Parallel evolutionary learning of fuzzy rule bases using the island injection genetic algorithm

In this paper the island injection genetic algorithm (iiGA) is applied to the evolutionary design of fuzzy rule bases. The iiGA is a parallel GA in which a hierarchy of subpopulations of candidate problem solutions employ representations at different resolutions. Emigration from one subpopulation to another occurs strictly down the hierarchy of increasing resolution, except at the lowest (i.e. highest resolution) level where subpopulations can exchange migrants. This arrangement allows subpopulations with more abstract representations to evolve approximate solutions, which are then “injected” into subpopulations at lower levels in the hierarchy for further refinement. We investigate the application of the technique to artificial evolution of fuzzy controllers. In applying the technique to this problem, different subpopulations use varying levels of abstraction in their representation of fuzzy set membership functions and number of fuzzy rules