Scribd
Upload
107 views

Ant Colony Optimization

Ant colony optimization is an algorithm inspired by how ants find food sources, where artificial ants communicate indirectly via pheromone trails to iteratively find optimal solutions; the ants probabilistically construct solutions using pheromone values that get continuously updated based on solution quality and evaporation, with the best solutions reinforcing their pheromone trails in a positive feedback loop. An example demonstrates how pheromone values guide ants in selecting the shortest path between points.

Uploaded by

Brian Nguyen
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
107 views

Ant Colony Optimization

Ant colony optimization is an algorithm inspired by how ants find food sources, where artificial ants communicate indirectly via pheromone trails to iteratively find optimal solutions; the ants probabilistically construct solutions using pheromone values that get continuously updated based on solution quality and evaporation, with the best solutions reinforcing their pheromone trails in a positive feedback loop. An example demonstrates how pheromone values guide ants in selecting the shortest path between points.

Uploaded by

Brian Nguyen
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 23

Ant Colony

Optimization
Intro - Ant Colony Optimization (ACO)
● Modeled after how ants behave
in the real world
● Multi-agent approach
○ Solve complex combinatorial
optimization problems

2
● Stochastic
Ant ● Induced through indirect
communication (leaving
Behavior pheromones behind)
○ Stigmergy = interacting
with environment
● Limited ability to sense local
environment

3
Pheromones

▪ As the ants travel, they leave pheromones behind


▪ Shorter paths will accumulate more pheromones
▫ Positive feedback loop
▪ Pheromones evaporate
▫ Helps with avoiding local optima

4
5
6
7
Let’s work through an
example
8
The Scenario

Landing
Zone

9
WP 1

Landing
Zone

WP 2

10
WP 1 WP 2 LZ 4 WP 1
3
∞
0 4 6 ∞
Landing
Cost 4 0 ∞ 3
WP 1 Zone
matrix 6

WP 2 6 ∞ 0 8 ∞
8
LZ ∞ 3 8 0 WP 2

WP 1 WP 2 LZ
1 WP 1
1
0
0 1 1 0
Pheromo Landing
ne matrix WP 1 1 0 0 1 1 Zone
(initial)
WP 2 1 0 0 1
0
1
LZ 0 1 1 0 WP 2
11
WP 1 WP 2 LZ 4 WP 1
3
∞
0 4 6 ∞
Landing
Cost 4 0 ∞ 3
WP 1 Zone
matrix 6

WP 2 6 ∞ 0 8 ∞
8
LZ ∞ 3 8 0 WP 2

WP 1 WP 2 LZ
8 WP 1
6
0
0 8 2 0
Pheromo Landing
ne matrix WP 1 8 0 0 6 2 Zone
(final)
WP 2 2 0 0 3
0
3
LZ 0 6 3 0 WP 2
12
Update Trails

13
Cost graph Pheromone graph

4 WP 1 1 WP 1
3 1
∞ 0

Landing Landing
Zone Zone
6 1

∞ 0
8 1
WP 2 WP 2

Let ⍴ = 0.5
𝛕xy = (1-⍴)𝛕xy + ∑𝛕kxy

14
Cost graph Pheromone graph

4 WP 1 1 WP 1
3 1
∞ 0

Landing Landing
Zone Zone
6 1

∞ 0
8 1
WP 2 WP 2

L1 = 4+3+3+4 = 14
Δ𝛕1xy = 1/14
Let ⍴ = 0.5
L2 = 4+3+3+4 = 14 𝛕xy = (1-⍴)𝛕xy + ∑𝛕kxy
Δ𝛕2xy = 1/14

15
Cost graph Pheromone graph

0.5*1+1/14+1/14+1/21
4 WP 1 WP 1
0.5*1+1/14+1/14+1/21
3
∞ 0

Landing Landing
Zone 0.5*1+1/21 Zone
6

∞ 0
8 0.5*1+1/21
WP 2 WP 2

L1 = 4+3+3+4 = 14
Δ𝛕1xy = 1/14
Let ⍴ = 0.5
L2 = 6+8+3+4 = 21 𝛕xy = (1-⍴)𝛕xy + ∑𝛕kxy
Δ𝛕2xy = 1/21

16
Calculate Path
Probabilities

17
4
3
∞

6
∞
8

1
1
0

0
1

18
4
3
∞

P = (1*1/4) / ((1*1/4)+(1*1/6)+(1*1/∞)) = 0.60


6
∞
8

1
1
0

0
1

19
4
3
∞

P = (1*1/4) / ((1*1/4)+(1*1/6)+(1*1/∞)) = 0.60


6
∞
8

P = (1*1/∞) / ((1*1/4)+(1*1/6)+(1*1/∞)) =
0.00 1
1
0

0
1

20
4
3
∞

P = (1*1/4) / ((1*1/4)+(1*1/6)+(1*1/∞)) = 0.60


6
∞
8

P = (1*1/∞) / ((1*1/4)+(1*1/6)+(1*1/∞)) =
0.00 1
1
0

0
1

P = (1*1/6) / ((1*1/4)+(1*1/6)+(1*1/∞)) = 0.40

21
● Advantages
○ Can be used in dynamic
applications
Advantages and ○ Rapid discovery of good
Disadvantages solutions
○ Distributed computation
● Disadvantages
○ Requires large amount of
computation
○ While convergence is
guaranteed, time to converge is
uncertain
22
Thank you for listening!

23

You might also like