Guided Bee Colony Algorithm Applied to the Daily Car Pooling Problem
Mouna Bouzid, Ines Alaya and Moncef Tagina
LARIA-ENSI, National School of Computer Sciences, University of Manouba, Manouba 2010, Tunisia
Keywords:
Metaheuristic, Bee Colony Optimization, The Daily Car Pooling Problem.
Abstract:
The foraging behavior of bees has been adapted in a Bee Colony Optimization algorithm (BCO). This approach
is a simple and an efficient metaheuristic that has been successfully used to solve many complex optimization
problems in different domains, mostly in transportation, location and scheduling fields.
In this study, we develop two algorithms for the Daily Car Pooling Problem based on the BCO approach. The
developed algorithms are experimentally tested on benchmark instances of different sizes. The computational
results show that the proposed approaches can produce good solutions when compared with an exact method.
1 INTRODUCTION
In recent decades, the increased human mobility and
the high use of private vehicles has caused many prob-
lems, such as air pollution, parking problem, traffic
congestion, noise pollution, toxic emissions, and in-
crease in the number of crashes and accidents.
Public transportation service can be a solution to
this problem but it cannot adequately cover all pas-
senger transportation needs. Car pooling has emerged
to be another solution for reducing private car usage
around the world.
Car pooling (Bruglieri et al., 2011; Manzini and
Pareschi, 2012; Correia and Viegas, 2011; Yan et al.,
2011; Vargas et al., 2008) is a collective transportation
system that encourage people to share a common car
to reach the same destination, in order to decrease the
number of cars on the road.
In literature, we distinguish two main ways of op-
erating the car pooling. It can be either a Daily Car
Pooling Problem (DCPP) or a Long-Term Car Pool-
ing Problem (LTCPP).
In the case of DCPP (Baldacci et al., 2004; Calvo
et al., 2004; Swan et al., 2013), on each day a number
of users (servers) are available to share their vehicle
with colleagues (clients) on that particular day. The
problem is to assign clients to servers and to iden-
tify the routes to be driven by the servers. The aim
is to minimize the total travel cost, with respect to
time window and car capacity constraints. The DCPP
can be considered as a special case of the Dial-a-
Ride Problem (DARP) or the Vehicle routing problem
(VRP). The DCPP is a NP-hard problem as it is a spe-
cial case of the VRP which is known to be NP- hard
Problem (Toth and Vigo, 2014).
However, in the case of LTCPP (Guo et al., 2012;
Yan et al., 2011; Bouzid et al., 2020), each user is
available to act both as a server and as a client. The
LTCPP requires to define crews or user pools
where each user will in turn, on different days, pick up
the remaining pool members. The objective here be-
comes that of maximizing pool sizes and minimizing
the total distance travelled by all users when acting
as servers, subject to car capacity and time window
constraints.
The specific problem addressed in this paper is the
DCPP. In spite of its hardness, only few researches
have been carried out in this problem.
Authors in (Baldacci et al., 2004) present both an
exact and a heuristic method for the DCPP based on
two integer programming formulations of the prob-
lem. The exact method is based on a bounding proce-
dure that combines three lower bounds derived from
different relaxations of the problem. A valid upper
bound is obtained by the heuristic method, which
transforms the solution of a Lagrangean lower bound
into a feasible solution.
In (Calvo et al., 2004), Calvo et al. give another al-
gorithm solving the DCPP. The main idea of this algo-
rithm is the use of greedy assignment alternating with
random perturbation. In fact, the greedy assignment
phase look to minimize a marginal quantity called re-
gret. Where the regret for each client i is given by the
difference of the length paths between the two servers
which have the least and the second least extra mile
when pick up client i.
Bouzid, M., Alaya, I. and Tagina, M.
Guided Bee Colony Algorithm Applied to the Daily Car Pooling Problem.
DOI: 10.5220/0010517504650472
In Proceedings of the 16th International Conference on Software Technologies (ICSOFT 2021), pages 465-472
ISBN: 978-989-758-523-4
Copyright
c
2021 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
465
Authors in (Swan et al., 2013) examine the effect
of varying the acceptance criterion. In particular, they
investigate the use of Late-acceptance Hillclimbing.
In this metaheuristic a new solution is accepted if it is
no worse than the k-th most recent solution.
After bibliographical study, we remarked that the
Bee Colony Optimization algorithm has been used
only to solve the LTCPP. Authors in (Bouzid et al.,
2020) developed a Bee Colony Optimization algo-
rithm for the Long-term Car Pooling problem where
all choices in back-ward pass are random.
In this article, we present a different algorithm
based on the Bee Colony Optimization to solve the
second type of car pooling: the Daily Car pooling
problem. In this algorithm, called the Guided-BCO
algorithm, all choices in the back-ward pass are based
on probability equations. This algorithm is com-
pared with an exact method (Baldacci et al., 2004)
and tested on two classes of benchmark instances
from (Christofides and Eilon, 1969; Christofides,
1979; Fisher, 1994). We have also developed another
version of BCO algorithm where all decisions and
choices in the backward pass are random, named Ran-
dom BCO-DCPP algorithm and we have compared
this version with the Guided BCO algorithm.
The paper introduces an adaptation of the Bee
Colony Algorithm (BCO) to the Daily Car Pooling
Problem ( DCPP) and it is organized as follows. The
DCPP, whose target is to share vehicles among users
(servers and clients ) to minimize the total travel cost
is expressed in Section 2. The BCO, in which an
artificial bee colony visits the search space to find
a feasible solution, is described in Section 3. BCO
is adapted to DCPP, by specializing the forward and
backward phases of BCO, in Section 4. Experimental
results and comparisons are the focus of Section 5.
2 PROBLEM FORMULATION
In this section, we present a mathematical formulation
of the DCPP using research in (Baldacci et al., 2004).
Mathematically, the DCPP can be described by
a directed graph G = {V {0},A}, where V =
{1,. . . ,n} is the set of employees, 0 represents the
destination, and A = {arc(i, j)/i V, j V {0}}
is the set of arcs. Every arc (i, j) A is associated
with a non-negative travel cost d
i j
and a travel time
t
i j
. Each employee i V is specified by an origin
(home), a node 0 represents the destination, a non-
negative value representing the earliest departure time
e
i
for leaving home; and a positive value denoting the
acceptable time l
i
for arriving at destination. The set
V is partitioned as V = V
s
V
c
, where V
s
= {1,...,n
s
}
is the subset of servers and V
c
= {n
s
+ 1, . . . ,n} is the
subset of clients. For each server k V
s
, we denote
by Q
k
the car capacity and by T
k
the maximum driv-
ing time he is willing to go from home to workplace.
Each client i V
c
is characterized by a penalty p
i
rep-
resenting its contribution to the total cost in case no
server picks him up. We denote by Γ
i
= { j : (i, j) A}
the set of successors of vertex i V {0}, and by
Γ
1
i
= {i : (i, j) A} the set of predecessors.
Before formulate our problem, we should present
some notations:
x
k
i j
: Binary variable equals to 1 if and only if arc
(i,j) is traveled by a server k, with (i, j) A and
k V
s
.
y
ik
: Binary variable equals to 1 if and only if i is
not picked by any server, where k V
s
and i V
c
.
S
i
: Positive variable indicating the pick-up time
of employee i V .
h
k
: Non-negative variable denoting the arrival
time of server k V
s
at the destination.
Objective function:
MinZ(F) =
(kV
s
)
(i, j)A
d
i j
x
k
i j
+
iV
c
p
i
y
i
(1)
Constraints :
jΓ
k
x
k
k j
= 1; k V
s
; (2)
jΓ
1
0
x
k
j0
= 1; k V
s
; (3)
jΓ
1
i
x
k
ji
jΓ
i
x
k
i j
= 0; k V
s
,i V
c
; (4)
(i, j)A
x
k
i j
Q
k
;k V
s
; (5)
(i, j)A
t
i j
x
k
i j
T
k
;k V
s
; (6)
S
j
S
i
t
i j
+ M(1
kV
s
x
k
i j
);(i, j) A; (7)
S
i
e
i
;i V, (8)
h
k
S
i
+t
i0
M(1 x
k
i0
);i V, k V
s
; (9)
h
k
l
i
+ M(1
jΓ
i
x
k
i j
);i V, k V
s
; (10)
kV
s
jΓ
i
x
k
i j
+ y
i
= 1; i V
c
(11)
x
k
i j
{0,1}; (i, j) A, k V
s
, (12)
y
i
{0,1}; i V
c
; (13)
h
k
0; k V
s
; (14)
S
i
0; i V ; (15)
ICSOFT 2021 - 16th International Conference on Software Technologies
466
Equation (2) shows that every server must leave its
house and equation (3) forces it to arrive at the des-
tination (workplace). Constraint (4) makes sure the
continuity of the path. The capacity and the maxi-
mum time constraints are translated in inequalities (5)
and (6), respectively. Equations (7) and (8), where M
is a big constant, define the arrival time variables S
i
,
while inequalities (9) and (10) set the arrival times h
k
, k V
s
, of the servers at the workplace and assume
that each employee i V should reach the workplace
before the latest arrival time l
i
, respectively. Equa-
tion (11) ensures that each client can be picked by a
server or is left unserved. Constraints (12) and (13)
are binary constraints, while (14) and (15) are posi-
tivity constraints.
3 THE BEE COLONY
OPTIMIZATION ALGORITHM
The Bee Colony Optimization is among the famous
natural inspired metaheuristic based on the collective
bee intelligence. It was proposed, first, by Lucic and
Teodorovic (Lu
ˇ
ci
´
c and Teodorovi
´
c, 2003) to deal with
the well hard combinatorial optimization problems
like the Vehicle Routing Problem (Lu
ˇ
ci
´
c and Teodor-
ovi
´
c, 2003; Teodorovi
´
c, 2008), the Job Shop Schedul-
ing Problem (Chong et al., 2006; Chong et al., 2007),
the p-Median Problem (Teodorovic and
ˇ
Selmic, 2007)
and the Traveling Salesman Problem (Wong et al.,
2010).
The main idea of this approach is to build an ar-
tificial bee colony, where bees investigate through
the search space looking for feasible solutions. Af-
ter that, they communicate, collaborate and exchange
information. Thanks to this communication systems
named ”Swarm Intelligence”, and using collective
knowledge and information sharing, artificial bees in-
crementally construct solution to the problem. The
BCO performs its search process in iterations until a
stopping condition is met.
Flying through the space, an artificial bee per-
forms either forward pass or backward pass. In
the case of forward pass, every bee makes a prede-
fined number of local moves, which slowly construct
and/or improve the solution, yielding a new solution.
Building partial solutions, bees start the second phase
called the backward pass. In this phase, bees return to
the hive. They exchange information about the qual-
ity of the partial solution created which is defined as
the current value of the objective function.
Having the evaluation of all partial solutions, each
bee decides with a certain probability whether to still
faithful to its created partial solution or not. It is obvi-
ous that bees with better solutions have greater chance
to keep and continue their own exploration.
Faithfull bees are considered as recruiters, and
their solutions would be considered by other bees.
However, if a bee chooses to abandon its solution it
becomes uncommitted and it has to select one solu-
tion from recruiters by the roulette wheel. Note that
better solutions have higher opportunities to be cho-
sen for further exploration. Forward and backward
pass, could be iteratively performed until each bee
completes the generation of its solution or a stopping
condition is satisfied. Among the possible stopping
conditions, we found the maximum total number of
forward/backward passes or the maximum total num-
ber of forward/backward passes without the improve-
ment of the objective function, etc.
4 THE PROPOSED BCO
ALGORITHM FOR THE DCPP
In this section, we present our algorithm based on the
Bee Colony Optimization called the Guided-BCO al-
gorithm.
4.1 Problem Representation
In order to maintain the simplicity of the BCO algo-
rithm, a rather straightforward solution representation
scheme is adopted. Let us represent each employee
by a node. Our problem is divided into stages where
the first stage represents servers and all others repre-
sent clients. In every stage, an artificial bee chooses
to visit one node. At the beginning of each new pool,
the selection of a new server (new initial node) was
represented by changing the location of a hive.
4.2 Forward Pass Phase
At the beginning of the BCO process, all artificial
bees are located in the hive. Bees depart from the
hive and fly to an unvisited client who satisfies con-
straints. It chooses a new node to be added to his
partial pool using the roulette wheel selection. This
technic is based on the probability values, which gives
the bee the likelihood to move from client i to client
j. To calculate the probability, we need the distance
between the current client i and the client to be vis-
ited j. It is obvious that, the shorter the distance,
the higher the probability to choose a client. There-
fore, the travel cost and the probability are inversely
proportional. Formally, the probability is defined in
equation (16) as follow:
Guided Bee Colony Algorithm Applied to the Daily Car Pooling Problem
467
P
i j
=
1
d
i j
jV
c
1
d
i j
(16)
4.3 Backward Pass Phase
During the backward pass, bees evaluate the quality of
all generated partial solutions by calculating the cur-
rent value of the objective function using equation (1).
Then, each bee b decides whether to stay loyal to its
partial solution or to abandon it. This choice was per-
formed in a probabilistic manner. This probability is
expressed as follows:
P
u+1
b
= e
O
max
O
b
u
;b = 1,2,..., B (17)
Where:
O
max
: denotes the maximum over all normalized
values of partial/complete solutions to be com-
pared;
O
b
: represents the normalized value for the objec-
tive function of partial/complete solution created
by the b-th bee;
U: is the number of forward pass;
Since we are in the case of minimization criterion, the
normalized value is calculated as follows:
O
b
=
C
max
C
b
C
max
C
min
;b = 1,2,.., B (18)
Where:
C
b
: is the value for the objective function of b-th
bee partial/ complete solution;
C
min
: represents the minimal objective function
value obtained by all engaged bees;
C
max
: denotes the maximal objective function
value obtained by all engaged bees;
Using a random number and equation (17) each bee
can make its decision. In fact, if the generated num-
ber is greater than the calculated probability the bee
is considered uncommitted else it is considered re-
cruiter. Once the bee becomes uncommitted, it must
choose which recruiter it will follow by the roulette
wheel. Also, this selection is guided by a probabil-
ity. The probability that b’s partial/complete solution
would be chosen by any uncommitted bee is equal to:
P
b
=
O
b
R
k=1
O
k
;b = 1,2,..., R; (19)
Where O
k
is the normalized value for the objective
function of the k-th solution and R denotes the num-
ber of recruiters.
4.4 Solution Construction
In order to build a pool, the two steps of the search al-
gorithm, forward and backward pass, are alternated it-
eratively until the total number of forward/ backward
passes reaches the car pool size. At this point, the
best among all B pools is determined. It is then used
to build a global solution and the construction of a
new pool begins. The iteration is considered finished,
when all clients are assigned to servers. Our algo-
rithm runs iteration by iteration until the maximum
number of iterations is reached. At the end, the global
best-found solution is reported as the solution of our
problem.
4.5 Guided BCO-DCPP Algorithm
The overall structure of the Guided BCO-
DCPP is outlined as shown in Algorithm 1.
Algorithm 1: Guided BCO-DCPP Algorithm.
Initialize parameters:- Number of bees B.
- Number of iterations.
Repeat
For every server i in the set of servers do{
Initialization: every bee is set to an empty pool;
Add server i to every pool;
While the car capacity of server i is not reached
do{
Forward pass
a) For every bee do {
i. Evaluate all unserved clients in the current
set of clients not yet pooled;
ii. Choose an unserved client c who satisfies
constraints using the roulette wheel selection
based on equation (16);
iii. Insert client c into the current car pool
and eliminate it from the current set of
clients not yet pooled;}
Backward pass
b) For every bee do {
Evaluate partial pool using equation (1); }
c) For every bee do {
Loyalty decision using equation (17); }
d) For every uncommitted bee do {
Choose a recruiter bee to follow by a roulette
selection based in equation (19); } }
Evaluate all pools, find the best one and add it to
the current partial solution; }
Update the best solution;
Until the number of iteration is reached;
Out put the best solution;
ICSOFT 2021 - 16th International Conference on Software Technologies
468
5 EXPERIMENTAL STUDY AND
DISCUSSION
In order to improve the performance and the effi-
ciency of our proposed algorithm and after biblio-
graphic research we have chosen to compare our ap-
proach with the exact algorithm described in (Bal-
dacci et al., 2004) since there aren’t other intelligent
algorithms carried out on this problem.
Also, we have developed another version of BCO
algorithm where all decisions and choices in the back-
ward pass are random, that is why it is named a Ran-
dom BCO-DCPP algorithm and we have compared
this version of algorithm with the Guided BCO algo-
rithm.
5.1 Benchmarks
We have tested both the Random BCO and the Guided
BCO algorithms on two classes of benchmark in-
stances: Class A and Class B. Class A includes in-
stances originally derived from dataset provided by
Christofides and Eilon (1969) (Christofides and Eilon,
1969), Christofides et al. (1979) (Christofides, 1979)
and Fisher (1994) (Fisher, 1994) for the VRP. The
number of users in each instance is ranging from 51
to 225.
For all instances of class A, we considered the
depot as the destination, while we retained the co-
ordinates of the customers, who become the employ-
ees. We randomly choose n/4 among employees to be
considered as servers of our problem (n
s
= n/4) and
others are considered as clients. The cost d
i j
was as-
sumed to be equal to the Euclidean distance between
user i and j. The travel time t
i j
were set to be equal to
the distance d
i j
.
For each server k V
s
, the car capacity Q
k
was set
with equal probability to be equal to 4 or 5 and the
maximum ride time T
k
was calculated as T
k
= 1.5t
k0
,
where t
k0
is the time needed to travel from the server
k’s home to the destination.
For each client i V
c
, the penalty p
i
was computed
as p
i
= 2d
i0
, where d
i0
is the travel cost from client’s
home directly to the destination.
The latest arrival times l
i
is an integrate value ran-
domly selected in the interval [510, 540], and the ear-
liest departure time e
i
was estimated to be equal to
e
i
= l
i
max(t
i0
+ 30, 2t
i0
), with i V .
Class B contains 23 problems composed of users
ranging from 50 to 250. This class of instance is
selecting from the real-world instance for a research
institution in Italy. In fact, the car capacity Q
k
, the
maximum ride time T
k
(k V
s
), the penalty p
i
(i V
c
),
the latest arrival times l
i
, the earliest departure time e
i
(i V ) are calculated exactly like in the class A.
5.2 Parameters Setting
All computational results were obtained on a 3317U,
1.70 GHz computer, and the proposed algorithms
were coded in java.
For the experiments, some common parameters
were set as follows:
- Number of iterations: IT = 1000,
- Number of bees: B = number of users of the in-
stance.
5.3 Comparative Results
In this section, we evaluate the developed algorithms
by comparing them with the exact algorithm. The re-
sults of the exact algorithm are gained directly from
(Baldacci et al., 2004). We run both algorithms 30
times for each instance.
Tables 1 and 2 present a summary result of our
study. For each test instance, the table indicates the
number of employees n, the number of servers n
s
, the
number of clients n
c
, the optimal solution value found
by the exact algorithm. We also report the best (Best)
and the average (AVG) values obtained by the devel-
oped algorithms. In addition, tables give the deviation
(DV%) of best BCO from the optimal solution. Fur-
thermore, to validate the statistical significance of the
proposed algorithms we have use the non-parametric
Wilcoxon rank signed test (Sheskin, 2003) (W-test).
Note that for W-Test, the level of significance con-
sidered is 0.05. We use (+) and (-) to denote if the
Guided BCO results, respectively, significantly or not
significantly better than the Random BCO algorithm.
Table 1 summarizes the results of instances of
class A. Through this table, we can say that the aver-
age deviation of the best solution for the Guided BCO
from the optimal solution is 0.43% and that the av-
erage deviation of the best solution for the Random
BCO from the optimal solution is 0.89%. So, we can
say that both algorithms can produce good results.
Indeed, the Guided BCO algorithm outperforms the
Random algorithm in 11 instances from 12 instances
in average solution and in the best-found solution.
Also, statistically, the Guided BCO algorithm is sig-
nificantly better than the Random BCO algorithm in
ten instances among the twelve instances.
Guided Bee Colony Algorithm Applied to the Daily Car Pooling Problem
469
Table 1: Comparison of Guided BCO and Random BCO algorithms with the exact method on set A instances.
Exact method Guided BCO Random BCO W-test
Instance n n
s
n
c
Optimal Best AVG DV% Best AVG DV%
A1 51 13 37 1202 1210 1237.5 0.67 1213 1241.4 0.92 (+)
A2 76 19 56 1490 1493 1526.5 0.2 1496 1547.67 0.4 (+)
A3 101 26 74 1378 1386 1461.75 0.58 1390 1479.5 0.87 (+)
A4 121 31 89 2348 2356 2389.6 0.34 2410 2453.86 2.64 (+)
A5 121 31 89 1949 1960 2010.14 0.56 1975 2014.36 1.33 (+)
A6 135 34 100 2271 2288 2325.26 0.75 2311 2415.23 1.76 (+)
A7 151 38 112 2091 2092 2118.72 0.05 2094 2152.37 0.14 (+)
A8 171 43 127 2872 2882 2922.67 0.35 2881 2923.88 0.31 (-)
A9 171 43 127 2668 2681 2738.38 0.49 2689 2759.37 0.79 (+)
A10 196 49 146 3187 3190 3232.25 0.09 3195 3231.17 0.25 (-)
A11 200 50 149 2355 2376 2417.35 0.89 2381 2454.29 1.1 (+)
A12 226 57 168 2343 2346 2377.46 0.13 2347 2396.12 0.17 (+)
Average 2179.5 2188.33 2229.8 0.43 2198.5 2255.77 0.89
Table 2: Comparison of Guided BCO and Random BCO algorithms with the exact method on set B instances.
Exact method Guided BCO Random BCO W-test
Instance n n
s
n
c
Optimal Best AVG DV% Best AVG DV%
B1 101 25 75 1509 1511 1521.5 0.13 1512 1548.6 0.2 (+)
B2 101 25 75 1484 1501 1537.2 1.35 1542 1573.71 3.9 (+)
B3 101 25 75 1445 1465 1496.7 1.38 1471 1499.3 1.8 (+)
B4 101 25 75 2218 2246 2266.14 1.26 2259 2273.21 1.85 (+)
B5 101 25 75 1593 1594 1627.34 0.06 1602 1640.53 0.56 (+)
B6 101 25 75 1282 1287 1298.42 0.39 1288 1300.86 0.47 (+)
B7 101 25 75 1333 1348 1371.87 1.12 1336 1377.42 0.23 (+)
B8 151 38 112 1985(a) 1799 1836.5 -9.37 1784 1817.13 -10.13 (-)
B9 151 38 112 1970 1998 2018.33 1.42 2016 2020.86 2.34 (+)
B10 151 38 112 2560(a) 2422 2453.16 -5.39 2431 2459.43 -5.03 (+)
B11 151 38 112 2010(a) 1867 1913.33 -7.11 1909 1951.11 -5.02 (+)
B12 151 38 112 1674 1687 1698.83 0.78 1693 1705.14 1.13 (+)
B13 201 50 150 2467(a) 2307 2335.11 -6.49 2309 2342.8 -6.4 (+)
B14 201 50 150 2289(a) 1807 1868.77 -21.06 1801 1862.2 -21.32 (-)
B15 201 50 150 3109(a) 2903 2943.1 -6.63 2957 2978.7 -4.89 (+)
B16 201 50 150 3172(a) 2927 2977.7 -7.72 3026 3048.5 -4.6 (+)
B17 201 50 150 3660(a) 3413 3435.89 -6.75 3410 3439.3 -6.83 (+)
B18 201 50 150 2521(a) 2386 2416.1 -5.36 2436 2480.9 -3.37 (+)
B19 251 63 187 3244(a) 3012 3040.25 -7.15 3011 3044.34 -7.18 (+)
B20 251 63 187 3312 3316 3336.34 0.12 3318 3339.55 0.18 (+)
B21 251 63 187 3676 3680 3690.12 0.11 3678 3689.32 0.05 (+)
B22 251 63 187 3323 3325 3348.11 0.06 3324 3351.33 0.03 (+)
B23 251 63 187 3000 3049 3079.3 1.63 3057 3083.56 1.9 (+)
Average 2139.92 2297.83 2326.53 -3.18 2311.74 2340.34 -2.61
Note. (a) Instances that had not been solved to optimality by Maniezzo.
From table 2, we can easily see that the Guided
BCO algorithm gets the best solution on 16 instances
from 23 and on 20 instances considering the average
solution quality comparing with the Random BCO al-
gorithm. Moreover, the average deviation percentages
for all set B instances are -3.18% for Guided BCO
and -2.61% for Random BCO from exact method.
Here, it should be noted that instances B8, B10, B11,
B13, B14, B15, B16, B17, B18 and B19 had not
been solved to optimality by Maniezzo. It means that
our proposed algorithms are more robust and more
efficient for these ten instances and that these algo-
ICSOFT 2021 - 16th International Conference on Software Technologies
470
rithms can produce very good solutions for all set B
instances. Remarkably, the Guided BCO algorithm is
significantly better than the Random BCO on 21 in-
stances of class B.
5.4 Comparison of Execution Times of
Guided BCO with Random BCO
The time taken to execute the proposed algorithms is
given in figure 1 and figure 2. Both algorithms were
executed on the same machine.
Based on figure 1 and figure 2 we can state that
for all instances the Random BCO algorithm curve is
always above the Guided BCO curve. So, we can con-
clude that the Random BCO algorithm requires very
little execution time and it is slightly more quickly
than the Guided BCO on all instances of different
classes.
Figure 1: Computing time needed by the Guided BCO al-
gorithm and the Random BCO algorithm for instances of
class A (in seconds).
Figure 2: Computing time needed by the Guided BCO al-
gorithm and the Random BCO algorithm for instances of
class B (in seconds).
All in all, we can say that the Guided BCO algo-
rithm outperforms the Random BCO algorithm, even
if it takes very slightly more CPU time.
6 CONCLUSIONS
In this paper, we presented two different intelligent
algorithms inspired from bee’s behavior to solve the
Daily Car Pooling Problem, one is called the Guided
BCO algorithm and the other is named the Random
BCO algorithm. To show the effectiveness of the de-
veloped algorithms we have tested them on different
benchmark instances. Experimental results show that
developed algorithms can produce very good results.
To the best of our knowledge, these are the first swarm
intelligent algorithms solving the DCPP.
These results motivate us to apply the proposed
BCO algorithm on other similar problems like vehicle
routing problem or travelling thief problem.
REFERENCES
Baldacci, R., Maniezzo, V., & Mingozzi, A. (2004). An
exact method for the car pooling problem based on
lagrangean column generation. Operations Research,
52(3), 422-439.
Bouzid, M., Alaya, I., & Tagina, M. (2020). A Bee Colony
Optimization Algorithm for the Long-Term Car Pool-
ing Problem. In 15th International Conference on
Software Technologies (ICSOFT 2020).
Bruglieri, M., Ciccarelli, D., Colorni, A., & Lu
`
e, A. (2011).
PoliUniPool: a carpooling system for universities.
Procedia-Social and Behavioral Sciences, 20, 558-
567.
Calvo, R. W., de Luigi, F., Haastrup, P., & Maniezzo, V.
(2004). A distributed geographic information system
for the daily car pooling problem. Computers & Op-
erations Research, 31(13), 2263-2278.
Chong, C. S., Low, M. Y. H., Sivakumar, A. I., & Gay, K.
L.(2006, December). A bee colony optimization algo-
rithm to job shop scheduling. In Proceedings of the
2006 winter simulation conference (pp. 1954-1961).
IEEE.
Chong, C. S., Low, M. Y. H., Sivakumar, A. I., & Gay, K.
L. (2007). Using a bee colony algorithm for neigh-
borhood search in job shop scheduling problems. In
21st European conference on modeling and simula-
tion (ECMS 2007).
Christofides, N. (1979). The vehicle routing problem. Com-
binatorial optimization.
Christofides, N., & Eilon, S. (1969). An algorithm for the
vehicle-dispatching problem. Journal of the Opera-
tional Research Society, 20(3), 309-318.
Correia, G., & Viegas, J. M. (2011). Carpooling and car-
pool clubs: Clarifying concepts and assessing value
enhancement possibilities through a Stated Preference
web survey in Lisbon, Portugal. Transportation Re-
search Part A: Policy and Practice, 45(2), 81-90.
Fisher, M. L. (1994). Optimal solution of vehicle rout-
ing problems using minimum k-trees. Operations re-
search, 42(4), 626-642.
Guided Bee Colony Algorithm Applied to the Daily Car Pooling Problem
471
Guo, Y. (2012). Metaheuristics for solving large size long-
term car pooling problem and an extension.
Lu
ˇ
ci
´
c, P., & Teodorovi
´
c, D. (2003). Vehicle routing problem
with uncertain demand at nodes: the bee system and
fuzzy logic approach. In Fuzzy sets based heuristics
for optimization (pp. 67-82). Springer, Berlin, Heidel-
berg.
Manzini, R., & Pareschi, A. (2012). A decision-support sys-
tem for the car pooling problem. Journal of Trans-
portation Technologies, 2(02), 85.
Sheskin, D. J. (2003). Handbook of parametric and non-
parametric statistical procedures. crc Press.
Swan, J., Drake, J.,
¨
Ozcan, E., Goulding, J., & Woodward,
J. (2013). A comparison of acceptance criteria for the
daily car-pooling problem. In Computer and informa-
tion sciences III (pp. 477-483). Springer, London.
Teodorovi
´
c, D. (2008). Swarm intelligence systems for
transportation engineering: Principles and applica-
tions. Transportation Research Part C: Emerging
Technologies, 16(6), 651-667.
Teodorovic, D., &
ˇ
Selmic, M. (2007). The BCO algorithm
for the p-median problem. Proceedings of the XXXIV
Serbian Operations Research Conferece, 417-420.
Toth, P., & Vigo, D. (Eds.). (2014). Vehicle routing: prob-
lems, methods, and applications. Society for Industrial
and Applied Mathematics.
Vargas, M. A., Sefair, J., Walteros, J. L., Medaglia, A. L., &
Rivera, L. (2008, April). Car pooling optimization: a
case study in Strasbourg (France). In 2008 IEEE Sys-
tems and Information Engineering Design Symposium
(pp. 89-94). IEEE.
Wong, L. P., Low, M. Y. H., & Chong, C. S. (2010). Bee
colony optimization with local search for traveling
salesman problem. International Journal on Artificial
Intelligence Tools, 19(03), 305-334.
Yan, S., Chen, C. Y., & Lin, Y. F. (2011). A model with
a heuristic algorithm for solving the long-term many-
to-many car pooling problem. IEEE Transactions on
Intelligent Transportation Systems, 12(4), 1362-1373.
ICSOFT 2021 - 16th International Conference on Software Technologies
472