Masahito
Yamamoto, Graduate School of Engineering, Hokkaido University,
Japan
Email: masahito@complex.eng.hokudai.ac.jp
K. Suzuki, Graduate School of Engineering, Hokkaido University, Japan
A. Ohuchi, Graduate School of Engineering, Hokkaido University, Japan
In Ant Algorithm, the behavior of colony consists of ant agents is
proceeded to the direction indicated by pheromone information as good
region on the solution space. The pheromone information is also
formed by results of feedback of their searches while ant agents
search its solutions. Then, the behavior of colony should converge at
a specific region, may be trapped on near global optima or be on near
local one, according as the time progresses, and they have not ability
to inspect other regions on solution space. That is, if colony is
trapped on the regions apart from global optima, it is impossible for
ants to find optimal solution.
To overcome this weakness, we introduce the multiple colonies approach based on original Ant Algorithm and constructed with several ants colonies. As a aspect of extended algorithm, the pheromone information on one colony has different effects to ants belonging to its colony and belonging to other colonies. For instance, the pheromone information generated by one colony is seemed for ants on other as "negative pheromone", and the regions indicated by "negative pheromone" are not preferred by ants. This mechanism enable the ant colony to search worth regions which other colonies can not find.