Scribd
Upload
38 views

Ant Colony Optimization: Nuno Abreu Muhammad Ajmal Zafeiris Kokkinogenis Behdad Bozorg

The document proposes an ant colony optimization approach called AntPacking for solving the one-dimensional bin packing problem. AntPacking uses a colony of ants that cooperatively construct and improve solutions by placing items into bins. The ants use a pheromone-based memory to probabilistically select which items to pack. Computational tests show AntPacking finds optimal solutions for benchmark problems as quickly as or faster than existing algorithms.

Uploaded by

Arshdeep
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
38 views

Ant Colony Optimization: Nuno Abreu Muhammad Ajmal Zafeiris Kokkinogenis Behdad Bozorg

The document proposes an ant colony optimization approach called AntPacking for solving the one-dimensional bin packing problem. AntPacking uses a colony of ants that cooperatively construct and improve solutions by placing items into bins. The ants use a pheromone-based memory to probabilistically select which items to pack. Computational tests show AntPacking finds optimal solutions for benchmark problems as quickly as or faster than existing algorithms.

Uploaded by

Arshdeep
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 26

Ant Colony Optimization

Nuno Abreu
Muhammad Ajmal
Zafeiris Kokkinogenis
Behdad Bozorg

FEUP-DEEC
2011-01-17
Outline

Swarm Intelligence
Introduction to Ant Colony Optimization (ACO)
Ant Behaviour
Stigmergy
Pheromones
Basic Algorithm
Example
Advantages and Disadvantages
References

2
Swarm Intelligence

Artificial intelligence technique based on the study of collective


behavior in decentralized, self-organized systems

Introduced by Beni & Wang in 1989

Collective system capable of performing complex tasks in


a dynamic environment

Model suited to distributed problem solving

Works without:
External guidance
Central coordination

Typically made up of a population of simple agents

3
Swarm Intelligence

4
Ant Colony Optimization

Proposed by Marco Dorigo in 1991

Inspired in the behavior of real ants

Multi-agent approach for solving complex combinatorial


optimization problems

Applications:

Traveling Salesman Problem


Scheduling
Network Model Problem
Vehicle routing

5
Ant Behavior

Nest Obstacle Food

6
Ant Behavior

Nest Obstacle Food

7
Ant Behavior

Nest Obstacle Food

Pheromones

8
Ant Behavior

Quest for food


Start

Walk randomly Follow Trail

Found food Continue

Lay pheromone Trail

9
Ant Behavior

Ant behavior is stochastic

The behavior is induced by indirect communication


(pheromone paths) - Stigmergy

Ants explore the search space

Limited ability to sense local environment

Act concurrently and independently

High quality solutions emerge via global cooperation

10
Stigmergy

Term coined by French biologist Pierre-Paul Grasse, means


interaction through the environment

Indirect communication via interaction with environment

Agents respond to changes in the environment

Allows simpler agents

Decreases direct communication

11
Pheromones

Ants lay pheromone trails while traveling

Pheromones accumulate with multiple ants using a path

This behavior leads to the appearance of shortest paths

Pheromones = long-term memory of an ant colony

Pheromones evaporate
Avoids being trapped in local optima

ρ small ⇒ low evaporation ⇒ slow adaptation

ρ large ⇒ high evaporation ⇒ fast adaptation

12
ACO Algorithm

Construct solutions
Explore the search space
Choose next step probabilistically according to the
pheromone model
Apply local search to constructed solutions (Optional)
Update pheromones (add new + evaporate)

13
Example: TSP

14
Example: TSP

15
Advantages and Disadvantages

Advantages
Can be used in dynamic applications
Positive Feedback leads to rapid discovery of good
solutions
Distributed computation avoids
premature convergence

Disadvantages
Convergence is guaranteed, but time to convergence
uncertain
Coding is not straightforward

16
References

Dorigo M., Blum C., Ant colony optimization theory: A survey,


Theoretical Computer Science, Volume 344, Issues 2-3,
November 2005
Blum C., Ant colony optimization: Introduction and recent
trends, Physics of Life Reviews, Volume 2, Issue 4, December
2005
Dorigo M., Stutzle T., Ant Colony Optimization, Ant Colony
Optimization, MIT Press 2004

17
AntPacking – An Ant Colony Optimization Approach for
the One-Dimensional Bin Packing Problem

by B. Brugger, K. Doerner , R. Hartl and M. Reimann


Introduction

Deals with the one-dimensional BPP

Meta-heuristic solution approach based on Ant Colony


Optimization

A set of ants repeatedly build and improve solutions

Ants update joint memory, guiding future searches

Memory update is based on solution quality

19
Pheromone Decoding

Items are grouped according to their size

Relates the size of an item to be packed to the filling degree


of the current bin

They only consider how much space is left in the bin

The importance of filling a bin is emphasized

20
Solution Construction and Local Search

A ant fills bins until all items are packed

The decision about which item to add is based on:


FFD rule
Pheromone information

Local search is performed when a ant finishes filling a bin


Tries to replace one bin item with an unpacked item if
this leads to less free space
Stops when the bin is full or no improving moves are
available
Bin Unpacked Items

21
Fitness Function and Pheromone Update

The fitness function guides the search

Fitness is calculated for each bin

Only full bins receive pheromone

Each ant is allowed to modify the memory (pheromone


update)

All the solution elements are subject to evaporation

22
Dataset and Preprocessing

Benchmark instances properties:


bin capacity C = 150
item sizes bounded by [20, 100]

Procedure was applied to reduce the ants search effort

Eliminates mainly large items

Problem size is reduced and consequently easier to solve

23
Computational Analysis

Algorithm coded in C and executed on a Pentium 3, 750 MHz

Number of ants fixed to 10

Time limit for termination of the algorithm set to 60 seconds

Computational test:
Benchmark comparison with Hybrid Grouping Genetic
Algorithm (HGGA) and the Hybrid ACO (HACO)

24
Algorithm Comparison

HACO HGGA AntPacking


Problem
class
# of optima seconds # of optima seconds # of optima seconds

u120 20 1 20 0.31 20 0.04


u250 18 52 20 0.75 20 0.58
u500 20 50 20 1.5 20 1.11
u1000 20 147 20 3.78 20 3.88

25
Conclusion

New pheromone decoding scheme and update strategy

Results show excellent performance

AntPacking approach performs at least as good as the HGGA


which is considered to be the best algorithm for BPPs

26

You might also like