Hybridization of Search Meta-Heuristics Bob Buehler.

20
Hybridization of Search Meta-Heuristics Bob Buehler
  • date post

    19-Dec-2015
  • Category

    Documents

  • view

    220
  • download

    1

Transcript of Hybridization of Search Meta-Heuristics Bob Buehler.

Page 1: Hybridization of Search Meta-Heuristics Bob Buehler.

Hybridization of SearchMeta-Heuristics

Bob Buehler

Page 2: Hybridization of Search Meta-Heuristics Bob Buehler.

A Recombination of Strengths

Genetic Algorithm High correlation

reproduction operators

Fast computation excluding fitness

Ant Colony Optimization Well suited in step-

wise solution creation

Strong local search using probabilistic pheromone model

EAnt

Page 3: Hybridization of Search Meta-Heuristics Bob Buehler.

EAnt

Page 4: Hybridization of Search Meta-Heuristics Bob Buehler.

The Power of Ants

The World The Ant The Pheromone The Dream

Page 5: Hybridization of Search Meta-Heuristics Bob Buehler.

Combinatorial Optimizers

Ant Colony Optimization Traveling Salesman Problem

S = The space of all possible solutions Τ = Pheromone model η = Heuristic values

Step-wise solution creation About to select the next component for

a partial solution cj = set of possible next components w(ci

j) = [τij]α[η(ci

j)]β

p(cij) = w(ci

j) / Σ w(cj)

Page 6: Hybridization of Search Meta-Heuristics Bob Buehler.

Basic ACO Algorithm

Initialize pheromones and heuristics Iterate until termination condition

Generate Solutions Update pheromones

Decay all Increase those present in

high fitness solutions

Page 7: Hybridization of Search Meta-Heuristics Bob Buehler.

EAnt

Evolving Pheromone Models Create random pheromone models as

arrays of real values Let k ants walk the pheromone and create

solutions Assign a fitness to the model equal to the

average of all solutions created Use GA reproduction operators Profit

Page 8: Hybridization of Search Meta-Heuristics Bob Buehler.

Testing

EA

vs ACO

vs EAnt

Page 9: Hybridization of Search Meta-Heuristics Bob Buehler.

Euclidean TSP

5

4

2

13

0

1 4 3 5 20 0

0 X

Y

Page 10: Hybridization of Search Meta-Heuristics Bob Buehler.

EA Representation

1 4 3 5 20 01 4 3 5 2 00

Page 11: Hybridization of Search Meta-Heuristics Bob Buehler.

EA Reproduction

3 4 2 1 50 0

1 4 3 2 50 0

1 4 3 5 20 0

1 4 3 2 50 0

Page 12: Hybridization of Search Meta-Heuristics Bob Buehler.

EAnt Representation

Pheromone Model is a two dimensional array M[n,m] where n is the node an ant is currently at and m is a node connected to n.

Every element is initialized with a random value in the range [0,5).

Page 13: Hybridization of Search Meta-Heuristics Bob Buehler.

EAnt Representation Example

4

2

13

0

54 11 2

033 1

12

4 5

4

23

03 1

10 1 2 3 4

01234

EAnt Genotype

1 4 3 2 00

Environment

Page 14: Hybridization of Search Meta-Heuristics Bob Buehler.

EAnt Reproduction

Parameterized Uniform Crossover Gaussian Mutation with σ = 1

Page 15: Hybridization of Search Meta-Heuristics Bob Buehler.

Results-Time Ranking

1. EA

2. ACO Step-wise cycle creation

3. EAnt Step-wise cycle creation O(n2) individual size and reproduction

Page 16: Hybridization of Search Meta-Heuristics Bob Buehler.

Results- EA and ACO Convergence

0

200

400

600

800

1000

1200

1400

1600

1800

200

800

1400

2000

2600

3200

3800

4400

5000

5600

6200

6800

7400

8000

8600

9200

9800

Cycles Generated

Cyc

le L

eng

th EA Local

EA Global

ACO Local

ACO Global

Page 17: Hybridization of Search Meta-Heuristics Bob Buehler.

Results- EAnt Convergence

0

200

400

600

800

1000

1200

1400

1600

200

800

1400

2000

2600

3200

3800

4400

5000

5600

6200

6800

7400

8000

8600

9200

9800

Cycles

Cyc

le L

eng

th

Eant(50,20,10)

Eant(50,40,5)

EAnt(100,10,5)

ACO

EA

(generations, individuals, fitness)

Page 18: Hybridization of Search Meta-Heuristics Bob Buehler.

Hope

0

200

400

600

800

1000

1200

1400

1600

200

800

1400

2000

2600

3200

3800

4400

5000

5600

6200

6800

7400

8000

8600

9200

9800

Cycles

Cy

cle

Le

ng

th

Eant(50,20,10)

Eant(50,40,5)

EAnt(100,10,5)

EAnt(1000,10,1)

ACO

EA

Page 19: Hybridization of Search Meta-Heuristics Bob Buehler.

Final Thoughts

Test for better final solution Different problem types EAnt pheromone model initialization

54 11 2

033 1

12

4 5

4

23

03 1

10 1 2 3 4

01234

55 01 5

052 0

21

2 5

2

21

01 0

10 1 2 3 4

01234

Improved?

Page 20: Hybridization of Search Meta-Heuristics Bob Buehler.

Questions?