Universiti Teknologi Malaysia Institutional Repository

A hybrid parallel genetic algorithm for solving job-shop scheduling problems

Gan, Teck Hui (2005) A hybrid parallel genetic algorithm for solving job-shop scheduling problems. Masters thesis, Universiti Teknologi Malaysia, Faculty of Electrical Engineering.


Official URL: http://dms.library.utm.my:8080/vital/access/manage...


The effort of searching an optimal solution for scheduling problems is important for real-world industrial applications especially for mission-time critical systems. In such an environment, an optimal result should be obtained within a relatively reasonable time. Genetic algorithms (GA) have long been applied to the many scheduling problems especially for solving job shop scheduling problems. However there are several problems when implementing GA into job shop scheduling problem (JSSP) such as slow and premature convergence. GA on one single personal computer is very time consuming while harder problems need bigger population and this has translate directly into higher computational costs. Therefore, this research proposes a different approach in solving JSSP using GA. Instead of conventional approach of using “serial” GA on a single PC, this research looks into the possibility of using parallel and distributed computing techniques and parallel genetic algorithm (PGA) in solving the same problems. Island model is chosen in this research to suit the implementation of distributed computing environment. The proposed PGA is a combination of both Asynchronous Colony Genetic Algorithm (ACGA) and Autonomous Immigration Genetic Algorithm (AIGA). This hybrid PGA is originally applied to symmetric multiprocessor machine (SMP) by Haruki Inoue. This type of modeling of PGA is based on the biologically reported observation that isolated environments, such as islands, often produce animal species that are more specifically adapted to the peculiarities of their environments than corresponding areas of wider surfaces. This theory also led to the hypothesis that several competing subpopulations could be more search-effective than a wider one in which all the members are held together. Each “island” is a different type of serial GA which represents one PC. Each island is using a combination of different crossover operator (GOX, Giffler and Thompson (GT) Crossover) and selection method (random selection, roulette wheel selection) into different types of serial GA. Every “island” has a certain convergence point where the improvement of the population fitness is almost stagnant. Therefore to solve this kind of undesired situation; a more sophisticated idea is used in this coarse grained PGA. This model of parallelization introduces a migration operator that is used to send some individual from one “island” to another “island”. The individual can migrate to any other “island” at their own judgment. Communication overhead during migration process is reduced by implementing a global mailbox method. Results show that PGA with a combination of different types of GA that creates a partially isolated environment for each sub demes, allowing a wider and more effective search, has outperformed PGA with a single type of GA. This research does not consider network topology in the scope as the system is linked via office network. To further improve the robustness of the system in the future, functions such as process migration and parallel input/output - by migrating intensive I/O processes to file servers (rather than the traditional way of bringing data to the processes) should be developed.

Item Type:Thesis (Masters)
Additional Information:Thesis (Sarjana Kejuruteraan (Elektrik)) -Universiti Teknologi Malaysia, 2005
Uncontrolled Keywords:genetic algorithms (GA)
Divisions:Electrical Engineering
ID Code:34719
Deposited By: Kamariah Mohamed Jong
Deposited On:24 Jul 2017 12:05
Last Modified:11 Oct 2017 04:41

Repository Staff Only: item control page