+ All Categories

Outline

Date post: 03-Jan-2016
Category:
Upload: fredricka-elijah
View: 13 times
Download: 0 times
Share this document with a friend
Description:
Anti-pheromone as a Tool for Better Exploration of Search Space by James Montgomery and Marcus Randall, Bond University, Australia. Outline. Abstract Introduction ACS Anti-pheromone Application Computation Experience Conclusion References. Abstract. - PowerPoint PPT Presentation
24
Anti-pheromone as a Tool for Better Exploration of Search Space by James Montgomery and Marcus Randall, Bond University, Australia
Transcript
Page 1: Outline

Anti-pheromone as a Tool for Better

Exploration of Search Space

by James Montgomery and Marcus Randall, Bond University, Australia

Page 2: Outline

Outline

• Abstract

• Introduction

• ACS

• Anti-pheromone Application

• Computation Experience

• Conclusion

• References

Page 3: Outline

Abstract

• Investigates an alternative form of ACS, using anti-pheromone, beside normal pheromone.

• Describes three variations of ACS

Page 4: Outline

Introduction

• Ants are known to use pheromones to communicate to coordinate activities like the location and collection of food.

• ACO (ant colony optimization). Based on the foraging behavior of ant colonies, ACO has generally used a single kind of pheromone to communicate between its ants.

• However, natural pheromone communication often consist of a number of different pheromones.

• ACO ’s reliance on positive feedback alone may make difficult for it to successfully escape local optima.

Page 5: Outline

Introduction

• Anti-pheromone which have the opposite effect of normal pheromone is describes in this paper.

• Three variations of an Ant Colony System (ACS) that use anti-pheromone in some form are described and compare with typical ACS

Page 6: Outline

ACS-TSP

• TSP: find the shortest path that traverses all cities in the problem exactly once, returning to the starting city.

• ACS-TSP: Using Ant Colony System to solve TSP

Page 7: Outline

ACS-TSP

Brief summary of ACS-TSP

(1)

(2)

Page 8: Outline

ACS-TSP

decay parameter 0 < < 1 0 amount of pheromone deposit at each edge

(3)

Page 9: Outline

ACS-TSP

deltaT(r, s) : reinforced pheromone L : length of the best (shortest) tour to date while Q is a problem dependent parameter gamma : the global pheromone decay parameter, 0 < gamma < 1.

(4)

(5)

Page 10: Outline

Anti-pheromone Applications

• Subtractive Anti-pheromone (SAP)

• Preferential Anti-pheromone (PAP) • Explorer Ants

Page 11: Outline

Subtractive Anti-pheromone (SAP)

• Idea: globally subtract pheromone in the worst solution.ɣ ‘ : pheromone removal rate vw is the iteration worst solution

gamma ‘ = 0.5 is optimal

(6)

Page 12: Outline

Preferential Anti-pheromone (PAP)

• Idea: explicitly using two types of pheromone, one for good solutions and one for poorer solutions.

= k-1/m-1

(7)

(8)

Page 13: Outline

Preferential Anti-pheromone (PAP)

’ : anti-pheromone• Local anti-pheromone update:

(9)

Page 14: Outline

Preferential Anti-pheromone (PAP)

• delta ’(r, s) reinforced pheromone on the edges of the iteration worst solution.

• Lw is the length of the worst tour from the iteration just ended.

(10)

(11)

Page 15: Outline

Explorer Ants

• Idea: explorer ants deposit pheromone in the same way as normal ants, but their preference for existing pheromone is reversed.

Page 16: Outline

Explorer Ants

• Tmax : the highest current level of pheromone in the system.

(12)

(13)

Page 17: Outline

Computational Experience

• 550 MHz Linux machine.• programs are written in the C

language. • 10 random seeds consisting of

3000 iterations. = -2, = 0.1, = 0.1, m = 10, q0

= 0.9.

Page 18: Outline

Computational Experience

inter-quartile range (“IQR”)

Page 19: Outline

Computational Experience

Page 20: Outline

Computational Experience

Page 21: Outline

Computational Experience

SAP produces better solutions than PAP

On problems with less than 100 cities, SAP produces better results than explorer ants

However, on problems with more than 200 cities, explorer ants performs better. Explorer ants also performs better than PAP across all problems.

Page 22: Outline

Conclusions

• We have proposed three variations of the Ant Colony System that employ anti-pheromone in some form.

• subtractive anti-pheromone. It works well on problems with less than 200 cities.

• preferential anti-pheromone, producing better results than the control on only the two smallest problems

• Explorer ants, It can produce better solutions than the control on small problems

Page 23: Outline

References

1. Vander Meer, R.K., Breed, M.D., Winston, M.L., Espelie, K.E. (eds.): PheromoneCommunication in Social Insects. Ants, Wasps, Bees, and Termites. WestviewPress, Boulder, Colorado (1997)2. Heck, P.S., Ghosh, S.: A Study of Synthetic Creativity through Behavior Modelingand Simulation of an Ant Colony. IEEE Intelligent Systems 15 (2000) 58–663. Kawamura, H., Yamamoto, M., Ohuchi, A.: Improved Multiple Ant Colonies Systemfor Traveling Salesman Problems. In Kozan, E., Ohuchi, A. (eds.): OperationsResearch/Management Science at Work. Kluwer, Boston (2002) 41–594. Kawamura, H., Yamamoto, M., Suzuki, K., Ohuchi, A.: Multiple Ant ColoniesAlgorithm Based on Colony Level Interactions. IEICE Transactions, FundamentalsE83-A (2000) 371–3795. Schoonderwoerd, R., Holland, O.E., Bruten, J.L., Rothkrantz, L.J.M.: Ant-BasedLoad Balancing in Telecommunications Networks. Adaptive Behavior 2 (1996)169–207

Page 24: Outline

The End

• Question?

• Comment?


Recommended