http://horos.rdsor.ro/ijcccv3n4Draft.pdf


Int. J. of Computers, Communications & Control, ISSN 1841-9836, E-ISSN 1841-9844

Vol. III (2008), No. 4, pp. 366-373

Ant Colony Solving Multiple Constraints Problem: Vehicle Route
Allocation

Sorin C. Negulescu, Claudiu V. Kifor, Constantin Oprean

Abstract: Ant colonies are successfully used nowadays as multi-agent systems
(MAS) to solve difficult optimization problems such as travelling salesman (TSP),
quadratic assignment (QAP), vehicle routing (VRP), graph coloring and satisfiability
problem. The objective of the research presented in this paper is to adapt an improved
version of Ant Colony Optimisation (ACO) algorithm, mainly: the Elitist Ant System
(EAS) algorithm in order to solve the Vehicle Route Allocation Problem (VRAP).
After a brief introduction in the first section about MAS and their characteristics, the
paper presents the rationale within the second section where ACO algorithm and its
common extensions are described. In the approach (the third section) are explained
the steps that must be followed in order to adapt EAS for solving the VRAP. The
resulted algorithm is illustrated in the fourth section. Section five closes the paper
presenting the conclusions and intentions.
Keywords: Ant Colony Optimisation, Vehicle Route Allocation Problem, Multi-
Agent Systems.

1 Introduction

Ant colonies are used nowadays as multi-agent systems to solve different optimization problems
from the NP-hard class. Examples of such problems are traveling salesman (TSP), quadratic assignment
(QAP), vehicle routing (VRP), graph colouring and satisfiability problem.

The multi-agent systems are copying some or all the characteristics of their biological counterparts
(depending on how much these characteristics are helpful to the MAS solving a particular type of prob-
lem) such as [6]:

• distributed society of autonomous individuals/agents;

• fully distributed control among the agents;

• localized communications among the individual;

• stochastic agent decisions;

• system-level behaviours transcending the behavioural repertoire of the single (minimalist) agent;

• simple interaction rules.

Consequently, the overall very important features of the system are: robustness, adaptability and scala-
bility.
The paper presents the technique of solving yet another difficult problem (VRAP) which belongs to the
NP-hard class of problems. Related work, regarding the ACO and EAS algorithms, is described recently
(2006-2007) in [1], [2], [3], [9]; to weaken redundancy, here details are skipped over. As a result, the rest
of the paper is structured as follows: Section 2 expounds the rationale where Ant Colony Optimisation
(ACO) and its common extensions are explained. Section 3 describes the steps that must be followed
in order to adapt EAS for solving the VRAP. The next section concentrates on presenting the resulted
algorithm. Conclusions and directions of future work (section 5) are closing the paper.

Copyright © 2006-2008 by CCC Publications



Ant Colony Solving Multiple Constraints Problem: Vehicle Route Allocation 367

2 Rationale and Approach

The ant colony optimization algorithm (ACO), introduced by Marco Dorigo in 1992 in his PhD
thesis, is a probabilistic technique for solving computational problems which can be reduced to finding
good paths through graphs and is inspired by the behaviour of ants in finding paths from the colony to
food [8].

In all early Ant System (AS) algorithms, ants are searching for (candidate) solutions based on two
main components: pheromone trails and problem-dependent heuristic information. These algorithms
have suffered frequent modifications in order to improve their efficiency. Thus, the AS [5] developed
into the Elitist Ant System (EAS) [4], because each ant that finds a better solution has the chance to
deposit more pheromone.

Other systems that emerged from the AS are the Max-Min Ant System (MMAS) [10] and the Ant-Q
[7] where the deposited pheromone amount is directly proportional to the quality of the found solution.
In the Rank-Based Ant System (ASrank) all solutions are ranked accordingly to their fitness. The amount
of pheromone deposited is then weighted for each solution, such that the more optimal solutions deposit
more pheromone than the less optimal solutions.

The basic VRAP consists in allocating vehicles from a fleet to different routes in such a way that the
average distance travelled in a month is (optimally) equal for all vehicles.

Before designing an ACO algorithm some questions regarding important aspects of the problem must
be answered first [6]:

• The representation of the problem (pheromone model): What are the most effective state features
to consider for learning "the good decisions"?

• Heuristic variables: What are they and what is their relative weight regarding the pheromone in
the decisions?

• Ant-routing table: How pheromone and heuristic variables are combined together to define the
goodness of a decision?

• Stochastic decision policy: How to make good exploration without sacrificing exploitation of
promising/good actions?

• Regarding pheromones: Is it useful to put limits in pheromone ranges? How should the pheromone
initialization and evaporation be dealt with? What is the best policy for pheromone updating:

online or offline?

• Scheduling of the ants: How and/or how many per iteration?

• Restarting the algorithm after stagnation: Would it be better to restart the algorithm when stagna-
tion is detected?

• Daemon components: What is the effect of local search?

The standard form of the EAS algorithm for solving the TSP is presented in figure 1.



368 Sorin C. Negulescu, Claudiu V. Kifor, Constantin Oprean

// Initialize pheromone trails
for (every edge i, j) {

τ = τ0
}

// Choose the starting town for every ant
for (k = 1; k ≤ m; k++) {

Place ant k on a randomly chosen city
}
// Initialize the best tour and length
T + = the shortest tour found from the beginning
L+ = the length of the best tour
// Main loop
for (t = 1;t ≤ T max;t++) {

// Compute a tour for every ant
for (k = 1; k ≤ m; k++) {

Build tour Tk(t) by applying n −1 times the following step:
Choose the next node (city) j with the probability

pki j(t) =
[τi j (t)]

α ·[ηi j (t)]β
∑

l∈Jk
i
[τil (t)]

α ·[ηil (t)]β
, if j ∈ J

pki j(t) = 0, if j 6∈ J
where i is the current city.

}
// Compute the tour lengths for all ants
for (k = 1; k ≤ m; k++) {

Compute the length Lk(t) of the tour Tk(t) produced by ant k
}
// Update the best tour if an improved tour is found
if (an improved tour is found) {

Update T + and L+

print T + and L+

}
// Global update for the pheromone trails
for (every edge i, j) {

Update the pheromone trails by applying the rule:
τi j(t) = (1− ρ) · τi j(t) + ∆τi j(t) + e · τ ei j(t), where
∆τi j(t) = ∑

m
k=1 ∆τ

k
i j(t)

∆τ ki j(t) =
{

Q/Lk(t),i f (i, j)∈T k(t)
0,otherwise

}

and

τ ei j(t) =
{

Q/L+,i f (i, j)∈T +
0,otherwise

}

}
// Calculate the intensity of the pheromone for next iteration
for (every edge i, j) {

τi j(t + 1) = τi j(t)
}

}

Figure 1: Elitist Ant System pseudocode for solving the Travelling Salesman Problem.



Ant Colony Solving Multiple Constraints Problem: Vehicle Route Allocation 369

In this paper is presented a problem that has more constraints than TSP in its basic form:

• The number of routes (destinations);

• The number of vehicles in the fleet;

• The route allocation history (the routes must be allocated in an equally distributed manner for each
month - TURNUS);

• The average distance travelled by vehicles in a month (this distance must be (optimally) equal for
all the vehicles - AVERAGE DISTANCE);

• The driver’s rest (of two days) between trips (REST).

Because of these constraints, new variables are necessary to be included in the algorithm for solving
VRAP. Moreover a different approach is necessary regarding the development of the graph used to solve
the problem. In figure 2 are depicted the graphs for solving TSP and VRAP accordingly.

Figure 2: Comparison between the graphs for solving TSP (a) and VRAP (b).

If in the case of TSP the nodes in the graph are representing the cities that the salesman must reach,
and the arches are standing for the roads (their cost being directly proportional to the distance between
cities), in the case of VRAP the nodes (seen as pairs node0 and noden) are equivalent to a certain route
(e.g. Bucharest - Paris) and the arches are representing the cost of allocating a vehicle to that route.
Another approach is that the node0 is a unique virtual node regardless of the name of the city that
corresponds to the departure station (e.g. node0 can be Bucharest, Paris, etc.).
The cost function must include the following factors:

• The vehicle history (the cost is greater if the vehicle departed from node0 and arrived at noden
more recently and smaller otherwise);

• The average distance ran by a vehicle in a month (the cost is directly proportional with the differ-
ence between the average distance and the actual distance).

• The rest (of two days) of the drivers (the cost is directly proportional with the difference between
the required resting days and the actual resting days).



370 Sorin C. Negulescu, Claudiu V. Kifor, Constantin Oprean

3 The resulted algorithm

The algorithm for solving VRAP by using the modified Elitist Ant System is presented in figure 3.

// Variables used
// Tmax - maximum numbers of iterations
// R - number of routes
// V - number of vehicles
// H - routes history matrix
// Di j - distance between cities i and j
// Pi j - pheromone intensity between nodes i and j
// a - number of ants used to solve the problem (a is also equal with the number of routes)
// Tkr(t) - the matrix tour of ant k at the iteration t
// C - the vector containing ant tours costs
// T + - vector containing the best tour
// C+ - the cost of the best tour
// α ≃ 0.7 - how much the ants takes in consideration the history
// β ≃ 0.3 - how much the ants takes in consideration the pheromone intensity
// γ ≃ 0.8 - how much the ants takes in consideration the routes distances
// ρ ≃ 0.5 - percent of the pheromone that evaporates
// ε ≃ 0.5 - the multiplication value for the pheromone intensity of the elitist ant
// p0 - minimum pheromone intensity on arches = 10

−6

// Initialize pheromone trails
for (every edge i, j) {

Pi j = p0
}
// Initialize the best tour and best tour cost
T + = the best tour found from the beginning
C+ = the length of the best tour

// Main loop
for (t = 1;t ≤ T max;t++) {

// For every vehicle (every ant colony)
for (v = 1; v ≤ V ; v++) {

// Choose the starting town for every ant
for (k = 1; k ≤ a; k++) {

Place ant k on node zero (start node)
}
// Compute a tour for every ant
for (k = 1; k ≤ a; k++) {

Build tour T r
k
(t) by applying R −1 times the following step:

Choose the next node i in witch the ant k will go
with the probability

pkvi(t) =
[Hvi(t)]

α ·[Pvi(t)]β ·[Vvi]γ
∑v∈T r

k
[Hvi(t)]α ·[Pvi(t)]β ·[Vvi]γ

, if i 6∈ T r
k
(t)

pkvi(t) = 0, if i ∈ T rk (t),
where v is the current vechicle

}



Ant Colony Solving Multiple Constraints Problem: Vehicle Route Allocation 371

// Compute the tour costs for all ants
for (k = 1; k ≤ a; k++) {

Compute the cost Ck(t) of the tour T
r

k
(t) produced by ant k

}
// Update the best tour if an improved tour is found
if (an improved tour is found) {

Update T + and C+

Print T + and C+

}
// Global update for the pheromone trails
for (every edge i, j) {

Update the pheromone trails by applying the rule:
Pi j(t) = (1− ρ) · Pi j(t) + ∆Pi j(t) + ε · Pei j(t)

}
// Calculate the intensity of the pheromone for next iteration
for (every edge i, j) {

Pi j(t + 1) = Pi j(t)
}

}
}

Figure 3: Elitist Ant System pseudocode for solving the Route Allocation Problem.

The α, β and γ parameters are controlling the relative weight of the history, pheromone intensity and
the distance in the process of choosing the next node in the itinerary. The formula for computing the
pheromone intensity between nodes i and j at the t moment is:

Pi j(t) = (1− ρ) · Pi j(t) + ∆Pi j(t) + ε · Pei j(t), where

• (1 − ρ) · Pi j(t) - the pheromone quantity that evaporates on the arch between nodes i and j at the
moment t;

• ∆Pi j(t) = ∑
a
k=1 ∆P

k
i j(t) - the total pheromone quantity that is deposited on the arch between nodes

i and j at the moment t as sum of pheromone quantity deposited by all the ants that have used this
arch;

• ∆Pki j(t) =

{

1−H
T i
k

i ,i f (i, j)∈T ik (t)
0,otherwise

}

- the pheromone quantity deposited by ant k on the arch between the

nodes i and j if that arch belongs to its itinerary (Tki(t)) at the moment t. This quantity is inversely
proportional with the history of the vehicle on the route associated to the node that points to that
arch;

• ∆Pei j(t) =

{

1−H
T e
k

i
,i f (i, j)∈T +(t)

0,otherwise

}

- the pheromone quantity deposited by the elitist ant k on the arch

between nodes i and j if that arch belongs to its itinerary (Tki(t) = T
+) at the moment t. This

quantity is inversely proportional with the history of the vehicle on the route associated to the node
that points to that arch.

This process repeats until a certain condition is met (number of iterations, time, the finding of an accept-
able solution, etc.).



372 Sorin C. Negulescu, Claudiu V. Kifor, Constantin Oprean

4 Conclusions and intentions

The self-organisation (as in the case of ant colonies) indirectly supports the quality of the allocation
service through the lack of the distortions or the deficiencies that may appear in a centralized coordina-
tion. The resulting implemented algorithm is robust (both reliable and error tolerant). The optimisation
algorithm (Elitist Ant System) proved to be perfectly adaptable to the Vehicle Route Allocation Problem.
The ant colony algorithms are showing an obvious potential as problem-solving tool, at least in the con-
text of simple MAS, and relevant results can be achieved on affordable hardware configurations.
The short-range intentions are:

• Improving the cost formula;

• Finishing the research regarding the control of pheromones and search paths;

• Speeding up the search by fine-tuning the algorithm.

The middle-range target is to apply the optimisation in a real-time manner by modifying the algorithm.

5 Acknowledgements

This research is supported by the Ceex 116 / 2006 project: eTransMobility - Agent Oriented System
for Modelling and Optimizing the Transport of Persons; coordinated by "Lucian Blaga" University of
Sibiu.

Bibliography

[1] Bărbat B.E, Moiceanu A., Pleşca S., Negulescu S.C. (2007).Affordability and Paradigms in Agent-
Based Systems. Computer Sc. J. of Moldova, 15, 2(44), pp.178-201.

[2] Bărbat B.E., Negulescu S.C. (2006). From Algorithms to (Sub-)Symbolic Inferences in Multi-
Agent Systems. International Journal of Computers, Communications and Control, 1, 3, pp.5-12.

[3] Bărbat B.E., Negulescu S.C., Zamfirescu C.B. (2005). Human-Driven Stigmergic Control. Moving
the Threshold. In N. Simonov (Ed.), Proc. of the 17th IMACS World Congress (Scientific Compu-
tation, Applied Mathematics and Simulation), pp.86-92. Paris: ISBN 2- 915913-02-01.

[4] Bonabeau E., Dorigo M., Theraulaz G. (1999). Swarm Intelligence: From Natural to Artificial
Systems. New York: Oxford University Press.

[5] Dorigo M., Maniezzo V., Colorni A. (1996). The Ant System: Optimisation by a Colony of Coop-
erating Agents. IEEE Transactions on Systems, Man, and Cybernetics, Part B, 26 (1), pp.29-42.

[6] Gambardella L.M., Di Caro G. (2005). The Ant Colony Optimization (ACO) Metaheuris-
tic: a Swarm Intelligence Framework for Complex Optimization Tasks. Retrived 2008,
from University of Bologna: First Summer School on Aspects of Complexity. Web site:
http://www.cs.unibo.it/˜fioretti/AC/AC2005/docs/slides_dicaro.pdf

[7] Gambardella L.M., Dorigo M. (1995). Ant-Q: A Reinforcement Learning Approach to the Travel-
ling Salesman Problem. In A. Prieditis and S. Russell (Ed.), Proceedings of the Eleventh Interna-
tional Conference on Machine Learning (pp.252-260). San Francisco, CA: Morgan Kaufmann.



Ant Colony Solving Multiple Constraints Problem: Vehicle Route Allocation 373

[8] Negulescu S.C., Bărbat B.E. (2004). Enhancing the effectiveness of simple multi-agent systems
through stigmergic coordination. In ICSC-NAISO (Ed.), Fourth International ICSC Symposium
on ENGINEERING OF INTELLIGENT SYSTEMS (EIS 2004), pp.149-156. Canada: ICSC-NAISO
Academic Press.

[9] Negulescu S.C., Zamfirescu C.B., Bărbat B.E. (2006). User-Driven Heuristics for nondeterminis-
tic problems. Studies in Informatics and Control (Special issue dedicated to the 2nd Romanian-
Hungarian Joint Symp. on Applied Computational Intelligence), 15, 3, pp.289-296.

[10] Stützle T., Hoos H.H. (2000). MAX-MIN Ant System. Future Generation Computer Systems, 16
(8), pp.889-914.

Sorin C. Negulescu
"Lucian Blaga" University of Sibiu

Faculty of Engineering
4, Emil CIORAN, IM 502

550025 Sibiu, Romania
E-mail: sorin_negulescu@yahoo.com

Claudiu V. Kifor
"Lucian Blaga" University of Sibiu

Faculty of Engineering
4, Emil CIORAN, IM 502

550025 Sibiu, Romania
E-mail: claudiu.kifor@ulbsibiu.ro

Constantin Oprean
"Lucian Blaga" University of Sibiu

Faculty of Engineering
4, Emil CIORAN, IM 502

550025 Sibiu, Romania
E-mail: constantin.oprean@ulbsibiu.ro