See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/281432201

Ant Colony Optimization: A Tutorial Review

Conference Paper · August 2015

CITATIONS READS
15 12,670

3 authors:

Sapna Katiyar Ibraheem Nasiruddin

ABES Institute of Technology Qassim University
11 PUBLICATIONS   64 CITATIONS    235 PUBLICATIONS   1,351 CITATIONS   

SEE PROFILE SEE PROFILE

Abdul Quaiyum Ansari

Jamia Millia Islamia
215 PUBLICATIONS   1,303 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Diagnostics & Therapeutics View project

Yes, I am working on SOGI based algorithm to mitigate the power quality View project

All content following this page was uploaded by Sapna Katiyar on 06 October 2015.

The user has requested enhancement of the downloaded file.

 Ant Colony Optimization: A Tutorial Review

Sapna Katiyar, Ibraheem, Abdul Quaiyum Ansari*

Department of Electrical Engineering

Jamia Millia Islamia, New Delhi- 110 025
Email IDs: [email protected], [email protected]
*
Corresponding Author: (Email: [email protected])

Abstract: The complex social behaviors of ants on the ground forms a pheromone trail, which allows
have been much studied, and now scientists are the ants to find good sources of food that have been
finding that these behavior patterns can provide previously identified by other ants. Ant behavior is
models for solving difficult combinatorial shown in Fig. 1. Using random walks and
optimization problems. The attempt to develop pheromones within a ground containing one nest and
algorithms inspired by one aspect of ant behavior, one food source, the ants will leave the nest, find the
the ability to find shortest paths, has become the food and come back to the nest. After some time, the
field of ant colony optimization (ACO). Ant way being used by the ants will converge to the
Colony Optimization (ACO) is a derivative of shortest path.
Swarm intelligence (SI). The ant colony
optimization algorithm (ACO), introduced by
Marco Dorigo, in the year 1992 and it is a
paradigm for designing meta heuristic algorithms
for optimization problems and is inspired by the
foraging behavior of ant colonies. Ant Colony
Optimization targets discrete optimization
problems and can be extended to continuous
optimization problems which is useful to find
approximate solutions. Now-a-days, a number of
algorithms inspired by the foraging behavior of
ant colonies have been applied to solve difficult
discrete optimization problems. In fact, ACO Fig.1. Ant Behavior
algorithm is the most successful and widely
recognized algorithmic based on the ant behavior. A. Ants in a pheromone trail between nest and food;
This paper gives an overview of growing research B. An obstacle interrupts the trail; C. Ants find two
field from theoretical inception to practical paths to go around the obstacle; D. A new pheromone
applications, ACO variants and some of the fields trail is formed along the shorter path.
where it can be applied.
B. Ant colony optimization: A metaheuristic
Key Words: Hybridization, Metaheuristic, The term metaheuristic is a combination of two
Parameters optimization, Pseudo-Random- Greek words. Heuristic derives from the verb
Proportional Action Choice Rule, Pheromone heuriskein which means “to find”, while the suffix
Meta means “beyond, in an upper level”. The new
I. INTRODUCTION heuristic has the following desirable characteristics:
• Versatile: It can be applied to similar versions of
A. Ant Behavior the same problem; for example, there is a
Ants communicate to one another by laying down straightforward extension from the traveling
pheromones along their trails, so where ants go salesman problem (TSP) to the asymmetric traveling
within and around their ant colony is a stigmergic salesman problem (ATSP).
system. In many ant species, ants walking from or to • Robust: It can be applied with only minimal
a food source, deposit on the ground a substance changes to other combinatorial optimization
called pheromone. Other ants are able to smell this problems such as the quadratic assignment problem
pheromone, and its presence influences the choice of (QAP) and the job-shop scheduling problem (JSP).
their path, that is, they tend to follow strong • Population based approach: This is interesting
pheromone concentrations. The pheromone deposited because it allows the exploitation of positive
feedback as a search mechanism, as explained later in greater concentration of pheromone. The algorithm is
the paper. It also makes the system amenable to based on adaptively adjusting the pheromone on
parallel implementations. routes at each node and is shown in Fig. 3. Choice of
this node is guided by a probability based selection
Ant system is the first member of ACO class of approach.
algorithms. This algorithm is inspired by the trail
laying and following behavior of natural ants. The
essential trait of ACO algorithms is the combination
of a priori information about the structure of a
promising solution with posterior information about
the structure of previously obtained good solutions.
The main underlying idea, loosely inspired by the
behavior of real ants, is that of a parallel search over
several constructive computational threads based on
local problem data and on a dynamic memory
structure containing information on the quality of
previously obtained result.

C. The Double Bridge Experiment

Deneubourg et al. verified the pheromones marking
of ants by the experience known as “double bridge
experiment”. This configuration is as shown in Fig. 2.
In the double bridge experiment, a nest of a colony of
Argentine ants is connected to a food source by two
bridges. The ants can reach the food source and
return to the nest using any of the two bridges. The
aim of the experiment is to observe the resulting
behavior of the colony.

It is observed that if the two bridges have the same

length, the ants tend to converge towards the use of
one of the two bridges. If this experiment is repeated
a number of times, it is observed that each of the two
bridges is used in about half of the cases. The fact is
while moving, ants deposit pheromone on the ground
and whenever they have to decide which path should
follow, their choice is based higher the pheromone
concentration found on a particular path, the higher is
the probability to follow that path.

Fig.2. Experimental Setup for Double Bridge

Experiment: A. Branches have equal length. B.
Branches have different length.

II. ANT COLONY OPTIMIZATION

ALGORITHM
When ants move from point A (source) to point B Fig. 3. ACO Algorithm
(destination), ants leave a chemical, pheromone, to
mark these paths. This helps the following ants to The ants are driven by a probability rule to choose
find the way of their team members as they detect their solution to the problem, known as a tour. The
pheromone and choose, in probability, paths having probability rule between two nodes j, called Pseudo-
Random-Proportional Action Choice Rule, and it
depends on two factors: the heuristic and In ant-density and ant-quantity ants deposit
metaheuristic. pheromone during building a solution, while in ant-
cycle ants deposit pheromone after the completion of
[ ij ] [ ij ] ----------------- (2.1) a complete tour.
pij   
 [
hs
ij ] [ ij ] m
k
 ij (t )  (1   ). ij (t  1)    ij ---------- (3.1)
Where τ is the pheromone, η is the inverse of k 1

the distance between the two nodes. Each ant Where ρ is the evaporation rate, m is the number of
modifies the environment in two different ways: k

ants and  ij is the quantity of pheromone laid on



i) Local trail updating: As the ant moves between edge (i, j) by ant k: τij (t) = Q / Lk, if ant k used edge
nodes it updates the amount of pheromone on the (i, j) in its tour otherwise it is zero. Where, Q is a
edge by the following equation: constant and Lk is the length of the tour constructed
by ant k.
 ij (t )  (1   ). ij (t  1)   . 0 ---------- (2.2)
B. MAX–MIN Ant System
Where, ρ is the evaporation constant. The value τ0 is MAX–MIN Ant System (MMAS) is another
the initial value of pheromone trails and can be improvement over the original ant system and it is
calculated as: proposed by Stützle and Hoos (2000). It introduces
τ0 = (n / Ln)-1 four main modifications with respect to ant system.
 Only the best ant adds pheromone trails i.e.
Where, n is the number of nodes and Ln the total either the ant that produced the best tour in
distance covered between the total nodes, produced the current iteration, or the best-so-far ant is
allowed to deposit pheromone. But this may
by one of the construction heuristics. lead to a stagnation situation.
 Above defined situation can be modified by
ii) Global trail updating: When all ants have MMAS i.e. the minimum and maximum
completed all the nodes that find the shortest path values of the pheromone are explicitly
updates the edges in its path using the following limited
equation:  The pheromone trails are initialized to the
 upper pheromone trail limit, which, together
 ij (t )  (1   ). ij (t  1)  ---------- (2.3) with a small pheromone evaporation rate
L
which increases the exploration of tours at
Where, L+ is the length of the best path generated by the start of the search.
one of the ants.  Finally, in MMAS, pheromone trails are
reinitialized each time the system
III. VARIANTS OF ACO ALGORITHM approaches stagnation or when no improved
tour has been generated for a certain number
In literature number of ACO algorithms have been of consecutive iterations.
proposed. The original ACO algorithm is known as When all ants have constructed a tour, pheromones
Ant System and was proposed in the early nineties. are updated by applying evaporation same as in the
Here we demonstrate the original Ant System (AS) case of ant system (AS), followed by the deposit of
and two other most successful variants: MAX-MIN new pheromone given by:
Ant System and Ant Colony System (ACS).
best
A. Ant System (AS)
 ij (t )   ij   ij -------------- (3.2)
Ant System (AS) was the first (1991) ACO algorithm best

Where  ij = 1/ Cbest, if the best ant used edge (i, j)


proposed in literature. Its importance is mainly
best
because of the prototype of a number of ant
in its tour.  ij = 0 otherwise, where Lbest is the

algorithms which have found many interesting and length of the tour of the best ant. The value of Lbest
successful applications in real time situations. Three
may be set either to Lib or to Lbs, or to a combination
AS algorithms have been defined and they differ by of both.
the way pheromone trails are updated. These
C. Ant Colony System (ACS)
algorithms are:
The difference between ACS and ant system is
 Ant-density because of three main aspects:
 Ant-quantity  The state transition rule provides a direct
 Ant-cycle way to balance between exploration of new
 edges and exploitation of a priori and  It is very easy to state and explain, so that
accumulated knowledge about the problem. the algorithm behavior is not obscured
 The global updating rule is applied only to by too many technicalities.
edges which belong to the best ant tour
 While ants construct a solution a local B. QUADRATIC ASSIGNMENT PROBLEN
pheromone updating rule is applied. (QAP)
The quadratic assignment problems (QAPs) are the
The most interesting contribution of ACS is the problem of assigning 'n' facilities to 'n' locations so
introduction of a local pheromone update in addition that the assignment cost is minimized, where the cost
to the pheromone update performed at the end of the is defined by a quadratic function. A concrete
construction process (called offline pheromone example would be a hospital planning optimization
update). The local pheromone update is performed by the location of various medical facilities in order to
all the ants after each construction step. minimize the overall travel distance of patients inside
the hospital.
IV. APPLICATIONS OF ANT COLONY
OPTIMIZATION C. VEHICLE ROUTING PROBLEM (VRP)
A direct extension of the TSP is Vehicle routing
ACO was first introduced for the application of problem (VRP). These are problems where a set of
travelling salesman problem (TSP). Since then, it has vehicles stationed at a depot has to serve a set of
been applied to many discrete optimization problems. customers before returning to the depot, and the
Initially it is applied to classical problems like objective is to minimize the number of vehicles used
assignment problems, graph coloring, the maximum and the total distance traveled by the vehicles.
clique problem, scheduling problems and vehicle Capacity constraints are imposed on vehicle trips, as
routing problems. Some of the examples of more well as possibly a number of other constraints
recent applications include the intelligent scheduling, deriving from real-world applications, such as time
design of communication networks, bioinformatics windows, backhauling, rear loading, vehicle
problems, cell placement problems arising in circuit objections, maximum tour length, etc.
design and machine learning. In recent years some
researchers have also focused on the application of D. GRAPH COLORING PROBLEM (GCP)
ACO algorithms to multi-objective problems and to One of the most well-studied combinatorial
dynamic or stochastic problems. Table 1 shows some optimization problems is the graph coloring problem.
successful applications of ACO. The problem is to find a coloring of the vertices with
minimum number of colors such that no pair of
A. TRAVELLING SALESMAN PROBLEM adjacent vertices has the same color. Graph coloring
(TSP) problem is expected to have a wide variety of
In the traveling salesman problem, a set of cities is applications such as:
given and the distance between each of them is  Scheduling,
known. The goal is to find the shortest tour that  Frequency assignment in cellular networks,
allows each city to be visited once and only once. In  Timetabling,
more formal terms, the goal is to find a Hamiltonian  Crew assignment
tour of minimal length on a fully connected graph.
In ant colony optimization, the problem is tackled by E. SEQUENTIAL ORDERING PROBLEM(SOP)
simulating a number of artificial ants moving on a The sequential ordering problem is closely related to
graph that encodes the problem itself: each vertex the asymmetric TSP, but additional precedence
represents a city and each edge represents a constraints between the nodes have to be satisfied.
connection between two cities. A variable called Gambardella and Dorigo have tackled this problem
pheromone is associated with each edge and can be by an extension of ACS enhanced by a local search
read and modified by ants. The traveling salesman algorithm (Gambardella & Dorigo, 1997). They
problem plays a central role in ant colony obtained excellent results and were able to improve
optimization because it was the first problem to be the best known solutions for many benchmark
attacked by ACO. instances.
The TSP was chosen because:
 It is relatively easy to adapt the ant colony F. JOB SCHEDULING PROBLEM (JSP)
metaphor to it. The job-shop scheduling problem is concerned with
 It is a very difficult problem (NP-hard). allocating limited resources to operations over time.
 It is one of the most studied problems in Although the job shop scheduling has an important
combinatorial optimization. role in various fields, it is one of the most difficult
problems in combinational optimization. The demand There are several advantages and disadvantages of
for scheduling is to achieve high performance ant colony optimization. Some of them are listed in
computing. Typically, it is difficult to find an optimal Table 2:
resource allocation for specific job that minimizes the TABLE-2
schedule length of jobs. The scheduling problem is S. Advantages Disadvantages
defined NP-hard problem and it is not trivial. No
1 Inherent parallelism Theoretical analysis is
G. ROUTING IN TELECOMMUNICATION difficult
NETWORK: 2 There is no central Sequences of random
ACO algorithms have shown to be a very effective control in the colony decisions (not
approach for routing problems in telecommunication independent)
networks where the properties of the system, such as
the cost of using links or the availability of nodes, 3 Positive Feedback Probability distribution
varies over time. ACO algorithms were first applied accounts for rapid changes by iteration
to routing problems in circuit switched networks (e.g. discovery of good
telephone networks) and then in packet-switched solutions
networks (e.g. LAN or the Internet). A well-known 4 Efficient for Traveling Time to convergence
example for network communication improved to the Salesman Problem and uncertain (but
point of being state-of-the-art in wired networks is similar problems convergence is
Ant-Net. Ant-Net has been extensively tested, in guaranteed)
simulation, on different networks and under different
traffic patterns, proving to be highly adaptive and 5 Can be used in dynamic Research is experimental
robust. applications (adapts rather than theoretical
to changes such as new
TABLE-1: A non-exhaustive list of successful ant distances, etc)
colony optimization algorithms
VI. CURRENT TRENDS IN ACO
S. Problem type Problem name
No It is a true metaheuristic, with dozens of application
1 Routing Traveling salesman (TSP) areas. While both the performance of ACO
Vehicle routing algorithms and our theoretical understanding of their
Sequential ordering working have significantly increased, as discussed
Connection-oriented earlier, there are several areas in which until now
network routing only preliminary steps have been taken and where
Connectionless network much more research will have to be done.
routing
Optical network routing One of these research areas is the extension of ACO
2 Assignment Quadratic assignment algorithms to more complex optimization problems
Course timetabling that include:
Graph coloring  Dynamic problems: in which the instance
Frequency Assignment data, such as objective function values,
3 Scheduling Project scheduling decision parameters, or constraints, may
Car sequencing change while solving the problem.
4 Subset Set covering  Stochastic problems: in which one has only
Multiple knapsack probabilistic information about objective
Maximum clique function value(s), decision variable values,
5 Machine Classification rules or constraint boundaries, due to uncertainty,
learning Bayesian networks noise, approximation, or other factors.
Neural networks  Multiple objective problems: in which a
Fuzzy Systems multiple objective function evaluates
6 Bioinformatics Protein folding competing criteria of solution quality.
DNA Sequencing
Active research directions in ACO include also the
effective parallelization of ACO algorithms and, on a
V. ADVANTAGES AND DISADVANTAGES OF
more theoretical level, the understanding and
ACO
characterization of the behavior of ACO algorithms
while applying it to any proposed system.
 [5] Maniezzo V, Colorni A. “The Ant System applied
VII. CONCLUSION to the quadratic assignment problem”, IEEE Trans.
Data Knowledge Engineering, 11(5), 769–78, 1999.
Ant Colony Optimization has been and continues to [6] M. Dorigo, G. Di Caro, and L. M. Gambardella,
be a fruitful paradigm for designing effective “Ant Algorithms for Discrete Optimization,”
combinatorial optimization solution algorithms. In Artificial Life, Vol. 5, No. 2, pp. 137–172, 1999.
this work we first gave a detailed description of the
origins and the basics of ACO algorithms. Then we [7] M. Dorigo, T. Stutzle, “Ant Colony
outlined the general framework of the ACO Optimization,” MIT Press, Cambridge, MA, 2004.
metaheuristic and presented some of the most [8] A. Q. Ansari, "The Basics of Fuzzy logic: A
successful ACO variants today. Generally majority of Tutorial Review," Computer Education – Stafford –
problems based on ACO are static and well defined Computer Education Group, U.K., No. 88, pp. 5 - 9,
combinatorial optimization problems where all February 1998.
necessary information related to the problem is [9] A. Q. Ansari, "Multiple Valued Logic Versus
available and does not change during problem Binary Logic," CSI Communications, India, Vol. 20,
solution. It has been also identified that ACO No. 5, pp. 30 - 31, November 1996.
algorithms can be applied in a systematic manner to
ill-structured problems for highly dynamic domains [10] A. Q. Ansari, Moinuddin, S. Deshpande, "Fuzzy
with only local information available or to such type Logic Control of Safety and Security Systems," The
of problems where it is not clear how to apply local Indian Police Journal, Vol. 45, No. 4, pp.56 – 62,
search. After listing some applications area of ACO, October–December 1998.
it has been found that the field of ant colony [11] S. A. Siddiqui, A. Q Ansari, Shikha Agarwal, "A
optimization algorithms is very dynamic in both Journey through Fuzzy Philosophy," Pranjana: The
theoretically and experimentally. Experimentally the Journal of Management Awareness, Vol. 6, No. 2,
direction of current research is to increase the number pp. 29 – 33, July – Dec. 2003.
of successfully solved problems based on ACO
[12] V. Kumar, A. Q. Ansari, T. Patki, A.B. Patki, S.
algorithm. It can be implemented in real-word Joshi, S. Chowdhary, "Applications of Evidence
industrial applications also. Based Software Engineering for I T Systems,"
Review of Business and Technology Research, Vol. 1,
Finally, a survey on very interesting recent trends in
No. 1, pp. 1 – 7, 2008.
ACO is discussed. The hybridization of ACO
algorithms with more classical artificial intelligence [13] Abdul Quaiyum Ansari, “Hierarchical Fuzzy
and operations research methods can be done. In the Control for Industrial Automation,” Scholar's Press.,
opinion of the author, this research direction offers Germany, 07/2013; ISBN: 978-3639515923.
many possibilities for valuable future research. [14] Mohammad Ayoub Khan, Abdul Quaiyum
Ansari, “Handbook of Research on Industrial
REFERENCES Informatics and Manufacturing Intelligence:
Innovations and Solutions,” Eds. Mohammad Ayoub
[1] M. Dorigo, “Optimization, Learning and Natural Khan, Abdul Quaiyum Ansari, IGI Global, USA,
Algorithms,” PhD thesis, Dipartimento di Elettronica, 2012. ISBN13: 9781466602946.
Politecnico di Milano, Italy, 1992.
[15] Abdul Quaiyum Ansari, Mohammad Ayoub
[2] M. Dorigo, L. M. Gambardella, “Ant Colony Khan, “Fundamentals of Industrial Informatics and
System: A Cooperative Learning Approach to the Communication Technologies,” in Handbook of
Traveling Salesman Problem,” IEEE Transactions on Research on Industrial Informatics and
Evolutionary Computation, 1(1), 53-66, 1997. Manufacturing Intelligence: Innovations and
[3] M. Dorigo, V. Maniezzo, and A. Colorni, “Ant Solutions, Chapter - 1, pages 1-19; IGI Global, USA,
System: Optimization by a Colony of Cooperating 2012.
Agents,” IEEE Transactions on Systems, Man, and [16] Sapna Katiyar, Avneesh Mittal, A. Q. Ansari, T.
Cybernetics – Part B, Vol. 26, No. 1, pp. 29–41, K. Saxena, "Ant Colony Algorithm Based Adaptive
1996. PID Temperature Controller," Proc. 7th Int. Conf. on
[4] M. Dorigo and G. Di Caro, “The Ant Colony Trends in Industrial Measurements and Automation
Optimization Meta-heuristic”, in New Ideas in (TIMA 2011), CSIR, Chennai, January 2011.
Optimization, D. Corne et al., Eds., McGraw Hill,
London, UK, pp. 11-32, 1999.

View publication stats