You are on page 1of 9

Ant Colony Optimization - A Novel Approach to Solve NP-Hard Problems

Many of numerous species of insects that


live in colonies manifest group integration
and division of labour. Swarm intelligence
are typically made up of a population of
simple agents interacting locally with one
Keywords: Ant colony, Swarm, TSP another and with their environment.
Abstract Although there is normally no centralized
control structure dictating how individual
The complex social behaviours of ants have agents should behave, local interactions
been much studied, and computer scientists between such agents often lead to the
are now finding that these behaviour patterns emergence of global behaviour.
can provide models for solving difficult The rest of the paper is divided as follows:
combinatorial optimization problems. The Sections 1.2 and 1.3 discuss the Ant Colony
attempt to develop algorithms inspired by and other Social Insect Behaviour
one aspect of ant behavior, the ability to find approaches. Section 2 gives a theoretical
shortest paths, has become the field of ant overview and an example of ACO. Variants
colony optimization (ACO). The ant colony of ACO are briefly explained in Section 3.
optimization algorithm (ACO), introduced by In Section 4, the application aspects of
Marco Dorigo, in the year 1992, is a ACO to real world problems is discussed
probabilistic technique for solving with emphasis on the TSP. The next section
computational problems which can be gives the merits and demerits of ACO.
reduced to finding good paths through Finally the conclusion and references are
graphs. This paper presents an overview of given.
this rapidly growing field, from its theoretical
inception to practical applications, including 1.2 Ant Colonies
descriptions of many available ACO
algorithms and their various applications to Ant Communication is accomplished
solve NP-Hard Problems. In this paper, we primarily through chemicals called
also discuss the Travelling Salesman pheromones. Ants communicate to one
Problem. another by laying down pheromones along
their trails. Other ants perceive the presence
1. Introduction. of pheromone and tend to follow paths
1.1 Behaviour of Social Animals where pheromone concentration is higher.
Over time, however, the pheromone trail
Insects are the most diverse group of starts to evaporate, thus reducing its
animals on the earth. Many insects possess attractive strength. The more time it takes
very sensitive organs of perception. for an ant to travel down the path and back
perception. Swarm Intelligence(SI) is an again, the more time the pheromones have
artificial technique based on the study of to evaporate. A short path gets marched
collective, self-organized systems. SI systems over faster, and thus the pheromone density
remains high as it is laid on the path as fast
as it can evaporate.
The distinctive behaviour of ants has been
extensively studied and has inspired a
number of methods and techniques among
which the most successful is the general
purpose optimization technique known as
Ant Colony Optimization (ACO). ACO
exploits a similar mechanism similar to that
of the foraging behaviour of some ant
species. From the early nineties, when the
first ant colony optimization algorithm was
Figure 1:Ant Behaviour first proposed, ACO attracted the attention
A. Ants in a pheromone trail between nest of increasing numbers of researchers and
and food; B. An obstacle interrupts the trail; many successful applications are now
C. Ants find two paths to go around the available. Moreover, a substantial corpus of
obstacle; D. A new pheromone trail is formed theoretical results is becoming available
along the shorter path. that provides useful guidelines to
researchers and practitioners in further
Pheromone evaporation has also the applications of ACO
advantage of avoiding the convergence to a
locally optimal solution. If there were no 2.2 Characteristics of ACO
evaporation at all, the paths chosen by the
first ants would tend to be excessively Stigmergy is a method of communication
attractive to the following ones. In that case, in systems in which the individual parts of
the exploration of the solution space would the system communicate with one another
be constrained. Thus, when one ant finds a by modifying their local environment. The
good path to a food source, other ants are two main characteristics of stigmergy that
more likely to follow that path, and positive differentiate it from other forms of
feedback eventually leaves all the ants communication are:
following a single path. Figure 1 describes (i) Stigmergy is an indirect, non-
this ant behaviour. symbolic form of communication
mediated by the environment:
1.3 Other Social Insects for insects exchange information by
Optimization modifying the environment; and
(ii) Stigmergic information is local: it
Examples of other optimisation systems can only be accessed by those
which can be found in nature include bird insects that visit the locus in which
flocking, animal herding, bacteria moulding it was released (or its immediate
and fish schooling, pedestrians, traffic. The neighbourhood)
beauty of the entire phenomenon lies in the
seemingly intelligent group behaviour that 2.3 The Double Bridge Experiment
emerges from multiple individual animals.
Deneubourg et al. thoroughly investigated
2. Ant Colony Optimization. the behaviour of ants. In an experiment
known as the “double bridge
2.1 The Theory of Ant Colonies experiment”( Figure 2) the nest of a
colony of Argentine ants was connected to
a food source by two bridges of equal
lengths. In such a setting, ants start to plays an important role: the ants choosing
explore the surroundings of the nest and by chance the short bridge are the first to
eventually reach the food source. Along their reach the nest. The short bridge, therefore,
paths, the ants deposit pheromone. Initially, receives pheromone earlier than the long
each ant randomly chooses one of the two one and this fact increases the probability
bridges. However, due to random that further ants select it rather than the
fluctuations, after some time, one of the two long one. A model of this observed
bridges presents a higher concentration of behaviour was developed.
pheromones than the other and , thus, attracts
more ants. This brings a further amount of 2.4 ACO Procedure - Pseudo Algorithm
pheromone on that bridge making it more
attractive with the result that after some time The ACO pseudo-algorithm is shown
the whole colony converges towards the use below. After initialization, the algorithm
of the same bridge. iterates over three phases: at each iteration,
This colony–level behaviour based on a number of solutions are constructed by
autocatalysis , that is , on the exploitation of the ants; these solutions are then improved
positive feedback , can be used by ants to through a local search (this step is
find the shortest path between a food source optional), and finally the pheromone is
and their nests. A variant of the “double updated.
bridge experiment”, in which one bridge is
significantly longer than the other was 1. (Initialization)
considered. In this case, the stochastic Set parameters, initialize pheromone
fluctuations in the initial choice of a bridge trails.
are much reduced and a second mechanism. Initialize τιψ, for each ι,ψ
2. (Construction of AntSolution)
For each ant k do
Repeat
Compute ηιψ for each ι,ψ.
Choose in probability the state to
move into
Append the chosen move to move
the kth ant’s set tabu.
Until ant k has completed its
solution.
[apply a local search(optional)]
3. (Pheromone Trail Update)
For each ant move ( ι,ψ) do
Compute ∆τιψ and update the trail
values
4. (Terminating condition)
If not (end_condition) go to step-2.
Figure 2: Experimental Setup for the
double bridge experiment (a) branches 3. Variants of the ACO Algorithm.
have equal lengths;(b) branches have
different lengths.
Different ant colony optimization algorithms algorithms can be useful for quickly
have been proposed .The original ACO finding high quality solution. Other
algorithm is known as Ant System and was popular applications are to dynamic
proposed in the early nineties. It’s main shortest path problems arising in
characteristic is that at each iteration, the telecommunication networks problems.
pheromone values are updated by all the ants The number of successful applications to
that have built a solution in the iteration academic problems has motivated people
itself. to adopt ACO for the solution of industrial
Since then a number of other ACO problems, proving that this computation of
algorithms, variants of the original Ant intelligence technique is also useful in real
System, were presented, of which the two world applications.
most successful algorithms were MAXMIN
and Ant Colony system (ACS). 4.1 Application to telecommunication
MAXMIN is an improvement over the networks
original AntSystem. Its characterizing
elements are that only the best ant updates ACO algorithms have shown to be a very
the pheromone trails and that the value of the effective approach for routing problems in
pheromone is bound. telecommunication networks where the
The most interesting contribution of ACS is properties of the system, such as the cost of
the introduction of a local pheromone update using links or the availability of the nodes,
in addition to the pheromone update vary over time. ACO algorithms were first
performed at the end of the construction applied to routing problems in circuit-
process. The local update done diversifies switched networks (such as telephone
the search performed by subsequent ants networks) and then in packet-switched
during an iteration: by decreasing the networks (such as local area networks or
pheromone concentration on the traversed the internet). A well known example is
edges, ants encourage subsequent ants to AntNet which has been extensively tested,
choose other edges an, hence, to produce in simulation, on different networks and
different solutions. This makes it less likely under different traffic patterns, proving to
that several ants produce identical solutions be highly adaptable and robust. A
during one iteration. comparison with state-of-the-art routing
algorithms has shown that, in most of the
4. Applications of ACO in NP-Hard considered situations, AntNet outperforms
Problems. its competitors.
In recent years, the interest of the scientific Ant-based algorithms have given rise to
community in ACO has risen sharply. In fact, several other routing algorithms, enhancing
several successful applications of ACO to a performance in a variety of wired network
wide range of different discrete optimization scenarios.
problems are now available. The large
majority of these applications are to NP- 4.2 Application to Industrial Problems
Hard problems; that is, to problems for
which the best known algorithms that The success of ACO on academic
guarantee to identify an optimal solution problems has raised the attention of a
have exponential time worst case number of companies that have started to
complexity. The use of such algorithms is sue ACO algorithms for real-world
often infeasible in practice, and ACO applications. They have applied ACO to a
number of scheduling problems such as a the following vertex is selected
continuous two-stage flow shop problem stochastically
with finite reservoirs. The problems modeled among the previously unvisited ones. In
included various real world constraints such particular, if j has not been previously
as setup times, capacity restrictions, resource visited, it can be selected with a probability
compatibilities and maintenance calendars. that is proportional to the pheromone
Some researchers have developed a set of associated with the edge (i,j). At the end of
tools for the solution of vehicle-routing the iteration, on the basis of the quality of
problems whose optimization algorithms are the solution constructed by the ants, the
based on ACO. Examples of companies who pheromone values are modified in order to
have successfully used ACO are EuroBios, bias ants in future iterations to construct
AntOptima, DYVOIL, AntRoute, Migros solutions similar to the best ones previously
(the main Swiss supermarket chain) and constructed. In Figure 3, ACO-TSP is
Barilla (the main Italian pasta maker). given.

4.3 Conventional ACO for the Travelling


Salesman Problem (TSP)

The traveling salesman problem (TSP) is a


problem in discrete or combinatorial
optimization. It is a prominent illustration of
a class of problems in computational
complexity theory which are hard to solve.
The problem states that : Given a number of
cities and the costs of traveling from any city
to any other city, what is the cheapest round-
trip route that visits each city exactly once
and then returns to the starting city?
In ant colony optimization, the problem is Figure 3 :An ant in city i chooses the next
tackled by simulating a number of artificial city to visit via a stochastic mechanism: If j
ants moving on a graph that encodes the has not been previously visited, it can be
problem itself: each vertex represents a city selected with a probability that is
and each edge represents a connection proportional to the pheromone associated
between two cities. A variable called with edge (i.j).
pheromone is associated with each edge and
can be read and modified by ants. Ant colony Many of the tackled NP-Hard problems can
optimization is an iterative algorithm. At each be considered as falling into one of the
iteration , a number of artificial ants are following categories: routing problems as
considered. Each of them builds a solution by they arise for example in the distribution of
walking from vertex to vertex on the graph goods; assignment problems, where a set of
with the constraint of not visiting any vertex items (objects, activities etc.) has to be
that she has already visited in her walk. At assigned to a given number of
each step of the solution construction, an ant resources(location, agents etc.)subject to
selects the following vertex to be visited some constraints; scheduling problems ,
according to a stochastic mechanism that is which –in the widest sense- concerned with
biased by the pheromone: when in vertex i, the allocation of scare resources to tasks
over time; and subset problems , where a
solution to a problem is considered to be a
selection of a subset of available items. In
addition, ACO has been successfully applied
to other problems emerging in fields such as
machine learning and bioinformatics.

5. The Merits and Demerits of ACO.

Some of the inherent advantages of ACO


algorithms are as follows :
• The Positive Feedback in ACO
accounts for rapid discovery of good
solutions
• ACO employs Distributed
computation, which avoids premature
convergence.
• The greedy heuristic used in ACO
helps find an acceptable solution in
the early solution in the early stages
of the search process.
Some of the disadvantages in ACO
algorithms are:
• These algorithms have slower
convergence than other Heuristics.
• They performed poorly for TSP
problems larger than 75 cities.
• There is no centralized processor to
guide the Ant System towards good
solutions.
SOCIAL INSECTS AND REAL ANT of the object with a 50%–50% probability
COLONIES distribution. All ants move roughly at the
In a colony of social insects, such as ants, same speed and deposit pheromone in the
bees, wasps, and termites, each insect usually trail at roughly the same rate. Therefore, the
performs its own tasks independently from ants that (by chance) go around the obstacle
other members of the colony. However, the by the shortest path will reach the original
tasks performed by different insects are track faster than the others that have
related to each other in such a way that the followed longer paths to circumvent the
colony, as a whole, is capable of solving obstacle. As a result, pheromone
complex problems through cooperation [2]. accumulates faster in the shorter path
Important survival related problems such as around the obstacle. Since ants prefer to
selecting and picking up materials, and follow trails with larger amounts of
finding and storing food, which require pheromone, eventually all the ants
sophisticated planning, are solved by insect converge to the shorter path.
colonies without any kind of supervisor or
centralized controller. This collective III. ANT COLONY OPTIMIZATION
behavior which emerges from a group of An ACO algorithm is essentially a system
social insects has been called “swarm based on agents that simulate the natural
intelligence” [2]. behavior of ants, including mechanisms of
Here, we are interested in a particular cooperation and adaptation. In [10], the use
behavior of real ants namely, the fact that of this kind of system as a new
they are capable of finding the shortest path metaheuristic was proposed in order to
between a food source and the nest (adapting solve combinatorial optimization problems.
to changes in the environment) without the This new metaheuristic has been shown to
use of visual information [9]. This intriguing be both robust and versatile—in the sense
ability of almost-blind ants has been studied that it has been applied successfully to a
extensively by ethologists. They discovered range of different combinatorial
that, in order to exchange information about optimization problems [11].
which path should be followed, ants
communicate with one another by means of ACO algorithms are based on the
pheromone (a chemical substance) trails. As following ideas.
ants move, a certain amount of pheromone is 1) Each path followed by an ant is
dropped on the ground, marking the path associated with a candidate solution for a
with a trail of this substance. The more ants given problem.
follow a given trail, the more attractive this 2) When an ant follows a path, the amount
trail becomes to be followed by other ants. of pheromone deposited on that path is
This process can be described as a loop of proportional to the quality of the
positive feedback, in which the probability corresponding candidate solution for the
that an ant chooses a path is proportional to target problem.
the number of ants that have already passed 3) When an ant has to choose between two
by that path [9], [11], [23]. or more paths, the path(s) with a larger
When an established path between a food amount of pheromone have a greater
source and the ants’ nest is disturbed by the probability of being chosen by the ant. As a
presence of an object, ants soon try to go result, the ants eventually converge to a
around the obstacle. First, each ant can short path, hopefully the optimum or a
choose to go around to the left or to the right near-optimum solution for the target
problem, as explained before for the case of
natural ants. In essence, the design of an
ACO algorithm involves the specification of
[2]:
1) an appropriate representation of the
problem, which allows the ants to
incrementally construct/modify solutions
through the use of a probabilistic transition
rule, based on the amount of pheromone in
the trail and on a local problem-dependent
heuristic;
2) a method to enforce the construction of
valid solutions, i.e., solutions that are legal in
the real-world situation corresponding to the
problem definition;
3) a problem-dependent heuristic function (
) that measures the quality of items that can
be added to the current partial solution;
4) a rule for pheromone updating, which
specifies how to modify the pheromone trail (
);
5) a probabilistic transition rule based on the
value of the heuristic function () and on the
contents of the pheromone trail () that is used
to iteratively construct a solution.
Artificial ants have several characteristics
similar to real ants, namely:
1) artificial ants have a probabilistic
preference for paths with a larger amount of
pheromone;
2) shorter paths tend to have larger rates of
growth in their amount of pheromone;
3) the ants use an indirect communication
system based on
the amount of pheromone deposited on each
path.
[3] Official website of the Ant Colony
6. Conclusion Metaheuristic: www.aco-metaheuristic.org

Ant Colony Optimization has been and [4] Other Related links:
continues to be a fruitful paradigm for www.en.wikipedia.org/wiki/Swarm_intellig
designing effective combinatorial ence
optimization solution algorithms. After more
than ten years of studies, both its application
effectiveness and its theoretical groundings
have been demonstrated, making ACO one of
the most successful paradigm in the
metaheuristic area.
Fifteen years ago, when the first ACO
algorithm was introduced, taking inspiration
from ants for designing optimization
algorithms seemed a crazy idea. The many

successful applications have changed our


perspective: what seemed a far-out idea is
now considered one of the most promising
approaches to the approximate solution of
difficult optimization problems.
This paper is a brief write-up of the ideas and
inspirations behind ACO. The effectiveness
of ACO is justified by the variety of
applications of ACO. This being an emerging
field of the Swarm Intelligence area, much
more exploration and experimentation can be
done in future.

7. References

[1] Marco Dorigo, Mauro Birattari, and


Thomas Stiitzli, “Ant Colony Optimization:
Artificial Ants as a Computational
Intelligence Technique”, IEEE Transaction
on Evolutionary Computation, November
2006, pp. 28-36.

[2] J.-L.Deneubourg, S.Aron, S.Goss, and J.-


M.Pasteels, “The Self Organizing
Exploratory Pattern of the Argentine Ant”,
Journal of Insect Behaviour, Vol.3, p.159,
1990.

You might also like