
ilk
, f
0
i
, f
0
il
, f
0
ilk
0 (23)
q
i
,q
il
,q
ilk
,q
0
il
,q
0
ilk
0 (24)
Objective function (15) minimizes the cost (total
duration time) of the combination. The starting time
of the first trip, i, must be within the operation hour of
the origin location of the trip (16). The finishing time
of this trip is the starting time plus the time needed to
go from its origin to its destination plus the mandatory
resting times (17). As for the next trip, l, it may start
after finishing the previous one, plus the repositioning
time from the destination of i to the origin of l plus the
resting times, if any, during this repositioning (18).
Constraints (19)-(21) do the same for other trips.
We run the above formulation for all triplets
i,l,k T . If the solution is feasible, we have the mini-
mal cost B
ilk
of the combination. Otherwise we have
that trips i.l.k cannot be combined sequentially.
The MILP formulation to define the minimal cost
for the combination of two trips i,l T is similar:
min C
il
= f
il
+ w
il
q
i
(25)
subject to all constraints except the ones including va-
riables with index ilk.
4.2 Pre-processing Heuristically
Here we propose a greedy constructive heuristic to
quickly define an upper bound on the cost and total
duration of a sequence of trips i,l,k, considering that
each trip departs and a arrives as early as possible.
For example, the starting time of the first trip is the
opening time of its origin point: q
i
= P
O
i
. The en-
ding time f
i
includes the service time, the distance
to the destination point, mandatory resting time, and
waiting time due to time windows. The starting time
of the following trips include also the repositioning
from the destination point of the previous trip and the
possible incurring resting and waiting time.
It is easy to see that this heuristic gives an up-
per bound on the cost/total duration because the se-
quence thereby constructed is feasible: all constraints
are considered. One may also notice that it may not
be optimal, because starting a trip as early as possible
may force waiting times in the future that may incre-
ase the total duration time. It is, however, very fast.
Therefore we have an exact ILP formulation that,
given a set of trips i,l,k determine the best star-
ting/end time of each trip in order to minimize the
cost (albeit costly in terms of computational time),
and a heuristic that quickly determines a possible
good value. Regardless the method used for this pre-
processing time, or even a mixed of both, we still have
to decide which trips to combine. This task is done by
the set-covering based ILP formulation given in the
next section.
4.3 Set-covering ILP Formulation
After pre-processing all combinations up to 3 trips, let
T
2
and T
3
be the sets of all feasible combinations of
Reducing Empty Truck Trips in Long Distance Network by Combining Trips
323
2 and 3 trips respectively, and let B
ilk
be the total cost
of combining trips i,l,k sequentially, C
il
the total cost
of combining trips i,l sequentially, and D
i
the cost of
a single travel i (in our case twice the distance because
it would be a round trip). We then define the following
compact ILP formulation, using variables v
il
and y
ilk
as binary decision variables equals to 1 if trips i,l and
i,l,k respectively are to be combined and 0 otherwise,
and x
i
a binary decision variable equals to 1 if the trip
i is to be performed as a single trip (i.e., not part of
any combination) and 0 otherwise.
min Z =
ilkT
3
B
ilk
y
ilk
+
ilT
2
C
il
v
il
+
iT
D
i
x
i
(26)
subject to:
ilkT
3
i= jl= jk= j
y
ilk
+
ilT
2
i= jl= j
v
il
+ x
j
= 1 j T (27)
y
ilk
,v
il
,x
i
{0, 1} (28)
Objective function (26) minimizes the overall
cost. Constraints (27) state that each trip may be at
most in one chosen combination or performed as a
single trip. The last constraints state that all variables
are binary.
5 EXPERIMENTAL RESULTS
The MILP formulations were implemented in C/C++
using the Concert Technology Library, and solved by
CPLEX 12.5 academic license. The heuristic was im-
plemented in C/C++. The experiments were run on an
IntelR Core TM i7-4790K CPU @ 4.00GHz x 8 with
32GB RAM, running Ubuntu 14.04 LTS 64 bits.
5.1 Instances
We generate a set of instances based on real data
from the Brazilian transportation network. Data were
collected from Fretebras
1
, a website containing thou-
sands of freightage offers, filled by drivers and com-
panies in real-time, covering all Brazilian states and
some nearby countries. One may freely consults in-
formation such as origin and destination of freightage,
distance between those sites, vehicle type, freight
type, and others.
We selected 10 cities, including Santos - SP, Cu-
bat
˜
ao - SP, Manhuac¸u - MG, Passos - MG, Arcos -
MG, Rondon
´
opolis - MT, Sorriso - MT, Dourados -
MS and Itumbiara - GO, which are among the main
1
http://www.fretebras.com.br
origin or destination points for export of grains (cof-
fee and soy), and import of agricultural inputs (ferti-
lizers and agricultural plaster), thus generating most
of the import and export trips on the roads network of
the southeast region of Brazil. Santos is a port city
that receive plenty of trucks everyday, both for import
and export freights. We created seven instances, ran-
ging from 7 to 124 trips, using data from selected trips
of the Fretebras website.
The distances of the trips range from 20 to 2246
km, while the estimated duration ranges from 15 min
to almost 30 hours (including the repositioning trips,
which can be very short). Table 1 shows a summary
of the data for each instance, considering the import,
export and inland trips.
Table 1: Summary of the instances.
ID # Trips
Duration (h)
min max avg
1 15
3.7 26.7
14.0
2 31 15.8
3 47 14.5
4 62 13.1
5 76 15.0
6 94 13.5
7 124 13.9
We consider the same operation time for all cities:
7 am as the opening time and 6 pm as the closing time.
5.2 Results
The instances were solved by the two approaches pro-
posed: pre-processing the combinations by the exact
ILP model or the constructive heuristic and then se-
lect a subset of combinations by the set-covering mo-
del. The results are presented in Table 2, for 4 types
of tests: trips allowed to combine in sequence of 2
and/or 3 trips; trips allowed to combine only in se-
quence of 3 trips; only in sequence of 2 trips; and no
combination allowed (last rows, combined type = 1).
Notice that for this last type, none of the formulations
or approaches are used, one truck is used for each trip,
and the total cost may be evaluated at practically zero
time.
For each type of test, we report the objective
function value (accumulated duration time of all
trips), the number of trucks used, and the total time for
each approach for pre-processing: MILP of Section
4.1 and Heuristic (Heur) of Section 4.2. We do not
report gaps of the formulations because all formulati-
ons were solved until a proved optimal is found. For
the case of no combination (type 1), each trip is con-
sidered as a round trip, then the duration of a trip is
twice the travel time and mandatory resting and wai-
ting times.
ICEIS 2018 - 20th International Conference on Enterprise Information Systems
324
From the results we can see that for all instan-
ces, the proposed approaches could combine the trips,
decreasing the overall duration time and the number
of trucks, compared to the case of no combinations.
This can be seen on the table, and on Figures 8 and 9,
which show respectively the objective value and num-
ber of trucks for all instances and combination types.
In Figure 8, the objective values were scaled to the
cost of the 1-type, which was always the highest. In
order to have a clearer visualization of the results for
the number of trucks, Figure 10 shows the results only
for types that allow combinations.
The combination type that reached the minimum
objective value was, as expected, the 2&3 combina-
tion type, when pre-processing is done exactly. This
can be seen on the table and on the graphic of Figure
8. However, the combination that minimizes the most
the number of trucks was the 3-combination type,
which can be seen on the table and on the graphic of
Figure 10.
We may conclude that, using only combinations
of 3 trips, we may minimize the number of trucks, but
this may not be the best option in terms of total time.
In some cases a third trip may force a long waiting
time, and the best option is to combine only two trips.
Therefore, we have to allow combinations of 2 and of
3 trips, if we want to reduce the overall time.
In general, results obtained by pre-processing
using heuristics have higher objective values, regard-
less the combination type (see Figure 8). This hap-
pens because not all trips are combined. In the other
hand, pre-processing the combinations by the exact
ILP formulation yields lower costs for each combina-
tion, and the result is that in the second phase almost
all trips are combined, reducing the overall costs.
The exact pre-processing, however, is computati-
onal costly. For the larger instances, the exponential
number of combinations and the high cost to solve all
of them, lead the algorithm to run for more than one
day, reaching almost one week for the largest instance
allowing 3 or 2&3 combinations. The running time of
the heuristic was at most 2 minutes.
The approach here proposed reached all the ex-
pected objectives, while satisfying the imposed con-
straints:
Minimization of costs
The combination of trips allows a given driver to
perform two or three trips. Doing this, we avoid to
use a new driver to cover some trips, reducing the
operational costs. Another reason is the smaller
distances traveled for repositioning trips, which
reduces empty truck trips.
Minimization of empty truck trips and maximiza-
tion of the use of trucks capacity
When choosing the best combined route, the mo-
del prefers low cost combinations. A low cost
combination has generally a small distance for re-
positioning or no repositioning at all. As those tra-
vel times are considered in the objective function,
a preference is given for smaller distance for repo-
sitioning (which is also empty truck trips), using
the capacity of trucks as long as possible.
Minimization of polluting gas emission
This is a direct consequence of the combination
of trips. As well as the reduction of the number
of drivers, lesser vehicles on the road reduces the
emission of polluting gases. Another cause is the
reduction on the number and distance travelled on
repositioning trips.
Minimization of waiting on port area
The proposed approach takes in account the arri-
ving and departure time on ports and cities. When
a trip arrives outside the time window, the waiting
time is added to the objective function. Therefore,
the waiting time is reduced in order to minimize
the overall time. Moreover, the formulation deci-
des also the best departure time of the first trip,
aiming to minimize the required waiting time.
Solutions conforme to Brazilian transport regula-
tion laws
The formulation include regulations of the Brazi-
lian transport law 13103/2015, adding the manda-
tory resting periods. This includes daily rests, of
30 minutes for each 5:30h of driving, and long rest
of 8h after 12h driving.
Figure 8: Objective value.
6 CONCLUSIONS
This paper proposes a mathematical programming
formulation to meet the needs of a transportation net-
work including export, import and inland trips. The
main objective is to reduce empty truck trips, i.e., the
number and distance traveled by empty trucks. This
Reducing Empty Truck Trips in Long Distance Network by Combining Trips
325
Table 2: Experimental results for all instances, approaches and combinations type.
Comb. Instance Total duration CPU time (s) #Trucks
type ID Size ILP Heur ILP Heur ILP Heur
1 15 318.5 339.2 28 <1 7 6
2 31 720.1 807.1 394 <1 15 14
3 47 1033.7 1082.6 2789 4 23 19
2&3 4 62 1241.6 1388.3 10792 12 31 25
5 76 1740.5 1841.0 44535 20 37 32
6 94 1832.3 1998.2 110329 49 44 38
7 124 2122.3 2783.5 591482 170 57 53
1 15 323.2 341.7 33 <1 7 7
2 31 790.0 829.9 402 <1 10 13
3 47 1058.0 1110.2 2840 2 17 21
3 4 62 1268.2 1379.7 10829 8 22 30
5 76 1767.8 1865.5 33358 13 26 34
6 94 1928.4 2033.1 112263 35 32 40
7 124 2642.4 2906.2 544535 104 42 62
1 15 320.7 372.5 <1 <1 7 10
2 31 774.3 813.3 4 <1 15 18
3 47 1035.9 1113.8 19 2 23 27
2 4 62 1241.6 1379.4 56 5 31 37
5 76 1742.7 1881.9 134 8 39 1/46
6 94 1876.6 2047.0 356 21 47 57
7 124 2584.6 2862.4 1329 76 62 72
1 15 594.3 - - 15
2 31 486.7 - - 31
3 47 2975.6 - - 47
1 4 62 2290.6 - - 62
5 76 3320.7 - - 76
6 94 3538.8 - - 94
7 124 4859.6 - - 124
Figure 9: Number of trucks for all combination types, in-
cluding no combinations.
task was accomplished by minimizing the total dura-
tion of trips, which forces the combination of 2 or 3
trips in order to avoid needless trips. The combina-
tion was done respecting regulation laws imposed to
drivers traveling long distances and also the operation
times of ports. Using real-based data of Brazilian sce-
nery, we show that the number of empty trucks was
reduced, which directly contributes to reduce the con-
gestion on port areas and polluting gases emission.
For a small number of trips, the MILP formulation
showed to be a useful tool for route planning. But as
the number of trips increase, the computational time
becomes impracticable. In such case, a fast heuristic
may find a reasonable solution, not far from the opti-
mal, in short computational time.
Figure 10: Number of trucks for all combination types.
As future works we plan to improve the model in
order to reduce the computational time, and use ot-
her strategies to reach the optimal solution, for exam-
ple dynamic programming. Other work would be to
improve the quality of the heuristic, maintaining the
short computational time. This would allow the set-
cover formulation to be used in almost real-time for
replanning the combination in case of eventualities.
A further step would be to assign the combinations or
sequence of combinations of trips to a set of drivers.
In this last case, we may have to use metaheuristics,
as the problem may become more complex and a sim-
ple greedy heuristic, despite the good results achieved
in this paper, may not give a good result in this more
complex scenery.
ICEIS 2018 - 20th International Conference on Enterprise Information Systems
326
ACKNOWLEDGEMENTS
The authors thank Coordenac¸
˜
ao de Aperfeic¸oamento
de Pessoal de N
´
ıvel Superior (CAPES) and Fundac¸
˜
ao
de Amparo
`
a Pesquisa do Estado de Minas Gerais
(FAPEMIG) for the financial support of this project.
The first author thanks Gilson F. Ataliba for his con-
tribution in the first ILP formulation.
REFERENCES
Audy, J.-F., DAmours, S., and Rousseau, L.-M. (2011).
Cost allocation in the establishment of a collabora-
tive transportation agreementan application in the fur-
niture industry. Journal of the Operational Research
Society, 62(6):960–970.
Caballini, C., Paolucci, M., Sacone, S., and Ursavas, E.
(2017). Towards the physical internet paradigm: A
model for transportation planning in complex road
networks with empty return optimization. In Interna-
tional Conference on Computational Logistics, pages
452–467. Springer.
Caballini, C., Rebecchi, I., and Sacone, S. (2015). Combi-
ning multiple trips in a port environment for empty
movements minimization. Transportation Research
Procedia, 10:694–703.
Caixeta Filho, J. V. (2010). Log
´
ıstica para a agricultura bra-
sileira. Revista Brasileira de Com
´
ercio Exterior.
Gavish, B. and Schweitzer, P. (1974). An algorithm for
combining truck trips. Transportation Science, 8.
Islam, S. (2017a). Empty truck trips problem at container
terminals: A review of causes, benefits, constraints
and solution approaches. Business Process Manage-
ment Journal, 23.
Islam, S. (2017b). Simulation of truck arrival process at
a seaport: evaluating truck-sharing benefits for empty
trips reduction. International Journal of Logistics Re-
search and Applications, pages 1–19.
¨
Ozener, O.
¨
O. and Ergun,
¨
O. (2008). Allocating costs in
a collaborative transportation procurement network.
Transportation Science, 42(2):146–165.
Schulte, F., Gonz
´
alez, R. G., and Voß, S. (2015). Redu-
cing port-related truck emissions: coordinated truck
appointments to reduce empty truck trips. In Interna-
tional Conference on Computational Logistics, pages
495–509, The Netherlands. Springer.
Reducing Empty Truck Trips in Long Distance Network by Combining Trips
327