Download as pdf or txt
Download as pdf or txt
You are on page 1of 22

PRESENTATION AND ANALYSIS OF THREE HYBRIDIZATION MODELS WITH TABOO SEARCH AND SIMILAR PARTICLE SWARM OPTIMIZATION HEURISTICS

FOR THE JOB SHOP SCHEDULING PROBLEM


1

Tatiana Balbi Fraga

SUMMARY OF THIS PRESENTATION


Problems presentation Solutions methods Local Search heuristics

Taboo Search Similar Particles Swarm Optimization


Hybridization models proposed


Hybrid Successive Application Hybrid Neighborhood Hybrid Improved Neighborhood

Results, conclusions and future work


2

GENERAL SCHEDULING PROBLEM

Tasks

Machines

Clothes manufacturing

EXAMPLE:

JOB SHOP SCHEDULING PROBLEM (JSSP)

Machine 1

Machine 2

Machine 3

Machine 4

Machine 5

Machine 6

SOLUTIONS METHODS FOR THE JSSP


Optimization

Approximation

Efficient methods

Johnsons method Mathematical


Mixed Integer Linear Programming Interger Linear Programming

Enumerative methods

Priority Dispatch Rules Bottleneck Based Heuristics Artificial Intelligence Local Search Heuristics
Taboo Search Simulated Annealing Genetic Algorithms Particle Swarm Optimization

Branch and Bound

LOCAL SEARCH HEURISTICS


Constructive method Initial parent solutions Neighborhood generation operator Neighborhood Selection method New parent solutions Best solution found

No

Yes

Stopping criterion is satisfied ?


7

HYBRIDIZATION MODELS
Constructive method Initial parent solutions Different neighborhood generation operators Hybrid neighborhood Selection method New parent solutions Best solution found

No

Yes

Stopping criterion is satisfied ?


8

HYBRIDIZATION MODELS
Hybrid Successive Aplication Hybrid Neighborhood Hybrid Improved Neighborhood

HYBRID SUCCESSIVE APPLICATION


Constructive method Initial parent solutions Best solution found

Operator 1
Operador 2 Operador N Hybrid neighborhood

No

Yes

Stopping criterion is satisfied ?

Selection method

New parent solutions

10

HYBRID IMPROVED NEIGHBORHOOD


Constructive method Initial parent solutions

Best solution found

No
Operator 1 Operator 2 Operator N

Yes

Stopping criterion is satisfied ? Hybrid neighborhood

Selection method

New parent solutions

11

HYBRID NEIGHBORHOOD
Constructive method Initial parent solutions

Best solution found

No

Yes

Operator 1

Operator 2

Operator N Stopping criterion is satisfied ?

Hybrid neighborhood

Selection method

New parent solutions

12

ANALYSIS
Taboo Search Similar Particle Swarm

Nowicki and Smutnicki (1996) neighborhood generating structure N5 Effective and efficient Improvement of each single solution Deterministic

Lian et al. (2006) New method, not yet applied in hybrid algorithms Iteration between the parent solutions Non-deterministic

13

TABOO SEARCH MAXIMUN NUMBER OF


ITERATIONS WITHOUT UPGRADE OF THE MAKESPAN

14

TABOO SEARCH TENURE

15

TABOO SEARCH TENURE

16

TABOO SEARCH TENURE

17

HYBRID SUCCESSIVE APPLICATION


Sensitivity analysis - tenure
Hybrid Successive Application (10 cycles) - Problem FT10
948 946 944 942 Makespan 940 938 936 934

932
930 928 6 7 8 9 10 Tenure HSA - best result Parallel Taboo Search HSA - mean Optimal solution HSA - worse result 11 12 13 14 15

18

HYBRID NEIGHBORHOOD
Sensitivity analysis - tenure
Hybrid Neighborhood (10 cycles) - Problem FT10
948 946 944 942 Makespan 940 938 936 934

932
930 928 6 7 8 9 10 Tenure HN - best result Parallel Taboo Search HN - mean Optimal solution HN - worse result 11 12 13 14 15

19

HYBRID IMPROVED NEIGHBORHOOD


Sensitivity analysis - tenure
Hybrid Improved Neighborhood (10 cycles) - Problem FT10
948 946 944 942 Makespan 940 938 936 934

932
930 928 6 7 8 9 10 Tenure HIN - best result Parallel Taboo Search HIN - mean Optimal solution HIN worse result 11 12 13 14 15

20

CONCLUSION
The

different operators applied to generate the hybrid neighborhood must present different and complementary aspects When combinig Taboo Search and Similar Particle Swarm heuristics, the hybrid models must be applied with a few iterations of Similar Particle Swarm. The performance of the hybrid algorithms is less sensitive to the tuning of its parameters. The Hybrid Successive Application was the model that presented the best results for all tests.

21

FUTURE WORKS
Test other neighborhood generation operators Make a sensitivity analysis for the population size Apply parallel programming

22

You might also like